Handbook of Fluidization and Fluid-Particle Systems

HANDBOOK of
FLUIDIZfiTION and
FLUID-PARTICLE SYSTEMS
edited by
Wen-Ching Yang
Siemens Westinghouse Power Corporation
Pittsburgh, Pennsylvania, U.S.A.
M A R C E L D E K K E R , INC . NE W Y O RK B A S E L
ila
D E K K E R
Copyright 2003 by Taylor & Francis Group LLC
Library of Congress Cataloging-in-Publication Data
A catalog record for this book is available from the Library of Congress.
ISBN: 0-8247-0259-X
This book is printed on acid-free paper.
Headquarters
Marcel Dekker, Inc.
270 Madison Avenue, New York, NY 10016
tel: 212-696-9000; fax: 212-685-4540
Eastern Hemisphere Distribution
Marcel Dekker AG
Hutgasse 4, Postfach 812, CH-4001 Basel, Switzerland
tel: 41-61-260-6300; fax: 41-61-260-6333
World Wide Web
http://www.dekker.com
The publisher offers discounts on this book when ordered in bulk quantities. For more information, write to Special Sales/
Professional Marketing at the headquarters address above.
Copyright # 2003 by Marcel Dekker, Inc. All Rights Reserved.
Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic or mechanical,
including photocopying, microlming, and recording, or by any information storage and retrieval system, without permission
in writing from the publisher.
Current printing (last digit):
10 9 8 7 6 5 4 3 2 1
PRINTED IN THE UNITED STATES OF AMERICA
CHEMICAL INDUSTRIES
A Series of Reference Books and Textbooks
Consulting Editor
HEINZ HEINEMANN
Berkeley, California
1. Fluid Catalytic Cracking with Zeolite Catalysts, Paul B Venuto and E. Thomas Habib, Jr.
2. Ethylene: Keystone to the Petrochemical Industry, Ludwig Kniel, Olaf Winter, and Karl Stork
3. The Chemistry and Technology of Petroleum, James G Speight
4. The Desulfurization of Heavy Oils and Residua, James G. Speight
5. Catalysis of Organic Reactions, edited by William R. Moser
6 Acetylene-Based Chemicals from Coal and Other Natural Resources, Robert J. Tedeschi
7 Chemically Resistant Masonry, Walter Lee Sheppard, Jr.
8 Compressors and Expanders: Selection and Application for the Process Industry, Heinz P. Bloch, Joseph A.
Cameron, Frank M. Danowski, Jr, Ralph James, Jr, Judson S. Sweanngen, and Marilyn E. Weightman
9. Metering Pumps. Selection and Application, James P. Poynton
10. Hydrocarbons from Methanol, Clarence D Chang
11. Form Flotation: Theory and Applications, Ann N. Clarke and David J. Wilson
12. The Chemistry and Technology of Coal, James G. Speight
13. Pneumatic and Hydraulic Conveying of Solids, O. A Williams
14. Catalyst Manufacture: Laboratory and Commercial Preparations, Alvm B. Stiles
15. Characterization of Heterogeneous Catalysts, edited by Francis Delannay
16. BASIC Programs for Chemical Engineering Design, James H. Weber
17. Catalyst Poisoning, L. Louis Hegedus and Robert W McCabe
18. Catalysis of Organic Reactions, edited by John R Kosak
19 Adsorption Technology: A Step-by-Step Approach to Process Evaluation and Application, edited by Frank L.
Slejko
20 Deactivation and Poisoning of Catalysts, edited by Jacques Oudar and Henry Wise
21 Catalysis and Surface Science: Developments in Chemicals from Methanol, Hydrotreating of Hydrocarbons,
Catalyst Preparation, Monomers and Polymers, Photocatalysis and Photovoltaics, edited by Heinz Heinemann
and Gabor A Somorjai
22 Catalysis of Organic Reactions, edited by Robert L. Augustine
23 Modern Control Techniques for the Processing Industries, T. H. Tsai, J W. Lane, and C. S. Lin
24 Temperature-Programmed Reduction for Solid Materials Characterization, Alan Jones and Brian McNichol
25 Catalytic Cracking: Catalysts, Chemistry, and Kinetics, Bohdan W Wojciechowski and Avelino Corma
26. Chemical Reaction and Reactor Engineering, edited by J. J. Carberry and A. Varma
27 Filtration: Principles and Practices: Second Edition, edited by Michael J Matteson and Clyde Orr
28 Corrosion Mechanisms, edited by Florian Mansfeld
29 Catalysis and Surface Properties of Liquid Metals and Alloys, Yoshisada Ogino
30 Catalyst Deactivation, edited by Eugene E Petersen and Alexis T. Bell
31. Hydrogen Effects in Catalysis: Fundamentals and Practical Applications, edited by Zoltan Paal and P G. Menon
32. Flow Management for Engineers and Scientists, Nicholas P. Cheremisinoff and Paul N. Cheremisinoff
33. Catalysis of Organic Reactions, edited by Paul N. Rylander, Harold Greenfield, and Robert L. Augustine
34. Powder and Bulk Solids Handling Processes: Instrumentation and Control, Koichi linoya, Hiroaki Masuda, and
Kinnosuke Watanabe
35. Reverse Osmosis Technology: Applications for High-Purity-Water Production, edited by Bipin S. Parekh
36. Shape Selective Catalysis in Industrial Applications, N. Y. Chen, William E. Garwood, and Frank G. Dwyer
37. Alpha Olefins Applications Handbook, edited by George R. Lappin and Joseph L. Sauer
38 Process Modeling and Control in Chemical Industries, edited by Kaddour Najim
39. Clathrate Hydrates of Natural Gases, E. Dendy Sloan, Jr
40 Catalysis of Organic Reactions, edited by Dale W Blackburn
41. Fuel Science and Technology Handbook, edited by James G Speight
42. Octane-Enhancing Zeolitic FCC Catalysts, Julius Scherzer
43. Oxygen in Catalysis, Adam Bielanski and Jerzy Haber
44. The Chemistry and Technology of Petroleum: Second Edition, Revised and Expanded, James G. Speight
45 Industrial Drying Equipment: Selection and Application, C. M. van't Land
46 Novel Production Methods for Ethylene, Light Hydrocarbons, and Aromatics, edited by Lyle F. Albright, Billy L.
Crynes, and Siegfried Nowak
47 Catalysis of Organic Reactions, edited by William E. Pascoe
48 Synthetic Lubricants and High-Performance Functional Fluids, edited by Ronald L. Shubkin
49 Acetic Acid and Its Derivatives, edited by Victor H Agreda and Joseph R. Zoeller
50 Properties and Applications of Perovskite-Type Oxides, edited by L G Tejuca and J. L. G. Fierro
51 Computer-Aided Design of Catalysts, edited by E Robert Becker and Carmo J. Pereira
52. Models for Thermodynamic and Phase Equilibria Calculations, edited by Stanley I Sandier
53 Catalysis of Organic Reactions, edited by John R Kosak and Thomas A Johnson
54 Composition and Analysis of Heavy Petroleum Fractions, Klaus H. Altgelt and Mieczyslaw M. Boduszynski
55. NMR Techniques in Catalysis, edited by Alexis T. Bell and Alexander Pines
56. Upgrading Petroleum Residues and Heavy Oils, Murray R. Gray
57. Methanol Production and Use, edited by Wu-Hsun Cheng and Harold H. Kung
58. Catalytic Hydroprocessing of Petroleum and Distillates, edited by Michael C. Oballah and Stuart S. Shih
59. The Chemistry and Technology of Coal: Second Edition, Revised and Expanded, James G Speight
60. Lubricant Base Oil and Wax Processing, Avilino Sequeira, Jr
61. Catalytic Naphtha Reforming- Science and Technology, edited by George J. Antos, Abdullah M. Aitani, and
Jose M. Parera
62. Catalysis of Organic Reactions, edited by Mike G. Scares and Michael L. Prunier
63. Catalyst Manufacture, Alvin B. Stiles and Theodore A Koch
64. Handbook ofGrignard Reagents, edited by Gary S. Silverman and Philip E. Rakita
65. Shape Selective Catalysis in Industrial Applications: Second Edition, Revised and Expanded, N. Y. Chen,
William E. Garwood, and Francis G. Dwyer
66. Hydrocracking Science and Technology, Julius Scherzer and A. J. Gruia
67 Hydrotreating Technology for Pollution Control- Catalysts, Catalysis, and Processes, edited by Mario L.
Occelli and Russell Chianelli
68. Catalysis of Organic Reactions, edited by Russell E. Malz, Jr.
69. Synthesis of Porous Materials: Zeolites, Clays, and Nanostructures, edited by Mario L. Occelli and Henri
Kessler
70. Methane and Its Derivatives, Sunggyu Lee
71. Structured Catalysts and Reactors, edited by Andrzej Cybulski and Jacob A. Moulijn
72. Industrial Gases in Petrochemical Processing, Harold Gunardson
73. Clathrate Hydrates of Natural Gases: Second Edition, Revised and Expanded, E Dendy Sloan, Jr
74. Fluid Cracking Catalysts, edited by Mario L. Occelli and Paul O'Connor
75. Catalysis of Organic Reactions, edited by Frank E. Herkes
76. The Chemistry and Technology of Petroleum: Third Edition, Revised and Expanded, James G. Speight
77 Synthetic Lubricants and High-Performance Functional Fluids- Second Edition, Revised and Expanded,
Leslie R. Rudnick and Ronald L. Shubkin
78 7776 Desulfurization of Heavy Oils and Residua, Second Edition, Revised and Expanded, James G. Speight
79. Reaction Kinetics and Reactor Design: Second Edition, Revised and Expanded, John B. Butt
80. Regulatory Chemicals Handbook, Jennifer M. Spero, Bella Devito, and Louis Theodore
81 Applied Parameter Estimation for Chemical Engineers, Peter Englezos and Nicolas Kalogerakis
82 Catalysis of Organic Reactions, edited by Michael E Ford
83 The Chemical Process Industries Infrastructure Function and Economics, James R Couper, O Thomas
Beasley, and W Roy Penney
84 Transport Phenomena Fundamentals, Joel L Plawsky
85 Petroleum Refining Processes, James G Speight and Baki Ozum
86 Health, Safety, and Accident Management in the Chemical Process Industries, Ann Mane Flynn and Louis
Theodore
87 Plantwide Dynamic Simulators in Chemical Processing and Control, William L Luyben
88 Chemicial Reactor Design, Peter Harriott
89 Catalysis of Organic Reactions, edited by Dennis G Morrell
90 Lubricant Additives Chemistry and Applications, edited by Leslie R Rudnick
91 Handbook of Fluidization and Fluid-Particle Systems, edited by Wen-Chmg Yang
92 Conservation Equations and Modeling of Chemical and Biochemical Processes, Said S E H Elnashaie and
Parag Garhyan
93 Batch Fermentation Modeling, Monitoring, and Control, All Qmar, Gulnur Birol, Satish J Parulekar, and
Cenk Undey
94 Industrial Solvents Handbook, Second Edition, Nicholas P Cheremisinoff
ADDITIONAL VOLUMES IN PREPARATION
Chemical Process Engineering Design and Economics, Harry Silla
Petroleum and Gas Field Processing, H K Abdel-Aal, Mohamed Aggour,, M A Nairn
Process Engineering Economics, James R Couper
Thermodynamic Cycles Computer-Aided Design and Optimization, Chin Wu
Re-Engineering the Chemical Processing Plant Process Intensification, edited by Andrzej Stankiewicz and
Jacob A Mouhjn
Preface
Every chemical engineer, whether a student or practicing, has looked up technical information in Perrys Chemical
Engineering Handbook. Its compilation was one of the most important contributions to the chemical engineering
education and profession. After more than six decades, it remains one of the elds most useful general-purpose
reference books. It was in this spirit of serving the profession that I undertook the task of compiling the Handbook
of Fluidization and Fluid-Particle Systems. Through future revisions and additions, I sincerely hope that this hand-
book will become an archivable reference volume for every practitioner in this eld, spanning the boundary of
various disciplines. Fluidization and uid-particle system engineering is being applied in industries as diverse as
basic and specialty chemicals, mineral processing, coal and biomass gasication and combustion for power gen-
eration, environmental technologies, resource recovery, FCC petroleum rening, pharmaceuticals, biotechnology,
cement, ceramics, and other solids handling and processing industries. The rst focused handbook ever published in
this extended eld, it collects all relevant and important information in a single volume. Both fundamentals and
applications are emphasized. Furthermore, all authors are internationally recognized practitioners in the area of
uidization and uid-particle systems.
This handbook contains 28 chapters and is authored by 34 internationally recognized experts from seven
countries; half of them are professors. Particle characterization and dynamicsimportant in all aspects of particle
production, manufacturing, handling, processing, and applicationsare discussed in Chapter 1. Chapter 2 presents
the ow through xed beds and summarizes packing characteristics of spherical and nonspherical particles, pres-
sure-drop correlations for ow through xed beds, and heat and mass transfer. Bubbling uidized beds are pre-
sented in detail in Chapter 3, which covers all important aspects including jetting phenomena and particle
segregation, topics not addressed extensively in other books on uidization. Other important design considerations
are treated in separate chapters: elutriation and entrainment in Chapter 4, eect of temperature and pressure in
Chapter 5, gas distributor and plenum design in Chapter 6, eect of internal tubes and baes in Chapter 7, attrition
in Chapter 8, and modeling in Chapter 9. Heat transfer (Chapter 10) and mass transfer (Chapter 11) are also
treated. The approaches for designing and scaling up uidized bed reactors are elucidated in Chapter 12, General
Approaches to Reactor Design, and Chapter 13, Fluidized Bed Scaleup.
Important industrial applications for uidized bed reactors are also discussed, including uid catalytic cracking
(Chapter 14), gasiers and combustors (Chapter 15), chemical production and processing (Chapter 16), coating and
granulation (Chapter 17), and uidized bed drying (Chapter 18).
The important variation of bubbling uidized bedsthe circulation uidized bedsare discussed in detail in
Chapter 19. Chapter 20 summarizes other nonconventional uidized beds, including spouted beds, recirculating
uidized beds with a draft tube, jetting uidized beds, and rotating uidized beds. The solids handling, transport
and circulating devices are described in Chapter 21, Standpipe and Nonmechanical Valves, and Chapter 22,
Cyclone Separators. Pneumatic transport is covered in Chapters 23 and 24. Instrumentation and measurement
requirements are reviewed in Chapter 25.
The last three chapters examine the uidized beds and uid-particle systems involving liquid.
This handbook took more than four years to complete. Along the way, content was altered, format was changed,
and chapters were revised to t the page limitation. The nal product is indeed one to be proud of by all who
participated. A monumental endeavor such as this could not have been possible without the cooperation and
dedication of all the authors, especially those who were asked to revise their chapters, sometimes several times. I
am truly indebted to them all for taking the time out of their busy schedule and for their cooperation, dedication,
and conscientious eort. The sta of the publisher, Marcel Dekker, Inc., also deserves credit for their patience and
tenacity in shepherding the project to its eventual completion. Finally, I thank my family, especially my wife, Rae,
for their continuous encouragement.
Wen-Ching Yang
Contents
Preface
Contributors
1 Particle Characterization and Dynamics
Wen-Ching Yang
2 Flow Through Fixed Beds
Wen-Ching Yang
3 Bubbling Fluidized Beds
Wen-Ching Yang
4 Elutriaton and Entrainment
Joachim Werther and Ernst-Ulrich Hartge
5 Eect of Temperature and Pressure
J. G. Yates
6 Gas Distributor and Plenum Design in Fluidized Beds
S. B. Reddy Karri and Joachim Werther
7 Eect of Internal Tubes and Baes
Yong Jin, Fei Wei, and Yao Wang
8 Attrition
Joachim Werther and Jens Reppenhagen
9 Modeling
Thomas C. Ho
10 Heat Transfer
John C. Chen
11 Mass Transfer
Thomas C. Ho
12 General Approaches to Reactor Design
Peijun Jiang, Fei Wei, and Liang-Shih Fan
13 Fluidized Bed Scaleup
Leon R. Glicksman
14 Applications for Fluid Catalytic Cracking
Ye-Mon Chen
15 Applications for Gasiers and Combustors
Richard A. Newby
16 Applications for Chemical Production and Processing
Behzad Jazayeri
17 Applications for Coating and Granulation
Gabriel I. Tardos and Paul R. Mort
18 Applications for Fluidized Bed Drying
Arun S. Mujumdar and Sakamon Devahastin
19 Circulating Fluidized Beds
John R. Grace, Hsiaotao Bi, and Mohammad Golriz
20 Other Nonconventional Fluidized Beds
Wen-Ching Yang
21 Standpipes and Nonmechanical Valves
T. M. Knowlton
22 Cyclone Separators
T. M. Knowlton
23 Dilute-Phase Pneumatic Conveying
George E. Klinzing
24 Electrostatics in Pneumatic Conveying
George E. Klinzing
25 Instrumentation and Measurements
Masayuki Horio, Rafal P. Kobylecki, and Mayumi Tsukada
26 LiquidSolids Fluidization
Norman Epstein
27 GasLiquidSolid Three-Phase Fluidization
Liang-Shih Fan and Guoqiang Yang
28 LiquidSolids Separation
Shiao-Hung Chiang, Daxin He, and Yuru Feng
Contributors
Hsiaotao Bi Department of Chemical and Biological Engineering, University of British Columbia, Vancouver,
British Columbia, Canada
John C. Chen Department of Chemical Engineering, Lehigh University, Bethlehem, Pennsylvania, U.S.A.
Ye-Mon Chen Shell Global Solutions US, Houston, Texas, U.S.A.
Shiao-Hung Chiang Department of Chemical and Petroleum Engineering, University of Pittsburgh, Pittsburgh,
Pennsylvania, U.S.A.
Sakamon Devahastin Department of Food Engineering, King Mongkuts University of Technology Thonburi,
Bangkok, Thailand
Norman Epstein Department of Chemical and Biological Engineering, University of British Columbia, Vancouver,
Liang-Shih Fan Department of Chemical Engineering, The Ohio State University, Columbus, Ohio, U.S.A.
Yuru Feng Department of Chemical and Petroleum Engineering, University of Pittsburgh, Pittsburgh,
Leon R. Glicksman Departments of Architecture and Mechanical Engineering, Massachusetts Institute of
Technology, Cambridge, Massachusetts, U.S.A.
Mohammad Golriz Department of Applied Physics and Electronics, Umea University, Umea , Sweden
John R. Grace Department of Chemical and Biological Engineering, University of British Columbia, Vancouver,
Ernst-Ulrich Hartge Technical University Hamburg-Harburg, Hamburg, Germany
Daxin He Department of Chemical and Petroleum Engineering, University of Pittsburgh, Pittsburgh,
Thomas C. Ho Department of Chemical Engineering, Lamar University, Beaumont, Texas, U.S.A.
Masayuki Horio Department of Chemical Engineering, Tokyo University of Agriculture and Technology, Tokyo,
Japan
Behzad Jazayeri Fluor Daniel, Inc., Aliso Viejo, California, U.S.A.
Peijun Jiang Department of Chemical Engineering, The Ohio State University, Columbus, Ohio, U.S.A.
Yong Jin Department of Chemical Engineering, Tsinghua University, Beijing, Peoples Republic of China
S. B. Reddy Karri Particulate Solid Research, Inc., Chicago, Illinois, U.S.A.
George E. Klinzing Oce of the Provost, University of Pittsburgh, Pittsburgh, Pennsylvania, U.S.A.
T. M. Knowlton Particulate Solid Research, Inc., Chicago, Illinois, U.S.A.
Rafal P. Kobylecki
Department of Chemical Engineering, Tokyo University of Agriculture and Technology,

Tokyo, Japan.
Paul R. Mort Procter & Gamble, Cincinnati, Ohio, U.S.A.
Arun S. Mujumdar Department of Mechanical Engineering, National University of Singapore, Singapore
Richard A. Newby Science & Technology Center, Siemens Westinghouse Power Corporation, Pittsburgh,
Jens Reppenhagen
y
Technical University Hamburg-Harburg, Hamburg, Germany
Gabriel I. Tardos Department of Chemical Engineering, The City College of the City University of New York,
New York, U.S.A.
Mayumi Tsukada Department of Chemical Engineering, Tokyo University of Agriculture and Technology,
Tokyo, Japan
Yao Wang Department of Chemical Engineering, Tsinghua University, Beijing, Peoples Republic of China
Fei Wei Department of Chemical Engineering, Tsinghua University, Beijing, Peoples Republic of China
Joachim Werther Technical University Hamburg-Harburg, Hamburg, Germany
Guoqiang Yang Department of Chemical Engineering, The Ohio State University, Columbus, Ohio, U.S.A.
Wen-Ching Yang Science & Technology Center, Siemens Westinghouse Power Corporation, Pittsburgh,
J. G. Yates Department of Chemical Engineering, University College London, London, United Kingdom
Current aliation: Energy Engineering Department, Czestochowa Technical University, Czestochowa, Poland.
y
Current aliation: BMH Claudius Peters, Buxtehude, Germany.
1
Particle Characterization and Dynamics
Wen-Ching Yang
Siemens Westinghouse Power Corporation, Pittsburgh, Pennsylvania, U.S.A.
1 INTRODUCTION
Particle characterization is important in all aspects of
particle production, manufacturing, handling, proces-
sing, and applications. Characterization of particles is
the rst necessary task required in a process involving
solid particles. The required characterization includes
not only the intrinsic static parameters (such as size,
density, shape, and morphology) but also their
dynamic behavior in relation to uid ow (such as
drag coecient and terminal velocity). In this chapter,
the characterization of single particles with dierent
available techniques is rst introduced. The dynamic
behavior of a single particle in the ow eld at Stokes
ow regime, the intermediate ow regime, and the
Newtons law regime is then discussed. The chapter is
concluded with coverage of multiparticle systems.
1.1 Characterization of Single Particles
The complete characterization of a single particle
requires the measurement and denition of the particle
characteristics such as size, density, shape, and surface
morphology. Because the particles of interest are
usually irregular in shape and dierent in surface
morphology, there are many dierent ways and tech-
niques to characterize the particles. Depending on the
methods employed, the results may not be completely
consistent. Some methods may be more appropriate
than others for certain selected applications.
1.1.1 Denitions of Particle Size
The particle size is one or more linear dimensions
appropriately dened to characterize an individual
particle. For example, an ideal particle like a sphere
is uniquely characterized by its diameter. Particles of
regular shapes other than spherical can usually be
characterized by two or three dimensions. Cubes can
be uniquely dened by a single dimension, while
cuboids require all three dimensions, length, width,
and height. Two dimensions are required for regular
isotropic particles such as cylinders, spheroids, and
cones.
Irregular particles of practical interest, most often,
cannot be uniquely dened. Their sizes are usually
dened based on certain reference properties. The
choice of any particular diameter for characterization
of an irregular particle depends, in many cases, on the
intended application. Unfortunately, in most cases, the
correct choice of a representative diameter is uncertain.
Many diameters have been dened to characterize the
irregular particles. The more common ones are sum-
marized below.
Volume Diameter
The volume diameter, d
v
, is dened as the diameter of
a sphere having the same volume as the particle and
can be expressed mathematically as
d
v
=
6V
p
p
_ _
1=3
where V
p
= volume of
the particle (1)
Surface Diameter
The surface diameter, d
s
, is dened as the diameter of a
sphere having the same surface area of the particle.
Mathematically it can be shown to be
d
s
=
S
p
p
_ _
1=2
where S
p
= surface area of
the particle (2)
SurfaceVolume Diameter
The surfacevolume diameter, d
sv
, also known as the
Sauter diameter, is dened as the diameter of a sphere
having the same external-surface-area-to-volume ratio
as the particle. This can be expressed as
d
sv
=
6V
p
S
p
=
d
3
v
d
2
s
(3)
Sieve Diameter
The sieve diameter, d
A
, is dened as the width of the
minimum square aperture in the sieve screen through
which the particle will pass. The sieve diameter will be
discussed in more detail when particle size analysis by
sieving is discussed later in the chapter.
Stokes Diameter
The Stokes diameter, d
st
, is the free-falling diameter of
the particle in the Stokes law region and can be calcu-
lated from
d
st
=
18mU
t
(r
p
r
f
)g
_
where U
t
=
terminal
velocity
of the
particle
(4)
Free-Falling Diameter
The free-falling diameter, d
f
, is the diameter of a sphere
having the same density and the same free-falling velo-
city (or terminal velocity) as the particle in a uid of
same density and viscosity. In the Stokes law region,
the free-falling diameter is the Stokes diameter dened
earlier.
Drag Diameter
The drag diameter, d
D
, is dened as the diameter of a
sphere having the same resistance to motion as the
particle in a uid of the same density and viscosity
and moving at the same velocity.
Perimeter Diameter
The perimeter diameter, d
c
, is the diameter of a circle
having the same perimeter as the projected outline of
the particle.
Projected Area Diameter
The projected area diameter, d
a
, is dened as the
diameter of a sphere having the same projected area
as the particle viewed in a direction perpendicular to
the plane of the greatest stability of the particle (see
Fig. 1).
Feret Diameter
The Feret diameter, d
F
, is a statistical diameter repre-
senting the mean value of the distances between pairs
of parallel tangents to a projected outline of the par-
ticle, as shown in Fig. 1. The Feret diameter is usually
used in particle characterization employing the optical
imaging technique to be discussed later.
Martin Diameter
The Martin Diameter, d
M
, is also a statistical diameter
dened as the mean chord length of the projected out-
line of the particle, which appropriately bisects the area
of the projected prole, as shown in Fig. 1. Like the
Feret diameter, the Martin diameter is also employed
most often during particle characterization using the
optical imaging technique to be discussed later.
Only four of the above particle size denitions are of
general interest for applications in packed beds and
uidized beds. They are the sieve diameter, d
A
, the
volume diameter, d
v
, the surface diameter, d
s
, and
the surfacevolume diameter, d
sv
. The most relevant
diameter for application in a uidized bed is the sur-
facevolume diameter, d
sv
. For applications in cataly-
tic reactors with dierent isometric catalyst shapes,
Rase (1990) suggested that we use the equivalent dia-
meters summarized in Table 1.
Figure 1 Illustration for projected area diameter, Feret dia-
meter, and Martin diameter.
1.1.2 Denitions of Particle Shape
Natural and man-made solid particles occur in almost
any imaginable shape, and most particles of practical
interest are irregular in shape. A variety of empirical
factors have been proposed to describe nonspherical
shapes of particles. These empirical descriptions of
particle shape are usually provided by identifying two
characteristic parameters from the following four: (1)
volume of the particle, (2) surface area of the particle,
(3) projected area of the particle, and (4) projected
perimeter of the particle. The projected area and
perimeter must also be determined normal to some
specied axis. For axisymmetric bodies, the reference
direction is usually taken to be parallel or normal to
the axis of symmetry.
All proposed shape factors to date are open to criti-
cism, because a range of bodies with dierent shapes
may have the same shape factor. This is really inevit-
able if complex shapes are to be described only by a
single parameter. Thus in selecting a particular shape
factor for application, care must be taken to assure its
relevance.
Sphericity
Wadell (1933) proposed the degree of true sphericity
be dened as
f =
Surface area of volume equivalent sphere
Surface area of particle
=
d
v
d
s
_ _
2
=
d
sv
d
v
(5)
For a true sphere, the sphericity is thus equal to 1. For
nonspherical particles, the sphericity is always less than
1. The drawback of the sphericity is that it is dicult to
obtain the surface area of an irregular particle and thus
it is dicult to determine f directly. Usually the more
the aspect ratio departs from unity, the lower the
sphericity. The sphericity, rst introduced as a measure
of particle shape, was subsequently claimed to be use-
ful for correlating drag coecient (Wadell, 1934).
There is some theoretical justication for the use of
sphericity as a correlating parameter for creeping
ow past bodies whose geometric proportions resem-
ble a sphere. But for other circumstances its use is
purely empirical (Clift et al., 1978). Leva (1959) and
Subramanian and Arunachalam (1980) suggested
experimental methods using the Ergun equation for
evaluation of the sphericity. This methodology will
be discussed when Ergun equation is introduced in
Chapter 2, Flow through xed beds. For regularly
shaped solids, the sphericities can be calculated from
Eq. (5), and they are presented in Table 2a. As for
commonly occurring nonspherical particles, their
sphericities are summarized in Table 2b.
Circularity
Wadell (1933) also introduced the degree of circular-
ity, dened as
,=
Circumference of circle having same cross-
sectional area as the particle
Actual perimeter of the cross-section
(6)
Table 1 Suggested Equivalent Particle Diameters for Catalysts in Catalytic
Reactor Applications
Shape Equivalent particle diameter, d
p
= d
sv
= 6=a
s
Sphere d
p
= diameter of a sphere
Cylinder with length (l
y
)
equal to diameter (d
y
)
d
p
= d
y
, the diameter of a cylinder
Cylinder, d
y
,= l
y
d
p
=
6d
y
4 2d
y
=l
y
Ring with outside diameter of d
o
,
and inside diameter of d
i
d
p
= 1:5(d
o
d
i
)
Mixed sizes
d
p
=
1
i
(x
i
=d
pi
)
Irregular shapes with f = 0:5 to 0.7 d
p
= d
v
; d
v
- d
A
Source: Rase (1990).
Unlike the sphericity, the circularity can be determined
more easily experimentally from microscopic or photo-
graphic observation. For an axisymmetric particle
projected parallel to its axis, , is equal to unity. Use
of , is only justied on empirical grounds, but it has
the potential advantage of allowing correlation of the
dependence of ow behavior on particle orientation
(Cli et al., 1978).
Operational Sphericity and Circularity
Since the sphericity and circularity are so dicult to
determine for irregular particles, Wadell (1933) pro-
posed that f and , be approximated by operational
sphericity and circularity:
f
op
=
Volume of particle
Volume of the smallest circumscribing
sphere
_
_
_
_
_
_
1=3
(7)
,
op
=
Projected area of particle
Area of the smallest circumscribing
circle
_
_
_
_
1=2
(8)
For the ellipsoids, the operational sphericity, f
op
, can
be expressed as
f
op
= e
1
e
2
( )
1=3
(9)
For the rounded particles, it can be approximated by
Eq. (9). The e
1
and e
2
are called the atness ratio and
the elongation ratio, respectively, and are dened as
Table 2a Sphericities of Regularly Shaped Solids
Shape Relative Proportions f = d
sv
=d
v
Spheroid 1:1:2 0.93
1:2:2 0.92
1:1:4 0.78
1:4:4 0.70
Ellipsoid 1:2:4 0.79
Cylinder Height = diameter 0.87
Height = 2 diameter 0.83
Height = 4 diameter 0.73
Height =
1
2
diameter 0.83
Height =
1
4
diameter 0.69
Rectangular parallelpiped 1:1:1 0.81
1:1:2 0.77
1:2:2 0.77
1:1:4 0.68
1:4:4 0.64
1:2:4 0.68
Rectangular tetrahedron 0.67
Regular octahedron 0.83
Source: Adapted from Geldart (1986).
Table 2b Sphericities of Commonly
Occurring Nonspherical Particles
Material Sphericity
Sand
Round sand 0.86
Sharp sand 0.66
Crushed sandstone 0.80.9
Coal
Pulverized coal 0.73
Crushed coal 0.630.75
Activated carbon 0.700.90
Mica akes 0.28
FischerTropsch catalyst 0.58
Common salt 0.84
Crushed glass 0.65
Silica gels 0.700.90
Tungsten powder 0.89
Sillimanite 0.75
Wheat 0.85
Source: Adapted from Geldart (1986).
e
1
=
b
t
Flatness ratio (10)
e
2
=
l
b
Elongation ratio (11)
where t = thickness, the minimum distance between
two parallel planes tangential to opposite surfaces.
One of the two planes is the plane of the maximum
stability. b = breadth, the minimum distance between
two parallel planes that are perpendicular to the planes
dening the thickness and tangential to opposite sur-
faces. l = length, projected on a plane normal to the
planes dening t and b.
The three characteristic dimensions have an increas-
ing order of magnitude t < b < l.
However, f
op
is not generally a good approxima-
tion to f. Aschenbrenner (1956) showed that a better
approximation to f is given by a working sphericity,
f
w
, obtained from the atness and elongation ratios:
f
w
=
12:8 e
1
e
2
2
_ _
1=3
1 e
2
(1 e
1
) 6
1 e
2
2
1 e
2
1
_ _
_ (12)
The working sphericity has been found to correlate
well with the settling behavior of naturally occurring
mineral particles.
Wadell (1935) suggested that ,
op
provides an esti-
mate of f based on a two-dimensional projection of a
particle, and thus is sometimes called the projection
sphericity. The operation circularity, however, does
not approximate f for regular bodies and has virtually
no correlation with settling behavior of natural irregu-
lar particles (Clift et al., 1978).
The Heywood Shape Factor
The Heywood shape factor is sometimes called the
volumetric shape factor. Heywood (1962) proposed a
widely used empirical parameter based on the pro-
jected prole of a particle as follows:
k =
V
p
d
3
a
(13)
where d
a
=
4A
p
=p
_
.
The projected area diameter, d
a
, is the diameter of
the sphere with the same projected area as the particle.
A number of methods have been suggested for obtain-
ing an estimate for d
a
. Even if d
a
is available, the
Heywood shape factor can only be evaluated if V
p
is
known. For naturally occurring particles or if a distri-
bution of particle sizes or shapes is present, V
p
may not
be readily available. Automatic techniques for charac-
terizing particle shape are under development. For a
review, see Kaye (1973).
Heywood (1962) suggested that k may be estimated
from the corresponding value, k
e
, of an isometric par-
ticle of similar form by
k =
k
e
e
1
e
2
_ (14)
Values of k
e
for some regular shapes and approximate
values for irregular shapes are given in Table 3.
Equation (14) is exact for regular shapes such as
spheroids and cylinders. Heywood suggested that k
be employed to correlate drag and terminal velocity,
using d
a
and the projected area to dene Re and C
D
,
respectively. There are justications for this approach
because many natural particles have an oblate shape,
with one dimension much smaller than the other two.
Over a large Reynolds number range in the intermedi-
ate regime, such particles present their maximum area
to the direction of motion, and this is the area char-
acterized by d
a
. There is also evidence that the shape of
this projected area, which does not inuence k, has
little eect on drag (Clift et al., 1978).
A more modern approach uses fractal analysis, or
Fourier transformation. The latter is a bit involved,
requiring several coecients for complex denition.
Table 3 Approximate Values of k
e
and k
k
e
for isometric irregular shapes
Rounded 0.56
Subangular 0.51
Augular
tending to a prismoidal 0.47
tending to a tetrahedron 0.38
k
e
for selected natural particles
Sand 0.26
Bituminous coal 0.23
Limestone 0.16
Gypsum 0.13
Talc 0.16
k for regular shape
Sphere 0.524
Cube 0.696
Tetrahedron 0.328
Cylinder with an aspect ratio of 1
viewing along axis 0.785
viewing normal to axis 0.547
Spheroids
with an aspect ratio of 0.5 0.262
with an aspect ratio of 2 0.370
Source: Heywood (1962).
1.1.3 Denitions of Particle Density
There are several particle density denitions available.
Depending on the application, one denition may
be more suitable than the others. For nonporous
particles, the denition of particle density is straight-
forward, i.e., the mass of the particle, M, divided by
the volume of the particle, V
p
, as shown in Eq. (15).
r
p
=
M
p
V
p
=
mass of the particle
volume that the particle would displace
if its surface were nonporous
(15)
For porous particles with small pores, the particle
volume in Eq. (15) should be replaced with the
envelope volume of the particle as if the particles
were nonporous as shown in Fig. 2. This would be
more hydrodynamically correct if the particle behavior
in the ow eld is of interest or if the bulk volume of
the particles is to be estimated. For total weight esti-
mation, then the skeleton density should be known.
The skeleton density is dened as the mass of the par-
ticle divided by the skeletal volume of the particle. In
practice, the pore volume rather than the skeletal
volume is measured through gas adsorption, gas or
water displacement, and mercury porosimetry. These
techniques will be discussed in more detail later. There
are also porous particles with open and closed pores.
The closed pores are not accessible to the gas, water or
mercury and thus their volume cannot be measured. In
this case, the calculated skeleton density would include
the volume of closed pores as shown in Fig. 2. For
nonporous particles, the particle density is exactly
equal to the skeleton density. For porous particles,
the skeleton density will be larger than the particle
density.
When a porous particle is broken into smaller
pieces, the particle density of the smaller pieces will
usually be larger than the original particle density by
virtue of the elimination of some pores. When the
particle size becomes ner, the particle density will
approach that of a nonporous particle because all the
pores will be completely eliminated. In the process of
handling porous particles, this tendency should be kept
in mind, and the particle density should be carefully
evaluated.
Knight et al. (1980) described a method of deter-
mining the density of nely divided but porous par-
ticles such as uid bed cracking catalyst. They
suggested to set a known mass and measurable
volume of powder in resin and then measure the voi-
dage of the sample sections of the resin. The results
were satisfactory, though the method was quite
tedious. Buczek and Geldart (1986) proposed to use
very nd dense powders as the pycnometric uid to
determine the density of porous particles. For a pow-
der to be a suitable pycnometric uid, it should be
nonporous and free-owing. Its density should be
much larger than the porous particles to be measured,
and its diameter should be much smaller than the
porous particles and greater than the biggest pores
of the porous particles. It should also be nonreactive
toward the porous particles. Akapo et al. (1989) also
proposed a novel method of determining particle den-
sity of porous aeratable powders. The method
depends on measurement of bed expansion of a gas
uidized bed of the powder in the region between the
minimum uidization and the minimum bubbling.
The Richardson and Zaki bed expansion equation
was then applied to back out the particle density.
The Richardson and Zaki bed expansion equation
and the minimum uidization and minimum bubbling
regimes will be discussed in Chapter 3, Bubbling
Fluidized Beds.
2 PARTICLE CHARACTERIZATION
TECHNIQUES
There are many techniques that can be employed to
characterize particles, some simple and primitive and
some complicated and sophisticated. Almost every
technique is associated with intrinsic experimental
errors and implicit assumptions. Thus care must be
exercised to select proper techniques for your specic
applications. The available techniques are reviewed in
this section. Figure 2 Skeleton particle density.
2.1 Methods for Direct Characterization of Particle
Size and Shape
2.1.1 Sieve Analysis
The most commonly used method for classifying
powders is to sieve the particles through a series of
screens with standardized mesh size by sifting, swirling,
shaking, or vibrating. Two standard mesh sizes, U.S.
sieve size and Tyler sieve size, are usually used in the
U.S. In the European practice, the British Standard
and German DIN sieve sizes are also employed. The
mesh number of a sieve refers to the number of parallel
wires per inch in the weave of the screen. The
American Society of Testing Materials (ASTM) speci-
cations for standard mesh sizes for those systems are
compared in Table 4. The mesh size was designed so
that the aperture or opening of the alternating mem-
bers in the series has a factor of 2
1=2
. For example, the
U.S. 12 mesh with an opening of 1.68 mm is 2
1=2
times
the U.S. 16 mesh with an opening of 1.19 mm. The
result of the sieve size analysis is commonly plotted
in a logarithmic-scale graph expressing the cumulative
weight percentage under size as the abscissa and the
particle size as the ordinate, as shown in Fig. 3. There,
the 2
1=2
factor in the mesh size arrangement allows the
size data points to be almost equally spaced on the
logarithmic scale. Also, experimentally, many crushed
materials yield a straight line if plotted as in Fig. 3.
The inaccuracies and uncertainties of sieve analysis
stem from the discrete steps of the mesh size arranged
at an approximate factor of 2
1=2
between successive
mesh sizes. Sieve analysis does not provide the infor-
mation for the largest and the smallest particle sizes.
The size cut provides an approximate value for the
mean particle size within the cut. Sieve analysis also
does not dierentiate the particle shape. A needle-
shaped particle can either pass through a mesh or be
retained on the screen, depending on its orientation
during sifting. The result of sieve analysis is also
dependent on the time of sieving action, the particle
loading on the sieve, and sieve blinding (also called
pegging). Enlargement of aperture due to wire erosion
of a sieve can cause discrepancy as well. For small
particles, agglomeration due to static electricity or
moisture can also occur.
The smallest mesh size for the Tyler Series is 400
mesh, equivalent to a 38 mm opening, while the smallest
mesh size for the U.S. Series is 635 mesh, equivalent to
a 20 mm opening. For particles ner than 20 mm, the
surface and electrostatic forces become important, and
particle classication by sieve analysis is not recom-
mended. Detailed discussion of sieve analysis techni-
ques can be found in standard textbook such as Allen
(1975) and Kaye (1981).
2.1.2 Imaging Technique
Direct measurement of particle dimensions is also pos-
sible from enlarged photographic or electronic images
of microscopes. There are three types of microscopes
commonly employed, i.e., the optical microscope, the
scanning electron microscope (SEM), and the trans-
mission electron microscope (TEM). The optical
microscope is employed for particles from 1 mm to
about 150 mm. Both SEM and TEM make use of elec-
tron beams and can be used for particles from 5 mm
down to as small as 0:01 mm. They are especially useful
for revealing the surface morphology of extremely
small particles.
Particles to be imaged in an optical microscope are
usually dispersed in a drop of viscous uid on a glass
slide. Their images are then visually compared with a
set of standard circles, geometric shapes, or linear grids
to derive their actual sizes and shapes. The Martin and
Feret diameters are also often used to characterize a
particle. The Martin diameter is dened as the magni-
tude of the chord that divides the image into two equal
areas with respect to a xed direction (see Fig. 1). The
Feret diameter of a particle image is dened as the
length of the image as projected with respect to a
specic reference direction (see Fig. 1). Both Martin
and Feret diameters are intended to be statistical dia-
meters, i.e., after characterization of many images.
Thus any slight departure from the true randomness
in the eld of view of particle images can produce bias
in the particle size characterization. Modern instru-
ments couple television cameras interfaced with com-
puters and sophisticated software can speed up the
imaging analysis considerably.
Martins statistical diameter is also referred to as the
mean linear intercept or mean chord and has been
shown to relate to the specic surface in the following
manner (Herdan, 1960):
Martin diameter =
4
rS
v
(16)
where r = the density of packing and S
v
= particle
surface area per unit volume of particle.
Based on the experimental evidence, on the average,
the Martin diameter is usually smaller than the mean
projected diameter and the Feret diameter larger. Since
both Martin and Feret diameters depend on the
particle shape, the ratio of Feret diameter to Martin
diameter is usually fairly constant for the same mate-
rial. For Portland cement, it is about 1.2; for ground
quartz and ground glass, approximately 1.3 (Herdan,
1960).
The microscopic measurement technique is most
suitable for particles relatively uniform in size and
granular in shape, because a large number of particles,
between 300 to 500, needs to be measured to minimize
statistical error.
2.1.3 Gravity and Centrifugal Sedimentation
The falling speeds of particles in a viscous uid under
the inuence of gravity are used to measure the particle
Table 4 Summary of Various Types of Standard Sieve
U.S.
mesh No.
Standard
opening, mm
Tyler
mesh No.
British
mesh No.
Standard
opening, mm
German
DIN No.
Standard
opening, mm
3
1
2
5.66 3
1
2
1 6.000
4 4.76 4
5 4.00 5
6 3.36 6 5 3.353 2 3.000
7 2.83 7 6 2.812

8 2.38 8 7 2.411 2
1
2
2.400
10 2.00 9 8 2.057 3 2.000
12 1.68 10 10 1.676 4 1.500

14 1.41 12 12 1.405

16 1.19 14 14 1.204 5 1.200

18 1.00 16 16 1.003 6 1.020
20 0.84 20 18 0.853
8 0.750
25 0.71 24 22 0.699

30 0.59 28 25 0.599 10 0.600
11 0.540
35 0.50 32 30 0.500 12 0.490
40 0.42 35 36 0.422 14 0.430
45 0.35 42 44 0.353 16 0.385

50 0.297 48 52 0.295 20 0.300
60 0.250 60 60 0.251 24 0.250
70 0.210 65 72 0.211 30 0.200
80 0.177 80 85 0.178
100 0.149 100 100 0.152 40 0.150

120 0.125 115 120 0.124 50 0.120
140 0.105 150 150 0.104 60 0.102
170 0.088 170 170 0.089 70 0.088

200 0.074 200 200 0.076 80 0.075
230 0.062 250 240 0.066 100 0.060
270 0.053 270 300 0.053
325 0.044 325
400 0.038 400
635 0.020
size in the gravity sedimentation technique. The mea-
sured speeds are then converted to Stokes diameters by
applying the Stokes equation [Eq. (4)], assuming that
the particles are all spherical in shape. Since the irre-
gularly shaped particles fall with dierent orientations
in the vertical direction and thus dierent settling velo-
cities, similar irregularly shaped particles can have a
range of Stoke diameters. Disturbances caused by the
presence of other particles, the concentration eect,
can also be important. Recommendations for the
upper limit particle concentration to be used during
the sedimentation analysis varies from 0.01 to 3.0%
(Kaye, 1981). An upper limit of 0.05% was recom-
mended by Kaye, and in certain practical cases up to
0.2% by volume is permissible. The probability of
forming clusters increases with particle suspension
concentration. The clusters tend to fall at a higher
speed and thus introduce measurement error.
The hindering eect of the containing wall on the
falling speed of the particles cannot be ignored either.
For a spherical particle, the eect can be expressed by
the Landenburg equation as
V
o
= V
m
1 2:4
d
p
D
_ _
(17)
According to Eq. (17), even if the column is 50 times
the diameter of the particle, there is still a 5% reduc-
tion in the falling speed of the particle.
Two basic suspension systems, the line start system
(or the two-layer sedimentation system) and the initi-
ally homogeneous system are employed in the gravity
sedimentation technique. In the line start system, a thin
layer of particles is placed at the top of the sedimenta-
tion column, and its settling behavior is analyzed by
dierent techniques. In the initially homogeneous sys-
tem, the column is homogenized rst and its settling
pattern is subsequently studied.
Classical techniques for measuring the sedimenta-
tion behavior include taking samples with a pipette,
measurement of height of sediment layer at the bot-
tom, and use of balance pan to measure the weight of
settled particles. Modern sedimentometers make use of
the diraction pattern of a light beam, the power loss
of an x-ray, or a Doppler shift of a laser beam. The
Figure 3 Graphic representation of sieve size analysislogarithmic-scale graph expressing cumulative % oversize or undersize.
modern techniques are primarily applied to monitor
the sedimentation kinetics of an initially homogeneous
suspension. The photosedimentometers measure the
beam with a small-angle detector. This application is
based on the LambertBeer law, which is
I
c
= I
o
e
aCL
(18)
where I
o
= the original light beam intensity, I
c
= the
intensity of a light beam after passing through a par-
ticle suspension of concentration C, L = the length of
the light beam path, and a = an empirical constant
depending on equipment and particle characteristics.
In the x-ray sedimentometers, the x-ray absorption
of a sedimentation suspension is used to measure the
concentration gradient in the suspension. The law out-
lined by Olivier et al. (1970/1971) is employed:
ln T = bW
s
(19)
where T = the measured transmittance, X
c
=X
o
, X
o
=
the intensity of the x-ray beam passing through the
sedimentation column lled with clear uid, X
c
= the
intensity of the x-ray beam passing through a particle
suspension of concentration C, W
s
= the weight frac-
tion of particles in the x-ray pass, and b = an empirical
constant depending on equipment, particle, and sus-
pending uid characteristics.
Alternatively, centrifugal force instead of gravita-
tional force can be created to enhance the sedimenta-
tion performance. Depending on the size of the
centrifugal arm and the speed of rotation, many
times the gravitational force can be applied.
The modern nonintrusive sedimentometers are e-
cient, can be adapted easily for hostile environments,
and can provide information in situ and thus are ideal
for quality control in manufacturing processes. The
resolution and sensitivity allow measurement of
Stokes diameters down to about 0:5 mm.
2.1.4 Characterization by Elutriation
Particle size characterization by elutriation makes use
of the same kind of principles employed by the sedi-
mentation. Instead of letting particles settle with grav-
ity, the particles are actually carried out against gravity
during elutriation. In vertical elutriators, the particles
with terminal velocities less than the vertical uid velo-
city will be elutriated out. By operating the elutriator
at dierent ow conditions, the particle size distribu-
tion of the sample can be calculated. The ow in the
elutriators is usually laminar ow, and the Stokes
equation is used to estimate the Stokes diameter of
the particle by assuming that the particle is spherical.
As in sedimentation, the concentration of the parti-
cles in the elutriators aects the results of measure-
ment. Also the velocity in the elutriators tends to be
parabolic rather than uniform and thus introduces
errors in measurement. Centrifugal force can also be
applied like that in the sedimentation to enhance the
performance. Dierent designs, including horizontal
elutriators, are available.
2.1.5 Cascade Impaction Technique
The cascade impaction technique is based on similar
principles employed in the elutriation technique and is
also based on the inertia of the particles. The particles
with terminal velocities smaller than the ow velocity
will be carried along by the gas ow. In addition,
smaller particles will tend to follow the streamlines of
the ow better than the larger particles owing to their
smaller inertia. When the ow changes direction
because of the presence of a plane surface, larger
particles with larger inertia will impact on the plane
surface and be collected. By successive decreases in
ow velocity from stage to stage, the particles can be
collected and classied into dierent particle size
fractions. Earlier designs employed slides coated with
adhesive to collect particles for microscopic analysis.
Modern devices have many more variations. The cas-
cade impaction technique is usually applied for particle
sizes between 0.1 and 100 mm.
2.1.6 Resistivity and Optical Zone Sensing
Techniques
The resistivity and optical zone sensing techniques
measure the particle size by measuring the changes of
resistivity or optical properties when the particles are
passed through the sensing zone of the instruments.
The well-known Coulter counter is a resistivity zone
sensing instrument (see Fig. 4). The particles to be
analyzed are rst suspended and homogenized in an
electrolyte and then are forced to pass through a
cylindrical orice placed between two electrodes. The
passages of particles through the orice generates vol-
tage pulses that are amplied, recorded, and analyzed
to produce a particle size distribution. The instrument
is usually calibrated with standard particles such as
latex spheres of known size. The data are analyzed
by assuming that the sensing zone is isotropic (i.e.,
the exact location of the particles in the sensing zone
is unimportant) and the pulse height is proportional to
the volume of the particle. The results from this ana-
lysis are thus the equivalent spherical diameters of the
particles. For the simple voltage-pulse-to-particle-
volume relationship to hold, the particle size to be
analyzed needs to be less than about 40% of the orice
diameter. Particles smaller than 3% of the orice
diameter do not produce a reliable result. Thus for a
sample of wide size distribution, changes of orices of
dierent diameters are usually necessary.
Possible errors during measurement arise from the
possibility of more than one single particle occupying
the sensing zone at the same time. This can easily be
solved by diluting the suspension or making multiple
measurements with increasingly dilute suspensions. If
the particles substantially deviate from the spherical
shape, the equivalent spherical diameters obtained
with this technique may not have any physical signi-
cance. Another potential error stems from the fact that
particles may not all pass through the axis of the ori-
ce. Similar size particles will generate dierent voltage
signals depending on whether they pass through the
orice at the axis or close to the cylindrical wall.
Modern instruments are increasingly capable of editing
the signals to reject those stray signals. The electrical
zone sensing technique can be applied for particles
ranging from 0.6 to 1200 mm.
The optical sensing technique measures the scat-
tered light from a particle passing through a sensing
volume illuminated by a light source, such as a white
light or a laser. The intensity of the scattered light is
then related to the size of the particle. In an ideal
situation, a monotonic relationship exists between the
intensity of the scattered light and the particle size and
thus allows unique determination of the size of the
particle. In reality, the light-scattering properties of a
particle depend in a very complex way on its refractive
index and shape, and on the wavelength of the light
used to illuminate the particle. Generally, there are two
basic designs: the scattered light can either be collected
in a narrow forward direction in the direction of the
illuminating light or be collected in a wide angular
range. The forward scattering systems are more suit-
able for sizing particles with light-absorbing proper-
ties, since for these particles there exists a monotonic
relationship between the intensity and the size. For
nonabsorbing particles, however, multiple values exist
for particles larger than 1 mm. Thus most of the exist-
ing particle-sizing instruments make use of the second
scattering geometry to collect the scattered light into a
large angular range oriented either perpendicularly or
axially with respect to the light beam direction, see Fig.
5. The instrument is usually calibrated with nonab-
sorbing spherical polystyrene latex particles, and the
scattered light intensity is a monotonic function of
the particle size. However, the presence of absorptivity
can reduce substantially the sizing sensitivity of the
instrument.
The light scattering theories employed for this tech-
nique are by Mie, Rayleigh, and Fraunhofer (Bohren
and Human, 1986). When the dimensions of a
particle are of the same order of magnitude of the
wavelength of the incident light, Mie theory is used.
When the particle is much smaller than the wavelength
of the light, the appropriate light-scattering theory is
that of Rayleigh. For particles much larger than the
wavelength of the light, Fraunhofer diraction is
employed. For more detailed discussions of principles
Figure 4 Electrical zone sensing (Coulter counter).
Figure 5 Laser-diraction spectrometry.
behind this application, see Fan and Zhu (1998). An
innovative instrumentation using three lasers to pro-
duce scattered light through an angular range from
close to 0
up to 169
in one continuous pattern was

described by Freud et al. (1993). Bonin and Queiroz
(1991) also described an application in an industrial-
scale pulverized coal-red boiler measuring local par-
ticle velocity, size, and concentration.
Besides for particle sizing, the optical sensing tech-
nique can also be employed for measurement of the
particle number concentration. Discussions mentioned
earlier regarding the measurement errors for the resis-
tivity sensing zone technique apply here as well. When
more than one particle is present in the sensing zone,
error can occur.
2.1.7 Techniques for Particle Surface
Characterization
In many applications related to chemical reactions, the
total surface area is more important than the particle
size and shape. To measure the particle surface area,
two commonly used techniques are the permeability
technique and the gas adsorption technique. The per-
meability technique is a method for the determination
of the power surface area by measuring the permeabil-
ity of a powder bed. The permeability is dened in
Chapter 2 in relation to Darcys law. The derivation
of pressure drop equations through power beds is quite
involved and will be discussed in Chapter 2, for mea-
suring the powder surface area the reader can refer to
Allen (1975).
Another technique commonly used to measure the
powder surface area and the pore size is the physical gas
adsorption technique based on the well-known BET
(BrunauerEmmettTeller) method on monolayer
coverage of adsorptives such as nitrogen, krypton,
and argon. The application is very well established,
and detailed discussions are available in Allen (1975).
2.2 Effect of Particle Shape on Size Distribution
Measured with Commercial Equipment
The eect of particle shape on particle size distribution
was investigated by Naito et al. (1998) with commer-
cial particle size analyzers based on ve dierent
measuring principles: electrical sensing zone, laser dif-
fraction and scattering, x-ray sedimentation, photose-
dimentation, and light attenuation. The particles used
are blocky aluminum oxide and barium titanate parti-
cles, aky boron nitride particles, and rodlike silicon
whisker ceramic powders. Altogether, four commer-
cially available models based on the electrical sensing
zone principle, eight commercially available models
based on the laser diraction and scattering principle,
three commercially available models based on the x-
ray sedimentation principle, seven commercially avail-
able models based on the photosedimentation princi-
ple, and one commercially available model based on
the light attenuation principle were employed in a
round-robin evaluation. They found that the eect of
anisotropic particles such as aky and rodlike particles
on the particle size measured by the commercial equip-
ment is much larger than that of the isotropic block-
shaped particles. In the x-ray sedimentation technique,
the equivalent volume diameter is usually measured.
The eect of particle shape is thus small, because the
transmittance intensity of x-rays is independent of the
particle shape and its orientation. Similarly, the eect
of particle shape was also found to be small when the
light attenuation technique was employed, because the
particles were dispersed at random in the agitated sam-
ple suspension. The laser diraction and scattering
technique, on the contrary, resulted in a large range
of size distributions, because the particles tended to
orient along the shear ow, especially the anisotropic
particles. The eect of particle shape also produced a
large range of size distributions in the coarse-size range
of the distribution in the photosedimentation techni-
que owing to the turbulence of liquid created by the
particle orientation, which resulted in intensity uctua-
tion of the transmitted light at the initial stage of grav-
itational sedimentation.
If the particle size distributions are characterized by
the average diameters at the cumulative masses of 10,
50, and 90%, the data scattering measured by dierent
techniques can be represented by the diameter ratios at
the cumulative masses of 10, 50, and 90%, or DR
10
,
DR
50
, and DR
90
. Since the particle diameter measured
by the electrical sensing zone technique is to be the
equivalent volume diameter and is independent of the
particle shape, its particle diameter is dened to be one.
The particle diameters measured by other techniques
are then ratioed with this diameter. The experimental
results of DR
10
, DR
50
, and DR
90
are summarized in
Table 5. It can be seen that the results of particle
analysis from dierent techniques can be quite dier-
ent.
Because of dierences in measuring techniques
based on dierent properties of irregular particles,
the particle size and size distribution obtained by dif-
ferent methods from the same sample are usually
dierent. The size and distribution measured by sieve
analysis will be dierent from that by laser diract-
ometer, x-ray sedimentation, and so on. Conversion
factors are available to convert particle size distribu-
tions obtained by one method to that measured by
another technique. Austin (1998) suggested a technique
to perform this correction and provided an equation
for conversion between laser diractometer and x-ray
sedigraph measurements.
The applicable particle size ranges for various par-
ticle sizing techniques are summarized in Table 6.
2.3 Measurement of Mechanical Properties of
Particles
2.3.1 Hardgrove Grindability Index (HGI)
The Hardgrove Grindability Index was originally
developed by Babcock and Wilcox to measure the rela-
tive ease of pulverizing coal. This ASTM D409-71
method has since been employed for characterizing
other particles as well. The method processes 50 g of
air-dried particles screened to a size of 16 30 mesh
(1180 600 mm) in a small ball-and-race mill for 60
revolutions. The amount of material (W
200
) passing
the 200 mesh (75 mm) screen is then measured. The
result is then compared with a standard index to arrive
at the HGI or the HGI is calculated from the equation
HGI = 13 6:93W
200
(20)
The values of the HGI usually range from 15 to 140.
The higher the HGI, the higher is the grindability of
the material. The HGI has been found to correlate
with the attrition characteristics of the particles in
uidized beds and in pneumatic transport lines
(Davuluri and Knowlton, 1998). The HGI does not
directly relate to hardness. For example, some materi-
als such as plastics are dicult to grind.
2.3.2 Attrition Index
For application in uidization and uidparticle
systems, the attrition index is probably the most
important particle characteristic. The particle attrition
can aect the entrainment and elutriation from a ui-
dized bed and thus subsequently dictate the design of
downstream equipment. The attrition in a pneumatic
transport line can change the particle size distribution
of the feed material into a uidized bed reactor and
thus alter the reaction kinetics. Davuluri and
Knowlton (1998) have developed standardized proce-
dures to evaluate the Attrition Index employing two
techniques, solids impaction on a plate and the
Davison jet cup. The two test units used are shown
in Figs. 6 and 7. They found that these two test tech-
niques are versatile enough to be applicable for a wide
range of materials, such as plastic, alumina, and lime-
Table 5 Experimental Results of DR
10
, DR
50
, and DR
90
Sensing technique
Aluminum
oxide
Barium
titanate
Boron
nitride
Silicon
nitride
whisker
DR
10
Electrical sensing zone 1.00 1.00 1.00 1.00
Laser diraction and scattering 0.61 0.77 0.94 0.61
X-ray sedimentation 0.82 1.10 0.69
Photosedimentation 0.82 0.95 0.58 0.75
Light attenuation 1.00 1.15 0.96 0.96
DR
50
DR
90
Source: Naito et al. (1998).
stone. For a more recent review on attrition, see
Werther and Reppenhagen (1999).
2.3.3 Abrasiveness Index
In pulverized coal combustion, the abrasiveness of the
particles severely limits the life of the pulverizer grind-
ing elements. The Abrasiveness Index in this applica-
tion is usually determined by the YanceyGeer Price
apparatus (Babcock and Wilcox, 1992). In this test,
four metal test coupons attached to a rotating shaft
are rotated at 1440 rpm for a total of 12,000 revolu-
tions in contact with a sample of 6350 mm0
(0:25 in: 0) coal. The relative Abrasiveness Index is
then calculated from the weight loss of the coupons.
Babcock and Wilcox has estimated the wear of full-
scale pulverizers on the basis of the YanceyGeer
Price Index.
2.3.4 Erosiveness Index
Babcock and Wilcox (1992) has also developed a
method of quantifying the erosiveness of coal by sub-
jecting a steel coupon to a stream of pulverized coal
under controlled conditions. The weight loss of the
coupon is an indication of the erosiveness of the parti-
cular coal and the potential damage to the processing
and handling equipment, and other boiler components.
3 FLUID DYNAMICS OF A SINGLE PARTICLE
For a particle moving in a uid, the force acting on the
surface of a particle depends only on the ow of the
uid in its immediate vicinity. For the simplest case, let
us consider a single particle moving at a velocity U
r
relative to its immediate uid around the particle. It is
also assumed that the uid is newtonian and that the
U
r
is constant. The uid dynamic parameters can then
be evaluated as follows.
3.1 Denition of Particle Drag Coefcient
The drag coecient is dened as the ratio of the force
on the particle and the uid dynamic pressure caused
by the uid times the area projected by the particles, as
shown in Eq. (21) and Fig. 8.
Table 6 Applicable Particle Size Ranges for Various Particle Sizing Methods
Particle sizing methods
Applicable particle size
range (mm) Measured dimension
Sieving
Dry >10 Sieve diameter
Wet 2500
Microscopic examination
Optical 1.0100 Length, projected area,
Electronic 0.01500 statistical diameters
Zone sensing
Resistivity 0.61200 Volume
Optical 1.0800
Elutriation
Laminar ow 375 Stokes diameter
Cyclone 850
Gravity sedimentation
Pipette and hydrometer 1100
Photoextinction 0.5100 Stokes diameter
X-ray 0.1130
Centrifugal sedimentation
Mass accumulation 0.525
Photoextinction 0.05100 Stokes diameter
X-ray 0.15
Centrifugal classication 0.550
Gas permeability 0.140
Gas adsorption 0.00550
Cascade impaction 0.0530
Source: Adapted from Pohl (1998) and Lloyd (1974).
C
D
=
F
(1=2)r
f
U
2
r
A
p
(21)
or
F =
1
2
C
D
r
f
U
2
r
A
p
(22)
The drag coecient, C
D
, is a function of particles
Reynolds number, (Re)
p
= U
r
d
p
r
f
=m, only. There are
three dierent regimes distinguishable depending on
the magnitude of the particles Reynolds number.
3.1.1 The Stokes Regime
C
D
=
24
(Re)
p
for (Re)
p
< 0:2 (23)
The Stokes law regime is also commonly known as the
creeping ow regime. In this regime, the viscosity of
the uid is dominating.
3.1.2 The Intermediate Regime
C
D
= f (Re)
p
for 0:2 < (Re)
p
< 500 (24)
In the intermediate regime, the drag coecient is a
function of the particles Reynold number.
3.1.3 The Newtons Law Regime
C
D
= 0:44 for (Re)
p
> 500 (25)
In the Newtons law regime, C
D
is relatively constant,
and the force F is largely due to the inertia of the uid
rather than to the viscosity of the uid.
3.2. The Stokes Law Regime
Only in the Stokes law regime, (Re)
p
< 0:2, have
theoretical methods of evaluating C
D
met with much
success. The theoretical analysis starts with the viscous
ow around a rigid sphere, which can be expressed as
F
s
= 3pmd
p
U
r
(26)
Figure 6 Schematic of attrition apparatus employing the
solids impaction principle.
Figure 7 Schematic of attrition apparatus employing
Davidson jet cup.
Figure 8 Denition of particle drag coecient.
Two-thirds of the drag force is due to the viscous shear
stresses acting on the particle surface (skin friction).
The other one-third is due to the dierence in pressure
on its surface (form drag).
Equating Eqs. (22) and (26), we have
F
s
= 3pmd
p
U
r
=
1
2
C
D
r
f
U
2
r
1
4
pd
2
p
(27)
or
C
D
= 24
m
d
p
U
r
r
f
_ _
=
24
(Re)
p
(28)
The linear relation between the F
s
and U
r
is an impor-
tant property of the Stokes law. It enables considerable
simplications in many problems.
3.3 The Stokes Free Fall Velocity
Equating the drag force and the gravitational force for
a single spherical particle, we have
F
s
=
p
8
C
D
r
f
U
2
r
d
2
p
=
pd
3
p
6
r
p
r
f
_ _
g (29)
or
C
D
=
4
3
d
p
r
p
r
f
_ _
g
r
f
U
2
r
(30)
or
U
r
= U
t
=
d
2
p
r
p
r
f
_ _
g
18m
(31)
Equation (4) for the Stokes diameter can be derived
from Eq. (31).
3.4 Wall Effect
When the uid is of nite extent, there are two eects.
The uid streamlines around the particle impinge on
the walls and reect back on the particle, causing
increasing drag. Also, since the uid is stationary at
a nite distance from the particle, this distorts the
ow pattern and increases drag. The simple correction
can be expressed as
F
s
= 3pd
p
mU
r
1 k
c
d
p
L
w
_ _
(32)
where L
w
= the distance from the center of the particle
to the walls, k
c
= 0:563 (for a single wall), k
c
= 1:004
(for two walls), and k
c
= 2:104 (for a circular cylinder).
3.5 Corrections to the Stokes Approximation
In 1910, Oseen provided the rst correction term to the
Stokes approximation as well as a mathematically self-
consistent rst approximation, as shown in Eq. (33):
C
D
=
24
(Re)
p
1
3(Re)
p
16
_ _
(33)
More recently, Proudman and Pearson (1957) pro-
vided the key to a more straightforward calculation
of further correction terms. The results for the drag
force on the sphere in accordance with Stokes,
Oseens, and Proudman and Pearsons calculations
are given in Table 7. A very large and important col-
lection of solutions of such problems can be found in
the book by Happel and Brenner (1965).
3.6 Empirical Drag Coefcient Expression
The preferred correlations to be used in dierent
ranges of particle Reynolds numbers were recom-
mended by Clift et al. (1978) and are summarized in
Table 8. In 1986, Turton and Levenspiel proposed a
single correlation applicable for the complete range of
Reynolds numbers and considerably simplied the cal-
culation of the single-particle drag coecient.
The Turton and Levenspiel (1986) equation is
C
D
=
24
(Re)
p
_
1 0:173(Re)
0:657
p
_
0:413
1 16:300(Re)
1:09
p
(34)
Haider and Levenspiel (1989) subsequently improved
the equation to cover the nonspherical particles and
proposed
Table 7 Summary of Theoretical Expressions for the
Drag Force [(Re)
p
_ 1]
Stokes (1851)
F
s
= 3pd
p
mU
r
Oseen (1910)
F = F
s
1
3
16
(Re)
p
O (Re)
2
p
_ _
_ _
Proudman and Pearson (1957)
F = F
s
1
3
16
(Re)
p

9
160
(Re)
2
p
ln
(Re)
p
2
_ _
O (Re)
2
p
_ _
_ _
C
D
=
24
(Re)
p
1 8:1716e
4:0655f
_ _
(Re)
0:09640:5565f
p
_ _
73:69 e
5:0748f
_ _
(Re)
p
(Re)
p
5:378e
6:2122f
(35)
For spherical particles, Eq. (35) reduces to
C
D
=
24
(Re)
p
3:3643(Re)
0:3471
p

0:4607(Re)
p
(Re)
p
2682:5
(36)
3.7 Corrections for Nonspherical Particles
The book by Clift et al. (1978) contains an extensive
review on this subject. The treatment is, however,
mostly for the axisymmetric particles such as spheroids
and cylinders and orthotropic particles such as rectan-
gular parallelepipeds. For particles of arbitrary shape,
no fully satisfactory method is available for correlating
the drag.
Pettyjohn and Christiansen (1948) determined the
free-settling rates of isometric particles of the following
shapes and sphericities: spheres (f = 1), cube octa-
hedron (f = 0:906), octahedron (f = 0:846), cube
(f = 0:806), and tetrahedron (f = 0:670). Their results
suggest that the correction factor should be
K = 0:8431 log
f
0:065
_ _
(Re)
t
< 0:05;
0:67 < f < 1
(37)
where K is the ratio of the settling velocity of the
volume-equivalent sphere to the settling velocity of
the particle. Results on other isometric particles have
also been published. These include the experimental
results of Heiss and Coull (1952) using solid cylinders
and rectangular parallelpipeds, Becker (1959) using
prisms and cylinders, Christiansen and Barker (1965)
using cylinders, prisms, and disks, Isaacs and Thodos
(1967) using cylinders, and Hottovy and Sylvester
(1979) using roundish irregularly shaped particles.
Table 8 Recommended Empirical Drag Correlations w = log
10
(Re)
p
(Re)
p
< 0:01
C
D
=
3
16
24
(Re)
p
0:01 < (Re)
p
_ 20
log
10
C
D
(Re)
p
24
1
_ _
= 0:881 0:82w 0:05w
2
C
D
=
24
(Re)
p
1 0:1315(Re)
(0:820:05w)
p
_ _
20 _ (Re)
p
_ 260
log
10
C
D
(Re)
p
24
1
_ _
= 0:7133 0:6305w
C
D
=
24
(Re)
p
1 0:1935(Re)
0:6305
p
_ _
260 _ (Re)
p
_ 1500 log
10
C
D
= 1:6435 1:1242w 0:1558w
2
1:5 10
3
_ (Re)
p
_ 1:2 10
4
log
10
C
D
= 2:4571 2:5558w 0:9295w
2
0:1049w
3
1:2 10
4
_ (Re)
p
_ 4:4 10
4
log
10
C
D
= 1:9181 0:6370w 0:0636w
2
4:4 10
4
_ (Re)
p
_ 3:38 10
5
log
10
C
D
= 4:3390 1:5809w 0:1546w
2
3:38 10
5
_ (Re)
p
_ 4 10
5
C
D
= 29:78 5:3w
4 10
5
_ (Re)
p
_ 10
6
C
D
= 0:1w 0:49
10
6
< (Re)
p
C
D
= 0:19
8 10
4
(Re)
p
Source: Adapted from Clift et al. (1978).
In the Stokes law regime, homogeneous symmetrical
particles can take up any orientation during their sett-
ling in a uid of innite extent. Ellipsoids of uniform
density and bodies of revolution with fore and aft sym-
metry can attain spin-free terminal states in all orienta-
tions. The terminal velocities of the ellipsoids will,
however, depend on their orientation. A set of identical
particles can therefore have a range of terminal velo-
cities, though the range is generally fairly small.
Asymmetric particles such as ellipsoids and discs do
not generally fall vertically unless they are dropped
with a principal axis of symmetry parallel to a gravity
eld. They tend to drift from side to side.
In the intermediate ow regime, particles adopt
preferred orientations. Particles will usually align
themselves with their maximum cross section normal
to the direction of relative motion. There is no appre-
ciable secondary motion in the intermediate ow
regime, so results for ow past xed objects of the
same shape can be used if the orientation corresponds
to the preferred orientation.
Secondary motion associated with wake shedding
occurs in the Newtons law regime, and C
D
is insensi-
tive to (Re)
p
. In this regime the density ratio, l, plays
an important role in determining the type of motion
and the mean terminal velocity. For particles of arbi-
trary shapes,
C
D
= l
1=18
(5:96 5:51f) for 1:1 < l < 8:6
(38)
Pettyjohn and Christiansen (1948) also suggested a
correction factor for nonspherical particles:
C
D
= 5:31 4:88f 2 10
3
< (Re)
t
< 2 10
5
;
0:67 < f < 1 (39)
C
D
here is based on the cross-sectional area of the
volume-equivalent sphere. The terminal velocity, U
t
,
can be calculated by
U
t
= 0:49l
1=36
g r
p
r
f
_ _
d
v
r
f
(1:08 f)
_ _1
2
(40)
3.8 Drag Coefcient for Particles with Density
Lighter than the Surrounding Fluid
All the above discussions are for particles with density
larger than the surrounding uid. For particles with
density smaller than the surrounding uid, it has
long been assumed that the free rising velocity is
governed by the same equations, the only dierence
being that of particle movement direction. However,
Karamanev and Nikolov (1992) and Karamanev
(1994) showed recently that the relationship between
C
D
and (Re)
p
for particles with density much smaller
than the uid follows the standard drag curve only
when (Re)
p
is less than 135. Otherwise the C
D
is a
constant equal to 0.95. Karamanev (1996) proceeded
to suggest the following drag coecient equations on
the basis of the Archimedes number rather than the
Reynolds number for calculation of the terminal
falling or rising velocities.
For free falling spheres,
C
D
=
432
Ar
1 0:0470Ar
2=3
_ _
0:517
1 154Ar
1=3
(41)
For free rising spheres,
C
D
=
432
Ar
1 0:0470Ar
2=3
_ _
0:517
1 154Ar
1=3
(42)
for Ar < 1:18 10
6
d
2
p
C
D
= 0:95 for Ar > 1:18 10
6
d
2
p
(43)
d
p
being in meters.
4 CALCULATION OF TERMINAL VELOCITY
FOR A SINGLE PARTICLE
The terminal velocity of a single particle is an intrinsic
characteristic of the particle, and its calculation and
measurement are as important as other intrinsic
particle properties, such as particle size and density.
Calculation of the terminal velocity of a single particle
used to be an iterative process. More recent develop-
ments allow direct calculations without trial and error.
These methods are introduced here.
4.1 Equation by Haider and Levenspiel (1989)
From Eq. (30), the terminal velocity for a single
spherical particle can be obtained as
U
r
= U
t
=
4
3
d
p
(r
p
r
f
)g
r
f
C
D
_
(44)
By combining Eq. (44) and the recommended drag
coecient correlations in Table 8 or drag coecient
equations proposed by Turton and Levenspiel (1986),
Eq. (34), and Haider and Levenspiel (1989), Eq. (35),
the terminal velocity can be calculated. Haider and
Levenspiel further suggested an approximate method
for direct evaluation of the terminal velocity by den-
ing a dimensionless particle size, d
+
p
, and a dimension-
less particle velocity, U
+
, by
d
+
p
= d
p
r
f
r
p
r
f
_ _
g
m
2
_ _
1=3
= Ar
1=3
=
3
4
C
D
(Re)
2
p
_ _
1=3
(45)
U
+
= U
r
2
f
m r
p
r
f
_ _
g
_ _
1=3
=
(Re)
p
Ar
1=3
=
4
3
(Re)
p
C
D
_ _
1=3
(46)
The terminal velocity of the irregular particles can then
be calculated by
U
+
t
=
18
d
+
p
_ _
2

2:335 1:744f
d
+
p
_ _
0:5
_ _
1
0:5 < f < 1
(47)
For spherical particles, Eq. (44) reduces to
U
+
t
=
18
d
+
p
_ _
2

0:591
d
+
p
_ _
0:5
_ _
1
f = 1 (48)
4.2 Terminal Velocity by Polynomial Equations
Fitted to Heywood Tables
Fouda and Capes (1976) proposed polynomial equa-
tions tted to the Heywood (1962) tables for calculat-
ing large numbers of terminal velocities. The Heywood
tables were recognized as being a simple and accurate
method for calculating both the terminal velocity and
the equivalent particle diameter, and they enjoyed wide
acceptance. The tables, however, are not very conven-
ient to use, especially for applications using a compu-
ter. The tables are now tted by Fouda and Capes
(1976) in the form of
Y =
5
n=0
a
n
X
n
(49)
where Y = log
10
(Pd
t
) or log
10
U
t
Q
_ _
(50)
and X = log
10
U
t
Q
_ _
or log
10
Pd
t
( ) (51)
The diameter d
t
is the particle diameter of a sphere
having the same terminal velocity as the particle. P,
Q, U
t
, and d
t
are related in the following two dimen-
sionless equations derived by Heywood:
C
D
Re
2
=
4
3
g
r
p
r
f
_ _
m
2
r
f
d
3
t
= P
3
d
3
t
(52)
Re
C
D
=
3
4g
r
2
f
r
p
r
f
_ _
U
3
t
m
=
U
3
t
Q
3
(53)
The resulting polynomial coecients are summarized
in Table 9.
The polynomial equations were reported to have an
average deviation less than 0.15% and a standard
deviation of less than 4.3%.
Similar types of equations were also proposed by
Hartman et al. (1994) for nonspherical particles.
They introduced the equations
Table 9 Polynomial Coefcients for Polynomial Equations Fitted to Heywood Tables
[see Eq. (49)]
Y X Polynomial coecients
Terminal velocity calculation log
10
(U
t
=Q) log
10
(Pd
t
) a
0
= 1.37323
a
1
= 2:06962
a
2
= 0:453219
a
3
= 0:334612 10
1
a
4
= 0:745901 10
2
a
5
= 0:249580 10
2
Equivalent diameter calculation log
10
(Pd
t
) log
10
(U
t
=Q) a
0
= 0:785724
a
1
= 0:684342
a
2
= 0:168457
a
3
= 0:103834
a
4
= 0:20901 10
1
a
5
= 0:57664 10
2
log(Re)
t
(Y; f) = log(Re)
t
(Y; 1) P(Y; f) (54)
and
log(Re)
t
(Y; 1) = 0:77481 0:56032 log Y
0:024246(log Y)
2
0:0038056
(log Y)
3
(55)
P(Y; f) = 0:10118(1 f) log Y 0:092944
(1 f)(log Y)
2
0:0098356(1 f)
(log Y)
3
0:12666(1 f)
2
log Y
(56)
where
Y =
4
3
_ _
g r
p
r
f
_ _
m
U
3
t
r
2
f
_ _
(57)
These equations are good for 0:01 < (Re)
t
< 16000
and 0:67 < f < 1. The equation tted the experimental
terminal velocities of limestone and lime particles to
better than 20%
4.3 Calculation of Terminal Velocity of Porous
Spheres
Calculation of the terminal velocity of a porous sphere
is useful and important in applications in water treat-
ment where settling velocities of a oc or an aggregate
are estimated. It is also important in estimation of
terminal velocities of clusters in uidized bed applica-
tions. The terminal velocity of a porous sphere can be
quite dierent from that of an impermeable sphere.
Theoretical studies of settling velocity of porous
spheres were conducted by Sutherland and Tan
(1970), Ooms et al. (1970), Neale et al. (1973),
Epstein and Neale (1974), and Matsumoto and
Suganuma (1977). The terminal velocity of porous
spheres was also experimentally measured by
Masliyah and Polikar (1980). In the limiting case of a
very low Reynolds number, Neale et al. (1973) arrived
at the following equation for the ratio of the resistance
experienced by a porous (or permeable) sphere to an
equivalent impermeable sphere. An equivalent
impermeable sphere is dened to be a sphere having
the same diameter and bulk density of the permeable
sphere.
=
2b
2
1 (tanh b)=b [ ]
2b
2
3 1 (tanh b)=b [ ]
(58)
where b is the normalized sphere radius expressed by
b =
R
k
_ (59)
where k is the permeability and R is the radius of the
sphere. The resistance ratio, , is normally less than
unity. For impermeable spheres, approaches unity
because b approaches innity. For innitely permeable
spheres, b tends to zero and so does .
Based on Eq. (58), the ratio of terminal velocities of
permeable and impermeable spheres can be written as
(U
t
)
p
(U
t
)
ip
=
1
(60)
and the ratio of dimensionless drag coecient can be
expressed as
(C
D
Re)
p
(C
D
Re)
ip
= (61)
where
(C
D
Re)
p
=
4d
2
p
g (1 e) r
p
r
f
_ _ _ _
3m(U
t
)
p
(62)
and
(C
D
Re)
ip
=
4d
2
p
g r
p
r
f
_ _
3m(U
t
)
ip
= 24 (63)
Equation (60) shows that a permeable sphere with the
same diameter and bulk density as the impermeable
sphere will have a higher terminal velocity than that
of the impermeable sphere. At higher Reynolds num-
ber, the experiments by Masliyah and Polikar (1980)
suggested the following equations:
For 15 < b < 33
C
D
=
24
(Re)
p
1 0:1315(Re)
(0:820:05w)
p
_ _
0:1 < (Re)
p
_ 7
(64)
and
C
D
=
24
(Re)
p
1 0:0853(Re)
(1:0930:105w)
p
_ _
7 < (Re)
p
< 120
(65)
where w = log
10
(Re)
p
.
This conrms the theoretical study of Neale et al.
(1973) that the drag experienced by a porous sphere
at low Reynolds numbers is less than that for an
impermeable sphere of similar diameter and bulky den-
sity. The eect of inertia at high Reynolds numbers is
higher for a porous sphere than for an impermeable
sphere of similar diameter and bulk density.
As far as the wall eect is concerned, the experi-
ments show that, at low Reynolds numbers, the wall
eect for a porous sphere is of the same order of mag-
nitude as that for an impermeable sphere. At high
Reynolds numbers, however, the wall eect become
smaller and less signicant for a porous sphere.
5 CHARACTERIZATION OF MULTIPARTICLE
SYSTEMS
In most applications, the systems and processes con-
tain large amounts of particles with size distribution;
each size may also possess a distinguished shape. To
describe properly these systems and processes for
design and analysis, they need to be adequately
characterized to reect their physical and chemical
potentials. In the following sections, dierent average
particle diameter denitions are introduced along with
statistical descriptions of particle size distribution.
Depending on applications, one denition may be
more suitable than others. Thus care must be exercised
to select the proper characterization for each process.
5.1 Different Denitions of Average Particle
Diameter
As mentioned earlier, naturally occurring particles are
usually irregular in shape and also dierent in size.
Dierent denitions of average particle diameter
have been used to characterize an assemblage of par-
ticles. Some may be hypothetical while others may
have physical signicance. They are summarized here.
5.1.1 Arithmetic Mean
The arithmetic mean is dened as the sum of all dia-
meters divided by the total number of particles
dd
av
=
i
n
i
d
pi
i
n
i
(66)
where n
i
can be the number of particles or the weight
percentage.
5.1.2 Surface Mean
The surface mean diameter is the diameter of a
hypothetical spherical particle whose surface multi-
plied by the total number of particles in the assemblage
would be equal to the total surface area of the assem-
blage, expressed mathematically as
dd
s
=
i
n
i
d
2
pi
i
n
i
_ (67)
5.1.3 Volume Mean
The volume mean diameter is the diameter of a
hypothetical spherical particle whose volume multi-
plied by the total number of particles in the assemblage
would be equal to the total volume of the assemblage.
dd
v
=
i
n
i
d
3
pi
i
n
i
3
_ (68)
5.1.4 VolumeSurface Mean
The volumesurface mean is also known as the Sauter
mean. This denes the average particle size based on
the specic surface area per unit volume or per unit
weight, as shown in Eq. (69):
dd
vs
=
i
n
i
d
3
pi
i
n
i
d
2
pi
=
1
i
x
i
d
pi
(69)
5.1.5 Weight Mean
The weight mean is also known as the DeBroucker
mean. This denes the average particle diameter
based on the unit weight of the particles. The weight
mean particle diameter is the diameter of a sphere
whose surface area times the total number of particles
equals the surface area per unit weight of the assem-
blage.
dd
w
=
i
v
i
d
pi
=
i
n
i
d
4
pi
i
n
i
d
3
pi
(70)
where
v
i
=
n
i
d
3
pi
i
n
i
d
3
pi
(71)
5.1.6 Length Mean
The length mean is the average diameter obtained by
dividing the total surface area of all particles with the
summation of the diameters, as shown here:
d d
l
=
i
n
i
d
2
pi
i
n
i
d
pi
(72)
5.1.7 Geometric Mean
The geometric mean is the logarithmic equivalent of
the arithmetic mean. Being a logarithmic average, the
geometric mean is always smaller than the arithmetic
mean.
dd
g
=
d
n
1
p1
d
n2
p2
d
n
n
pn
n
_
(73)
dd
g
=
i
n
i
log d
pi
_ _ _ _
i
n
i
(74)
5.1.8 Harmonic Mean
This is an average diameter based on the summation of
the reciprocals of the diameters of the individual com-
ponents of an assemblage and can be expressed math-
ematically as
dd
h
=
i
n
i
i
n
i
d
pi
=
1
i
x
i
d
pi
(75)
The harmonic mean is actually related to the average
spherical particle corresponding to the particle surface
per unit weight. Mathematically, the harmonic mean is
similar to the volumesurface mean or Sauter mean.
5.2 Statistical Characterization of Particles with a
Size Distribution
Statistically, the particle size distribution can be
characterized by three properties: mode, median, and
mean. The mode is the value that occurs most
frequently. It is a value seldom used for describing
particle size distribution. The average or arithmetic
mean diameter,

dd
av
, is aected by all values actually
observed and thus is inuenced greatly by extreme
values. The median particle size, d
1=2
, is the size that
divides the frequency distribution into two equal areas.
In practical application, the size distribution of a typi-
cal dust is typically skewed to the right, i.e., skewed to
the larger particle size. The central tendency of a
skewed frequency distribution is more adequately
represented by the median rather than by the mean
(see Fig. 9). Mathematically, the relationships among
the mean, median, and mode diameter can be
expressed as
dd
av
= log
1
log

dd
g
1:151 log
2
s
g
_ _
(76)
d
1=2
=

dd
g
(77)
d
m
= log
1
log

dd
g
2:303 log
2
s
g
_ _
(78)
Because of its mathematical properties, the standard
deviation s is almost exclusively used to measure the
dispersion of the particle size distribution. When the
skewed particle size distribution shown in Fig. 9 is
replotted using the logarithm of the particle size, the
skewed curve is transformed into a symmetrical bell-
shaped curve as shown in Fig. 10. This transformation
is of great signicance and importance in that a sym-
metrical bell-shaped distribution is amenable to all the
statistical procedures developed for the normal or
gaussian distribution.
In the log-normal particle size distribution, the
mean, median, and mode coincide and have an identical
value. This single value is called the geometric median
particle size,

dd
g
, and the measure of dispersion, the geo-
metric standard deviation, s
g
. Thus the log-normal
particle size distribution can be described completely
by these two characteristic values. To determine
whether the particles have a distribution close to log-
normal distribution, the particle cumulative frequency
data can be plotted on a logarithmic probability graph
paper. If the particle size distribution is log-normal, a
straight line will result. The geometric median particle
size is the 50% value of the distribution as shown in
Fig. 11. The geometric standard deviation is equal to
Figure 9 Skewed particle distribution of typical dust.
the ratio of 84.1% value divided by the 50% value or
50% value divided by the 15.9% value. Although
68.26% of the particles will lie in the particle size
range between

dd
g
s
g
and

dd
g
s
g
. If the particle size
reduction is due to comminution such as crushing,
milling, and grinding, the resulting particle size distri-
bution very often tends to be log-normal distribution.
Pulverized silica, granite, calcite, limestone, quartz,
soda, ash, sodium bicarbonate, alumina, and clay all
were observed to t the log-normal distribution.
Mathematically the log-normal distribution can be
expressed as
f =
1
s
g
2p
_ exp
(z zz)
2
2s
2
g
_ _
(79)
where
f =
dc
dz
; z = ln(d
pi
); zz =
i
z dc
i
dc
(80)
and c can be number, surface, or weight.
Mathematically, the log-normal distribution can
also be derived from the arithmetic distribution, by
substituting x by ln x = z to give Eq. (81):
f =
1
s
2p
_ exp
(x xx)
2
2s
2
_ _
(81)
s
2
=
(x xx)
2
n 1
(82)
An important characteristic of the log-normal distri-
bution is that the transformation among the various
particle size denitions and the statistical diameters
can be performed analytically and graphically (Smith
et al., 1929). The Hatch-Choate (1929) transformation
equations are summarized here:
log

dd
gm
= log

dd
gc
6:908 log
2
s
g
(83)
log

dd
av
= log

dd
gc
1:151 log
2
s
g
(84)
log

dd
s
= log

dd
gc
2:303 log
2
s
g
(85)
log

dd
v
= log

dd
gc
3:454 log
2
s
g
(86)
log

dd
vs
= log

dd
gc
5:757 log
2
s
g
(87)
log

dd
w
= log

dd
gc
8:023 log
2
s
g
(88)
log

dd
gc
= log

dd
gm
6:908 log
2
s
g
(89)
log

dd
av
= log

dd
gm
5:757 log
2
s
g
(90)
log

dd
s
= log

dd
gm
4:605 log
2
s
g
(91)
log

dd
v
= log

dd
gm
3:454 log
2
s
g
(92)
log

dd
vs
= log

dd
gm
1:151 log
2
s
g
(93)
log

dd
w
= log

dd
gm
1:151 log
2
s
g
(94)
Figure 10 Log-normal particle size distribution.
Figure 11 Log-normal distribution plotted on a logarithmic
probability graph paper.
5.3 The RosinRammler Distribution
The RosinRammler (1933) distribution was rst
developed for broken coal, but it has since been
found to be applicable to many other materials, such
as cement, gypsum, magnetite, clay, dyestus, quartz,
int, glass, and ores. From probability considerations,
the authors obtained
dW
f
(S
a
)
dS
a
= 100 nbS
n1
a
exp bS
n
a
( ) (95)
where n and b are constants: b is a measure of the range
of particle sizes present, and n is a characteristic of the
substance. Integrating Eq. (95), we have
W
f
= 100 exp bS
n
a
( ) (96)
or
loglog
100
W
f
_ _
= constantn log(S
a
) (97)
If loglog(100=W
f
) is plotted against log S
a
, a straight
line results. The peak of the distribution for n = 1 is at
100=e = 36:8%. This is used to characterize the degree
of comminution of the material. In Eq. (97), W
f
is the
weight percent of material retained on the sieve of
aperture S
a
. If the mode of the particle size distribution
curve is (S
a
)
m
, Eq. (97) gives b = 1=(S
a
)
m
. Since the
slope of the line on the RosinRammler graph depends
on the particle size range, the ratio of tan
1
(n) and (S
a
)
m
is a form of variance. This treatment following Rosin
and Rammler is useful for monitoring grinding opera-
tions for highly skewed distributions. It should be
employed carefully, however, since taking the loga-
rithm always reduces scatter, taking the logarithm
twice as in this case, tends to obscure the actual scatter
in the distribution.
5.4 Measurement of the Angle of Repose and the
Angle of Internal Friction
Two important characteristics of powder rheology are
the angle of repose and the angle of internal friction.
Simple devices can be constructed to measure both.
Figure 12 depicts a simple two-dimensional bed with
transparent walls and a small orice at the bottom of
the bed. After lling the bed with the powder to be
examined, the powder is allowed to ow out of the
test device to the surface of the test stand. The angle
of the powder-free surface measured from the at sur-
face of the test stand, the angle b in Fig. 12, is called
the angle of repose. This angle is an intrinsic charac-
teristic of the powder and should be close to a constant
when the same powder is pulled on top of a at sur-
face. Normal powders have angles of repose of around
35
. The angles of repose for commonly occurring

powders are summarized in Table 10.
The angle made by the free surface of the powder
still remaining in the test device with the at bottom of
the test device, the angle a in Fig. 12, is called the angle
of internal friction of the powder. The angle of internal
friction is also an intrinsic property of the powder and
can be considered as a shearing plane of the powder. In
order for the powder to ow, the angle has to be higher
than the angle of internal friction of the powder.
Normal powders have angles of internal friction of
around 70
. The angles of internal friction for com-

monly occurring powders are also listed in Table 10.
Both the angle of repose and the angle of internal
friction can be modied and reduced by ow aids
such as aeration, vibration, and the addition of other
powders.
Figure 12 A two-dimensional device for management of
angle or the repose and the angle of internal friction.
Table 10 Typical Angles of Repose and Angles
of Internal Friction
Material
Angle of repose
degree
Angle of
internal friction
degree
FCC catalyst 32 79
Sand 36 64
Resin 29 82
Wet ashes 50
Wheat 55
Oats 21
NOMENCLATURE
A
p
= projected area of a particle
Ar = Archimedes number, Ar =
d
3
p
r
f
(r
p
r
f
)g
m
2
a
s
= exterior surface area/volume of catalyst pellet
b = breadth
C = concentration of particle suspension
C
D
= drag coecient
d
a
= projected-area diameter
d
A
= sieve diameter
dd
av
= arithmetic mean particle diameter
d
c
= perimeter diameter
d
D
= drag diameter
d
f
= free-falling diameter
d
F
= Feret diameter
dd
g
= geometric mean particle diameter or geometric
median particle diameter
dd
gc
= geometric median diameter by count
dd
gm
= geometric median diameter by weight
dd
h
= harmonic mean particle diameter
d
i
= inside diameter of a ring
dd
l
= length mean particle diameter
d
M
= Martin diameter
d
m
= mode particle diameter
d
o
= outside diameter of a ring
d
p
= equivalent particle diameter
d
pi
= particle diameter of size i
d
+
p
= dimensionless particle diameter
d
s
= surface diameter
dd
s
= surface mean particle diameter
d
st
= Stokes diameter
d
sv
= surfacevolume diameter
d
t
= particle diameter of a sphere having the same
terminal velocity as the particle
d
v
= volume diameter
dd
v
= volume mean particle diameter
dd
vs
= volumesurface mean particle diameter
dd
w
= weight mean particle diameter
d
y
= diameter of a cylinder
D = sedimentation column diameter
DR
10
= particle diameter ratio at the cumulative mass of
10%
DR
50
50%
DR
90
90%
d
1=2
= median particle diameter
e
1
= atness ratio
e
2
= elongation ratio
f = frequency of observation
F = force exerting on a particle
F
s
= force exerting on a sphere
g = gravitational acceleration
I
c
= intensity of light beam after passing through a
particle suspension of concentration C
I
o
= original light beam intensity
k = Heywood shape factor
k = permeability
K = ratio of the settling velocity of the volume-
equivalent sphere to the settling velocity of the
particle
k
c
= a constant
k
e
= Heywood shape factor for an isometric particle
of similar form
l = length
l
y
= length of a cylinder
L = length of light beam path
L
w
= the distance from the center of a particle to the
walls
M = mass of a particle
n
i
= number of particles of size i or weight percent
of particles of size i
R = radius of a sphere
(Re)
p
= Reynolds number based on the particle diameter
(Re)
t
= Reynolds number based on the terminal velocity
of the particle
S
a
= sieve aperture
(S
a
)
m
= sieve aperture at mode of particle size
distribution
S
p
= surface area of a particle
S
v
= particle surface per unit volume of particle
t = thickness
T = measured transmittance, X
c
=X
o
U
+
= dimensionless particle velocity
U
r
= relative velocity between the particle and the
uid
U
t
= terminal velocity of a single particle
(U
t
)
ip
= terminal velocity of an impermeable particle
(U
t
)
p
= terminal velocity of a permeable (or porous)
particle
V
m
= measured velocity in a column of diameter D
V
p
= volume of a particle
V
o
= falling speed of a sphere in an innite uid
W
f
= weight fraction
W
s
= weight fraction of particles in the x-ray pass
W
200
= weight of material passing through 200 mesh
screen in grams
x
i
= weight fraction of particle size d
pi
X
c
= intensity of x-ray beam passing through a
particle suspension of concentration C
X
o
= intensity of x-ray beam passing through the
sedimentation column lled with clear uid
m = viscosity of uid
r = packing density
r
f
= density of uid
r
p
= density of particle
f = sphericity
f
op
= operational sphericity
f
w
= working sphericity
, = circularity
,
op
= operational circularity
a = an empirical constant depending on equipment
and particle characteristics
= angle of internal friction
b = an empirical constant depending on equipment,
particle, and suspending uid characteristics
b = normalized sphere radius
b = angle of repose
l = density ratio
= ratio of resistance experienced by a porous
sphere to an equivalent impermeable sphere
= number, surface, or weight
s = standard deviation
s
g
= geometric standard deviation
REFERENCES
S Akapo, TM Khong , CP Simpson , B Toussaint , JG Yates.
A novel method for determining the particle density of
porous aeratable powders. Powder Technol 58:237242,
1989.
T Allen. Particle Size Measurement. London: Chapman and
Hall, 1975.
B Aschenbrenner. A new method of expressing particle
sphericity. J Sediment Petrol 26:1531, 1956.
LG Austin. Conversion factors to convert particle size dis-
tributions measured by one method to those measured by
another method. Part Part Syst Charact 15:108111, 1998.
Babcock & Wilcox. Steam. 40
th
ed. Barberton: The Babcock
& Wilcox Company, 1992, pp 12-812-9.
HA Becker. The eects of shape and Reynolds number on
drag in the motion of a freely oriented body in an innite
uid. Can J Chem Eng 37:8591, 1959.
CF Bohren, RD Human. Absorption and Scattering of
Light by Small Particles. New York: Wiley-Interscience,
1986.
MP Bonin, M Queiroz. Local particle velocity, size, and con-
centration measurements in an industrial-sale, pulverized
coal-red boiler. Comb and Flame 85:121133, 1989.
B Buczek, D Geldart. Determination of the density of porous
particles using very ne dense powders. Powder Technol
45:173176, 1986.
EB Christiansen, DH Barker. The eect of shape and density
on the free settling of particles at high Reynolds numbers.
AIChE J 11:145151, 1965.
R Clift, JR Grace, ME Weber. Bubbles, Drops, and
Particles. New York: Academic Press, 1978.
RP Davuluri, TM Knowlton. Development of a standardized
attrition test procedure. In: LS Fan, TM Knowlton, eds.
Fluidization IX. New York: Engineering Foundation,
1998, pp 333340.
N Epstein, G Neale. On the sedimentation of a swarm of
permeable spheres. Chem Eng Sci 29:18411842, 1974.
LS Fan, C Zhu Principles of GasSolid Flows. Cambridge:
Cambridge University Press, 1998, pp 1316.
AE Fouda, CE Capes. Calculation of large numbers of
terminal velocities or equivalent particle diameters using
polynomial equations tted to the Heywood tables.
Powder Technol 13:291293, 1976.
PJ Freud, MN Trainer, AH Clark, HN Frock. Unied scatter
technique for full-range particle size measurement. Proc.
Pittsburgh Conference, Atlanta GA, March 812, 1993.
A Haider, O Levenspiel. Drag coecient and terminal
velocity of spherical and non-spherical particles. Powder
Technol 58:6370, 1989.
J Happel, H Brenner. Low Reynolds Number Hydro-
dynamics. New Jersey: Prentice Hall, 1965.
M Hartman, O Trnka, K Svoboda. Free settling of non-
spherical particles. Ind Eng Chem Res 33:19791983,
1994.
T Hatch, S Choate. Statistical description of the size proper-
ties of non-uniform particulate substances. J Franklin Inst
207:369387, 1929.
JF Heiss, J Coull. Eect of orientation and shape. Chem Eng
Prog 48:133140, 1952.
G Herdan. Small Particle Statistics. 2
nd
ed. New York:
Butterworths, 1960.
H Heywood. Uniform and non-uniform motion of particles
in uids. In: Proc Symp Interaction Fluids and Parts.
London: Inst Chem Eng, 1962, pp 18.
JD Hottovy, ND Sylvester. Drag coecients for irregularly
shaped particles. Ind Eng Chem Process Des Devel
18:433436, 1979.
JL Isaacs, G Thodos. The free-settling of solid cylindrical
particles in the turbulent region. Can J Chem Eng
45:150155, 1967.
DG Karamanev. On the rise of gas bubbles in quiescent
liquids. AIChE J 40:14181421, 1994.
DG Karamanev. Equations for calculation of the terminal
velocity and drag coecient of solid spheres and gas
bubbles. Chem Eng Comm 147:7584, 1996.
DG Karamanev, LN Nikolov. Free rising spheres do not
obey Newtons law for free settling. AIChE J 38:1843
1846, 1992.
BH Kaye. Automated decision-taking in ne particle science.
BH Kaye. Direct Characterization of Fine Particles. New
York: John Wiley, 1981.
MJ Knight, PN Rowe, HJ Macgillivray, DJ Cheesman.
On measuring the density of nely divided but porous
particles such as uid bed cracking catalyst. Trans Instn
Chem Engrs 58:203207, 1980.
M Leva. Fluidization. New York: McGraw-Hill, 1959.
PJ Lloyd. Particle characterization. Chem Eng 120122,
April 29, 1974.
JB Masliyah, M Polikar. Terminal velocity of porous
spheres. Can J Chem Eng 58:299302, 1980.
K Matsumoto, A Suganuma. Settling velocity of a permeable
model oc. Chem Eng Sci 32:445447, 1977.
M Naito, O Hayakawa, K Nakahira, H Mori, J Tsubaki.
Eect of particle shape on the particle size distribution
measured with commercial equipment. Powder Technol
100:5260, 1998.
G Neale, N Epstein, W Nader. Creeping ow relative to
JP Olivier, GK Hickin, C Orr Jr. Rapid automatic particle
size analysis in the sub-sieve range. Powder Technol
4:257263, 1970/1971.
G Ooms, PF Mijnlie, HL Beckers. Frictional forces exerted
by a owing uid on a permeable particle, with particular
reference to polymer coils. J Chem Phys 53:41234134,
1970.
CW Oseen. U
ber die Stokessche Formel und u ber die

verwandte Aufgabe in der Hydrodynamik. Arkiv for
Mathematik Astronomi och Fysik. Vol. 6, No. 29,
1910.
ES Pettyjohn, EB Christiansen. Eect of particle shape on
free-settling rates of isometric particles. Chem Eng Prog
44:157172, 1948.
M Pohl. Technology update: particle sizing moves from
the lab to the process. Powder and Bulk Eng. 3946,
February 1998.
I Proudman, JRA Pearson. Expansions at small Reynolds
numbers for the ow past a sphere and a circular cylinder.
J Fluid Mech 2:237262, 1957.
HF Rase. Fixed-Bed Reactor Design and Diagnostics.
Boston: Butterworths, 1990, pp 118119.
P Rosin, EJ Rammler. The laws governing the neness of
powdered coal. J Inst Fuel 7(31):2936, 1933.
WO Smith, PD Foote, PF Busang. Capillary rise in sands of
uniform spherical grains. Phys Rev 34:12711279, 1929.
P Subramanian, Vr Arunachalam. A simple device for the
determination of sphericity factor. Ind Eng Chem
Fundam 19:436437, 1980.
DN Sutherland, CT Tan. Sedimentation of a porous sphere.
Chem Eng Sci 25:19481950, 1970.
R Turton, NN Clark. An explicit relationship to predict
spherical particle terminal velocity. Powder Technol
53:127129, 1987.
R Turton, O Levenspiel. A short note on the drag correlation
for spheres, Powder Technol 47:8386, 1986.
H Wadell. Sphericity and roundness of rock particles. J Geol
41:310331, 1933.
H Wadell. The coecient of resistance as a function of
Reynolds number for solids of various shapes. J
Franklin Inst 217:459490, 1934.
H Wadell. Volume, shape, and roundness of quartz particles.
J Geol 43:250280, 1935.
J Werther, J Reppenhagen. Attrition in uidized beds and
pneumatic conveying lines. In: WC Yang, ed.
Fluidization, Solids Handlings, and Processing
Westwood, NJ: Noyes Data Corporation, 1999,
pp 435491.
2
Flow Through Fixed Beds
Wen-Ching Yang
Fixed beds, also called packed beds, play a vital role in
chemical processes. Their simplicity induced applica-
tions in many unit operations such as adsorption, dry-
ing, ltration, dust collection, and other catalytic and
noncatalytic reactors. The primary operating cost for a
xed bed is the pressure drop through the packed bed
of solids. Thus to understand the design and operation
of a packed bed requires study of packing character-
istics of particles and their eect on pressure drop
through the bed.
1 PACKING CHARACTERISTICS OF
MONOSIZED SPHERES
The simplest case of a packing system is a bed that
consists of only uniform and regular monosized sphe-
rical particles. For this simple system, the voidage of
an ordered arrangement of monosized spheres can be
derived satisfactorily by mathematical considerations.
Random packing of monosized spheres is more com-
plicated and can be described mathematically through
a coordination number. However, the voidage deter-
mination of random packing of monosized spheres
should be done through experiments.
1.1 Regular Packing of Uniform-Sized Spheres
For convenience, the regular packing can be consid-
ered to be constructed from regular layers and rows.
The two basic layers are the square layer with a 90
angle and the triangular or the simple rhombic layer of

an angle of 60
(see Fig. 1). The vertical stacking of

these layers yields six possible regular packings. For
the six possible regular packings of spherical particles,
the voidage is only a function of packing arrangement.
The voidage is also independent of the particle size.
The six arrangements are cubic, two orientations of
orthorhombic, tetragonal-spheroidal, and two orienta-
tions of rhombohedral. They are graphically presented
in Fig. 2. Formulae for calculating the voidage of these
six arrangements have been developed and the packing
characteristics of these ordered arrangements are sum-
marized in Table 1 (Herdan, 1960). The closest packing
is the rhombohedral with a voidage of 0.2595, and the
loosest, the cubic with a voidage of 0.4764. The stabi-
lity of the systems increases as the voidage decreases.
The systems tend towards the orthorhombic state,
especially if a mechanical disturbance such as vibration
is applied to the systems. The voidage of a bed of
monosized particles after prolonged shaking is usually
about 0.395, approaching the characteristic of an
eight-point contact packing of orthorhombic.
1.2 Random Packing of Uniform Monosize Spheres
Random packings of uniform monosize spheres are
created by irregular and random arrangements of par-
ticles. According to Scott (1960), two reproducible
states of random packing, dense random packing and
loose random packing, can be experimentally created.
The dense random packing can be developed by
Figure 1 Two basic packing layers of monosize spheres
square layer (90
) and simple rhombic layer (60
).
Figure 2 Six possible arrangements of monosize spheresone cubic (Figure 2a), two orthorhombic (Figures 2b and 2c), one
tetragonalspheroidal (Figure 2d), two rhombohedral (Figure 2e).
pouring the spheres into cylindrical containers and
then shaking for several minutes. The loose random
packing is created by tipping the container horizontally
and rotating slowly about its axis, and then returning
slowly to its vertical position. The dense random pack-
ing density obtained by several authors ranged from
0.61 to 0.63, equivalent to a voidage of 0.39 and 0.37,
corresponding closely to orthorhombic packing with a
voidage of 0.3954. The loose random packing obtained
has a voidage of 0.40 to 0.42, corresponding to a pack-
ing between the cubic packing (voidage = 0:4764) and
orthorhombic packing. Haughey and Beveridge (1969)
classied the packing into four dierent modes: very
loose random packing, loose random packing, poured
random packing, and close random packing. The loose
random packing and poured random packing corre-
spond to Scotts loose and dense random packings.
The very loose random packing corresponds to the
state where the bed is rst uidized and then the gas
is slowly reduced until it is below the minimum uid-
ization. The bed so formed usually has a voidage of
about 0.44. The close random packing is formed by
vigorously shaking or vibrating the bed, and then the
voidage is usually approaching 0.359 to 0.375.
1.3 Properties of Regular and Random Packing of
Uniform Monosized Spheres
The coordination number, dened as the number of
spheres in contact with any neighboring spheres, is
used to characterize the voidage of packing. Table 2
lists the coordination numbers from 3 to 12 and its
associated voidage. The coordination numbers of 4,
6, 8, 10, and 12 correspond to arrangements that are
regular as shown in Table 1.
A model derived by Haughey and Beveridge (1966)
relates the average coordination number with the mean
bulk voidage as
n = 22:47 39:39e 0:259 _ e _ 0:5 (1)
A parameter called the layer spacing, bd
p
, where d
p
is
the diameter of the sphere, has also been used to char-
acterize the packing. For most common packings, b is
between 0.707 and 1.0, the limits corresponding to the
rhombohedral packing and cubic packing, respectively.
Though the layer spacing has little physical meaning in
random packing, the concept is useful for describing
the packing. It is actually related to the bed voidage in
the following way:
b =
2
3
_
p
3
2
_
(1 e)
_ _
1=3
(2)
1.4 Specic Surface Area of the Bed
The specic surface area of the bed is dened as the
ratio of total particle surface area to the total bed
volume. Since the number of spheres per unit volume
of bed is
N
p
=
6(1 e)
pd
3
p
(3)
Table 1 Packing Characteristics of Ordered Uniform Monosized Spheres
Packing
Points of
contact Porosity, %
Surface of spheres
per cm
3
Cross-sectional
pore area per cm
2
Cubic 6 47.64 1.57/R 0.2146
Orthohombic
(two orientations) 8 39.54 1.81/R 0.2146
Tetragonal-Spheroidal 10 30.19 2.10/R 0.093
Rhombohedral
(two orientations) 12 25.95 2.22/R
Table 2 Correspondence of
Coordination Number and
Voidage
Coordination Number Voidage
3 0.7766
4 0.6599
5 0.5969
6 0.4764
7 0.4388
8 0.3955
9 0.3866
10 0.3019
11 0.2817
12 0.2595
The specic surface area can be evaluated as
S =
6(1 e)
d
p
(4)
2 PACKING CHARACTERISTICS OF BINARY
MIXTURES OF SPHERICAL AND
NONSPHERICAL PARTICLES
The packing of binary mixtures of spheres is the sim-
plest system and thus has received a lot of studies. The
packing of binary mixtures of spheres depends on their
diameter ratio and the percentage of large particles in
the mixtures. Following McGearys (1967) develop-
ment, the minimum fraction of the coarse particles
for closest packing is 73%, and the closest packing
voidage is 0.14.
The dependence of maximum observed binary
mechanical packing of spheres on the ratio of dia-
meters between the large and small spheres shows
that when the size ratio, or diameter ratio, decreases,
the changes in voidage increase. With a diameter ratio
larger than about 7 to 1, the change in voidage is only
very slight. This ratio is very close to the critical
diameter ratio of 6.5 for the smaller sphere to pass
through the triangular opening formed by three larger
spheres in close packing (see Fig. 7).
Abe et al. (1979) carried out theoretical analysis of a
packed bed of binary mixture and proposed equations
to estimate voidage at dierent degree of mixing of
binary mixtures of spherical and nonspherical parti-
cles. If cohesive forces are absent, the binary mixture
will yield a packed bed of minimum voidage at X
min
, as
shown in Fig. 3. In the region X
b
< X
min
, large parti-
cles are distributed randomly and evenly in the packed
bed. This state is called complete mixing. In the region
X
b
> X
min
, there are two dierent states possible. In
the rst state, the small particles are distributed in
the interstices of the big particles. When the fraction
of the small particles decreases, segregation of small
particles becomes inevitable, as shown in Fig. 4.
Equations were proposed to calculate the average
voidage of packed bed with binary mixtures.
For 0 _ X
b
<

XX
b
,
e
av
= 1
1 e
s
(1 X
b
) aX
b
(1 e
s
)
(5)
where
a = 1 f
c
d
s
d
b
_ _
(6)
f
c
= 1 for
d
s
d
b
= 0:5 (7)
f
c
= 1:2 for
d
s
d
b
= 0:25 (8)
f
c
= 1:4 for
d
s
d
b
_ 0:125 (9)
For

XX
b
_ X
b
< X
min
,
e
av
= 1
(1 e
s
)(X
b

XX
b
)
X
b
1

XX
b
1 (1 e
s
)a [ ]
_ _
2
XX
b
(1 e
s
)
X
b
1

XX
b
1 (1 e
s
)a [ ]
_ _
(10)
where
XX
b
=
a
a (1 aa)(1 e
s
)
(11)
and
a =
(1 e
b
)
1 1:5(d
s
=d
b
) [ ]
3
(12)
For X
min
_ X
b
_ 1;
e
av
= 1
(1 e
b
)
X
b
(1 b
s
b
c
)
3
(13)
where
b
c
=
1
X
b
3
_
1 (14)
and
b
s
=
d
s
d
b
_ _
n
(15)
n equals 1=3 for crushed stones and 1=2 for round
sand.
Figure 3 Packed bed of minimum voidage for binary par-
ticle mixtures.
The intersection of Eqs. 10 and 13 gives the binary
mixture of large and small particles where the voidage
is the minimum, or X
min
and e
min
:
e
av
= Average voidage of the binary mixture
e
min
= The minimum voidage of the binary mix-
ture
e
s
= Voidage of a packed bed with small parti-
cles alone
e
b
= Voidage of a packed bed with large parti-
cles alone
X
b
= Volume fraction of large particles
X
min
= Volume fraction of large particles where the
voidage of the binary mixture is the mini-
mum
d
s
= Diameter of the small particles
d
b
= Diameter of the large particles
A common approach to predict the voidage of binary
particle mixtures is to make use of the empirical corre-
lation developed by Westman (1936) based on the total
bed volume occupied by the specic volume of solid
material V = 1=(1 e). The equation was later modi-
ed by Yu et al. (1993) as shown in the following
equation to apply to both spherical and non-spherical
particles.
V V
b
X
b
V
s
_ _
2
2G
V V
b
X
b
V
s
_ _
V X
b
V
s
X
s
V
b
1
_ _
V X
b
V
s
X
s
V
b
1
_ _
2
= 1
(16)
The empirical constant G is independent of the com-
position of the mixtures but depends on the size ratio
of the particles. Yu et al. (1993) gave the following
values for G:
1
G
= 1:355r
1:566
p
for r
p
_ 0:824 (17)
and
1
G
= 1 for r
p
_ 0:824 (18)
where
r
p
=
d
s
d
b
(19)
For the limiting case where r
p
= 1, the voidage of the
packed bed will not change through mixing of the
binary particles. When r
p
approaches 0, the interstices
between large particles can be lled with the small
particles as discussed earlier. The voidage and the
specic solid volume of the packed bed become
e = e
b
e
s
(20)
V =
V
b
V
s
V
b
V
s
1
(21)
X
b
=
1 e
b
1 e
b
e
s
(22)
The eect of changing the particle size ratio on the
packing of binary particles is summarized in Fig. 5.
For nonspherical particles, Yu et al. (1993) sug-
gested to substitution of the particle diameter by the
packing equivalent particle diameter calculated by
d
pe
= 3:1781 3:6821
1
f
1:5040
1
f
2
_ _
d
ve
(23)
where d
ve
is the volume equivalent particle diameter
and f is the Wadells sphericity described in Chapter
1. This approach provides good agreement for packing
characteristics of both spherical and nonspherical
binary particles, though evaluation of the packing
equivalent particle diameter is somewhat cumbersome.
Figure 4 States of binary particle mixtures.
An approach employing the structural ratio to
predict the voidage of binary particle mixtures for
both spherical and nonspherical particles is also pro-
posed recently by Finkers and Homann (1998). The
structural ratio based on packing of spherical and non-
spherical particles is dened as
r
str
=
1
e
b
1
_ _
r
3
p
1 e
s
(24)
The structural ratio is then used to calculate the
empirical constant G for use in Eq. (16).
G = r
k
str
1 e
k
b
_ _
(25)
A value of 0:63 was suggested for k in Eq. (25). The
proposed approach is good for both spherical and
nonspherical particles. For binary particles with size
distribution in each fraction, the approach did not
fare as well. By changing the value k to 0:345, the
proposed approach gave excellent results for data by
Sohn and Moreland (1968). It was suggested that k
relates to the particle size distribution in each fraction
to provide an even more general equation of particle
packing in a packed bed.
Packings of ternary and quaternary mixtures of
solid particles, considerably more complex systems,
have also been studied by Ouchiyama and Tanaka
(1989) and by Homann and Finkers (1995). The
packing characteristics of spheres of unequal sizes
have also been investigated (Herdan, 1960). In rhom-
bohedral packings of dierent spherical particle sizes,
the voidage may be reduced to less than 0.15.
3 CRITICAL RATIO OF ENTRANCE AND
CRITICAL RATIO OF OCCUPATION IN
BINARY SYSTEMS
When the monosized spheres are arranged in normal
loosest square or tightest rhombohedral packings,
there are critical smaller spheres, which can pass
through the openings formed by the larger monosized
spheres, as shown in Figs. 6 and 7. These critical dia-
meter ratios are called the critical ratio of entrance.
Figure 5 Eect of changing particle size ratio on the packing of binary particles. (Adapted from Finkers and Homann, 1998).
Figure 6 Critical ratio of entrance for square packing.
For the loosest square packing, this critical diameter
ratio can be expressed as
d
b
d
S
=
1
2
_
1
= 2:414 (26)
For the tightest rhombohedral packing, the critical
ratio can be found from
d
b
d
S
=
1
(2=
3
_
) 1
= 6:464 (27)
There also exists a critical size sphere that, though it
cannot pass through the opening made by the mono-
sized spheres, can occupy the volume enclosed by the
monosized spheres without disturbing the basic pack-
ing. To make this possible, the smaller sphere has to be
in the position already during packing of the larger
spheres. This critical diameter ratio is named the
critical ratio of occupation. For the loosest square
packing, the ratio can be calculated as (Cumberland
and Crawford, 1987)
d
b
d
M
=
1
0:732
= 1:366 (28)
For the tightest rhombohedral packing, there are two
values:
d
b
d
M
=
1
0:414
= 2:415 (29)
d
b
d
M
=
1
0:225
= 4:444 (30)
where d
M
is the particle diameter required during
critical ratio of occupation.
4 PACKING OF NON-SPHERICAL PARTICLES
There has been very little theoretical and experimental
work performed in this area owing to the complexity of
such a system. Oman and Watson (1944) conducted
experiments using particles of various shapes and
found the voidage of the packing increased in the
following order: cylinders, spheres, granules, Raschig
rings, and Berl saddles. Coulson (1949) studied pack-
ings of cubes, cylinders, and plates and found that the
results depend on the height of fall of the particles. His
results are shown in Table 3a. For random packed
beds of uniform-sized particles, Brown (1966)
suggested that the packing porosity depends on the
sphericity of the particles and can be related as
shown in Fig. 8 and Table 3b based on experimental
ndings. For small grains observable under the micro-
scope, the sphericity can be obtained following Wadell
(1935) and the average values obtained from the fol-
lowing equation after observing a sample of particles.
f =
d
c
D
c
(31)
where d
c
= diameter of a circle equal in area to the
projected area of the particle when resting on its larger
face, D
c
= diameter of the smallest circle circumscrib-
ing the projection of the particle.
When the packed bed is made up of a mixture of
particles with dierent shapes, the sphericity is usually
calculated from the arithmetic mean of the various
sphericities.
Zou and Yu (1996) studied both loose and dense
random packing of monosized nonspherical particles
and found that the porosity was strongly dependent on
both the particle shape and the packing method. The
initial porosity of loose and dense random packings
can be expressed as
For loose random packing:
ln e
01;cylinder
= f
5:58
exp 5:89(1 f) [ ] ln 0:40 (32)
ln e
01;disk
= f
0:60
exp 0:23(1 f)
0:45
_ _
ln 0:40 (33)
For dense random packing,
ln e
0d;cylinder
= f
6:74
exp 8:00(1 f) [ ] ln 0:36 (34)
ln e
0d;disk
= f
0:63
exp 0:64(1 f)
0:54
_ _
ln 0:36 (35)
The initial porosity of the nonspherical particles can be
approximated by the proper use of the packing results
of cylinders and disks shown in Eqs. (31) through (34)
and expressed as
e
0
=
I
d
I
c
I
d
e
0;cylinder
I
c
I
c
I
d
e
0;disk
(36)
where I
c
is called the cylindrical index and I
d
, the disk
index. They are dened as follows.
I
c
= f f
c
(37)
I
d
= f f
d
(38)
Figure 7 Critical ratio of entrance for rhombohedral pack-
ing.
where f
c
is an equivalent sphericity or aspect ratio
dened as the ratio of the maximum length to the
diameter of the circle having the same projected area
normal to the maximum length, and f
d
, a sphericity
dened as the ratio of the shortest length to the dia-
meter of the circle having the same projected area.
4.1 Relationship Between the Hausner Ratio and the
Sphericity
The Hausner ratio is dened as the ratio of tapped
density to loose density, it is a measurement of the
compressibility and cohesiveness of the powder.
Based on the work of Zou and Yu (1996), the
Hausner ratio (HR) and the sphericity have the rela-
tionship
HR = 1:478 10
0:136f
(39)
5 FACTORS AFFECTING PACKING DENSITY
IN PRACTICE
Particles, containers, and lling and handling methods
contribute to packing density in practice. The impor-
tant factors can be summarized as
Table 3a Rough Estimate of Bed Void Fraction
Particle of pellet Normal charge Dense packed Multiplier for small tubes
Tablets 0.36 0.31 1 0:43 d
p
=D
Extrudates
Short 0.40 0.33 1 0:46 d
p
=D
Long 0.46 0.40 1 0:46 d
p
=D
Spheres
Uniform 0.40 0.36 1 0:42 d
p
=D
Mixed sizes 0.36 0.32
Irregular 0.42 (average) 0.42 (average) 1 0:3 d
p
=D
Table 3b Voidage of Randomly Packed
Beds with Uniformly Sized Particles
Larger than 500 mm
Sphericity
Voidage
Loose packing Dense packing
0.25 0.85 0.80
0.30 0.80 0.75
0.35 0.75 0.70
0.40 0.72 0.67
0.45 0.68 0.63
0.50 0.64 0.59
0.55 0.61 0.55
0.60 0.58 0.51
0.65 0.55 0.48
0.70 0.53 0.45
0.75 0.51 0.42
0.80 0.49 0.40
0.85 0.47 0.38
0.90 0.45 0.36
0.95 0.43 0.34
1.00 0.41 0.32
Source: Adapted from Brown, 1966.
Particle shape, particle size, particle size distribu-
tion, particle coecient of restitution, particle
surface properties (friction)
Container shape, container size, container surface
properties (friction)
Deposition method, deposition intensity, velocity of
particle deposition
Vibratory compaction, pressure compaction
6 CORRELATIONS FOR FIXED BED BULK
VOIDAGE
The voidage is the parameter most frequently
employed to characterize the pressure drop in a xed
bed. For accurate determination, experimental tech-
nique such as the water displacement is usually used.
In the application of catalytic reactors, the catalysis
packing is important not only for pressure drop but
also for heat and mass transfer. Furnas (1929) was
probably the rst to study the eect of wall on the
packing of the particles. He proposed the following
correlations for voidage close to the wall and at the
core.
e
w
= 1 0:6
d
p
D
_ _ _ _
e 0:6
d
p
D
_ _
(40)
e
c
= e 0:3(1 e) [ ] 1 0:6
d
p
D
_ _ _ _
0:6
d
p
D
_ _
(41)
where e is the average voidage experimentally mea-
sured or the ratio of void volume to the total volume
of the packed bed. When the ratio of particle diameter
to the bed diameter is less than 0.02, the correction for
wall eect becomes negligible.
Figure 9 is a plot obtained by Benenati and
Brosilow (1962) in a bed of spheres showing the radial
oscillations of voidage away from the containing
cylindrical wall. The observation was conrmed by
Propster and Szekely (1977). Experiments performed
by Goodling et al. (1983) also indicated that for
uniformly sized spherical particles, the oscillations in
voidage can be up to 5 particle diameters from the
wall. The oscillations in voidage are down to 2 to 3
diameters from the wall, for a mixture of two spherical
sizes, and down to 1 particle diameter from the wall for
a mixture of three particle sizes.
Propster and Szekely (1977) also found that a
marked local minimum in voidage existed in the inter-
Figure 8 Relation between packing voidage and sphericity. (Adapted from Zou and Yu, 1996).
facial region when small particles were placed on top of
a bed of larger particles because of penetration of small
particles into the interstices of large particles.
However, when the large particles were placed on top
of the smaller particles, the local minimum in voidage
was not as pronounced. When the particle ratio of the
lower layer to the upper layer is less than 2, relatively
little particle penetration was observed. Ready pene-
tration of small particles could be observed when the
ratio equals 6, corresponding to a critical ratio of
entrance of 6.464 for the tightest rhombohedral pack-
ing (see Sec. 3). This observation will have practical
application in the iron blast furnace or cupola where
alternate layers of coke and iron ore of dierent
particle sizes are charged into the reactors.
Fixed beds of very low tube-toparticle diameter
ratio have also been proposed and studied. For these
reactors, the eect of wall and particle shape on bulk
voidage becomes important. It will be convenient to
have correlations for xed bed bulk voidage for com-
monly used catalysts such as spheres, cylinders, and
rings. The following correlations are those proposed
by Dixon (1988). A more comprehensive review was
conducted by Haughey and Beveridge (1969).
For spherical particles
e = 0:4 0:05
d
p
D
_ _
0:412
d
p
D
_ _
2
d
p
D
_ 0:5 (42)
e = 0:528 2:464
d
p
D
0:5
_ _
0:5 _
d
p
D
_ 0:536
(43)
e = 1 0:667
d
p
D
_ _
3
2
d
p
D
_ _
1
_ _
0:5
d
p
D
_ 0:536
(44)
For full cylinders
Figure 9 Schematic radial oscillations of voidage away from the containing cylindrical wall. (Adapted from Propster and
Szekely, 1977.)
e = 0:36 0:10
d
pv
D
_ _
0:7
d
pv
D
_ _
2
d
pv
D
_ 0:6
(45)
e = 0:677 9
d
pv
D
0:625
_ _
2
0:6 _
d
pv
D
_ 0:7
(46)
e = 1 0:763
d
pv
D
_ _
2
d
pv
D
_ 0:7 (47)
where d
pv
is the diameter of a sphere having the same
volume as the cylinder.
For hollow cylinders
(1 e
hc
) = 1 2
d
yi
d
yo
0:5
_ _
2
1:145
d
pv
D
_ _
_ _
1
d
2
yi
d
2
yo
_ _
(1 e
sc
)
d
yi
d
yo
_ 0:5
(48)
where subscripts hc and sc denote the hollow and solid
cylinders respectively, and d
yi
and d
yo
are the inside
and outside diameter of the hollow cylinder, respec-
tively.
7 FLOW THROUGH PACKED BEDS
There have been numerous investigations into ow
through packed beds. Scheidegger (1961) critically
reviewed the earlier models. A more recent review is
by Molerus (1993). There were two main approaches,
the discrete particle model and the pipe ow analogy.
Both approaches gave reasonable predictions of pres-
sure drop of ow through packed beds of spherical and
near-spherical particles but were inadequate for beds
with particles substantially dierent from the spheres.
The pressure drop through the packed bed is due not
only to the frictional resistance at the particle surface
but also to the expansion and contraction of ow
through the interstices among the particles
The most popular approach is the pipe ow analogy
model, also called the capillary tube model or channel
model, which approximates the ow through the
packed bed by the ow through a bundle of straight
capillaries of equal size. Further renement produced
the constricted tube model. In this model, an assembly
of tortuous channels of varying cross sections simu-
lates the varying dimensions and curvatures of pores
in the packed bed. The major contributions following
this approach include Blake (1922), Kozeny (1927),
Carman (1937), and Ergun (1952). The discrete particle
model assumes that the packed bed consists of an
assembly of discrete particles that possess their own
boundary layer during the ow through the packed
bed. The primary developments are due to Burke
and Plummer (1928), Ranz (1952), Happel (1958),
Galloway and Sage (1970), and Gauvin and Katta
(1973). Conceptually, the discrete particle model is clo-
ser to the physical description of the ow through the
packed bed, but the pipe ow analogy is historically
more widely employed. The correlations developed
through the pipe ow analogy are usually applicable
for particles with sphericities larger than about 0.6.
The correlations developed through the discrete parti-
cle model have wide applications, including particles of
sphericities less than 0.6.
7.1 Darcys Law
The theory of laminar ow through homogeneous por-
ous media is based on a classical experiment originally
performed by Darcy in 1856. Darcys experiment is
described in Fig. 10. The total volume of the uid
percolating through the xed bed, Q, can be expressed
in terms of the height of the bed and the bed area as
Q =
KA(h
2
h
1
)
h
(49)
where K is a constant depending on the properties of
the uid and of the porous medium. The minus sign
indicates that the ow is in the opposite direction of
increasing height, h. Darcys law can be restated in
terms of the pressure P and the density r
f
of the liquid.
Assuming that the liquid density is constant, we have
Q =
K
/
A P
2
P
1
r
f
gh
_ _
h
(50)
and
P
1
= r
f
g(h
1
z
1
) P
2
= r
f
g(h
2
z
2
) (51)
Equations (49) and (50) are equivalent statements of
Darcys law. It is valid for a wide domain of ows. The
ow of newtonian uid at low Reynolds number is
known to follow Darcys law. Thus it is valid for arbi-
trary small pressure dierentials for liquids. It has been
used to measure ow rates by determining the pressure
drop across a xed porous bed. For liquids at high
velocities and for gases at very low and very high velo-
cities, Darcys law becomes invalid.
Darcys law in its original form is rather restricted in
its usefulness. The physical signicance of the constant
K
/
, known as the permeability constant and having
dimensions M
1
L
3
T
1
, has to be elucidated. The depen-
dence of K
/
on the porous medium and on the liquid
has to be separated for practical applications. Nutting
(1930) proposed that
K
/
=
k
s
m
(52)
where k
s
is the specic permeability, having dimensions
L
2
, and the darcy is used as a unit for specic perme-
ability (1 darcy = 9:87 10
9
cm
2
). By letting h
becomes innitesimal, Darcys law can be written in
the dierential form
q
=
Q
A
=
k
s
m
_ _
grad P r
f
g
_ _
(53)
The dierential form of Darcys law is by itself not
sucient to determine the ow pattern in a porous
medium for given boundary conditions, as it contains
three unknowns, q
, P, and r
f
. Two more equations
are required to specify the problem completely. One is
the dependence of the density on the pressure:
r
f
= r
f
(P) (54)
and the other is the continuity equation
e
@r
f
@t
= div r
f
g
_ _
(55)
By eliminating all unknowns except the pressure, we
have
e
@r
f
@t
= div
r
f
k
s
m
_ _
grad P r
f
g
_ _
_ _
(56)
7.2 Blakes Correlation
Blakes (1922) may be the rst one to suggest using a
modied dimensionless group incorporating the voi-
dage, e, in a particulate system. He proposed the use
of the interstitial velocity, instead of the supercial
velocity, and the reciprocal of the total particle surface
area per unit volume, instead of the particle diameter,
as the characteristic length. The Reynolds number and
friction factor proposed by Blake are
Figure 10 Darcys experiment.
(Re)
B
=
Ud
p
r
f
m(1 e)
(57)
f
B
=
P
L
d
p
r
f
U
2
g
e
3
(1 e)
(58)
These dimensionless groups have been used ever since
as the basis for the pipe ow analogy in almost all
packed bed correlations.
From dimensionless analysis, the pressure drop
through a packed bed of particles can be obtained as
P
L
o m
2n
r
n1
f
U
e
_ _
n
D
n3
(59)
The interstitial uid velocity in the average direction of
uid motion is used here. The D is a length analogous
to the hydraulic radius of a conduit and is dened as
D =
Mean cross-sectional area of flow
channels through bed
Mean wetted perimeter of flow channels
(60)
Multiplying the numerator and the denominator by L,
the height of the bed, we have
D =
(Total bed volume) (voidage)
(Total bed surface)
(61)
or
D =
e
S
=
ed
p
6(1 e)
(62)
where S is the total surface of solids per unit bed
volume assuming spherical particles, it can be
expressed as
S =
6(1 e)
d
p
(63)
Equation (63) is essentially similar to Eq. (4).
Substituting into Eq. (59), we obtain
P
L
o m
2n
r
n1
f
U
n
(1 e)
3n
e
3
d
n3
p
(64)
For conventional pressure drop through a packed bed
of particles, n = 2. Equation (64) becomes
P
L
=
2f r
f
U
2
gd
p
(65)
The friction factor, f , is proportional to
f o m
2n
r
n2
f
d
n2
p
U
n2
(1 e)
3n
e
3
(66)
or
f o
(1 e)
3n
e
3

1
(Re)
2n
p
(67)
At low Reynolds numbers, where the pressure drop
does not depend on the uid density, n = 1, and
f o
(1 e)
2
e
3
(Re)
p
(68)
At high Reynolds numbers where the viscosity is not
important, n = 2, and
f o
(1 e)
e
3
(69)
Because of the large dependence on the voidage, a 30-
to-70-fold increase in f is predicted as the voidage
changes from 0.3 to 0.7 in the turbulent and viscous
ranges, respectively. The expression found above for f
at low Reynolds numbers has been experimentally
proved to be correct. At high Reynolds numbers,
however, the expression seems to predict a somewhat
stronger dependence of the friction factor on the
voidage than is actually found experimentally. For a
single particle, voidage becomes 1 by denition, and
the Blakes Reynolds number approaches innity.
This is not surprising, since the Blake analogy based
on a capillary ow analogy breaks down in this
range, because such an analogy has no physical mean-
ing.
7.3 The Brownell and Kats Correlation
Another correlation incorporating the voidage is that
by Brownell and Katz (1947). They introduced a
Reynolds number dened as
(Re)
pe
=
Ud
p
r
f
me
m
(70)
where the exponent m depends on the ratio of the
sphericity to the porosity and ranges from 2 to 20.
The use of dimensionless analysis in correlating the
data is justied as long as the identied dimensionless
groups represent true similarity. The Brownell and
Kats correlation clearly does not represent true simi-
larity, because it predicts that the Reynolds number
decreases with increasing voidage, contradicting the
experimental ndings.
7.4 The Carman and Kozeny Correlations
Carman (1937) studied extensively the uid ow
through various packings in the viscous range and
found that
f =
90(1 e)
2
e
3
(Re)
p
for 0:26 < e < 0:89 (71)
He also found that Eq. (71) can be applied to other
regular shapes as long as their surfaces can be deter-
mined accurately and provided that d
p
is expressed as
6V
p
=S
p
. V
p
is the particle volume, and S
p
is the surface
of the particle. For mixtures of various sizes and
shapes, V
p
and S
p
should be taken as averages of all
particles in the bed.
The Kozeny (1927) equation is usually expressed as,
by combining Eqs. (65) and (71),
P
L
=
180(1 e)
2
mU
ge
3
d
2
p
=
5(1 e)
2
mU
ge
3
(V
p
=S
p
)
2
(72)
Kozeny derived the equation by assuming that a
granular bed is equivalent to a group of parallel simi-
lar channels, commonly known as the pipe ow ana-
logy. He started from the general equation for
streamline ow through a uniform channel and
assumed that the hydraulic radius R
h
of the channel
is e=S. This is equivalent of assuming that the total
internal surface and the total internal volume of the
group of parallel similar channels are equal to the
particle surface and the void volume of the bed. To
nd the ratio of the surface and volume of the parti-
cles, Carman proposed to measure the pressure drop
through the bed of the irregular particles experimen-
tally. The specic surface, S
p
=V
p
, can then be evalu-
ated from
S
p
V
p
=
g Pe
3
5(1 e)
2
mUL
_
(73)
The pipe ow analogy breaks down beyond the creep-
ing ow range. Thus beyond the creeping ow range,
purely empirical models must be employed, the only
remaining link with the pipe ow is the structure of the
dimensionless groups.
7.5 The Ergun Correlation
There are many other correlations such as that by
Oman and Watson (1944) and Happels correlation
(1958). The most widely used empirical correlation of
this type is by Ergun (1952) employing Blakes deni-
tion of drag coecient and the Reynolds number, as
shown in Eqs. (57) and (58).
P
L

gd
p
f
2r
f
U
2
e
3
(1 e)
= 75
(1 e)
f(Re)
p
0:875 (74)
For general applications, including irregular particles,
the Ergun equation shown in Eq. (74) is expressed with
sphericity by substituting fd
p
for d
p
, where d
p
is the
diameter of the irregular particle obtained by particle
measurement techniques, such as sieving or the Coulter
counter, described in Chapter 1. It can be seen that the
Ergun equation reduces to the BlakeKozenyCarman
equation at low Reynolds number, and at high
Reynolds number, to the BurkePlummer equation
for turbulent ow. Many extensions and modications
of the Ergun equation have been proposed, such as
that by Handley and Heggs (1968), Hicks (1970),
Tallmadge (1970), Leva and coworkers (1947), and
Rose and Rizk (1949). Among them the correlations
suggested by Tallmadge and Leva et al. are for high
Reynolds numbers, where the Ergun equation fails to
t the data well.
7.5.1 Use of Ergun Equation to Determine
Sphericity Factor
Subramanian and Arunachalam (1980) suggested a
technique making use of the Ergun equation to deter-
mine the sphericity of irregular particles. For very low
ow rates, the viscous forces predominate, and the
Ergun equation in Eq. (74) can be rearranged explicitly
for the sphericity as
f =
150L(1 e)
2
Bm=(d
2
p
e
3
r
f
g)
(1 L=H
1
)
_ _
1=2
(75)
with
B =
ln H
0
ln H
1
t
H
0
> H
1
> L (76)
By carrying out the experiment with a packed bed of
irregular particles and by draining the liquid in a
laminar ow from the height H
0
to H
1
, the time, t,
required can be determined. Substituting into
Equations (75) and (76), the sphericity of the irregular
particles can be obtained. A simple and accurate
device was described in Subramanian and
Arunachalam (1980).
7.6 Modied Ergun Equation
Gibilaro et al. (1985) modied the Ergun equation and
proposed an alternative pressure drop equation on the
basis of theoretical considerations:
P =
17:3
(Re)
p
0:336
_ _
r
f
U
2
L
d
p
(1 e)e
4:8
(77)
The proposed equation compared well with published
experimental data obtained from high-voidage xed
beds of spheres; it represented a signicant improve-
ment over that of Ergun. The equation is a combina-
tion of two equations, one for the laminar regime and
the other for fully turbulent ow. The laminar ow
regime equation is derived to match the Blake
Kozeny equation at e equal to 0.4 and can be expressed
as
P =
17:3
(Re)
p
0:336
_ _
r
f
U
2
L
d
p
(1 e)e
4:8
(78)
Equation (78) gives an accurate prediction of uidized
bed expansion characteristics for the laminar regime
and applies equally well to xed and suspended particle
systems.
The equation for the fully turbulent ow regime was
also derived as
P = 0:336
r
f
U
2
L
d
p
(1 e)e
4:8
(79)
The constant 0.336 is the result of matching e = 0:4 in
the BurkePlummer (1928) equation, Eq. (80), which
describes the normal packed bed pressure drop well:
P = 1:75
r
f
U
2
L
d
p
(1 e)
e
3
(80)
Equation (79) well represents the steady-state expan-
sion characteristics of a turbulent regime in uidized
beds.
7.7 General Friction Factor Correlations
The pressure drop equations can be converted to
general friction factor correlations as follows. From
the Ergun equation, the friction factor correlations
will be
f
e
=
Pd
p
e
3
r
f
U
2
L(1 e)
=
150(1 e)
(Re)
p
1:75 (81)
The friction factor correlation resulting from the
equation by Gibilaro et al. (1985) becomes
f
p
=
Pd
p
e
4:8
r
f
U
2
L(1 e)
=
17:3
(Re)
p
0:336 (82)
Equation (82) gives a signicantly better representa-
tion for data at higher Reynolds numbers, where
Ergun equation consistently overestimates the
observed friction factor. Wentz and Thodos (1963)
also proposed a general equation for friction factor
for packed and distended beds of spheres:
f
w
=
0:351
Re
0:05
1:2
(83)
Equation (83) is good for voidage between 0.354 and
0.882 and Reynolds numbers between 2,600 and
64,900.
7.8 Drag Coefcient for a Particle in an Array
The drag coecient for a particle in an array was also
derived by Gibilaro et al. (1985) as
C
D1
=
4
3
f
p
e
3:8
C
D
e
3:8
(84)
where C
D
= drag coecient for a particle in an
innite uid.
Equation (84) provides accurate quantitative predic-
tions of particulate uidized bed expansion character-
istics in both laminar and turbulent ow regimes. For
the intermediate ow regime, only a qualitative trend
was observed. Equation (84) is slightly dierent in
dependence of voidage from the drag coecient sug-
gested by Wen and Yu (1966) as shown here:
C
D1
= C
D
e
4:8
(85)
7.9 The General Correlation by Barnea and Mednick
A general correlation for the pressure drop through
xed beds of spherical particles, based on a discrete
particle model corrected for particle interaction, was
proposed by Barnea and Mednick (1978). They
extended the standard C
D
versus Re curve for single
spheres to multiparticle systems by incorporating
proper functions of the volumetric particle concen-
tration. The modied drag coecient and Reynolds
number they suggested are
(Re)
f
= (Re)
p
1
e exp
5(1 e)
3e
_ _
_
_
_
_
(86)
(C
D
)
f
=
8f e
2
3(1 e) 1 K(1 e)
1=3
_ _ ;
f =
P
L

gd
p
2r
f
U
2
(87)
and
(C
D
)
f
= 0:63
4:8
(Re)
f
_ _
2
(88)
where K is a constant
The correlation allows the prediction of pressure
drop or drag in single-particle and multiparticle system
with a single curve. The correlation provides good
agreement with data in the creeping ow and inter-
mediate regimes. Deviation is observed in the turbulent
regime. For the data in the distended bed, the correla-
tion also provides a good t in the highly turbulent
range. The range of applicability may be extended by
the application of average particle size denitions,
shape factors, and wall eect correlations.
7.10 The Concept of Stagnant Voidage
Happel (1958) introduced the interesting concept of a
stagnant voidage, a part of the voidage in the packed
bed that is occupied by the wake of the particles and is
thus not available for the uid ow. The concept is
primarily employed in the discrete particle model but
may be useful for other modeling eort.
Kusik and Happel (1962) derived this equation for
estimating the stagnant voidage for the packing of
spheres:
e
s
= 0:75(1 e)(e 0:2) (89)
Gauvin and Katta (1973) suggested a slightly dierent
equation for a bed packed with spheres:
e
s
= 1:6(1 e)(e 0:2) (90)
They also proposed equations for packing of other
isometric particles.
For packing of cylinders,
e
s
= 1:95K
s
(e 0:2)(1 e) (91)
For packing of ellipsoids,
e
s
= 2:5K
s
(e 0:2)(1 e) (92)
For packing of prisms and wafers,
e
s
= 0:93K
d
e(1 e) (93)
For wood chips,
e
s
= 0:803e(1 e) (94)
where K
s
is the ratio of the mean projected area of a
particle (sphere, cylinder, ellipsoid) to that of a sphere
of the same volume, and K
d
is the ratio of the mean
projected area of a prism to that of a disc with the base
having the same area as the larger face of the prism.
Galloway and Sage (1970) reported that the stag-
nant voidage, e
s
, varied from 0.172 to 0.157 when the
Reynolds number was varied from 10,000 to 35,000
during their experiments.
7.11 Permeability of Packed Beds
Permeability of a packed bed can usually be estimated
from the rearranged form of KozenyCarman equa-
tion:
R
2
k
=
75(1 e)
2
2e
3
(95)
where R is the radius of particle and k is the perme-
ability.
Equation (95) has been found to provide a good
estimation of permeability for packed beds of voidage
between 0.26 and 0.80. Carman (1956) also found that
if the hydraulic radius was used to replace the particle
radius in Eq. (95), the equation was also good for
mixtures of dierent particle sizes. The hydraulic
radius for a bed of spherical particles can be calculated
from
R
h
=
ed
p
6(1 e)
(96)
To account for wall eect, Mehta and Hawley (1969)
modied the equation for hydraulic radius to give
R
h
=
ed
p
6(1 e)M
(97)
and
M = 1
4d
p
6D(1 e)
(98)
where D is the diameter of packed column.
The permeability of a packed bed with polydisperse
spherical or nonspherical particles can also be esti-
mated using Eq. (95) with particle size calculated
through the harmonic mean or the Sauter mean if
the size distribution is not very broad. For wide size
distributions, Li and Park (1998) have proposed equa-
tions for calculating the permeability for both spherical
and nonspherical particles.
8 GAS VELOCITY DISTRIBUTION IN PACKED
BEDS
Because of the uneven distribution of voidage across
the packed bed created during the lling process, the
radial gas velocity distribution in the packed bed is
neither parabolic, as in an empty pipe, nor uniform.
As pointed out earlier, the voidage close to the wall is
usually higher, owing to the wall eect; the gas ow
tends to be higher close to the wall. Schwarz and Smith
(1953) measured the radial gas velocity proles in a
2 in. pipe and a 4 in. pipe lled with 1/8 in. cylinders
and found it indeed was the case (see Fig. 11). When
the packed bed diameter becomes larger, the inuence
of the wall eect decreases. Thus for a large-diameter
bed, the assumption of uniform radial velocity distri-
bution can be a rst approximation. Theoretical pre-
diction of radial velocity distribution is not possible for
a randomly packed bed. The radial gas distribution
can also be profoundly aected by the design of gas
inlet and outlet regions of the packed bed due to
bypassing. Szekely and Poveromo (1975) employed
the vectorial dierential form of the Ergun equation
to predict the ow maldistribution in a packed bed.
The experimental measurements were found to be in
reasonable agreement. They also found that a uni-
formly packed bed with a height/diameter ratio larger
than one could also be used as a ow straightener
because it evened out the nonuniformities introduced
upstream of the bed.
Cohen and Metzner (1981) indicated that for new-
tonian uids, wall eect corrections could be neglected
if the bed-to-particle-diameter ratio is larger than 30.
For nonnewtonian uids, it is 50. They also proposed a
model dividing the bed into three regionsa wall
region, a transition region, and a bulk region. The
wall region extends a distance of one particle diameter
from the wall. The transition can be up to six particle
diameters from the wall where appreciable voidage
oscillation occurs, as discussed in Sec. 6. In the bulk
region, the remainder of the bed region, the voidage is
essentially constant. For packed beds of small bed: -
to : diameter ratios, the use of the single-region model
based on the average voidage tends to over-predict
the average mass ux. The triregional model proposed
by Cohen and Metzner (1981) tted the experimental
data much better. For bed-to-diameter ratios less than
30, the transition region represent the largest fraction of
the total bed cross-sectional area. The mass uxes in the
wall and transition regions were estimated to be larger
than that of the bulk region by as much as 50% for
uids with power indexes of 0.25 and by 10% for uids
with power indexes of 1. Saunders and Ford (1940),
employing a pitot tube, found the velocity in a ring
about one particle diameter from the wall about 50%
higher than the bulk gas velocity. Schwartz and Smith
Figure 11 Radial gas velocity proles in a 50.8 mm and 101.6 mm diameter bed. (Adapted from Schwarz and Smith, 1953.)
(1953), in a bed with a bed-to-particle-diameter ratio
less than 30, also found a peak velocity at about one
particle diameter away from the wall approximately
30% to 100% higher than the bulk velocity. For bed-
to-particle-diameter ratio larger than 30, the bulk
region has the largest cross-sectional area and thus
has the largest contribution to the ow. The wall region
is actually larger than the bulk region for packed beds
with the bed to diameter ratios less than 18. This phy-
sical division is an important consideration for packed
bed design in the laboratory and for attempting to use
the laboratory data for scale-up. Since the voidage in
the bulk region is constant and independent of the bed-
to-particle-diameter ratio, its average ux decreases
with increasing bed to a particle diameter ratio.
9 HEAT TRANSFER IN PACKED BEDS
As for the soliduid heat transfer coecient for ow
through a randomly packed bed, large variations of up
to two- to ve-fold have been observed (Barker, 1965).
The primary diculties have been the attempt to
model the packed bed with simple average parameters
that describe the complex local variations in packing
voidage and the eects of particle shape, distribution,
and velocity. Heat transfer in gassolid packed bed
systems was critically reviewed by Balakrishnan and
Pei (1979). The overall heat transfer in the packed
bed is quite complex and consists of the following
mechanisms: (1) the conduction heat transfer between
particles in both radial and axial directions, (2) the
convective heat transfer between the bed particle and
the ow gas, (3) the interaction of mechanisms (1) and
(2), (4) heat transfer due to radiation between the bed
particles, between the particles and the owing gas,
and between the owing gas and the bed wall, and
(5) heat transfer between the bed wall and bed parti-
cles. Most of the studies in the literature were directed
toward developing correlations for the total heat trans-
fer rate in terms of dimensionless parameters such as
Reynolds number.
9.1 Rase Heat Transfer Correlations
Based on the recommendations by Rase (1990), the
heat and mass transfer correlations to be used for
packed bed calculations are summarized here.
9.1.1 Particle to Fluid Heat Transfer
Heat transferred to a single particle can be expressed as
q
p
= h
s
a
p
(T
s
T) (99)
The heat transfer coecient can be evaluated from the
equation
h
s
d
p
k
g
= 2 1:1Pr
1=3
(Re)
0:6
p
accuracy to 25%
(100)
where (Re)
p
= d
p
Ur
f
=m and d
p
is the diameter of a
sphere or an equivalent sphere. U is the supercial
velocity in the packed bed.
9.1.2 Heat Transfer Through WallOne-
Dimensional ModelAxial
For a cylindrical vessel with spherical particle packing,
q = h
i
A
i
(T T
w
) (101)
where A
i
= inside surface of the cylindrical vessel and
T
w
= wall temperature of the uid at axial position of
interest. For the homogeneous model, T of uid and of
bed are assumed identical.
h
i
D
k
g
= 2:26(Re)
0:8
p
Pr
0:33
exp
6d
p
D
_ _
20 _ (Re)
p
_ 7600 and 0:05 _
d
p
D
_ 0:3
(102)
For a cylindrical vessel with cylindrical particles
packing,
h
i
D
k
g
= 1:40(Re)
0:95
p
Pr
0:33
exp
6d
p
D
_ _
20 _ (Re)
p
_ 800 and 0:03 _
d
p
D
_ 0:2
(103)
9.1.3 Wall Heat Transfer CoefcientTwo-
Dimensional
q = h
w
A
i
(T
R
T
w
) (104)
were T
R
= temperature at inside radius of the vessel.
For spherical particle packing,
h
w
d
p
k
g
= 0:19(Re)
0:79
p
Pr
0:33
20 _ (Re)
p
_ 7600 and 0:05 _
d
p
D
_ 0:3
(105)
For cylindrical particle packing,
h
w
d
p
k
g
= 0:18(Re)
0:93
p
Pr
0:33
20 _ (Re)
p
_ 800 and 0:03 _
d
p
D
_ 0:2
(106)
9.1.4 Effective Radial Thermal Conductivity
q
A
i
= k
g
@T
@r
_ _
r=R
(107)
h
w
d
p
k
g
e
1 e
= 0:27 500 <
d
p
Ur
f
m(1 e)
< 6000
and 0:05 <
d
p
D
< 0:15 (108)
9.2 Heat Transfer Correlations Recommended by
Molerus and Wirth (1997)
In a recent book by Molerus and Wirth (1997), the
recommended heat transfer correlations for packed
beds can be summarized as follows. For fully devel-
oped laminar ow, an approximation formula for
the mean Nusselt number, derived from the pipe ow
analogy, was proposed as
(Nu)
D
=
hD
p
k
g
=
3:66
3
1:61
3
(Pe)
p
D
p
L
p
3
_
for 0:1 <
(Pe)
p
D
p
L
p
< 10
4
(109)
where (Pe)
p
is the pipe ow Peclet number and L
p
is
the pipe length, equivalent to the bed depth. The D
p
and (Pe)
p
can be expressed as
D
p
=
2e
3(1 e)
d
p
(110)
(Pe)
p
=
r
f
C
pg
D
p
U
k
g
(111)
where U is supercial ow velocity.
For heat transfer from a single particle inside a
particle array,
Nu =
hd
p
k
g
= 5:49
1 e
e
_ _
(112)
For voidage values between 0.35 and 0.5, Eq. (112)
gives a range of Nusselt numbers between 5 and 10.
For uid heating by percolating through a particle
array,
(Pe)
d
d
p
L
p
= 7:83
(1 e)
2
e
(113)
where
(Pe)
d
=
r
f
C
pg
d
p
U
k
g
(114)
Equation (113) indicates that the heat transfer between
the particles and the percolating uid is very fast.
Assuming e = 0:4 and L
p
= nd
p
and substituting into
Eq. (113) we have
n = 0:14(Pe)
d
(115)
For (Pe)
d
< 10, Eq. (115) implies that just one particle
layer is enough to heat the percolating gas to the sur-
face temperature of the particles.
9.3 Analytical Models for Heat Transfer with
Immersed Surfaces
When the heat removal is entirely due to the owing
gas and there is no convective particle movement,
Gabor (1970) proposed the following simple model.
For heat transfer from a at plate of length L
h
immersed in the packed bed,
@T
@z
=
k
e
C
pg
G
@
2
T
@y
2
(116)
with the boundary conditions
T = T
w
at y = 0 and 0 < z < L
h
(117)
T = T
o
at y = o (118)
T = T
o
at z = 0 (119)
For heat transfer from a cylindrical heater of length L
h
and radius R
y
immersed in the packed bed,
@T
@z
=
k
e
C
pg
G
@
2
T
@r
2

1
r
@T
@r
_ _
(120)
with the boundary conditions
T = T
w
at r = R
y
and 0 < z < L
h
(121)
T = T
o
at r = o (122)
T = T
o
at z = 0 (123)
The solutions for the average heat transfer coecients
for the at plate and cylinder cases were found by
Gabor (1970).
For at plate case,
h
av
=
4
p
k
e
C
pg
G
L
h
_
(124)
For the cylindrical heater case,
h
av
=
4
p
k
e
C
pg
G
L
h
_

1
2
k
e
R
y
(125)
The eective thermal conductivity of the packed bed,
k
e
, can be estimated from
k
e
= k
o
e
0:1 C
pg
d
p
G
_ _
(126)
The eective thermal conductivity of the packed bed
expressed in Eq. (126) includes two terms, the conduc-
tivity term with no gas ow, k
o
e
, and the convective
term. The factor 0.1 was recommended by Yagi and
Kunii (1957) for spherical particles. The conductivity
term with no gas ow can be calculated below, follow-
ing that suggested by Swift (1966) for orthorhombic
particle packing with a voidage of 0.395.
k
o
e
= 0:9065
2
(1=k
g
) (1=k
s
)
_ _
k
s
k
s
k
g
ln
k
s
k
g
_ _
1
_ _
0:0935k
g
(127)
Botterill and Denloye (1978) conducted an even more
detailed analysis of the heat transfer between an
immersed cylindrical heater and the packed bed by
dividing the heat transfer into two regions, a region
of higher voidage within half a particle diameter
from the immersed cylindrical surface, and the region
of constant voidage outside the wall region. The radial
transfer of heat for the wall region (region 1) is
governed by
r
f
U
1
C
pg
@T
1
@z
= k
e1
@
2
T
1
@r
2

1
r
@
2
T
1
@r
2
_ _
at z > 0
and R
t
< r < R
t
d
p
2
_ _
(128)
In the region outside the wall region, region 2, the
governing equation is
r
f
U
2
C
pg
@T
2
@z
= k
e2
@
2
T
2
@r
2

1
r
@
2
T
2
@r
2
_ _
(129)
Heat transfer in region 1 was assumed to take place by
three dierent mechanisms, neglecting the heat transfer
through radiation. The three mechanisms are (1) heat
transfer by turbulent diusion in the radial direction,
k
a
, (2) heat transfer by molecular conduction in the
uid boundary layer, k
b
, and (3) heat transfer
through the thin lm near particle contact points,
k
c
. Mechanisms (1) and (2) operate in series while
mechanism (3) operates in parallel. The resulting
eective bed conductivity can thus be expressed as
k
e1
k
g
=
k
c
k
g
1
(k
g
=k
a
) (k
g
=k
b
)
(130)
k
a
k
g
= 0:045(Re)
1
Pr (131)
where
(Re)
1
=
d
p
U
1
r
f
m
and Pr =
C
pg
m
k
g
The gas velocity in region 1, the wall region, is taken to
be 50% larger than that in region 2, the bulk region.
h
2
d
p
k
g
= 1:7(Re)
0:5
1
(132)
k
c
k
g
= e
w
(1 e
w
)
2f
w
(2=3)(k
g
=k
s
)
(133)
where
f
w
=
1
4
(K 1)=K [ ]
2
ln K (K 1)=K
_ _
=
1
3K
K =
k
s
k
g
(134)
e
w
= 1
(1 ee)(0:7293 0:5139Y)
1 Y
(135)
where Y = d
p
=2R.
The eective conductivity in region 2 can be evalu-
ated as
k
e2
k
g
=
k
o
e
k
g
0:075(Re)
2
Pr for (Re)
2
< 100
(137)
k
e2
k
g
=
K
o
e
k
g
0:125(Re)
2
Pr for (Re)
2
> 100
(138)
Equations (137) and (138) need to be solved numeri-
cally. The agreement with experimental data was
reported to be within 25%. The prediction of
Gabors model was consistently higher than the
experimental data. The theoretical predictions show
an almost linear dependence of the heat transfer coef-
cient on the particle Reynolds number. Because of
its simplicity, Gabors model may be used as a rst
approximation.
9.4 Mass Transfer in Packed Beds
Again based on recommendation by Rase (1990), the
following correlations for a particle to uid mass
transfer have accuracy to within 25%
Mass transfer to a single particle can be calculated
from
m
p
= k
sj
a
p
C
js
C
j
_ _
(139)
The mass transfer coecient can be obtained from
m
r
f
q
j
= 2 1:1Sc
1=3
Re
0:6
(140)
with accuracy to 25%.
The eective radial diusivity is
N
jr
= q
r
@C
j
@r
_ _
z;r
(141)
eq
r
Ud
p
=
1
m
0:38
Re
(142)
For d
p
=D > 0:1;
m = 11 for Re > 400 (143)
m = 57:85 35:36 log Re 6:68(log Re)
2
for 20 < Re < 400 (144)
For d
p
=D < 0:1, divide q
r
calculated from the above
equation by
1 19:4
d
p
D
_ _
2
_ _
(145)
For more general equations in terms of Re, tortuosity,
and e, see Wen and Fan (1975).
NOMENCLATURE
A
i
= inside surface of the cylindrical vessel
a
p
= surface area of a particle
C
D
= drag coecient of a particle in an innite uid
C
D1
= drag coecient of a particle in an array
(C
D
)
f
= drag coecient for multiparticle systems
C
j
= concentration of component j in bulk region
C
js
= concentration of component j at particle
surface
C
pg
= heat capacity of gas at constant pressure
D = column diameter
D = diameter of the cylindrical vessel
D = dened in Eq. (60)
d
b
= diameter of the large particles
d
c
= diameter of a circle equal in area to the
projected area of the particle when resting on
its larger face
D
c
= diameter of the smallest circle circumscribing
the projection of the particle
d
M
= particle diameter required during critical ratio
of occupation, see Sec. 3
d
p
= equivalent particle diameter
d
pe
= packing equivalent particle diameter
d
pv
= diameter of sphere with equivalent volume of
the cylinder
d
s
= diameter of the small particles
d
ve
= volume equivalent particle diameter
d
yi
= inside diameter of the hollow cylinder
d
yo
= outside diameter of the hollow cylinder
f = friction factor
f
B
= friction factor proposed by Blake, Eq. (58)
f
e
= general friction factor from Ergun (1952)
f
p
= general friction factor from Gibilaro et al.
(1985)
f
w
= general friction factor from Wentz and
Thodos (1963)
g
= vector form of gravitational acceleration

G = mass ow rate of gas
G = an empirical constant
h
av
= average heat transfer coecient, averaged over
the length of the heater
h
i
= heat transfer coecient at inside surface of the
cylindrical vessel
h
s
= heat transfer coecient to a single particle
h
w
= heat transfer coecient at the wall
h
1
= uid column height 1
h
1
= heat transfer coecient in region 1
h
2
= uid column height 2
h
2
= heat transfer coecient in region 2
I
c
= cylindrical index
I
d
= disk index
k = permeability
k = an empirical constant
K = ratio of thermal conductivity of solid and gas,
K = k
e
=k
g
K = a constant
K
/
= permeability constant
k
s
= heat transfer by turbulent diusion in radial
direction
k
b
= heat transfer by molecular conduction in uid
boundary layer
k
c
= heat transfer through the thin lm near
particle contact point
K
d
= ratio of mean projected area of a prism to that
of a disc with the base having the same area
as the larger face of the prism
k
e
= eective thermal conductivity of packed bed
k
e1
= eective bed thermal conductivity in region 1
k
e2
= eective bed thermal conductivity in region 2
k
o
e
= eective thermal conductivity of the packed
bed with zero gas ow
k
g
= thermal conductivity of gas
k
s
= specic permeability
k
s
= thermal conductivity of solid
k
sj
= mass transfer coecient of component
K
s
= ratio of mean projected area of a particle to
that of a sphere of the same volume
L = bed height
L
h
= length of heat transfer surface
L
p
= pipe length
m = exponential coecient
m
p
= mass transfer rate
n = coordination number
n = an exponential coecient
N
p
= number of spheres per unit volume of bed
Nu = Nusselt number
(Nu)
D
= pipe ow Nusselt number
(Pe)
d
= pipe ow Peclet number dened in Eq. (114)
(Pe)
p
= pipe ow Peclet number dened in Eq. (111)
P = pressure
P
1
= pressure at point 1
P
2
= pressure at point 2
P = pressure drop
Pe = Peclet number
q = heat transfer rate
q
= vector quantity of volumetric ow rate per

unit bed area
Q = volumetric uid ow rate
q
p
= heat transfer rate to a single particle
r = radial coordinate
r
p
= diameter ratio of small particle to large
particle
r
str
= structural ratio dened in Eq. (24)
R = radius of sphere
R
t
= radius of the heater
R
h
= hydraulic radius
Re = Reynolds number, = DUr
f
=m
(Re)
B
= Reynolds number proposed by Blake, Eq. (57)
(Re)
p
= Reynolds number based on particle diameter,
d
p
Ur
f
=m
(Re)
pe
= Reynolds number proposed by Brownell and
Kats, and dened in Eq. (70)
(Re)
f
= Reynolds number for multiparticle systems
(Re)
1
= Reynolds number in region 1
(Re)
2
= Reynolds number in region 2
R
y
= radius of cylindrical heater
S = specic particle surface area; total particle
surface area per unit volume of bed
Sc = Schmidt number, = m=q
j
r
f
S
p
= particle surface
T = temperature
T
o
= initial gas temperature
T
R
= temperature at inside radius of the vessel
T
s
= temperature of a single particle
T
w
= temperature at the heater wall
T
w
= temperature at wall
T
1
= temperature in region 1
T
2
= temperature in region 2
U = supercial uid velocity
U
1
= supercial uid velocity in region 1
U
2
= supercial uid velocity in region 2
V = specic volume of solid material,
V = 1=(1 e)
V
b
= volume of big particles
V
p
= particle volume
V
s
= volume of small particles
X
b
= volume fraction of big particles
X
min
= volume fraction of large particles where the
voidage of the binary mixture is the minimum
X
s
= volume fraction of small particles
y = horizontal coordinate
z = vertical coordinate
z = axial distance along the heat transfer surface
z
1
= vertical coordinate at point 1
z
2
= vertical coordinate at point 2
q
j
= diusivity
b = layer spacing
m = uid viscosity
r
f
= uid density
r
p
= particle density
f = Wadells sphericity, the ratio of the surface
areas of a volume equivalent sphere and the
actual particle
f
c
= an equivalent sphericity or aspect ratio dened
as the ratio of the maximum length to the
diameter of the circle having the same
projected area normal to the maximum length
f
d
= a sphericity dened as the ratio of the shortest
length to the diameter of the circle having the
same projected area
e = voidage of the xed bed
e
av
= average voidage of the binary mixture
e
b
= voidage of a packed bed with large particles
alone
e
c
= voidage at core region
e
hc
= voidage of the hollow cylinder packing
e
min
= the minimum voidage of the binary mixture
e
o
= initial packing voidage
e
o;cylinder
= voidage of initial packing of cylinders
e
o;disk
= voidage of initial packing of disks
e
o1;cylinder
= voidage of initial loose packing of cylinders
e
o1;disk
= voidage of initial loose packing of disks
e
od;cylinder
= voidage of initial dense packing of cylinders
e
od;disk
= voidage of initial dense packing of disks
e
s
= voidage of a packed bed with small particles
alone
e
s
= stagnant voidage
e
sc
= voidage of the solid cylinder packing
e
w
= voidage at wall region
ee = average voidage
REFERENCES
E Abe, H Hirosue, A Yokota. Pressure drop through a
packed bed of binary mixture. J Chem Eng Japan
12:302, 1979.
AR Balakrishnan, Pei DCT. Heat transfer in gassolid
packed bed systems, a critical review. Ind Eng Chem
Process Des Dev 18:3040, 1979.
JJ Barker. Heat transfer in packed beds. Ind Eng Chem
57(4):4351, April 1965.
E Barnea, RL Mednick. A generalized approach to the uid
dynamics of particulate systems part III: general correla-
tion for the pressure drop through xed beds of spherical
particles. Chem Eng J 15:215227, 1978.
RF Benenati, CB Brosilow. Void fraction distribution in beds
of spheres. AIChE J 8:359361, 1962.
FC Blake. The resistance of packing to uid ow. Trans Am
Inst Chem Eng 14:415, 1922.
JSM Botterill, AOO Denloye. A theoretical model of heat
transfer to a packed or quiescent uidized bed. Chem Eng
Sci 33:509515, 1978.
GG Brown. Unit Operations. New York: John Wiley, 1966.
LE Brownell, DL Katz. Flow of uids through porous media
I. Single homogeneous uids. Chem Eng Prog 43:537
548, October 1947.
SP Burke, WB Plummer. Gas ow through packed columns.
Ind Eng Chem 20:11961200, 1928.
PC Carman. Fluid ow through granular beds. Trans Inst
Chem Eng 15:150, 1937
PC Carman. The ow of Gases Through Porous Media. New
York: Academic Press, 1956.
Y Cohen, AB Metzner. Wall eects in laminar ow of uids
through packed beds. AIChE J 27:705715, 1981.
JM Coulson. The ow of uids through granular beds: eect
of particle shape and voids in streamline ow. Trans Inst
Chem Eng 27:237257, 1949.
DJ Cumberland, RJ Crawford. The Packing of Particles.
Amsterdam: Elsevier, 1987, p 60.
H Darcy. Les Fontaines publiques de la ville de Dizon. Paris:
Dalmont, 1856.
AG Dixon. Correlations for wall and particle shape eects on
xed bed bulk voidage. Can J Chem Eng 66:705708,
1988.
S Ergun. Fluid ow through packed columns. Chem Eng
Prog 48:8994, February 1952.
HJ Finkers, AC Homann. Structural ratio for predicting
the voidage of binary particle mixtures. AIChE J
44:495498.
CC Furnas. Flow of gases through beds of broken solids.
Bull 307. US Bureau of Mines, 1929.
JD Gabor. Heat transfer to particle beds with gas ows less
than or equal to that required for incipient uidization.
Chem Eng Sci 25:979984, 1970.
TR Galloway, BH Sage. A model of the mechanism of trans-
port in packed, distended, and uidized beds. Chem Eng
Sci 25:495516, 1970.
WH Gauvin, S Katta. Momentum transfer through packed
beds of various particles in the turbulent ow regime.
AIChE J 19:775783, 1973.
L Gibilaro, RDI Delice, SP Waldram, PU Foscolo.
Generalized friction factor and drag coecient correla-
tions for uidparticle interactions. Chem Eng Sci
40:18171823, 1985.
JS Goodling, RI Vachon, WS Stelpug, SJ Ying, MS
Khader. Radial porosity distribution in cylindrical beds
packed with spheres. Powder Technol 35:2329, 1983.
D Handley, PJ Heggs. Momentum and heat transfer mechan-
isms in regular shaped packings. Trans Inst Chem Eng
46:T251T264, 1968.
J Happel. Viscous ow in multiparticle systems: slow motion
of uids relative to beds of spherical particles. AIChE J
4:197201, 1958.
DP Haughey, GG Beveridge. Local voidage variation in a
randomly packed bed of equal-sized spheres. Chem Eng
Sci 21:905916, 1966.
DP Haughey, GG Beveridge. Structural properties of packed
bedsa review. Can J Chem Eng 47:130140, 1969.
G Herdan. Small Particle Statistics. New York: Academic
Press, 1960.
RE Hicks. Pressure drop in packed beds of spheres. Ind Eng
Chem Fundam 9:500502, 1970.
AC Homann, HJ Finkers. A relation for the void fraction
of randomly packed particle beds. Powder Technol
82:197203, 1995.
J Kozeny. U
ber Kapillare Leitung des Wassers im Boden.

Sitzungsber Akad Wiss Wien 136:271, 1927.
CL Kusik, J Happel. Boundary layer mass transfer with
heterogeneous catalysis. AIChE J 8:163, 1962.
M Leva. Pressure drop through packed tubes: Part I. A
general correlation. Chem Eng Prog 43:549554, 1947.
Y Li, CW Park. Permeability of packed beds lled with poly-
disperse spherical particles. Ind Eng Chem Res 37:2005
2011, 1998.
RK McGeary. Mechanical packing of spherical particles. J
Am Ceram Soc 44:513522, 1961.
D Mehta, MC Hawley. Wall eects in packed columns. Ind
Eng Chem Proc Design Dev 8:280282, 1969.
O Molerus. Principles of Flow in Disperse Systems. London:
Chapman and Hall, 1993.
O Molerus, KE Wirth. Heat Transfer in Fluidized Beds.
London: Chapman and Hall, 1997.
PG Nutting. Bull Amer Assoc Petrol Geo 14:1337, 1930.
AO Oman, KM Watson. Pressure drops in granular beds.
Renery Mat Petrol Chem Tech 36:R795, 1944.
N Ouchiyama, T Tanaka. Predicting the densest packings of
ternary and quaternary mixtures of solid particles. Ind
Eng Chem Res 28:15301536, 1989.
M Propster, J Szekely. The porosity of systems consisting of
layers of dierent particles. Powder Technol 17:123138,
1977.
WE Ranz. Friction and transfer coecients for single parti-
cles and packed beds. Chem Eng Prog 48:247253, 1952.
HF Rase. Fixed-Bed Reactor Design and Diagnostics.
HE Rose, MA Rizk. Further researches in uid ow through
beds of granular material. Proc Inst Mech Eng London
160:493503, 1949.
OA Saunders, N Ford. Heat transfer in the ow of gas
through a bed of solid particles. J Iron Steel Inst
141:291316, 1940.
AE Scheidegger. The Physics of Flow Through Porous
Media. New York: MacMillan, 1961.
CE Schwartz, JM Smith. Flow distribution in packed beds.
Ind Eng Chem 45:12091218, 1953.
GD Scott. Packing of spheres Nature 188:908911, 1960.
HY Sohn, C Moreland. The eect of particle size distribution
on packing density. Can J Chem Eng 46:162167, 1968.
P Subramanian, Vr Arunachalam. A simple device for deter-
mination of sphericity factor. Ind Eng Chem Fandam
19:436-437, 1980.
DL Swift. The thermal conductivity of spherical metal pow-
ders including the eect of an oxide coating. Int J Heat
Mass Transfer 9:10611074, 1966.
J Szekely, JJ Poveromo. Flow maldistribution in packed
beds: a comparison of measurements with predictions.
AIChE J 21:769775, 1975.
JA Tallmadge. Packed bed pressure dropan extension to
higher Reynolds numbers. AIChE J 16:10921093, 1970.
H Wadell. Volume, shape, and roundness of quartz particles.
J Geol 43:250280, 1935.
CY Wen, LT Fan. Models for Flow Systems and Chemical
Reactors. New York: Marcel Dekker, 1975.
CY Wen, YH Yu. Mechanics of uidization. Chem Eng Prog
Symp Ser 62(62):100111, 1966.
CA Wentz Jr, G Thodos. Pressure drops in the ow of gases
through packed and distended beds of spherical particles.
AIChE J 9:8184, 1963.
AER Westman. The packing of particles: empirical equations
for intermediate diameter ratios. J Am Ceram Soc 19:127
129, 1936.
S Yagi, D Kunii. Studies on eective thermal conductivities
in packed beds. AIChE J 3:373381, 1957.
AB Yu, N Standish, A McLean. Porosity calculation of
binary mixtures of nonspherical particles. J Am Ceram
Soc 76:28132816, 1993.
RP Zou, AB Yu. Evaluation of the packing characteristics of
monosized nonspherical particles. Powder Technol 88:71
79, 1996.
3
Bubbling Fluidized Beds
Wen-Ching Yang
1 INTRODUCTION
For chemical reactor applications, the xed (or
packed) beds described in Chapter 2 have some
major disadvantages. If the reactions are fast and
highly exothermic or endothermic, hot or cold spots
will form in the packed beds and render the reactor
ineective; or it increases the chance or unwanted
byproduct production. By nature, the xed bed opera-
tion has to be in batch mode, which is much less e-
cient than a continous operation. Sintering, plugging,
and uid maldistribution can also occur much more
readily in packed beds. Comparing to xed beds, ui-
dized beds have many advantages. Once the solids in
the bed are uidized, the solids inside the bed will
behave just like liquid (Gelperin and Einstein, 1971;
Davidson et al., 1977). The bed surface of a bubbling
uidized bed resembles that of a boiling liquid, and it
can be stirred easily just like a liquid. Objects with a
density lighter than the bulk density of the bed will
oat and those heavier will sink. If there is a hole on
the side of a uidizing bed, the solids will ow out like
a liquid jet. The gas bubble size, shape, formation,
rising velocity, and coalescence in the uidized beds
have quantitative similarity with those of gas bubbles
in liquids.
The liquid like behavior of a uidized bed thus
allows the solids to be handled like a uid, and
continous feeding and withdrawal therefore become
possible. The rigorous mixing in a uidized bed results
in a uniform temperature even for highly exothermic or
endothermic reactions. This in turn provides an easier
reactor control as well. The rigorous mixing also
improves solids and uid contacting, and it enhances
heat and mass transfer. However, uidized beds also
possess some serious disadvantages. Rigorous solids
mixing in the bed produces solid nes through
attrition. Operating at high udiization velocities,
nes elutriation and entrainment can become a serious
operational problem. Also, because of the rigorous
mixing in the bed, a uidized bed is essentially a
continous stirred tank reactor with varying solids resi-
dence time distribution. These deciencies may or may
not be resolved through design. In practice, uidiza-
tion is still an empirical science (Yang, 1998a). Care
must be exercised to select proper correlations for
design and scale-up. A recent review of hydrodynamics
of gassolid uidization can be found in Lim et al.
(1995).
There are many dierent variations of uidized beds
in practice, which are covered in dierent chapters in
this handbook. For conventional bubbling uidized
beds, the essential elements are depicted in Fig. 1.
Depending on the applications, not all elements
shown in Fig. 1 will be necessary. The design of the
individual element can also vary substantially from one
design to another design. For example, the cyclone can
be arranged either internally or externally; the heat
transfer tubes can be either vertical or horizontal,
etc. These design aspects and various phenomena
occurring in a bubbling uidized bed will be discussed
throughout this book.
2 POWDER CLASSIFICATIONS AND TYPES
OF GAS FLUIDIZATION
The uidization phenomena of gassolids systems
depend very much on the types of powders employed.
There are several classications available in the litera-
ture, all based on the original work by Geldart (1973).
2.1 Geldart Classication of Powders
Geldart (1973) was the rst to classify the behavior of
solids uidized by gases into four clearly recognizable
groups characterized by the density dierence between
the particles and the uidizing medium, (r
p
r
f
) and
by the mean particle size, d
p
. Geldarts classication
has since become the standard to demarcate the types
of gas uidization. The most easily recognizable fea-
tures are
Group A The bed particles exhibit dense phase
expansion after minimum uidization
and before the beginning of bubbling.
Gas bubbles appear at the minimum
bubbling velocity.
Group B Gas bubbles appear at the minimum
uidization velocity.
Group C The bed particles are cohesive and dicult
to uidize.
Group D Stable spouted beds can be easily formed
in this group of powders.
Demarcation for Groups A and B Powders. For
Group A,
U
mb
U
mf
_ 1
U
mf
=
8 10
4
gd
2
p
(r
p
r
f
)
m
(1)
d
p
=
1
i
x
i
d
pi
_ _ (2)
U
mb
= Kd
p
(3)
Thus the powders are Group A powders if
8 10
4
gd
p
(r
p
r
f
)
Km
_ 1 (4)
For air at Room Temperature and Pressure, K =
100
For Group D Powders, U
B
_
U
mf
e
mf
,
U
B
_
8 10
4
(r
p
r
f
)gd
2
p
me
mf
(5)
Equation (5) is the demarcation between Group B and
D powders. The powder classication diagram for the
uidization by air at ambient conditions was presented
by Geldart, as shown in Fig. 2.
Minimum Bubbling Velocity. For gas uidization
of ne particles, the uidization velocity at which the
gas bubbles rst appear is called the minimum bub-
bling velocity, U
mb
. Geldart and Abrahamsen (1978)
observed that
U
mb
U
mf
=
4:125 10
4
m
0:9
r
0:1
f
(r
p
r
f
)gd
p
(6)
where the units are in kg, m, and s. Equation (6) gives
U
mb
< U
mf
, which is generally true for Groups B and
D powders. For ne particles (Group A), the ratio
U
mb
=U
mf
> 1 was predicted and observed to increase
with both temperature and pressure. The ratio also
increases with smaller particles and lighter particles.
2.2 Molerus Interpretation of Geldarts
Classication of Powders
Molerus (1982) classied the powders by taking into
account the interparticle cohesion forces. Free particle
motion for Group C powder is suppressed by the
Figure 1 Essential elements of a bubbling uidized bed.
dominant inuence of cohesion forces, and its demar-
cation can be expressed by the equation
10
(r
p
r
f
)d
3
p
g
F
H
= K
1
10
2
(7)
where F
H
is the adhesion force transmitted in a single
contact between two particles. For polypropylene
powder, F
H
= 7:71 10
7
newtons, and for glass
beads and cracking catalyst, F
H
= 8:76 10
8
new-
tons. K
1
is found experimentally.
For Group A and Group B transition, the following
equation can be used.
p
6
(r
p
r
f
)d
3
p
g
F
H
= K
2
0:16 (8)
For Group B and Group D, the transition equation is
(r
p
r
f
)d
p
g
r
f
U
2
mf
15 (9)
2.3 Particle Classication Boundaries Suggested by
Grace
Based on additional data beyond that analyzed by
Geldart, Grace (1986) suggested new boundaries
between Groups A and B, and between Groups B
and D of Geldarts classication. The new boundaries
are good also for gases other than air and for tempera-
ture and pressure other than ambient.
Boundary Between Group A and Group B
Ar = 1:03 10
6
r
p
r
f
r
f
_ _
1:275
(10)
For (r
p
r
f
)=r
f
1000 to 2000, Eq. (10) reduces to
Ar 125 for air at atmospheric pressure. This com-
pares to a value of Ar = 88.5 as suggested by Goossens
(1998).
Boundary Between Group B and Group D
Ar = 1:45 10
5
(11)
This Group B/D boundary can be compared with Ar =
176,900, as suggested by Goossens (1998).
2.4 Goossens Classication of Particles by
Archimedes Number
Based on the hypothesis that the relative importance of
laminar and turbulent phenomena governs the uid-
ization behavior, Goossen (1998) classied the
powders on the basis of Archimedes number. The
proposed classication is of general application, apply-
ing equally well in both liquid uidization and gas
uidization. The four boundaries he suggested are as
follows:
Figure 2 Geldart classication of powders.
Group C boundary Ar = 0.97 (12)
Group A/C boundary Ar = 9.8 (13)
Group A/B boundary Ar = 88.5 (14)
Group B/D boundary Ar = 176,900 (15)
The demarcations between groups of powder compare
well with that of Geldarts and Molerus when air is
used as the uidization medium, except for the Group
A/B boundary.
2.5 Powder Characterization by Bed Collapsing
The bed collapsing technique has been employed to
study uidization for various objectives (Rietema,
1967; Morooka et al., 1973; Geldart, 1986). Kwauk
(1992) has instrumented the bed for automatic surface
tracking and data processing to characterize powders.
When the gas is abruptly turned o for an operating
uidized bed, the bed collapses in three distinct stages:
(1) bubble escaping stage, (2) hindered settling stage,
and (3) solids consolidation stage. The bed collapsing
stage can be conveniently represented grahically in Fig.
3. Mathematically, the bed collapsing stages can be
described as follows.
2.5.1 Bubble Escaping Stage, 0 < t < t
b
During the bubble escaping stage, the change of bed
surface is linearly dependent on the time, or
H
1
= H
0
U
1
t (16)
Figure 3 Bed collapsing stages during bed collapsing characterization of powders. (Adapted from Kwauk, 1992.)
where U
1
=
f
B
(1 f
w
e
e
)U
B
e
e
(1 f
B
F
w
f
B
)U
e
(1 f
B
)(1 e
e
)
(17)
and f
B
=
H
0
H
e
H
0
(18)
2.5.2 Hindered Settling Stage, t
b
< t < t
c
During the hindered settling stage, the bed surface set-
tles at a constant velocity, and the settled layer at the
bottom, layer D in Fig. 3, also increases in height. The
change in bed height for both layers can be expressed
as
H
2
= H
e
U
2
t (19)
H
D
=
H
o
H
e
H
o
_ _
U
2
t
1 exp K
3
H
e
H
o
H
e
_ _
t
_ _ _ _
U
2
H
e
(H
e
H
o
)K
3
1 e
e
e
e
e
c
H
e
H
e
H
o
_ _
(20)
where K
3
is the rate constant for the increasing of
height in layer D expressed as a rst-order process:
dH
dt
= K
3
(H
D
H
o
) (21)
2.5.3 Solids Consolidation Stage, t
c
< t < t
o
The consolidation stage is essentially the consolidation
of layer D and thus the change in bed height can be
expressed as
dH
3
dt
=
dH
D
dt
= K
3
(H
D
H
o
) (22)
or
H
3
= H
D
= (H
c
H
o
) exp[K
3
(t t
c
)] H
o
(23)
Not all powders exhibit all three stages during bed
collapsing experiments. Only Geldarts Group A
powders exhibit all three stages. For Group B and D
powders, the rst bubble escaping stage and the second
hindered settling stage are practically instantaneous.
The transition between the hindered settling stage
and the solids consolidation stage, the point (t
c
; H
c
)
in Fig. 3, is called by Yang et al. (1985) the critical
point. The better to qualify the powders through the
bed collapsing tests, Yang et al. (1985) also dened a
dimensionless subsidence time of a powder, y
s
.
For small powders whose terminal velocity can be
calculated by the Stokes law, y
s
can be expressed by
y
s
=
t
c
d
p
(r
p
r
f
)H
o
(24)
The dimensionless subsidence time of a powder was
found to be related to the ratio of minimum bubbling
and minimum uidization velocities
1n
U
mb
U
mf
_ _
= 4y
0:25
s
(25)
Thus for Group B and D powders, U
mb
=U
mf
= 1, the
corresponding subsidence time is zero, and hence t
c
is
zero. Stages 1 and 2 occur almost instantaneously. For
larger values of y
s
, the ratios of U
mb
and U
mf
become
larger, and the bed exhibits the particulate uidization
more prominently.
Yang et al. (1985) have performed extensive experi-
ments employing the bed collapsing technique to study
the modication of uidization behavior by addition of
nes. They found that improvement in uidization
increases monotonically with increases in nes concen-
tration in the bed. A more interesting study is the series
for binary mixtures of Group BC powders. They
found that addition of cohesive Group C powder
into Group B powder will improve the uidization
behavior of Group B powder. Conversely, addition
of Group B powder into the cohesive Group C powder
will also improve the uidization quality of a cohesive
powder. There is, however, some limitations on the
eective amount to be added in the mixture.
Addition beyond the maximum amount will adversely
aect the uidity and uidization behavior of the
mixture.
3 DIFFERENT REGIMES OF FLUIDIZATION
Fluidization regimes can be classied into two broad
categoriesparticulate (smooth) and aggregative
(bubbling) (Harrison et al., 1961). Most liquid uidized
beds under normal operation exhibit the particulate
uidization. In particulate uidization, the solid
particles usually disperse relatively uniformly in the
uidizing medium with no readily identiable bubbles.
Thus the particulate uidization sometimes is also
called homogeneous uidization. In the heterogeneous
or aggregative uidization, voids containing no solids
are usually observed. Those voids are called bubbles.
Those voids can be well dened as in a bubbling ui-
dized bed or in a slugging bed, or they can be small
voids where particle clusters dart to and from like in a
turbulent bed or in a fast uidized bed. There are
a number of criteria available to determine whether
a particular system will exhibit particulate or aggre-
gative uidization. Those criteria are summarized in
Table 1.
It was suggested by Harrison et al. (1961) that
aggregative uidization may be expected if the ratio
of maximum stable bubble size to particle diameter,
(D
B
)
max
=d
p
, is larger than 10, and particulate uidiza-
tion, if the ratio is less than or equal to unity. A transi-
tion region exists with the ratio between 1 and 10. They
proposed to calculate the ratio as follows:
(D
B
)
max
d
p
= 71:3
m
2
gd
3
p
r
2
f
_ _
(r
p
=(r
p
r
f
)) e
mf
1 e
mf
_ _
(26)
1
gd
3
p
r
f
54m
2
(r
p
r
f
)
_ _
1=2
1
_
_
_
_
2
For large particles, Eq. (26) reduces to
(D
B
)
max
d
p
= 1:32
r
p
r
f
r
f
_ _
(r
p
=(r
p
r
f
)) e
mf
1 e
mf
_ _
(27)
Eqs. (26) and (27) are only applicable when the
maximum stable bubble size is much smaller than the
bed diameter, i.e., without wall eect. Experimental
evidence of maximum bubble size was observed by
Matsen (1973).
For gassolid systems, there are at least ve distin-
guishable regimes of uidization observable experi-
mentally: xed bed, particulate uidization, bubbling
uidization, slugging uidization, and turbulent uid-
ization. When the operating velocity is higher than the
transport velocity such that recycle of entrained parti-
cles is necessary to maintain a bed, additional uidiz-
ing regimes are possible. The regimes of uidization for
circulating uidized beds is discussed in more detail in
Chapter 19, Circulating Fluidized Beds. The dier-
ent regimes for gassolid uidization are summarized
in Table 2 and Figs. 4 and 5. Not all of these regimes
can be observed in all systems, however, because some
regimes are also dependent on the size of the equip-
ment employed. For a recent review of uidization
regimes, see Bi and Grace (1995a).
3.1 Transition among Fixed Bed, Particulate
Fluidization, and Bubbling Regime
For Geldarts Group B and Group D powders, the bed
transfers from the xed bed into a bubbling uidized
bed when the gas velocity is increased beyond the mini-
mum uidization velocity of the system. For Group A
powders, no bubbles will be observed, instead the bed
will expand homogeneously. The bubbles only appear
when the gas velocity is increased beyond the minimum
bubbling velocity. Thus the transition point from the
Table 1 Criteria for Transition Between Particulate and Aggregative
Fluidization
Criteria for particulate uidization Reference
Fr =
U
2
mf
gd
p
< 0:13 Wilhelm and Kwauk (1948)
(D
B
)
max
d
p
< 30; (D
B
)
max
=
2U
2
t
g
Harrison et al. (1961)
U
mf
< 0:2 cm/s Rowe (1962)
U
3
mf
(
p
f
)H
mf
gmD
< 100 Romero and Johanson (1962)
(
p
f
)
m
gd
3
p
_
< C
1
depending on e and e
mf
Verloop and Heertjes (1974)
gd
3
p
r
f
(r
p
r
f
)
m
2
_ _
m
r
mf
r
f
_ _
0:5
< C
2
Doichev et al. (1975)
m and C
2
depend on range of operation
xed bed to the bubbling regime is the minimum uid-
ization velocity for Group B and Group D powders
while for Group A powders, it is the minimum bub-
bling velocity. The homogeneous expansion, also
called particulate uidization, occurs only in Group
A powders for gassolids systems. The particulate
uidization is especially important for liquid uidized
beds because where most of the occurrence is observed.
Liquidsolids uidization will be discussed in Chapter
26, LiquidSolids Fluidization.
3.2 Transition Between Bubbling and Slugging
Regimes
A slugging regime occurs only in beds with bed height
(H) over bed diameter ratio (D) larger than about 2.
With large H=D ratios, the bed provides enough time
for bubbles to coalesce into bigger ones. When the
bubbles grow to approximately 2/3 of the bed dia-
meter, the bed enters the slugging regime with periodic
passing of large bubbles and regular large uctuation
of bed pressure drop corresponding to the bubble
frequency. There are several correlations available to
predict this transition, they are discussed in Sec. 10;
Slugging Beds.
3.3 Transition Between Bubbling and Turbulent
Regimes
When the gas velocity is continuously increased, the
bubbles grow bigger owing to coalescence, and the
bubbling bed can transfer into a slugging bed if the
bed diameter is small and the particle diameter is
large, or into a turbulent bed if the bed diameter is
large and the particle diameter is small. If the standard
deviation of pressure uctuation is measured and
plotted against the supercial uidization velocity,
two characteristic velocities, U
c
and U
k
, rst suggested
by Yerushalmi and Cankurt (1979), can be identied.
The velocity U
c
corresponds to the bed operating con-
ditions where the bubbles or slugs reach their maxi-
mum diameter and thus have the largest standard
deviation of pressure ucuation. Continuing increases
beyond this velocity, large bubbles start to break up
into smaller bubbles with smaller pressure uctuation,
and eventually the standard deviation of the pressure
uctuation reaches a steady state. This velocity is
denoted as U
k
, also a characterization velocity for
transition from the bubbling regime to the turbulent
regime. Subsequent study by Chehbouni et al. (1994),
employing dierential and absolute pressure transdu-
cers and a capacitance probe, concluded that the onset
Table 2 Summary Description of Different Regimes of Fluidization
Velocity range Fluidization regime Fluidization features and appearance
0 _ U < U
mf
Fixed bed Particles are quiescent; gas ows through interstices
U
mf
_ U < U
mb
Particulate regime Bed expands smoothly and homogeneously with small-scale
particle motion; bed surface is well dened
U
mb
_ U < U
ms
Bubbling regime Gas bubbles form above distributor, coalesce and grow; gas
bubbles promote solids mixing during rise to surface and
breakthrough
U
ms
_ U < U
c
Slug ow regime Bubble size approaches bed cross section; bed surface rises
and falls with regular frequency with corresponding
pressure uctuation
U
c
_ U < U
k
Transition to turbulent
uidization
Pressure uctuations decrease gradually until turbulent
uidization regime is reached
U
k
_ U < U
tr
Turbulent regime Small gas voids and particle clusters and streamers dart to
and fro; bed surface is diused and dicult to distinguish
U > U
tr
Fast uidization Particles are transported out of the bed and need to be
replaced and recycled; normally has a dense phase region
at bottom coexisting with a dilute phase region on top; no
bed surface
U U
tr
Pneumatic conveying Usually a once-through operation; all particles fed are
transported out in dilute phase with concentration varying
along the column height; no bed surface
of turbulent uidization was at U
c
, and the velocity U
k
actually did not exist. The existence of U
k
is an artifact
due to the use of dierential pressure transducers for
experiments. They maintained that there is only one
transition velocity, i.e., U
c
. The turbulent uidization
starts at U
c
and ends at the transport velocity, U
tr
, the
velocity capable of transporting all particles out of the
reactor. There is still controversy in the literature on
the actual transition boundary between the bubbling
and the turbulent regimes.
In 1986, Horio proposed the following equations to
calculate U
c
and U
k
(see also Horio, 1990).
(Re)
c
=
d
p
r
f
U
c
m
= 0:936Ar
0:472
(28)
(Re)
k
=
d
p
r
f
U
k
m
= 1:46Ar
0:472
for Canada et al. (1978) data (29)
(Re)
k
=
d
p
r
f
U
k
m
= 1:41Ar
0:56
for Yerushalmi et al. (1978) data (30)
The extensive literature data based on absolute pres-
sure uctuation and bed expansion measurements up
to 1989 were correlated by Cai et al. (1989) to be
(Re)
c
=
d
p
r
f
U
c
m
= 0:57Ar
0:46
(31)
Figure 4 Regimes of uidization for nontransporting systems.
Perales et al. (1990) also suggested equations to calcu-
late U
k
and U
tr
as follows.
(Re)
k
=
d
p
r
f
U
k
m
= 1:95Ar
0:453
(32)
(Re)
tr
=
d
p
r
f
U
tr
m
= 1:41Ar
0:483
(33)
Since the U
k
and U
tr
are very similar, they suggest that
the following equation alone may be used to calculate
both U
k
and U
tr
for simplication:
(Re)
k;tr
=
d
p
r
f
U
k
m
=
d
p
r
f
U
tr
m
= 1:45Ar
0:484
(34)
Bi and Fan (1992) armed the existence of turbulent
regimes in gassolid uidization and suggested the fol-
lowing criteria for transition to turbulent uidization.
(Re)
k
=
d
p
r
f
U
k
m
= 16:31Ar
0:136
U
t
gD
_
_ _
0:941
for Ar _ 125 (35)
(Re)
k
=
d
p
r
f
U
k
m
= 2:274Ar
0:419
U
t
g
_
D
_ _
0:0015
for Ar _ 125 (36)
(Re)
tr
=
d
p
r
f
U
tr
m
= 2:28Ar
0:419
(37)
Bi and Grace (1995b) correlated the literature data
based on the dierential pressure uctuation and
arrived at the following transition equation:
(Re)
c
=
d
p
r
f
U
c
m
= 1:24Ar
0:45
for 2 < Ar < 1 10
8
(38)
For gas uidization of large particles, Catipovic et al.
(1978) further subdivided the regimes into slow bub-
bles, fast bubbles, and rapidly growing bubble regimes.
The transition equations are summarized here.
Regime Between Fast and Slow Bubbles
U
mf
e
mf
= U
B
=
gD
B
_
2
(39)
Regime of Rapidly Growing Bubbles
In this regime, the bubble growth rate is of the same
magnitude as the bubble rising velocity, and for
shallow beds,
d(D
B
)
dt
=
d(D
B
)
dh
dh
dt
= U
B
(40)
or
d(D
B
)
dh
= 1 (41)
3.4 Transition to Fast Fluidization
Continuing increases in operating velocity beyond that
required at turbulent uidization, a critical velocity,
commonly called the transport velocity U
tr
, will be
reached where a signicant particle entrainment
occurs. Beyond this point, continuing operation of
the bed will not be possible without recycle of the
entrained solids. The bed is now said to be in the fast
uidization regime. The transition velocity has been
correlated by Bi et al. (1995) as
U
tr
= 1:53Ar
0:5
for 2 < Ar < 4 10
0:5
(42)
Figure 5 Regimes of uidization for transporting and non-
transporting systems.
For Group A and B particles, the transition velocity,
U
tr
, calculated from Eq. (42) is larger than the terminal
velocity of the individual particles, while for Group D
particles, the transition velocity equals essentially the
terminal velocity of the individual particles.
However, a uidized bed operating at a high gas
velocity alone (>U
tr
) does not make it a fast uidized
bed. A generally accepted denition of a fast uidized
bed is the coexistence of a dilute phase and dense phase
regimes (see review by Rhodes and Wang, 1998).
4 THEORETICAL AND EMPIRICAL
PREDICTIONS OF MINIMUM
FLUIDIZATION VELOCITY
The phenomenon of uidization can best be character-
ized by a P=L versus U plot such as the one shown in
Fig. 6. Below a characteristic gas velocity known as the
minimum uidization velocity, a packed bed of solid
particles remains xed, though a pressure drop across
the bed can be measured. At the minimum uidization
velocity, all the particles are essentially supported by
the gas stream. The pressure drop through the bed is
then equal to the bed weight divided by the cross-
sectional area of the bed, P = W=A. Further
increases in gas velocity will usually not cause further
increases in pressure drop. In actual practice, however,
pressure drop at minimum uidization velocity is actu-
ally less then W=A because a small percentage of the
bed particles is supported by the wall owing to the less
than perfect design of the gas distributor, to the nite
dimension of the containing vessel, and to the possibi-
lity of channeling. At the point of minimum uidiza-
tion, the voidage of the bed corresponds to the loosest
packing of a packed bed. The loosest mode of packing
for uniform spheres is cubic, as discussed in Chapter 2
or e
mf
= (6 p)=6 = 0:476. Substituting into the origi-
nal Carman equation, we obtain
P
L
=
72
cos
2
(g)
_ _
mU(1 e)
2
d
2
p
e
3
(43)
and assuming
72
cos
2
(g)
= 180; g is usually from 48 to 518 (44)
we have
P
L
= 459
mU
mf
d
2
p
(45)
At the point of minimum uidization, the pressure
drop is enough to support the weight of the particles
and can be expressed as
P
L
= (r
p
r
f
)(1 e
mf
) (46)
Combining Eqs. (45) and (46) with the voidage at the
minimum uidization e
mf
= 0:476, we nd
U
mf
= 0:00114
gd
2
p
(r
p
r
f
)
m
(47)
Figure 6 Pressure drop vs. uidization velocity plot for determination of minimum uidization velocity.
Leva (1959) employed the experimental values of
voidage at minimum uidization and arrived at the
equations
U
mf
= [0:0007(Re)
0:063
mf
]
gd
2
p
(r
p
r
f
)
m
(Re)
mf
=
d
p
U
mf
r
f
m
(48)
Equations (47) and (48) are comparable because
(Re)
0:063
mf
is on the order of unity for most uidized
systems where the Reynolds number at minimum ui-
dization, (Re)
mf
, is generally between 10
2
to 10
2
.
Rowe (1961) performed experiments on water ow
through a regular array of spheres and found that the
force on the single sphere in the array was 68.5 times
the force on an isolated sphere at the same supercial
velocity. If we assume that the same factor of 68.5 is
also applicable at the minimum uidization condition
when the drag force of the uid on the particles is just
balanced by the net downward force on the particles,
we have, for the low-Reynolds number case,
F = 3pmUd
p
(49)
and
68:5 3pmU
mf
d
p
=
(r
p
r
f
)pd
3
p
g
6
(50)
or
U
mf
=
0:00081(r
p
r
f
)gd
2
p
m
(51)
Another widely employed approach is by simplica-
tion of the Ergun equation for the packed bed. By
combining the original Ergun equation as shown in
Eq. (52) and the pressure drop equation at minimum
uidization as shown in Eq. (46)
P
L
gfd
p
2r
f
U
2
e
3
(1 e)
= 75
(1 e)
f(Re)
p
0:875 (52)
we obtain the equation for calculating the minimum
uidization velocity, including both the kinetic energy
term and the viscous term, as
Ar = 150
(1 e
mf
)
f
2
e
3
mf
(Re)
mf
1:75
1
fe
3
mf
(Re)
2
mf
(53)
Wen and Yu (1966a, 1966b) suggested a simplied
form of the Ergun equation by assuming the followings
based on experimental data:
1
fe
3
mf
14 and
(1 e
mf
)
f
2
e
3
mf
11 (54)
In a simplied form, the Ergun equation, in terms of
Reynolds number and Archimedes number, can be
reduced to
(Re)
mf
=
C
2
1
C
2
Ar
_
C
1
(55)
There are many sets of values suggested for the con-
stants C
1
and C
2
based on dierent databases used for
correlation. A simplied set is shown in the table.
Reference C
1
C
2
Wen and Yu (1966a) 33.7 0.0408
Richardson (1971) 25.7 0.0365
Saxena and Vogel (1977) 25.3 0.0571
Babu et al. (1978) 25.25 0.0651
Grace (1982) 27.2 0.0408
Chitester et al. (1984) 28.7 0.0494
According to Grace (1982), a change of C
1
from the
33.7 suggested by Wen and Yu to 27.2 improves the t
for ne particles. For the limiting cases for small par-
ticles where the kinetic energy term is not important,
and for large particles where it is dominant, the sim-
plied equations become
For Ar<10
3
,
(Re)
mf
= 7:5 10
4
Ar or
U
mf
= 0:00075
(r
p
r
f
)gd
2
p
m
(56)
For Ar>10
7
,
(Re)
mf
= 0:202
Ar
_
or
U
mf
= 0:202
(r
p
r
f
)gd
p
r
f
_
(57)
It is worthwhile to note that at both small and large
Archimedes numbers, the minimum uidization velo-
city, U
mf
, is directly proportional to the corresponding
terminal velocity, U
t
, of a single spherical particle of
diameter d
p
in an innite medium. For small spheres
where the Stokes law applies, U
t
=U
mf
approaches 74.
For large spheres in the Newtons law regime, U
t
=U
mf
approaches 8.6.
Since in practical applications the particles are
rarely in narrow distribution, the average particle size
used in all equations dicussed so far for minimum
uidization velocity is recommended to be the
surface-volume mean:
d
sv
=
1
i
x
i
d
pi
(58)
where x
i
is the weight fraction of particle size d
pi
. All
correlations available for calculating the minimum
uidization velocity have been reviewed by Babu et
al. (1978). A method of estimating the minimum
uidization velocity at elevated temperatures and pres-
sures was proposed by Yang (1998a). The eect of
temperature and pressure on minimum uidization
velocity is discussed in more detail in Chapter 5,
Eect of Temperature and Pressure.
5 THE RATIO OF TERMINAL VELOCITY TO
MINIMUM FLUIDIZING VELOCITY
For spherical particles, Bourgeois and Grenier (1968)
proposed semitheoretical correlations relating the ratio
of terminal velocity to minimum uidizing velocity
with the Archimedes number, Ar, as shown in Eqs.
(59) through (64). Depending on whether the uidized
bed is uidized by air or water, the velocity ratios are
slightly dierent. The terminal velocity of a solid
particle has been discussed in Sec. 4 of Chapter 1,
Particle Characterization and Dynamics.
For uidization by air and 10
2
< Ar < 4 10
4
,
R =
U
t
U
mf
= 135:7 45:0(log Ar) 4:1(log Ar)
2
s = 0:69 for 20 < R < 60 (59)
For uidization by air and 4 10
4
< Ar < 8 10
6
,
R =
U
t
U
mf
= 26:6 2:3(log Ar)
s = 0:52 for 10 < R < 20 (60)
For uidization by air and Ar > 8 10
6
,
R =
U
t
U
mf
= 10:8 s = 0:5 (61)
For uidization by water and 50 < Ar < 2 10
4
,
R =
U
t
U
mf
= 132:8 47:1(log Ar) 4:6(log Ar)
2
s = 2:9 for 20 < R < 60 (62)
For uidization by water and 2 10
4
< Ar < 1 10
6
,
R =
U
t
U
mf
= 26:0 2:7(log Ar)
s = 0:67 for 9 < R < 20 (63)
For uidization by water and Ar> 1 10
6
,
R =
U
t
U
mf
= 90 s = 0:3 (64)
The above equations are valid for spherical isometric
particles with negligible wall eect. If the terminal
velocity of a particle is available, the above equations
can be used to estimate the minimum uidization velo-
city as well.
6 TWO-PHASE THEORY OF FLUIDIZATION
The two-phase theory of uidization was rst pro-
posed by Toomey and Johnstone (1952). The model
assumed that the aggregative uidization consists of
two phases, i.e., the particulate (or emulsion) phase
and the bubble phase. The ow rate through the emul-
sion phase is equal to the ow rate for minimum uid-
ization, and the voidage is essentially constant at e
mf
.
Any ow in excess of that required for minimum uid-
ization appears as bubbles in the separate bubble
phase. Mathematically, the two-phase theory can be
expressed as
G
B
A
= U U
mf
(65)
where G
B
is the average visible volumetric bubble ow
across a given cross section of the bed. For some sys-
tems, the two-phase theory is good and simple approx-
imation, but there is also much experimental evidence
that correction may be necessary, especially at high
pressures. Experimental evidence indicated that the
equation should be
G
B
A
= U K
4
U
mf
with K
4
> 1:0 (66)
The reason for this deviation was suggested by Grace
and Clift (1974) to be due primarily to two factors: an
increase in interstitial gas velocity in the emulsion
phase above that required for minimum uidization
and possibly through ow inside the bubbles. This
led to their modied or n-type two-phase theory.
G
B
A
= U U
mf
(1 ne
B
)
= U [U
mf
(1 e
B
) (n 1)U
mf
e
B
] (67)
where e
B
is the average volume fraction of visible
bubble phase in the bed. The ow through particulate
phase is
Flow through particulate phase = U
mf
(1 e
B
)A
(68)
and the through ow relative to bubbles is
Through flow relative to bubbles = (n 1)U
mf
e
B
A
(69)
For the classical two-phase theory, n = 0. The model
implicitly assumes that the supercial velocity in the
emulsion phase remains at U
mf
, while the velocity
inside the rising voids in (n 1)U
mf
relative to the
boundary of the bubbles. It was hoped by the authors
that n would turn out to be a universal positive con-
stant and thus the magnitude of n would become a
measure of the overall through ow in freely bubbling
beds. Experimental evidence available so far indicates
that n is not universal. It is dierent from system to
system and may be even dierent from observer to
observer. Within a given system, n is also dependent
on height and supercial gas velocity. The diculty
arises from the inability to measure the invisible ow
components of the total ow, since the invisible ow
component is substantial and its distribution remains
unknown. The possibility of substantial through ow
through the uneven distribution of bubbles was ana-
lyzed by Valenzuela and Glicksman (1981) theoreti-
cally.
7 VISCOSITY OF A FLUIDIZED BED
A number of investigators have measured the apparent
viscosity of uidized bed using methods available for
ordinary liquids. For example, Matheson et al. (1949)
and Furukawa and Ohmae (1958) employed a rotating
paddle viscometer and found that a fully uidized bed
had a viscosity from 0.5 to 20 poises. The viscosity
increases with the size of the particles in the bed.
Other researchers, Kramers (1951), Dickman and
Forsythe (1953), and Schugerl et al. (1961), used visc-
ometers of slightly dierent designs and obtained prac-
tically similar results. For small shear stress, the fully
uidized beds behave as newtonian uids. The viscos-
ity of the uidized bed is very high at close to minimum
uidization condition and decreases sharply with
increases in gas ow.
As early as 1949, Matheson et al. reported that the
addition of relatively small amounts of nes to a coarse
bed will decrease the viscosity of the bed substantially.
It remained for Trawinski (1953) to nd an analytical
explanation for this phenomenon. He assumed that the
viscosity in a uidized bed is primarily due to the rub-
bing of coarse particles among one another. The addi-
tion of nes will reduce the friction between coarse
particles by coating a thin layer of nes on each coarse
particle and thus decrease the viscosity of a uidized
bed. The minimum eective amount of nes required
to reduce the viscosity is thus that quantity necessary
to coat each coarse particle with just a single layer of
ne particles, as shown in Fig. 7.
In an actual uidized bed, the bed is usually
expanded during operation, and the voids occupied
by the gas bubbles must be subtracted from the volume
to give the weight fraction of the lm, which repre-
sents more closely the true minimum weight required.
Assuming similar density for d
s
and d
B
, the weight
fraction of nes can be calculated as
(W)
d
s
=
(1 e) 3
d
S
d
B
_ _
d
s
d
B
_ _
2
_ _
(H H
0
)
H
0
_ _
1 (1 e) 3
d
S
d
B
_ _
d
S
d
B
_ _
2
_ _
(H H
0
H
0
_ _
(70)
where H
0
and H represent the initial and the expanded
bed height. Assuming the particles to be perfect
spheres, the optimum two-component system with
the maximum surface area has been computed to be
the one in which the diameter of the nes is 22.5% that
of the larger particles, and in which the smaller parti-
cles constitute 25% of the mixture by weight. A study
by Geldart (1972) employing the surfacevolume par-
ticle diameter seemed to conrm this as well. Some of
the data by Matheson et al. (1949) and replotted by
Geldart are presented in Fig. 8.
The importance of nes content in the uidized bed
reactors was also investigated by Grace and Sun (1990)
from a dierent aspect. They found that nes spent a
much longer time in the bubble phase and considerably
enhanced the concentration inside the bubbles due to
the through ow of uidizing gas through the bubbles.
This enhanced particlesgas contact and improved the
reactor performance.
Figure 7 Schematic of coating of ne particles on coarse
particlesTrawinskis model of reduction in uidized bed
viscosity upon addition of nes.
Grace (1970) reviewed the literature available in this
area in 1970 and concluded that the direct
measurement techniques commonly used to determine
the viscosity of newtonian liquids tended to alter the
behavior of the bed, e.g., the regions underneath and
above the probe tended to have high and low voidages.
Thus the viscosity obtained by those methods may not
be the true viscosity of a uidized bed. He, in turn,
proposed an indirect method based on the behavior
of bubbles in uidized beds.
The spherical-cap bubbles observed in uidized beds
can best be characterized by the included angle as
shown in Fig. 9. The angle was found to be dependent
only on the bubble Reynolds number, as shown in Eq.
(71)
(Re)
e
=
D
e
U
B
r
f
m
= 23e
0:004y
2008 < y < 2608
(71)
where D
e
is the equivalent bubble diameter and can be
calculated as
D
e
=
6V
B
p
_ _
1=3
(72)
The viscosity of the uidized bed can then be estimated
by the equation
m =
D
e
U
B
r
p
(1 e
mf
)
(Re)
e
(73)
The estimated range of the viscosity of uidized beds is
from 4 to 13 poises by using this method. These values
are remarkably similar to that obtained by others using
a rotating cylinder viscometer. For more discussion on
the viscosity of uidized beds, see that by Schugerl
(1971).
The dense phase viscosity of uidized beds at ele-
vated pressure was studied by King et al. (1981) at
pressures up to 20 bar by measuring the velocity of a
falling sphere. They found that for particles less than
about 100 mm, increases in pressure caused substantial
decreases in viscosity. For larger particles, however,
the viscosity is almost independent of pressure.
Figure 8 Mathesons data of uidized bed viscosity. (Replotted by Geldart, 1972.)
Figure 9 Schematic of spherical-cap bubbles for determina-
tion of uidized bed viscosity.
8 BED EXPANSION
8.1 Expansion of an Aggregative Fluidized Bed
The bed expansion of an aggregative uidized bed can
be calculated using the equation
L
L
mf
=
(1 e
mf
)
(1 e)
(74)
Since the bed surface is oscillating because of constant
bubble breaking through the bed surface, the bed
height and the voidage are the time-averaged values.
8.1.1 Bed Expansion with a Constant Bubble Size
The bed height oscillation for a bed with a constant
bubble size can be calculated from the absolute bubble
velocity expressed as in Eq. (75), and the time it
requires for the bubble to travel through the bed
shown in Eq. (76) (Xavier et al., 1978).
U
A
= (U U
mf
) U
B
(75)
T =
H
max
U
A
(76)
The bed expands from its minimum height, H
mf
, to its
maximum height, H
max
, with a total bubble ow of
(U U
mf
)A, assuming that the two-phase theory
holds, or
(H
max
H
mf
)A = (U U
mf
)AT (77)
The bed expansion equation is obtained by combining
Eq. (75) through (77) to be
(H
max
H
mf
)
H
mf
=
(U U
mf
)
U
B
(78)
If the bubble velocity, U
B
, is known, the maximum bed
expansion can be calculated from Eq. (78).
8.1.2 Bed Expansion with a Constant Bubble Size
and an Array of Rods in the Bed
From volumetric balance, it can be written that
H
max
H
mf
= (U U
mf
)T
H
max
V
Tmax
H
mf
V
Tmf
(79)
where V
Tmax
= the fraction of bed occupied by rods
when bed height is at the maximum, H
max
, and V
Tmf
=
the fraction of bed occupied by rods when bed height is
at the minimum, H
mf
.
Eliminating the U
A
and T among Eqs. (75), (76),
and (79), we nd
H
max
(1 V
Tmax
) H
mf
(1 V
Tmf
)
H
mf
(1 V
Tmf
) H
max
V
Tmax
=
U U
mf
U
B
(80)
8.1.3 Bed Expansion for the Case with Increases in
Bubble Size Due to Coalescence
Xavier et al. (1978) employed the bubble coalescence
correlation by Darton et al. (1977) and suggested to
calculate the bed expansion from the following equa-
tion. The bubble coalescence will be discussed in more
detail in a later section.
H = H
mf

5b
3
_ _
(H B)
0:6
B
0:6
_ _
5b
2
(H B)
0:2
B
0:2
_ _
(81)
5b
2:5
tan
1
(H B)
0:2
b
0:5
_ _
tan
1
B
0:2
b
0:5
_ _ _ _
where
b = 1:917
(U U
mf
)
0:8
g
0:4
and B = 4
A
o
_
(82)
The A
o
is the so-called catchment area for a bubble
stream at the distributor plate. It is usually the area
of distributor plate per orice.
Geldart (1975) also developed an equation for bed
expansion in freely bubbling beds with sandlike
powders as follows:
H
H
mf
1 =
2
d
2
g
_
Y(U U
mf
)
(c dH)
1=2
c
1=2
H
mf
_ _
(83)
where d = 0:027(U U
mf
)
0:94
and c = 0:915 (U
U
mf
)
0:4
for porous distributor plates. Y is a correction
for deviation from the two-phase theory. Quantitative
values of Y are shown in Fig. 26.
For a distributor with N holes per cm
2
, c can be
calculated by
c =
1:43
g
0:2
U U
mf
N
_ _
0:4
(84)
All units in Eqs. (83) and (84) are in cm and second.
9 BUBBLE PHASE IN FLUIDIZED BEDS
The bubbles in a uidized bed have two basic proper-
ties. In general, they rise at a nite velocity, and they
usually grow in size owing primarily to static pressure
or to coalescence. It has been found that there is a
striking similarity between the behavior of large gas
bubbles in liquids and those in a uidized bed
(Davidson et al., 1977).
9.1 Bubbles in Liquids
In a liquid of small viscosity, the rate of rise of large
bubbles depends primarily on inertial forces and
surface tension. The viscous eect is negligible in com-
parison. The shape of the bubble will adjust itself to
maintain the pressure inside the bubble constant. An
approximate solution by Dumitrescu (1943) for a long
bubble in a tube gives
U
B
= 0:35
gD
_
(85)
Equation (85) was veried experimentally by Nicklin
et al. (1962) for application to a nite bubble or to a
slug rising in a tube. Davies and Taylor (1950) also
provided a solution with a slightly dierent empirical
constant.
U
B
= 0:711
gD
e
_
(86)
where D
e
is the diameter of the sphere having the same
volume as the bubble and can be calculated from Eq.
(72).
Uno and Kintner (1956) measured the rising
velocity of gas bubbles in liquid contained in tubes of
various diameters and found that the wall eect pre-
dominates when D
e
is more than 1/3 of the bed
diameter. The wall eect becomes negligible only
when D
e
is less than 0.1 of the bed diameter. The
regime where the wall eect is dominant is generally
called the slugging regime.
In the case of a stream of bubbles in a vertical tube
generated continously by blowing air in at the bottom,
the absolute upward rising velocity of each bubble is
greater than the velocity of a similar size single bubble
rising in a stagnant liquid. By making a simple material
balance, it is possible to derive the absolute bubble
velocity as
U
A
=
G
A
0:35
gD
_
(87)
Nicklin et al. (1962) found experimentally that the
absolute bubble velocity is
U
A
= 1:2
G
A
0:35
gD
_
(88)
The factor 1.2 stems from the fact that the peak velo-
city at the middle of the tube is about 1.2 times the
average velocity, owing to the nonuniform velocity
prole. The bubbles evidently rise relative to the fastest
moving liquid in the middle of the tube.
For a swarm of bubbles, the same thing applies:
U
A
= U U
B
(89)
From the continuity of gas ow it can be derived that
U
B
U
=
H
0
(H H
0
)
(90)
where H is the liquid height when gas velocity is U,
and H
0
is the liquid height when gas velocity is zero.
Equation (90) provides a relationship for estimating
U
B
from experimentally observable quantities.
Equations (89) and (90) can be applied to a bubbling
gassolid uidized bed by replacing U with U U
mf
,
assuming that the two-phase theory applies.
9.2 Bubbles in GasSolid Fluidized Beds
Bubbles in gassolid uidized beds usually are sphe-
rical-capped as shown in Fig. 9 with the included angle
equal to 2408 found by experiments as compared to
1208 derived theoretically. The bubbles in airwater
systems have an angle of 1008.
The rising velocity of a single bubble in a quiescent
bed has been found experimentally to be
U
B
= 0:71g
1=2
V
1=6
B
(91)
This compares with the experimental value of Davis
and Taylor (1950) for bubbles in liquids shown in
Eq. (86).
Equation (91) implies that the bubble rising velocity
is independent of the type of bed materials used in the
uidized beds. This has been experimentally proven by
Yasui and Johanson (1958).
A spherical-cap bubble with an included angle, y, as
shown in Fig. 9, has a volume of
V
B
= pR
3
B
2
3
cos(y)
1
3
cos
3
(y)
_ _
(92)
If Eq. (91) is written in a general form, it can be
expressed as
U
B
= Cg
1=2
V
1=6
B
(93)
Combining Eqs. (92) and (93), we have
C =
2
3
p
1=6
2
3
cos(y)
1
3
cos
3
(y)
_ _
1=6
(94)
For a gassolid uidized bed where the bubble velocity
can be expressed as in Eq. (91), i.e., C = 0:71, the
included bubble angle is 1208.
When the bubbles are large enough to exceed
approximately 25% of the column diameter, the bub-
ble velocity is aected by the presence of the vessel
wall. The bubble velocity can then be approximated
by Eq. (85) for a long bubble in a tube containing
liquid. The relative regime of application of Eqs. (85)
and (91) for gassolid bubbling uidized beds is shown
in Fig. 10.
9.3 Davidsons Isolated Bubble Model
In 1961, Davidson developed a simple theory that was
capable of explaining a lot of phenomena relating to
bubbles in uidized beds observed experimentally. His
development involves the following assumptions.
1. The particulate phase is an incompressible uid
with a bulk density similar to that of a uidized bed at
minimum uidization. The continuity equation of the
particles can thus be expressed as
@V
x
@x

@V
y
@y
= 0 (95)
2. The relative velocity between the particles and
the uidizing medium is assumed to be proportional to
the pressure gradient within the uid, and thus
DArcys law is applicable.
U
x
= V
x
K
@P
@x
and U
y
= V
y
K
@P
@y
(96)
Since DArcys law (see Chapter 2) is good only for low
Reynolds numbers, the same limitation also applies
here.
3. Fluidizing uid is assumed to be incompressi-
ble, and thus the continuity equation can be written as
@U
x
@x

@U
y
@y
= 0 (97)
with the assumption that the voidage is constant at e
mf
.
Eliminating the velocities from Eqs. (95) through
(97), we have
@
2
P
@x
2

@
2
P
@y
2
= 0 (98)
Equation (98) is exactly similar to that for a xed bed,
and thus the pressure distribution in a uidized bed is
unaected by the motions of the particles in a uidized
bed.
4. The pressure throughout the bubble is constant.
5. The particulate phase behaves as an inviscid
liquid.
6. The bubble has a circular cross section.
Davidson solved these equations in terms of particle
motion, the pressure distribution within the uidizing
uid, the absolute velocities of the uidizing uid, and
the exchange between the bubble and the particulate
phase. He derived the uid stream function as
c
f
= U
B
U
mf
e
mf
_ _
1
R
r
_ _
3
_ _
r
2
sin
2
y
2
(99)
Figure 10 Relative regions of applications for bubble velocity equations.
R
R
B
_ _
3
=
U
B
2(U
mf
=e
mf
)
U
B
(U
mf
=e
mf
)
(100)
He found that the geometry of the stream function is
crucially aected depending whether the bubble velo-
city U
B
is larger or smaller than interstitial minimum
uidization velocity, U
mf
=e
mf
.
9.3.1 Fast Bubbles RegimeWhen U
B
> U
mf
=e
mf
The uidizing uid in this case moves downward rela-
tive to the bubble motion. The uid ows past the
ctitious sphere of radius R with velocity (U
B
U
mf
=e
mf
) at r = o. Inside the sphere of penetration
of radius R, the uid leaves the roof of the buble and
recirculates back to the base of the bubble as shown in
Fig. 11. The radius of the penetration for a 3-D bed can
be calculated from Eq. (101).
R =
U
B
(U
mf
=e
mf
)
2
U
B
(U
mf
=e
mf
1
_
_
_
_
_
_
1=3
R
B
(101)
In this case, the bubbles in the uidized bed are accom-
panied by a cloud while rising through the bed. For
fast bubbles where the bubble velocity is large, or for
uidized beds of ne powders when the minimum
uidization velocity is small, the cloud is usually very
thin. In most of the uidized beds of pratical interests,
this is the case. The theoretical ndings here has been
experimentally veried.
The stream function for the particles was obtained
to be
c
p
= U
B
1
R
B
r
_ _
3
_ _
r
2
sin
2
y
2
(102)
In this case, the particulate phase streams past a sphere
of radius R
B
as shown in Fig. 11. The relative velocity
of the particulate ow to the void is U
B
at r = o.
9.3.2 Slow Bubbles RegimeWhen U
B
< U
mf
=e
mf
In this case, the uidizing uid moves upward relative
to the bubble motion. This case is usual for beds of
large particles and small bubbles. Since R is less than
zero in this case, the majority of uidizing uid enters
the void at the base and leaves from the roof. The
uidizing uid, in essence, uses the bubble void as
the shortcut. The uidizing uid penetrates the parti-
culate phase freely from the bubble except for a small
fraction of uid in the shaded area, as shown in Fig. 12
in a circle of radius a
/
, expressed in the following equa-
tion.
a
/
=
[R[
2
1=3
= 0:8[R[ (103)
Figure 11 Davidsons bubble modelfast bubbles.
The stream function for the particles in this case can be
similarly expressed as in Eq. (102).
In both these cases, i.e., U
B
> U
mf
=e
mf
and
U
B
< U
mf
=e
mf
, the gas exchange between bubble and
particulate phase was obtained by Davidson through
an analogy to a xed bed:
q = 3pR
2
B
U
mf
for a 3-D bed (104)
The gas exchange rate q expressed in Eq. (104) is the
total volumetric ow rate of gas passing through the
bubble void. It can either pass through the bubble void
or be recirculated back to the bottom of the bubble,
depending on the relative magnitude of the bubble
velocity and the minimum uidizing velocity.
There are other models for the bubble motion in
uidized beds of more complexity, primarily those by
Jackson (1963) and Murray (1965). They are more
exact than the Davidsons model but give essentially
the same results.
9.4 Bubble Formation in a Fluidized Bed
Bubble formation in a uidized bed was found experi-
mentally to be very similar to that in an inviscid liquid.
At a very low gas ow rate, the frequency and size of
the bubbles formed are primarily governed by a bal-
ance between the surface tension of the liquid and the
buoyancy force of the bubble. The inertia of the liquid
moved by the rising bubbles becomes more improtant
than the surface tension at higher gas rates. It is in this
regime that the similarity between the formation of
bubbles in a uidized bed and that in an inviscid liquid
is most applicable, because the surface tension in a
uidized bed is zero. At still higher gas rates, the
momentum of gas issuing from the orice can manifest
as a jet before breakup into bubbles. The existence of a
jet in a uidized bed, however, has become a con-
troversial issue in the literature that will be discussed in
more detail later.
Using the analogy of bubble formation in an in-
viscid liquid, Davidson and Harrison (1963) derived
equtions for both the bubble frequency and the bubble
size (volume), assuming there is no gas leakage from
the bubble to the emulsion phase.
t =
1
g
0:6
6G
p
_ _
0:2
(105)
V
B
= 1:138
G
1:2
g
0:6
(106)
At high gas ow rates where the bubble sizes are inde-
pendent of the bed viscosity, the inviscid liquid theory
can predict the bubble sizes satisfactorily. At low gas
ow rates where the viscosity eects are quite pro-
nounced, the inviscid liquid theory underestimates
the bubble sizes. In this case, the following equation
by Davidson and Schuler (1960) should be used.
V
B
= 1:378
G
1:2
g
0:6
(107)
Equation (107) ts the data by Harrison and Leung
(1961) and the data by Bloore and Botterill (1961)
fairly well. There is, however, increasing evidence
showing that the gas leakage from the bubble to the
emulsion phase can be substantial, especially for large
particles. Nguyen and Leung (1972) performed experi-
ments in a 2-D bed with a uidizing velocity of 1.2
times the minimum uidizing velocity and found that
the bubble volume could be better approximated by
the following equation:
V
B
= 0:53
G
f
n
f
n
= bubble frequency (108)
From the computer enhanced video images of rising
bubbles in uidized beds, Yates et al. (1994) observed
that the bubbles are surrounded by a region of emul-
sion phase in which the solids concentration is lower
than that in the emulsion phase far from the bubbles.
This region of increasing voidage was called the shell
by Yates et al. The volumetric gas in the bubble and in
the shell can be correlated as
Figure 12 Davidsons bubble modelslow bubbles.
V
s
= 16V
0:42
B
(109)
Based on this equation, the volume of void following
coalescence of two bubbles is about 28% larger than
the volumes of the two constituent voids, because of
the incorporation of gas from the shells.
9.5 Coalescence of Bubbles in Fluidized Beds
The bubbles in uidized beds grow in size due prim-
arily to three factors:
1. The eective hydrostatic pressure descreases
toward the top of the uidized bed.
2. Bubbles coalesce in the vertical direction with
the trailing bubble catching up the leading
bubble, and
3. Bubbles coalesce in the horizontal direction
with the neighboring bubbles.
The eect of the hydrostatic pressure is usually small,
and the bubbles grow in size owing largely to coales-
cence. There are a number of bubble coalescence
models in the literature. A few of the more well-
known ones are discussed here.
Geldart (1972) found that the uidization behavior
of Group B powders was independent both of the
mean particle size and of particle size distribution. In
particular, the mean bubble size was found to depend
only on the type of the distributor, the distance above
the distributor plate, and the excess gas velocity above
that required at the minimum uidization condition,
U U
mf
. Mathematically, it can be expressed as
D
B
= D
Bo
KH
n
(U U
mf
)
m
(110)
According to the theory by Davidson and Harrison
(1963), the size of a bubble issuing from a single orice
in a uidized bed at the minimum uidization condi-
tion can be calculated from Eq. (106). The same idea
was extended by Geldart to multiorice distributor
plates by replacing G with (U U
mf
)=N
o
and by
replacing V
B
with (1 f
w
)pD
3
Bo
=6. By assuming the
wake fraction f
w
to be 0.25, we have
D
B
= 1:43
U U
mf
N
o
_ _
0:4
g
0:2
KH
n
(U U
mf
)
m
(111)
Experimentally, it has been found that for the porous
plates, the following equation applies.
D
B
= 0:915(U U
mf
)
0:4
0:027H(U U
mf
)
0:94
(112)
For orice plates, the following equation can be used:
D
B
= 1:43
(U U
mf
)=N
o
[ ]
0:4
g
0:2
0:027H(U U
mf
)
0:94
(113)
The constant 0.027 is a dimensional constant with
units of (cm/s)
0:94
. Equation (113) gives reasonable
agreement with data from uidized beds using indus-
trial types of orice distributor plates. Porous distribu-
tor plates, as expressed in Eq. (112), behave as though
they contained approximately 1 hole per 10 cm
2
of bed
area. The principal eect of adding nes to a uidized
bed of group B powders is the reduction of the mean
particle size. At equal values of excess (U U
mf
), this
results in increased bed expansion and solid circulation
rates but produces no decrease in mean bubble size.
Mori and Wen (1975) assumed that all gas above
the minimum uidizing velocity went to form a single
train of bubbles rising along the center line of the bed
and calculated the diameter of bubble that would exist
as
D
BM
= 0:652[A(U U
mf
)]
2=5
(114)
The bubble diameter at bed height H can then be esti-
mated as
D
BM
D
B
D
BM
D
Bo
= exp 0:3
H
D
_ _
(115)
For perforated plates, the initial bubble diameter D
Bo
is expressed as
D
Bo
= 0:347
A(U U
mf
)
N
o
_ _
2=5
(116)
where A is the area of the bed and N
o
is the total
number of orices. For porous plates, the following
expression should be used to estimate the initial bubble
sizes.
D
Bo
= 0:00376(U U
mf
)
2
(117)
The validity of the above equations has been tested
within the ranges of the following parameters:
30 < D < 130 cm 0:5 < U
mf
< 20 cm=s
0:006 < d
p
< 0:045 cm U U
mf
< 48 cm=s
Rowe (1976) suggested the following equation to esti-
mate the bubble size in a uidized bed:
D
B
=
(U U
mf
)
1=2
(H h
o
)
3=4
g
1=4
(118)
Here h
o
is an emperical constant and is a characteritics
of the distributor plate. The h
o
is eectively zero for a
porous plate buy may be more than a meter for large
tuyeres.
Darton et al. (1977) assumed that the bubbles are
lined up as close together as possible, as shown in Fig.
13. They also dened a so-called catchment area for
each particular bubble track. The bubble frequency
can then be calculated by U
B
=2R
B
with the bubble
velocity U
B
= 0:711(gD
e
)
1=2
. The bubble ow in each
track can be calculated as follows assuming the two-
phase theory.
(U U
mf
) =
pD
3
e
6
_ _
U
B
2R
B
_ _
=
pD
3
e
12R
B
_ _
(0:711
gD
e
_
) (119)
If the bubbles are hemispheres, the volume of each
individual bubble can be calculated by the equation
2R
3
B
3
_ _
=
pD
3
e
6
_ _
(120)
Substituting Eq. (120) into Eq. (119), we have
D
e(o)
= 1:63
(U U
mf
)A
c
g
_
_ _
2=5
; A
c
= catchment area
(121)
The catchment area is dened as the area of distributor
plate per hole. When two bubbles of equal volume
from the nth stage coalesce to form a bubble of the
(n 1)th stage, we have
D
3
e(n1)
= 2D
3
e(n)
(122)
If we further assume that the height of each bubble
coalescence stage is proportional to the diameter of
the catchment area, we have
A
c
=
p
4
D
2
c
and h = lD
c
(123)
where D
c
is the diameter of a circular catchment area
for each bubble stream. We have from Eq. (123) that
h
n
= lD
co
lD
c1
lD
c(n1)
(124)
Figure 13 Daltons bubble coalescence model.
Since D
e(n)
= 2
n=3
D
e(o)
, the height of the nth stage is
h
n
=
0:62g
0:25
l
(U U
mf
)
0:5
D
5=4
e(o)
n1
n=0
2
5n=12
(125)
The total bed height can thus be expressed as
H =
1:85g
0:25
l
(U U
mf
)
0:5
(D
5=4
e(n)
D
5=4
e(o)
) (126)
It was experimentally found by Darton et al. that
l = 1:17.
With the initial bubble diameter shown in Eq. (121),
we have
D
e(n)
=
0:54(U U
mf
)
2=5
(H 4:0

A
c
_
)
4=5
g
1=5
(127)
If the bubbles grow to the size of the vessel diameter,
the bed becomes a slugging bed. From this analysis,
this occurs at the following conditions:
H
D
> 3:5 1
1
N
o
_
_ _
(128)
Zenz (1977) assumed the bubble growth in the uidized
bed resembles the well-known Fibonacci series (Zenz,
1978) and propsoed the following equation for bubble
growth.
D
B
D
Bo
= 0:15
H
D
Bo
_ _
0:85 (129)
Bubble growth corresponding to the Fibonacci series is
depicted in Fig. 14.
Choi et al. (1998) proposed a generalized bubble-
growth model on mean bubble size and frequency for
Geldarts Group A, B, and D particles. The model
made use of empirical correlations for volumetric bub-
ble ux and bubble splitting frequency. The proposed
model correlated well with the extensive data reported
in the literature on mean bubble size and frequency.
They also found that the equilibrium bubble diameter
increased linearly with the ratio of volumetric bubble
ux to the splitting frequency of a bubble.
All the models cited for the growth of bubbles in
uidized beds have assumed some type of ordered pro-
gression in the bubble coalescence mechanism indepen-
dent of solid movement in the beds. Whitehead (1979)
pointed out that the pattern of bubble coalescence and
solid circulation in large industrial uidized beds
depended on both the bed depth and the operating
velocity. Figure 15 shows the patterns observed in
large uidized beds of Group B powders with indus-
trial-type distributor plates (Whitehead and Young,
1967; Whitehead et al., 1977). Unless the solids move-
ment in the bed is also taken into account in develop-
ing the bubble coalescence model, the model may not
accurately predict the actual phenomena occurring
in industrial uidized beds. Bubble growth in large
uidized beds was also studied by Werther (1967a).
9.5.1 Bubble Coalescence from Mutiple Entry
Nozzles
Bubble coalescence from mutiple entry nozzles was
studied by Yates et al. (1995) for a Group A powder
by means of x-rays. The simple correlation expressed
here was found to correlate the data well:
h
c
= 39:6
U
U
mf
_ _
1=3
l
s
(130)
where h
c
is the average height above the orice at
which coalescence was complete and l
s
is the orice
separation distance. The volume of the bubble void
and its associated gas shell following coalescence can
be correlated as
V
s
= 31V
0:42
B
(131)
This is approximately twice as large as the bubble
volumn in a bubbling uidized bed obtained by the
same authors (Yates et al., 1994) as shown in Eq. (109).
Figure 14 Bubble growth by Fibonacci series.
10 SLUGGING BEDS
The slugging phenomenon in uidized beds has been
extensively studied and is described in detail in
Hovmand and Davidson (1971). A slugging bed is
characterized by gas slugs of sizes close to reactor
cross section that rise at regular intervals and divide
the main part of the uidized bed into alternate regions
of dense and lean phase. The passage of these gas slugs
produces large pressure uctuations inside the uidized
bed. The occurrence of slugging is usually accompa-
nied by deterioration in quality of bed mixing and
gassolid contacting. The slugging generally occurs in
reactors of laboratory and pilot plant scale. There are
basically two types of slugging uidized beds. Type A
slugging beds consists of axisymmetric round-nosed
gas slugs, the solids ow past the gas slugs in an annu-
lar region close to the wall, as shown in Fig. 16. This
type of slugging bed usually occurs with bed materials
that uidize easily, such as Group A and B powders.
The type B slugging beds have slugs that are essentially
square-nosed (see Fig. 16). The gas slugs occupy the
complete bed cross section. The only way the solids can
pass through the gas slug is by raining down through
the slugs as solids streamers. For cohesive and particles
of angular shape, this type of slugging bed is most
prevalent. Slugging beds with combination of types A
and B have also been observed, depending on the oper-
Figure 15 Patterns of bubble coalescence and solids circulation in large industrial uidized beds. (Adapted from Whitehead,
1967.)
ating gas velocity and the bed depth. The so-called wall
slugs, which appear as half of a round-nosed slug, can
also be classied as a variation of the type A slug.
10.1 Slugging Criteria
The Stewart and Davidson criterion on slugging
(Stewart and Davidson, 1967) can be simply expressed
as
U U
mf
> 0:2U
s
= 0:2(0:35
gD
_
) for slugging
(132)
The criterion was developed based on a number of
assumptions. The height of the slugs was assumed to
be equivalent to the diameter of the bed. The distance
between each gas slug was assumed to be two times the
diameter of the bed and no coalescence of slugs
occurred. The volume was assumed to be pD
3
=8, and
the slug rises at a velocity of 0.35(gD)
0:5
relative to the
surrounding solids by analogy to the gasliquid sys-
tems. If the two-phase theory applies, we can write
(U U
mf
)A = U
SA
e
S
A (133)
where U
SA
is the absolute slug velocity and e
S
is the
fraction of the bed occupied by the slugs, which can be
expressed as
U
SA
= (U U
mf
) 0:35
gD
_
(134)
e
S
=
pD
3
=8
(pD
2
=4)3D
=
1
6
(135)
Hence at the onset of slugging,
(U U
mf
=
1
6
_ _
(U U
mf
) 0:35
gD
_
_ _
(136)
which reduces to Eq. (132).
In beds of height less than 30 cm, Baeyens and
Geldart (1974) found that it was much more dicult
to induce slugging. They proposed a modied slugging
criterion as
(U U
mf
) > 0:07
gD
_
1:6 10
3
(60D
0:175
H
mf
)
2
(137)
Equation (137) is essentially an empirical correlation,
and c.g.s. units are to be used.
10.2 Slugging Bed Expansion
The bed surface of a slugging bed oscillates alternately
between a maximum and a minimum following the
eruption of each slug. The bed expansion equation
was rst derived by Matsen et al. (1969) following
the two-phase theory:
(H
max
H
mf
)
H
mf
=
(U U
mf
)
0:35

gD
_ (138)
The variation of the bed height of a slugging bed is
shown in Fig. 17. Through same kind of argument, the
total uctuation of bed height in a slugging bed has
been suggested by Kehoe and Davidson (1973a):
H
D
=
(U U
mf
)
U
B
_ _
N
T
(139)
where N
T
is the number of bed diameters between the
rear of a leading slug and the nose of a trailing one.
Combining Eqs. (138) and (139) gives the minimum or
collapsed bed height of a slugging bed:
H
min
H
mf
= 1 1
N
T
D
H
mf
_ _
(U U
mf
)
U
B
_ _
(140)
10.3 Slug Length and Slug Frequency
From the material balance of solids and gas, it can be
calculated that the slug length is
L
S
=
[(U U
mf
)N
T
D
_
]
0:35

g
_ (141)
Figure 16 Type A and Type B slugging beds, and wall slugs.
Consequently, the slug frequency can be expressed as
f
S
=
(U U
mf
)
L
S
=
0:35

g
_
N
T
D
_ (142)
This means that the slug frequency is independent of
the operating velocity. The only unknown in Eq. (142)
is the spacing between the slugs, which has to be
determined experimentally. Values of N
T
from 2 up
to 8 have been observed experimentally (Kehoe and
Davidson, 1973b).
Hovmand and Davidson (1968) derived an equation
for calculating the slug length by assuming an interslug
spacing to be two times the diameter of the bed.
L
S
D
0:495
L
S
D
_ _
0:5
(1 B) 0:061 1:94B = 0
(143)
The equation was later modied by Kehoe and
Davidson (1973b) to read as
L
S
D
0:495
L
S
D
_ _
0:5
(1 B) 0:061
(T
S
0:061) B = 0
(144)
where
B =
(U U
mf
)
U
B
> 0:2 (145)
and T
S
is called the stable slug spacing factor, the value
of N
T
at which a following slug ceases to be attracted
to a leading one.
10.4 Minimum Bed Height Required for Slugging
From the analysis by Baeyens and Geldart (1974),
there are three separate zones in a deep bed operating
with an excess velocity. In Zone 1, there is a freely
bubbling bed. There is a slugging bed in Zone 2, but
the slug grows with continous coalescence. Only in
Zone 3 is the slug coalescence complete and stable
slug spacing is established. They suggest to calculate
the minimum bed height for stable slugging using the
following equation:
H
L
= 60D
0:175
D in cm (146)
The maximum bed height below which the bed will be
freely bubbling can be calculated from
H
fb
=
(D2:51D
0:2
)
0:13D
0:47
D in cm (147)
Figure 17 Slugging bed height.
By using the bubble coalescence theory of Darton et al.
(1977), it can be derived that the limiting minimum bed
height required for slugging can be expressed as in Eq.
(128).
10.5 Wall Slugs
Wall slugs usually form with large particles. The wall
slugs move faster than the round-nosed slugs and its
motion can be calculated as
U
S
= 0:35
2gD
_
(148)
The wall slugs think they are in a bed of diameter 2D.
10.6 Effect of Expanded Section on Slugging Bed
Height
An eective way to suppress the slugging and to reduce
the maximum slugging bed height is to expand the bed
cross-sectional area toward the top of a uidized bed.
Yang and Keairns (1980b) developed a correlation to
predict the eect of an expanded section on the slug-
ging bed height. They divided the expansion of a slug-
ging uidized bed with an expanded section into three
dierent stages, as shown in Fig. 18.
For H
mf
< h, the maximum bed height can be cal-
culated by the following two equations for Case I and
Case II. For Case I, the gas leakage from the gas slug
to the surrounding emulsion phase in the expanded
section is assumed to be negligible. In Case II, the
gas leakage from the gas slug to the surrounding emul-
sion phase in the expanded section is assumed to be
instantaneous, so that the emulsion phase in the
expanded section is always minimally uidized.
For Case I where H
mf
< h,
H
max
H
mf
=
h
H
mf
D
1
D
2
_ _
2
1
D
1
D
2
_ _
2
(U
1
U
mf
)
U
B2
_ _
1
h
H
mf
1
1
(U
1
U
mf
)
U
B1
_ _
_
_
_
_
(149)
For Case II where H
mf
< h,
H
max
H
mf
=
h
H
mf
(U
2
U
mf
)
(U
1
U
mf
)
1
(U
2
U
mf
)
U
B2
_ _
1
h
H
mf
1
1 (U
1
U
mf
)=U
B1
[ ]
_ _
(150)
Similar equations were also developed for the design
where H
mf
> h, as is shown here:
For Case I where H
mf
> h,
H
max
H
mf
= 1
D
1
D
2
_ _
2
h
H
mf
(U
1
U
mf
)
U
B1
1
h
H
mf
_ _
(U
1
U
mf
)
U
B2
_ _
(151)
For Case II where H
mf
> h,
H
max
H
mf
= 1
h
H
mf
(U
2
U
mf
)
U
B1
_
1
h
H
mf
_ _
(U
2
U
mf
)
U
B2
_ (152)
Experimental results from Yang and Keairns (1980b)
indicated that the best results were obtained when the
bubble velocities were assumed to be
U
B1
= U
B2
= 0:35
gD
1
_
(153)
This led to the conclusion that the gas leakage from the
gas slug to the emulsion phase at the expanded section
is very small, and the emulsion phase at the expanded
section is actually less than minimally uidized. By the
time the bed surface expands into the expanded sec-
tion, the bed tends to deuidize and cause bridge form-
ation at the transition section, especially at high bed
heights. On the basis of the experimental evidence, the
expanded section will be eective in reducing slugging.
Care must be exercised, however, in locating the
expanded section at a proper distance from the gas
distributor plate or to provide aeration at the transi-
tion to reduce the bridging tendency at the conical
transition.
11 JETTING PHENOMENA IN FLUIDIZED
BEDS
Jets in uidized beds appear usually in the gas entry
region of the bed, such as a gas distributor plate. The
construction of a gas distributor plate can be variable
and usually diers from application to application.
The gas distributor plate designs for uidized beds
are discussed in Chapter 6, Gas Distributor and
Plenum Design. The gas distribution arrangment
can be a simple perforated or orice plate, a pipe
sparger, or more complicated tuyeres and caps.
When gas velocity through the orices is low, the
gas issued from the orices appears as discrete bub-
bles. However, at a high orice velocity, a standing jet
with periodically truncated bubbles is formed. The
jets from a gas distributor plate are generally single
phase, i.e., gas phase, and small in diameter, usually
less than 15 mm. The study of jetting phenomena at
the distributor plate is of interest because the intense
gassolid mixing, and the heat and mass transfer
induced by the jets, are important for the overall per-
formance of uidized bed reactors. The jetting zone
immediately above a grid is characterized by a time-
averaged, sharp, vertical variation of bed density. A
large uctuation of bed density occurs in this zone,
indicating extensive mixing and contacting of solids
and gas. For chemical reactors where the reaction
rate is fast, much of the conversion may occur in
this jetting region. To achieve uniform gas distribu-
tion, a minimum pressure drop equivalent to at least
30% of the uidized bed pressure drop is required.
This necessitates a small orice diameter in practice,
generally less than 15 mm. Thus most of the existing
jet correlations were developed on the basis of data
generated from small jets.
When solids are fed pneumatically into operating
uidized beds, they generate jets as well. In this appli-
cation, they are gas-solids two-phase jets. The gas and
solids entrainment into the jets and the extent of the jet
region (jet penetration depth) may characterize the per-
formance of the reactor and sometimes dictate the
reactor design. Because of the sizes of solids to be
fed continously into the beds, the feeding nozzles and
thus the jets are usually larger than 15 mm.
There is another category of jets in uidized beds
that tend to be substantially larger in size. For exam-
ple, a large central jet is applied in a jetting uidized
bed, similar to a spouted bed, to induce solids mixing
and circulation, to facilitate heat and mass transfer
from one region of a uidized bed to an adjacent
region, and to promote granulation and agglomera-
tion. This type of jetting uidized beds were used for
solids mixing and drying, coal gasication, and pow-
der granulation. The study of large jets, substantially
larger than 15 mm in diameter, is not as common
(Yang, 1998b). The jetting uidized beds are dis-
cussed in Chapter 20, Other Nonconventional
Fluidized Beds.
An ambiguity in studying the jetting phenomenon is
the lack of consensus on what constitutes a jet (Rowe
et al., 1979; Yang and Keairns, 1979). The gas issuing
from an orice might be in the form of bubbles, a
pulsating jet (a periodic jet), or a permanent amelike
Figure 18 Eect of expansion sections on slugging bed height.
jet, depending on the relative properties of the gas, the
bed material, and the operating conditions.
11.1 Different Jetting Regimes Observed in Fluidized
Beds
Dierent jetting regimes were observed experimentally
by dierent researchers using dierent bed congura-
tions, dierent bed materials, and dierent operating
conditions (see review by Massimilla, 1985).
When the gas velocity is low, the bed material is
dense, and the particle size is small, the gas jet issuing
from the orice or the nozzle tends to be truncated into
bubbles right at the orice, a phenomenon very similar
to that observed when gas is injected into a liquid
medium. The chain of bubbles, at a frequency of 5 to
20 Hz, resembles a bubbling plum and can be called a
bubbling jet although some authors, e.g., Rowe et al.
(1979), prefer to reserve the word jet for the cases
where a permanent amelike void exists. The bubbling
jet has been observed experimentally by Rowe et al.
(1979) using x-rays in a three-dimensional bed and
by Clift et al. (1976).
Markhevka et al. (1971) observed a jet in a ui-
dized bed located close to the wall formed elongated
cavities, which were periodically truncated to
become bubbles at the orice. During the formation
of the elongated voids, solids were observed to
become entrained into the voids. Merry (1975)
later proposed a correlation for jet penetration
based on this observation. Tsukada and Horio
(1990) studied the gas motion and bubble formation
at the distributor of a semicircular uidized bed
with a transparenet front wall. They found that
there were two dierent regions above the grid
zone. One they called the jet stem region, where
stable jets existed, and the other, the bubble form-
ing region. The height of the bubble forming
region is of the same order of magnitude as the
initial bubble size. The fraction of the bubble form-
ing region is larger when ne particles are used as
bed material compared to the coarse particles.
With increasing gas density, such as operating
under high pressure, and when the gas stream has a
high momentum, such as delivering through a pneu-
matic transport line containing solids, the jet issuing
into the uidized bed exhibits itself as an oscillating
jet plume, as shown in Figs. 19a and 19c. There is a
high rate of coalescence immediately above the orice
or jet nozzle, and the truncation of the jet no longer
occurs at the opening of the orice, moreover, there is
always a permanent void present at the minimum jet
penetration. At the maximum jet penetration depth,
the jet plume is constructed through a series of bub-
bles with interconnecting roofs and bottoms. Solids
were observed to penetrate through the bubble roof
into the bottom of the bubble immediately above it.
A bubble is formed at the end of the jet plume after
the momentum of the jet dissipates. The oscillating jet
plume was observed by Knowlton and Hirsan (1980)
in a 1-foot diameter pressurized uidized bed oper-
ated up to 53 atm; by Yang and Keairns (1978) using
hollow epoxy spheres of density 210 kg/m
3
in a 1-foot
diameter uidized bed operated at atmospheric pres-
sure; and by Yang and Keairns (1980a) and
Kececioglu et al. (1984) in a jet delivered via a pneu-
matic transport line containing solids of dierent
loadings.
Permanent amelike jets have been observed pri-
marily in two-dimensional beds, in uidized beds
with spoutable bed materials, and the emulsion phase
around the jet is less than minimally uidized. The
permanent amelike jet was observed by Kozin and
Baskakov (1967), Zenz (1968), and Wen et al. (1977)
in two-dimensional beds, and by Yang and Keairns
(1980a) in a semicircular uidized bed with a large
spoutable bed material.
From the above classication, the word jet is used
loosely to represent a region in the uidized bed cre-
ated mainly by the momentum of the gas stream or
solids-carrying gas stream issuing from an orice.
The important parameters inuencing the jetting
modes are the characteristics of the bed material
(particle size, particle density and perhaps particle
size distribution), the gas density/solid density ratio,
the gas velocity from the orice, the solid loading in
the gas stream, and the design congurations (such as
nozzle or orice; single or multiple orices). A more
objective technique for measurement of the jet penetra-
tion height by pressure signal analysis has recently
being suggested by Vaccaro et al. (1997).
Grace and Lim (1987) proposed a simple criterion
based on stable spouting correlation by Chandnani
and Epstein (1986), as shown in Eq. (154), as a neces-
sary but not sucient condition for permanent ame
jet formation in uidized, spout-uid, and spouted
beds. They also observed that increasing temperature
has a signicant destabilizing inuence on jet forma-
tion, and that a lower ratio of orice and particle dia-
meter is required for permanent jet formation at high
temperatures. Larger auxiliary gas ow can also lead
to unstable jets and absence of permanent ame jets.
For jets other than in the vertical direction, the forma-
tion of permanent ame jets is less likely.
d
or
d
p
_ 25:4 (154)
In 1993, Roach proposed a critical Froude number,
(Fr)
c
, as a demarcation for jetting and bubbling. For
systems above the critical Froude number, bubbles are
formed; below it, jets are present. The critical Froude
number is expressed as follows:
(Fr)
c
=
U
gd
p
_ = 520b
1=4
d
p
=d
or
_ _
r
p
=r
f
_ _
_ _
1=2
(155)
where U is the supericial uidization velocity and b is
the distributor porosity. The criterion thus relates the
bubbling and jetting phenomena to the ow through
mutliple orices on a distributor plate.
11.2 Characteristics of Jets in Fluidized Beds
The jets in uidized beds have the following properties:
the jet penetration depth, the jet expansion angle (or
the jet half angle), gas and solids entrainment, initial
bubble size, and frequency issuing from the jets. They
will be discussed now.
11.2.1 Jet Momentum Dissipation
The jet momentum dissipation can be calculated from
the following equation (Yang, 1998b):
U
j
U
m
= 0:26
x
d
o
or
U
m
U
j
= 3:84
d
o
x
(156)
This compares to the velocity scale proposed for a
homogeneous circular jet in an innite medium as
shown in Eq. (157).
U
j
U
m
= 0:16
x
d
o
or
U
m
U
j
= 6:30
d
o
x
(157)
This means that the maximum jet velocity at the axis,
U
m
, for a jet in a uidized bed, dissipates faster than
that in a homogeneous medium by a factor of
6:3=3:84 = 1:64. Equation (156) applies only in the
region beyond the potential core.
11.2.2 Jet Potential Core
For small particles, like the cracking catalyst, experi-
ments by Behie et al. (1970, 1975, 1976) indicated that
the potential core, where the properties of the jet are
essentially similar to that at the jet nozzle, is within
about half of the nozzle diameter. Donsi et al. (1980)
found that the potential core length depended on the
particle characteristics and nozzle size in a very com-
plex way. Because of the lack of understanding, the
potential core is usually neglected in the analysis of
jets in uidized beds.
11.2.3 Jet Penetration Depth
There are at least a dozen proposed correlations in the
literature for calculating the jet penetration depth for a
vertical jet, they are summarized in Table 3. The jet
penetration correlations were also reviewed by Blake et
al. (1984), Massimilla (1985), and more recently
Kimura et al. (1994, 1995). The discrepancy between
the prediction from dierent correlations sometimes
can be more than an order of magnitude on the depen-
dency of gas velocity or operating pressure. The pri-
mary reason is because of the complex jetting modes
observed experimentally, which create conicting de-
nitions of jet penetration depth by dierent research-
ers. Those by Merry (1975), Knowlton and Hirsan
(1980), and Kececioglu et al. (1984) are presented in
Fig. 19.
The recommended approach to estimate the jet
penetration depth is to make use of the two-phase
Froude number dened as
(Fr)
0:5
=
r
f
(r
p
r
f
)
U
2
j
gd
o
_ _
0:5
(158)
The two-phase Froude number was rst suggested by
Yang and Keairns (1978) to calculate the jet penetra-
tion depth in the equation
L
j
d
o
= 6:5
r
f
(r
p
r
f
)
U
2
j
gd
o
_ _
0:5
= 6:5 (Fr)
0:5
(159)
There was no corresponding theorectical basis pro-
posed at the time. In fact, the dependence of jet pene-
tration on the two-phase Froude number can be
derived theoretically from the buoyancy theory of
Turner (1973) as shown by Yang (1998a). The impor-
tance of the two-phase Froude number was also
recently veried by Vaccaro (1997). Equation (159)
was later modied for high temperature and high pres-
sure applications (Yang, 1981).
L
j
d
o
= 7:65
1
R
cf
r
f
(r
p
r
f
)
U
2
j
gd
o
_ _
0:472
(160)
where
R
cf
=
(U
mf
)
p
(U
mf
)
atm
(161)
Table 3 Summary of Jet Penetration Correlations
Jet penetration correlations References
Vertical single jet
L
j
d
o
= 6:5
r
f
r
p
r
f
_ _
U
2
j
gd
o
_ _ _ _
0:5
; L
j
=
(L
min
L
max
)
2
Yang and Keairns (1978); Yang (1998a)
L
B
d
o
= 26:6
r
f
U
j
r
p
gd
p
_
_ _
0:67
U
U
cf
_ _
0:24
Hirsan et al. (1980)
L
max
d
o
= 19:3
r
f
U
j
r
p
gd
p
_
_ _
0:83
U
U
cf
_ _
0:54
L
max
d
o
= 7:65
(U
cf
)
atm
(U
cf
)
p
r
f
r
p
r
f
_ _
U
2
j
gd
o
_ _ _ _
0:472
for U = U
cf
Yang (1981); Yang (1998a)
L
max
d
o
= C
(U
cf
)
atm
(U
cf
)
p
r
f
r
p
r
f
_ _
U
2
j
gd
o
_ _ _ _
0:835
for U > U
cf
Yang (1981); Yang (1998a)
1:81 _ C _ 4:21 depending on solid properties
Vertical multiple distributor jets
L
j
d
o
1
2
cot any
_ _
= 13
r
f
U
j
r
p
gd
p
_
_ _
; L
j
=
(L
min
L
max
)
2
Shakhova (1968)
0:0144
L
max
d
o
1:3 = 0:5 log(r
f
U
2
j
); r
f
in lb/ft
3
and U
j
in ft/s
Zenz (1968)
L
max
d
o
=
0:919d
p
0:007 0:566d
p
_ _
U
0:35
j
d
0:3
o
; units in cgs system
Basov et al. (1969)
L
max
d
o
= 5:2
r
f
d
o
r
p
d
p
_ _
0:3
1:3
U
2
j
gd
o
_ _
0:2
1
_
_
_
_
Merry (1975)
L
j
d
o
= 814:2
r
p
d
p
r
f
d
o
_ _
0:585
r
f
d
o
U
j
m
_ _
0:654
U
2
j
gd
o
_ _
0:47
Wen et al. (1977)
L
j
d
o
= 15:0
r
f
r
p
r
f
_ _
U
2
j
gd
o
_ _ _ _
0:187
Yang and Keairns (1979)
L
max
d
o
= 1:3
U
2
j
gd
p
_ _
0:38
r
f
d
p
U
j
m
_ _
0:13
r
f
r
p
_ _
0:56
d
o
d
p
_ _
0:25
Wen et al. (1982)
This fundamental relationship has been clouded over
the years by dierent experimental arrangements, dif-
ferent denitions of jet penetration depth, and subjec-
tive observations of uctuating jet penetration depth,
which result in many jet penetration correlations that
are inconsistent and of limited applicability. Equation
(162) summarized the dependence of jet penetration
depth on the design and operating parameters for the
existing correlations. The wide range of dependence is
not very comforting for practical applications in multi-
million dollar projects.
L
j
o r
k
1
f
r
k
2
p
d
k
3
o
d
k
4
p
U
k
5
j
m
k
6
g
k
7
(U
mf
)
atm
(U
mf
)
p
_ _
k
8
(162)
Jet penetration correlations References
Horizontal single and multiple jets
L
j
d
o
= 24 10
4
q
r
f
r
p
_ _
0:15
gd
3
p
r
2
f
m
2
_ _
0:30
U
j
d
o
r
f
m
Kozin and Baskakov (1967)
0:5 _ q _ 1:5; for a single jet q = 0:5; for multiple jets q = 1:5
L
j
d
o
= 7:8
r
f
U
j
r
p
gd
p
_
_ _
Shakhova (1968)
0:044
L
max
d
o
1:48 = 0:5 log(r
f
U
2
j
); r
f
in lb/ft
3
and U
j
in ft/s
Zenz (1968)
L
j
d
o
4:5 = 5:25
r
f
U
2
j
(1 ")r
p
gd
p
_ _
0:4
r
f
r
p
_ _
0:2
d
p
d
o
_ _
0:2
Merry (1971)
Table 3 Continued
Figure 19 Dierent jet penetration depth denitions in the literature.
where
k
1
= 0:30 to 0:83 k
2
= 0:3 to 0:83
k
3
= 0 to 1:25 k
4
= 0 to 0:5
k
5
= 0:4 to 0:944 k
6
= 0 to 0:13
k
7
= 0:2 to 0:472 k
8
= 0 to 0:472
The proper advice for estimating jet penetration at
high temperature and high pressure is that, if in
doubt, carry out your experiments at atmospheric
pressure and room temperature by simulating the
gas/particle density ratio at high temperature and
high pressure by using a low-density bed material.
The available correlations for horizontal jets are
also summarized in Table 3. The correlation proposed
by Zenz (1968, 1971) for horizontal jets was also
claimed to be suitable for vertical downward jets.
11.2.4 Jet Half Angle
Merry (1975) correlated the experimental jet half angle
reported in the literature and suggested the equation
coty = 10:4
r
p
d
p
r
f
d
o
_ _
0:3
(163)
Wu and Whiting (1988) recommended to change the
coecients to t the available data better. Their pro-
posed equation is
coty = 8:79
r
p
d
p
r
f
d
o
_ _
0:236
(164)
Based on the denition by Merry, the jet half angle can
be calculated if the initial bubble size and the jet pene-
tration depth are known.
y = tan
1
(D
B
d
o
)
2L
j
_ _
(165)
It is worthwhile to note that the jet half angle, in the
most rigorous sense, only applies to the permanent
amelike jets. In other jetting modes, the jet half
angle represents only the average behavior of the jet-
ting boundary dened geometrically by Eq. (165). The
existence of the jet half angle may not be generally
accepted, but the concept considerably simplies the
analysis of the jetting region in uidized beds. The
same approach was also adapted for the bubbling jet
created by injecting gas into a liquid medium where a
jet half angle of about 108 was proposed (Anagbo,
1980).
A recent study by Vaccaro (1997) indicated that the
jet half angle could be correlated with the two-phase
Froude number quite well for d
o
=d
p
> 7:5.
Experimental jet half angle in the literature over a
wide range of particle size (0:107 10
3
to
2:9 10
3
m), particle density (1117 to 11300 kg/m
3
),
jet diameter (0:5 10
3
to 17:5 10
3
m), operating
pressure (1, 10, 15, and 20 bar), and operating tem-
perature (20, 650, and 8008C) gave a value ranging
from approximately 3.5 degree to close to 25 degree.
11.2.5 Gas Velocity Proles in the Jet
The velocity proles of the gas in the jet were usually
obtained by pitot tubes. Depending on whether the gas
momentum can be isolated, the results of the pitot tube
measurement can either be translated into the velocity
proles or remain as the momentum distribution of the
gassolid mixture in the jet. Generally, the velocity
proles in the jet were found to be similar and could
be approximated by the typical proles of turbulent
jets in homogeneous media (Abramovich, 1963). The
Schlitchting type of similarity was obtained by
Shakhova and Minaev (1972), De Michele et al.
(1976), and Donadono et al. (1980); the Tollmien
type of similarity was observed by Yang and Keairns
(1980a), Yang et al. (1984a), and Yang (1998b). Both
types of similarity are presented in Fig. 20. When the
jet is not a permanent amelike jet, the similarity repre-
sents the average properties at the particular axial
cross section in question.
11.3 Initial Bubble Size and Frequency
When the momentum of the jets dissipates, bubbles are
formed at the end of the jets. The initial bubble size
was studied by Basov et al. (1969) and Merry (1975).
They suggested the equations
D
B
d
o
= 0:41
U
0:375
j
d
0:25
o
_ _
units in cm and cm/s (166)
and
D
B
d
o
= 0:33
U
0:4
j
d
0:2
o
_ _
(167)
The experimental bubble frequency was found by
Rowe et al. (1979), Sit and Grace (1981), and
Kececioglu et al. (1984) to be around 5 to 10 per sec-
ond, smaller than that predicted theoretically by
Davidson and Harrison (1963). This discrepancy was
believed to be due to gas leakage from the jet before
forming visible bubbles. Large leakage has been
reported by Nguyen and Leung (1972) and Buevich
and Minaev (1976) with the corresponding equations
G = 0:53V
B
f
n
(168)
and
G = 0:37V
B
f
n
(169)
To describe the jet adequately, especially a large jet, the
bubble size generated by the jet needs to be studied. A
substantial amount of gas leaks from the bubble to the
emulsion phase during the buble formation stage, par-
ticularly when the bed is less than minimally uidized.
A model developed on the basis of this mechanism
predicted the experimental bubble diameter well
when the experimental bubble frequency was used as
an input (Yang et al., 1984b). The model is described in
Yang (1999) and in Chapter 20, Other Nonconven-
tional Fluidized Beds.
11.4 Two-Phase Jets and Concentric Jets
The inlet jet velocity for a single-phase (gas) jet is cal-
culated based on the cross-sectional area of the jet
nozzle and the total volumetric ow rate of the jet. If
the jet also carries solids, as in the case of feeding solids
pneumatically into a uidized bed, the inlet jet velocity
should be calculated from the equation
r
f
(U
j
)
2
s
=
M
g
U
g
M
s
U
s
A
t
(170)
Since the voidage in the inlet jet is close to 1, the gas
velocity, U
g
, can be calculated based on the cross-
sectional area of the jet nozzle and the volumetric jet
ow rate. The solid particle velocity, U
s
, can be calcu-
lated assuming the gas/solid slip velocity to be the
terminal velocity of a single particle of the average size.
For a concentric jet with the inner jet carrying solid
particles, the inlet jet velocity can be calculated based
on the equation
r
f
(U
j
)
2
c
=
(M
g
)
i
(U
g
)
i
(M
s
)
i
(U
s
)
i
(M
g
)
o
(U
g
)
o
A
t
(171)
11.5 Gas and Solids Entrainment
The interchanges of gas between the jetting region and
the outside emulsion phase can be studied by integra-
tion of the measured velocity proles in the jet or by
tracer gas analysis. The velocity proles in a jet in a
uidized bed can be taken to be similar and can be
approximated by either Schlitchting or Tollmien simi-
larity. The turbulent jet equations already developed in
the homogeneous medium can then be integrated to
give the entrainment of gas into the jet region in a
uidized bed. It is generally agreed that gas is
entrained from the emulsion region into the jet at
close to the jet nozzle. This exchange reverses direction
at larger distance from the jet nozzle (Donadono and
Massimilla, 1978; Yang and Keairns, 1980a; Filla et
al., 1981; Filla et al., 1983b; Gbordzoe et al., 1988,
Yang, 1998b).
The rate of solids entrainment was studied by
Donadono et al. (1980), Massimilla et al. (1981),
Figure 20 Schlitchting type and Tollmien type similarities for gas velocity proles in jets in uidized beds.
Yang and Keairns (1982a), Filla et al. (1981), and Filla
et al. (1983b) by following solid tracer particles into the
jet with movies. It was also measured by Donadono et
al. (1980) and by Donsi et al. (1980) using an impact
probe located inside the jetting region. The particle
velocity and collision frequencies were obtained.
Fluid and particle entrainment into vertical jets in
uidized beds was also studied by Merry (1976) by
using a two-dimensional bed of lead shot uidized by
water. Although the system employed is liquidsolid,
the appearance of the jet and the motion of particles
are very similar to those observed for gassolid sys-
tems. The author considered that the results were
equally applicable for gassolid systems.
Merry (1976) also postulated a dividing streamline
in the uid. All the uid inside this dividing streamline
is entrained into the jet, while the uid outside
bypasses the jet. All uid or particles intended to be
entrained into the jet should be injected within this
dividing streamline to be eective. A similar dividing
streamline was also observed experimentally by Yang
(1998b).
11.5.1 Model for Solid Entrainment into a
Permanent Flamelike Jet
A mathematical model for solid entrainment into a
permanent amelike jet in a uidized bed was pro-
posed by Yang and Keairns (1982a). The model was
supplemented by particle velocity data obtained by
following movies frame by frame in a motion analyzer.
The particle entrainment velocity into the jet was
found to increase with increases in distance from the
jet nozzle, to increase with increases in jet velocity, and
to decrease with increases in solid loading in the gas
solid two-phase jet. High-speed movies indicated that
the entrained particles tended to bounce back to the jet
boundary more readily under high solid loading con-
ditions. This may explain why the entrainment rate
decreases with increases in solid loading in a two-
phase jet. A ready analogy is the relative diculty in
merging into a rush-hour trac as compared to mer-
ging into a light trac.
The simple model by Yang and Keairns (1982a) for
solid entrainment into a permanent amelike jet shown
in Fig. 21 resulted in the equation
W
j
= 2pr
p
(1 e
z
)
C
1
tany
3
L
3
j

1
2
C
1
d
o
2
C
2
tany
_ _
L
2
j

C
2
d
o
2
L
j
_ _
(172)
The solids entrainment rate into a jet in a uidized bed
can be calculated from Eq. (172), if the empirical con-
stants C
1
and C
2
and the jet half angle y are known.
For the rst approximation, the jet half angle y can be
taken to be 108 as suggested by Anagbo (1980), a value
very close to 7.58 obtained from solid particle tranjec-
tories reported for a semicircular column. The jet half
angle can also be calculated from Eqs. (163) and (164).
The model as formulated in this section cannot be
used to predict a priori the solid entrainment rate into
the jet because of the two empirical constants in Eq.
(172). Lefroy and Davidson (1969) have developed a
theoretical model based on a particle collision mechan-
ism for entrainment of solid particles into a jet. The
resulting equation for particle entrainment velocity is
V
jz

(1 e
j
)
(1 e
z
)
(e
j
e
z
)
(1 e
z
)

p
2
e(1 e)d
p
16r
V
j
(173)
Equation (173) predicts correctly the increase in solid
entrainment into the jet with increases in jet velocity
and the decrease with increases in solid loading in a
two-phase jet.
11.6 Interacting Jets in Fluidized Beds
When there is more than one jet in a uidized bed, jet
interaction is to be expected in most cases. Thus jet
interaction is an important phenomenon to study for
operation and design. Unfortunately, not too much
information is available in the literature. Wu and
Whiting (1988) studied the interaction of two adjacent
Figure 21 A model for solids entrainment into a jet in a
uidized bed.
jets and found that the jets behave like two isolated jets
only at low nozzle velocities. At high nozzle velocities,
the jets interact, and the jet penetration depth becomes
constant. The jet interaction is divided into three dif-
ferent regions, as shown in Fig. 22.
Yang (unpublished data) also has studied the jet
interaction for two and three jets in a two-dimensional
beds. His observation of the jet interaction, shown in
Fig. 23, is very much similar to that reported by Wu
and Whiting. Yang also developed two correlations for
the stagnant emulsion regions between the adjacent jets
based on the pitch and jet diameter ratio. The height of
the stagnant emulsion regions between the jets is inver-
sely proportional to the interaction between the jets.
The higher the jet velocity, the more intense the jet
inaction, which in turn results in lower height of the
stagnant emulsion regions between the jets, as evident
in Fig. 23. The proposed equations are
Y
h
P
j
= 1:1660 0:02876
U
j
U
t
_ _
for two jets (174)
Y
h
P
j
= 0:4445 0:01220
U
j
U
t
_ _
for three jets (175)
For multiple-jet systems such as that above a gas dis-
tributor plate, Wu and Whiting (1988) suggested to
make use of the jet half angle discussed earlier to derive
an equation for jet penetration depth. Their proposed
equation is
L
m
= 8:79
r
p
d
p
r
f
d
o
_ _
0:236
P
j
d
o
2
_ _
(176)
Equation (176) predicts that the jet velocity does not
aect the jet penetration depth in a multiple-jet system
and infers that the jet region above the gas distributor
has a constant height independent of the jet velocity.
Yang and Keairns (1979) correlated the jet penetra-
tion depth data for multiple-grid jets and proposed an
equation to estimate the jet penetration depth employ-
ing the two-phase Froude number. The resulting equa-
tion, as shown below, did show a decreasing, but not
negligible, eect of jet velocity on jet penetration depth
for multiple-jet systems. This obviously is due to the jet
interaction observed experimentally.
L
m
d
o
= 15:0
r
f
r
p
r
f
U
2
J
gd
o
_ _
0:187
(177)
12 PARTICLE MIXING AND SEGREGATION
IN A GAS FLUIDIZED BED
Particle mixing and segregation phenomena are impor-
tant in industrial uidized beds, where particles of wide
size distribution or particles of dierent densitites are
usually handled. Studies have indicated that the uid-
ized bed reactors can be operated in dierent modes
either to promote particle mixing or to enhance the
Figure 22 Schematic of jet interaction of multiple jets in uidized beds. (Adapted from Wu and Whiting, 1988.)
particle segregation. It is not unusual to have one part
of the uidized bed reactor operated in mixing mode
while the other part of the same reactor is in segrega-
tion mode. It is important to point out that the segre-
gation pattern or the particle distribution prole in the
bed is set up by the dynamic equilibrium between the
two competing mechanisms of solids mixing and par-
ticle segregation. Particle segregation can usually be
prevented by operating a uidized bed at a suciently
high uidizing velocity. On the other hand, a bed with
particle size ratio between the largest and the smallest
as small as 3 can be made to segregate by operating at
a small enough uidizing velocity (Wen and Yu,
1966b).
Past studies on particles in a gas uidized bed have
concentrated primarily on the mixing aspect of the
phenomenon, notably those by Rowe and Nienow
(1976) using two separate layers of otsam and jetsam
as a starting mixture. The otsam is the lighter or
smaller components; which tend to remain at the top
layer of the bed upon uidization, while the jetsam is
those heavier or larger components, which tend to stay
close to the bottom part of the uidized bed. These
words were coined originally by Rowe et al. (1972)
and now have become widely accepted terminology.
The works performed by Professor Rowe and his
associates have been reviewed in Rowe and Nienow
(1976). A quantitative analysis was proposed for the
mixing of two segregating powders of dierent densi-
ties (Nienow et al., 1978a). In a separate study by
Burgess et al. (1977), the initial condition of the bed
was found to be important. The well-mixed initial con-
dition (as compared to the unmixed initial condition of
two separate layers of otsam and jetsam) led to less
segregation at all gas ow rates. Unfortunately, only
limited experimental data are available with a well-
mixed starting mixture (Chen and Keairns, 1975,
1978).
There are two primary objectives for investigating
the particle segregation phenomenon in gas uidized
beds. In one respect, the uidized beds are studied to
determine the operating conditions required to pro-
mote bed mixing and eliminate or minimize particle
segregation. A mixing index can generally be dened
in this case to measure the closeness to perfect mixing.
On the contrary, the other objective is to study the
optimum conditions under which clean separation
can be accomplished between dierent materials (or
components) in the bed. For this case, both the degree
and the rate of particle separation are important
aspects of investigation.
It is generally recognized that both particle mixing
and segregation in a gas uidized bed require the pre-
sence of gas bubbles. No mixing or segregation of par-
ticles can occur in a xed bed, because relative motion
between the particles is not possible. The only way the
segregation can happen in xed beds is for one of the
components to be suciently small, usually less than
Figure 23 Schematic of jet interaction of multiple jets in uidized beds. (Adapted from Yang, unpublished data.)
about 1/6 the diameter of the larger ones, to sieve
through the interstices (Rowe et al., 192; see also
Sec. 3 of Chapter 2). Both particle mixing and segrega-
tion occur simultaneously in a gas uidized bed to
maintain a dynamic equilibrium in particle distribu-
tion. Depending on the controlling mechanism under
a specic operating condition, a gas uidized bed can
be operated under a wide range of characteristics, from
a perfectly mixed to a completely segregated uidized
bed.
12.1 Particle Mixing in a Gas Fluidized Bed
The particle mixing in a gas uidized bed is entirely
induced by the passage of gas bubbles. Particles are
picked up by the bubbles in the bubble wakes at
close to the gas distributor plate and are carried up
to the top of the uidized bed. Along the way up, the
bubble wakes exchange their content with the particles
in the rest of the bed in a phenomenon commonly
known as wake shedding. The gas bubbles also draw
up a spout of the surrounding particles to produce an
upward drift of particles. To balance this overall
upward movement of the bed material, there is a
mass circulation downward in the bubble-free region
of the bed. The global circulation induced by the bub-
bles thus ensues, and the rapid mixing of bed material
quickly follows. This mixing mechanism is qualita-
tively depicted in Fig. 24. The existence of bubble
wake (usually expressed as the volumetric ratio of
wake volume and bubble volume, b
w
= V
w
=V
B
), the
phenomenon of drift (expressed as b
d
= V
d
=V
B
), and
wake shedding have been discussed in detail by Rowe
et al. (1965) and reviewed by Rowe (1971). According
to Rowe, the amount of solids induced upward by a
single bubble is equal to about 0.6 times the bubble
volume, of which about 60% is due to drift action
(i.e., b
w
= 0:24 and b
d
= 0:36). These values agree
reasonably well with the ndings by Fane and
Nghiem (1983), who found that the volume of solids
set in motion by a single bubble is about 0.5 to 0.8
times the bubble volume and approximately 75% was
due to drift. Using beds initially segregated into two
pure horizontal layers, Burgess et al. (1977) found that
the particle mixing between these two layers started at
the moment the gas velocity exceeded the minimum
uidization velocity of the otsam, (U
mf
)
F
, for binary
particle systems of equal density. The mixing was
accomplished by gathering jetsam in the bubble
wakes at the interface between the two layers. For
binary systems of dierent densities, however, the mix-
ing did not start until the gas velocity was above the
minimum uidization velocity of the jetsam, (U
mf
)
J
. In
this case, the bubbles must exist in the jetsam layer to
induce mixing between the two layers.
Baeyens and Geldart (1973) also studied the solid
circulation and found that for most Group B powders,
the wake fraction, b
w
, had an average of about 0.35
Figure 24 Solids mixing mechanism in uidized bedswake and drift.
For Group D powders, b
w
is about 0.1. For Group A
solids, it is l.0. Geldart (1986) correlated the wake and
drift fractions with the Archimedes number, as shown
in Fig. 25.
12.1.1 Lateral Mixing of Solids in GasSolid
Fluidized Beds
The lateral mixing of particles in gassolid uidized
beds is induced by bubble movement through the
bed, bubble bursting at the bed surface, and gross par-
ticle circulation inside the bed. Through the random
walk theory, Kunii and Levenspiel (1969) have derived
an equation for the lateral dispersion coecient, D
sr
,
for lateral mixing of solids in gassolid uidized beds:
D
sr
=
3
16
"
B
1 "
B
_ _
U
mf
D
B
"
mf
(178)
where e
B
is the volume fraction of bubble in the
uidized bed.
The lateral particle dispersion coecient was also
studied by Shi and Fan (1984) and Subbarao et al.
(1985). A one-dimensional diusion model was used
by Shi and Fan (1984) to characterize lateral mixing
of solids. Through dimensional analysis and nonlinear
regression analysis of the literature data, they arrived
at an equation for the lateral dispersion coecient for
general application.
D
sr
(U U
mf
)H
mf
= 0:46
(U U
mf
)d
p
r
f
m
_ _
0:21
h
mf
d
p
_ _
0:24
r
p
r
f
r
f
_ _
0:43
(179)
12.1.2 Convective Solids Transport and Mixing
The mixing in gassolid uidized beds is primarily due
to two processes created by the passage of bubbles,
bubble wake and bubble drift, as discussed earlier.
The convective solids transport and mixing can be esti-
mated based on the bubble properties, as suggested by
Geldart (1986). The solid circulation ux in a uidized
bed can be calculated from the equation
J = r
p
(1 "
mf
)(U U
mf
)Y(b
w
0:38b
d
) (180)
where Y is a correction for deviation from the two-
phase theory, and the factor 0.38 for drift ux is
from the experimental evidence that the particles car-
ried up in the drift travel on average at about 38% of
the bubble velocity. The value of wake and drift frac-
tion, b
w
and b
d
, can be obtained from Fig. 25, while
the correction for deviation from the two-phase theory,
Y, can be found from Fig. 26, also due to Geldart
(1986). Values of b
w
, b
d
, and Y for common materials
are summarized in Table 4.
Figure 25 Dependence of wake and drift fraction on particle characteristics. (Adapted from Geldart, 1986.)
The bubble properties have also been measured by
Werther (1976b) in large-diameter uidized beds, 1 m
in diameter; and an approach for calculating the
convective transport of solids was proposed. The semi-
empirical correlation suggested is
J = 0:67(1 "
mf
)r
p
1
f
B
1
_ _
(U U
mf
) (181)
where f
B
is the bubble shape factor and can be calcu-
lated from
f
B
= {1 0:3 exp[8(U U
mf
)]]e
oh
(182)
and
o = 7:2(U U
mf
) exp 4:1(U U
mf
)] [ (183)
The dimensions for h, the height above the distributor,
and (U U
mf
) in Eqs. (182) and (183) are m and m/s.
The suggested approach was said to be good over the
following range of variables.
Solid density 1115 _ r
p
_ 2624 kg/m
3
Mean particle diameter 100 _ d
p
_ 635 mm
Minimum uidization 0:02 _ U
mf
_ 0:24 m/s
velocity
Excess gas velocity U U
mf
_ 0:80 m/s
12.1.3 Bed Turnover Time
In actual practice where particles are continuously
being fed and withdrawn, the bed turnover time should
be as short as possible compared to the particles resi-
dence time in the bed. Geldart (1986) suggested a ratio
of residence time to turnover time of about 5 to 10.
With the known solid circulation rate, the bed turnover
time can be calculated easily.
t
T
=
H
mf
Ar
p
(1 "
mf
)
JA
(184)
Substituting Eq. (180) into Eq. (184), we nd
t
T
=
H
mf
Y(U U
mf
)(b
w
0:38b
d
)
(185)
Figure 26 Correction for deviation from the two-phase theory. (Adapted from Geldart, 1986.)
Table 4 Wake Fraction, Drift Fraction,
and Correction Factor for Two-Phase
Theory for Some Common Materials
Powder
Size
(mm) b
w
b
d
Y
Catalyst 47 0.43 1.00 1.00
Angular sand 252 0.26 0.42 0.50
470 0.20 0.28 0.25
Round sand 106 0.32 0.70 0.82
195 0.30 0.52 0.65
Source: Baeyens and Geldart, 1973.
In reality, particles spend vastly dierent amounts of
time in the bed, depending very much on the locations
of feed and withdrawal nozzles, on ow and mixing
patterns in the bed, and on many other factors. The
particles residence time is an even more important
consideration in uidized bed design.
12.1.4 Solids Residence Time Distribution in a
Fluidized Bed
Particle residence time distribution in a uidized bed is
more close to that of a stirred tank reactor (CSTR) or
ideal backmix reactor than that of a plug ow reactor
(Yagi and Kunii, 1961). In a perfect plug ow reactor,
all particles have the same residence time, which is
equivalent to the mean residence time of particles
and can be calculated by
t
R
=
W
F
o
(186)
For a completely mixed stirred tank reactor, the resi-
dence time distribution can be expressed following
ideal backmix reactor model.
R(t) =
1
t
R
_ _
e
t=t
R
(187)
R(t) dt is the fraction of solids staying in the bed for the
time period between t and t dt. Yagi and Kunii
(1961) found experimentally that Eq. (187) represented
the particle residence time in a uidized bed quite well.
The fraction of solids spending less than time t can
then be calculated from
f = 1 e
t=t
R
(188)
From Eq. (188), it can be seen that a signicant
amount of solids, about 18.2%, spends less than
20% of the average particle residence time in the bed.
The wide residence time distribution for an ideal
backmix reactor is detrimental to high conversion of
solids. However, the residence time distribution can be
considerably narrowed by staging. The residence time
distribution for a system with a N stages of uidized
beds of equal size can be estimated by
R(t) =
1
(N 1)!t
i
t
t
i
_ _
N1
e
t=t
i
; (189)
In some applications, a simple CSTR is not sucient
to describe the residence time distribution of particles.
Habermann et al. (1998) described a metallurgical pro-
cess where the uidized bed showed a stagnant zone
and a two-zone model with a CSTR and a stagnant
dead-zone tted the data much better. For the resi-
dence time distribution of systems with various combi-
nations of series and parallel reactors, the reader can
consult Levenspiel (1979).
12.2 Particle Segregation in a Gas Fluidized Bed
The mechanisms of particle segregation in a gas
uidized bed were rst studied by Rowe et al. in a
classical publication in 1972 using binary systems of
near-spherical particles in both two-dimensional and
three-dimensional cylindrical beds. The terminology
of jetsam and otsam, now widely accepted for the
component that settles to the bottom and the compo-
nent that oats to the top, respectively, was also rst
suggested in this paper. Three distinctly dierent
mechanisms were found to be important in creating
the relative movement of particles in the bed. The lift-
ing of particles in the wake of a rising gas bubble,
which was found to be the primary particle mixing
mechanism discussed earlier, was also identied as
the most important particle segregation mechanism.
This is practically the only way the otsam can be
transported to the upper part of the bed. For the larger
and denser particles known as jetsam, the migration
down to the bottom of the bed involves two dierent
mechanisms. The larger and denser particles usually
descend by falling through the bubbles, while the smal-
ler, denser particles percolate downward interstitially.
The second mechanism, however, was found to be not
an overly important one. It only occurs when the
small, dense particles are suciently small. This inter-
particle percolation is restricted only to the regions
recently disturbed by a passing bubble.
Tanimoto et al. (1980) investigated the movement of
a small aggregate of jetsam induced by passage of a
single gas bubble in a two-dimensional bed using
cinema-photographic technique and found that the
aggregate moved downward intermittently. The aver-
age descending distance of the aggregate was propor-
tional to the bubble diameter and the distance from the
bubble center. Chiba and Kobayashi (1982) proposed
a segregation model based on this observation.
In a bed containing a small proportion of large par-
ticles in considerably smaller ones (with a size ratio of
more than an order of magnitude), the sink and oat
behavior of the large particles depends on the ratio of
absolute density of the large particles relative to the
bulk density of the smaller ones (Nienow et al.,
1978b). In practice, this is usually the case in uidized
bed combustion of coal, where the large coal particles
constitute only a small proportion in a bed of pre-
dominantly smaller particles of ash or limestone. The
particle shape was found to be not an important
variable during separation unless the particles were
platelike.
Nienow and Cheesman (1980) studied the mixing
and segregation of a small proportion of large platelike
particles in a gas uidized bed of regular smaller par-
ticles using x-ray cinemaphotography. They concluded
that the large at particles with sphericity larger than
0.5 behaved essentially like regular particles with
sphericity larger than 0.8. Extremely at particles,
those with sphericity less than 0.5, exhibited as jetsam
over a small range of uidization velocity, even when
they should be otsam by pure particle density consid-
eration. They also discovered that the at particles
mixed better than the spheres owing to their ability
to y through the bed like a frisbee.
12.2.1 Analogy to GasLiquidSolid Phase
Equilibrium
Kondukov and Sosna (1965) and Gelperin et al. (1967)
(see also Gelperin and Einstein, 1971) were among the
rst to construct the phase equilibrium diagrams for
binary and ternary systems in a gas uidized bed. A gas
uidized bed behaves like a liquid in many ways. Its
liquidlike behavior was reviewed by Gelperin and
Einstein (1971) and by Davidson et al. (1977). A uid-
ized system goes through the phase transformations
with changes in uidizing velocity just like those
experienced by a liquid during changes in temperature.
The packed-uidized-dilute phase states in a uidized
system correspond to the solid-liquid-gas phases in a
liquid, with the minimum uidization and the terminal
velocities being the equivalent of the melting and boil-
ing temperatures respectively.
The phase equilibrium diagrams are constructed
with the beginning uidization velocity, U
bf
, the velo-
city at which the ner and lighter particles begin to
uidize, and the total uidization velocity, U
tf
, the
velocity at which all particles including the large and
heavy particles start to uidize. Figure 27 shows the
determination of U
bf
and U
tf
from the uidization
curve of a mixture suggested by those authors.
Typical phase equilibrium diagrams constructed fol-
lowing this method are shown in Fig. 28 (Chen and
Keairns, 1975). With increases in size ratio or density
ratio of the components or decreases in operating pres-
sure, the area between the curves of U
bf
and U
tf
increases as well, signifying the increase in segregation
tendency of the particles in the bed.
Vaid and Sen Gupta (1978) proposed the only two
equations for calculating U
bf
and U
tf
as follows:
Figure 27 Determination of beginning uidization velocity
(U
bf
) and total uidization velocity (U
tf
).
Figure 28 Typical phase equilibrium diagrams. (Adapted
from Chen and Keairns, 1975.)
(Re)
bf
=
DU
bf
r
f
m
= [(18:1)
2
0:0192Ar]
0:5
18:1
(190)
(Re)
tf
=
DU
tf
r
f
m
= [(24:0)
2
0:0546Ar]
0:5
24:0
(191)
In a study on the phase equilibrium diagram of acrylic
and dolomite particles, Yang and Keairns (1982b) con-
cluded that the total uidization velocity U
tf
obtained
from Fig. 27 or calculated from Eq. (191) had no phy-
sical signicance in term of the state of particle mixing
or separation. At U
tf
the bed may be in total uidiza-
tion, but the bed is far from perfectly mixed.
12.2.2 Classication of Flotsam and Jetsam
In a gas uidized bed where the bed particles are of
dierent densities, it will be benecial to know which
component will sink (jetsam) and which will oat (ot-
sam). In most cases, especially the two-component sys-
tems, the classication of otsam and jetsam is
obvious. In some isolated cases, whether the particular
component will behave as a otsam or a jetsam will
have to be determined experimentally. This is espe-
cially true for a bed of multicomponent mixture with
a wide size and density distribution. For a two-compo-
nent binary system, Chiba et al. (1980) suggested the
following general rules:
In a gas uidized bed of a mixture of wide size and
density distribution, the distinction between a otsam
and a jetsam becomes less clear because the individual
components are distributed axially into an equilibrium
distribution governed primarily by hydrodynamics.
One particular component may be a otsam with
respect to some components in the bed, while simulta-
neously it is also a jetsam relative to other components.
Thus the distinction between a jetsam and a otsam is
less important in a multicomponent system with den-
sity dierences. The component distribution at equili-
brium is of primary concern. There are mathematical
models available in the literature which enable the cal-
culation of this equilibrium concentration prole in a
gas uidized bed, to be discussed later.
12.2.3 Minimum Fluidization Velocity of a Binary
Mixture
The minimum uidization velocity of a uidized bed
with a single component bed material, i.e., a bed mate-
rial with particles of relatively narrow particle size dis-
tribution and of similar particle density, is well dened.
For mixtures of particles of dierent sizes or densities,
especially for those highly segregating system, the de-
nition and determination of the minimum uidization
velocity are not as straightforward. Though the mini-
mum uidization velocity of a segregating mixture can
still be dened conventionally following the procedure
suggested for a single component system, the minimum
uidization velocity dened in this way loses its physi-
cal meaning. The particles in the bed are far from com-
pletely supported by the uidizing gas at this velocity,
as was observed experimentally by Knowlton (1977)
and by Chiba et al. (1979).
A study by Chiba et al. (1979) on the minimum
uidization velocity of binary particle mixtures indi-
cated that the uidization curve shown in Fig. 27 was
Case I d
B
=d
S
_ 10
I
a

B
=
S
Jetsam = bigger component
I
b

B
,=
S
Jetsam = heavier component
Case II d
B
d
S
and bed material 100% smaller component
II
a

B
> (
B
)
S
II
b

B
< (
B
)
S
Jetsam = smaller component
Case III d
B
d
S
and bed material 100% bigger component
III
a

B
>
S
III
b

B
<
S
Jetsam = either component may be jetsam
Case IV The minor component is platelike with f < 0:5
IV
a
Platelike particle is denser Jetsam = platelike component
IV
b
Platelike particle is lighter Jetsam = either component may be jetsam
not typical, and U
bf
could not generally be determined
with accuracy, especially for strong segregating sys-
tems. Since the segregating particle mixture will start
to segregate when the gas velocity is higher than the
minimum uidization velocity of the mixture, the des-
cending portion of the PU curve obtained with
decreasing gas velocity is that of a partially segregated
bed. Depending on the rate of particle separation and
the time spent to obtain the complete uidization
curve, the descending portion will assume dierent
paths.
For an ideal system where the particles are of small
size dierence and of equal density, both the ascending
and descending portions of the uidization curve will
coincide, shown as curve (a) in Fig. 29. The conven-
tional procedure will yield a minimum uidization
velocity, shown as (U
mf
)
M
. For a highly segregating
mixture where the particle separation rate is fast, the
uidization curve will follow curve (a) for the ascend-
ing portion but descend along curve (b). This is
because the mixture will already have separated into
two distinct layers of otsam and jetsam when the gas
velocity is higher than the minimum uidization velo-
city of the jetsam. Curve (b) can be constructed a priori
by adding together the uidization curves of pure jet-
sam and pure otsam, as shown in Fig. 29. The con-
ventional procedure of determining the minimum
uidization velocity will give a velocity at (U
mf
)
S
,
dened in Chiba et al. (1979) as the apparent minimum
uidization velocity. The descending portion of the
uidization curve of any real systems lying between
(a) and (b), such as curve (c), depends on the rate of
separation and the time spent on obtaining the uid-
ization curve. It is clear then that a unique U
bf
may not
be generally obtainable from the uidization curve as
shown in Fig. 27. The minimum uidization velocity
determined following the conventional procedure,
(U
mf
)
I
shown in Fig. 29, is not unique. It should be
noted that though the uidized bed will be completely
uidized (i.e., no static region at the bottom) at U
tf
, a
value generally very close to the U
mf
of the jetsam,
(U
mf
)
J
in Fig. 29, the uidized bed is far from comple-
tely mixed. As pointed out by Yang and Keairns
(1982b) that U
tf
does not adequately represent the equi-
librium boundary determined experimentally for segre-
gating systems.
The reported minimum uidization velocities for
binary systems in the literature are probably the velo-
cities of either (U
mf
)
M
or (U
mf
)
J
. For binary systems of
small particle size and density dierence, these two
minimum uidization velocities may be taken to be
similar. Some of the equations proposed for calculat-
ing the minimum uidization velocity of a binary mix-
ture are summarized below.
Goossens et al. (1971) modied the equation by
Wen and Yu (1966b) on the minimum uidization
velocity for single particle size systems by substituting
with the mixture particle density, r
p
, and the mixture
particle size, d
p
, of a binary mixture as follows:
d
p
U
mf
r
f
m
= (33:7)
2
0:0408
d
3
p
r
f
(r
p
r
f
)
m
2
_ _1=2
33:7
(192)
where
1
r
p
=
x
F
r
F
(1 x
F
)
r
J
d
p
=
R
o
R
_ _
(d
p
)
F
(d
p
)
J
(193)
R = (1 x
F
)r
F
(d
p
)
F
x
F
r
J
(d
p
)
J
R
o
= (1 x
F
)r
F
x
F
r
J
(194)
They found that the Wen and Yu equation as modied
could be applied to binary mixtures dierent in both
size and density for the minimum uidization velocity,
(U
mf
)
M
.
Cheung et al. (1974) proposed an equation for bin-
ary systems with particles of similar density and with
particle size ratios of less than 3. The totally empirical
Figure 29 Determination of minimum uidization velocity
of a binary mixture.
equation for the minimum uidization velocity of the
mixture is expressed as
(U
mf
)
M
= (U
mf
)
F
(U
mf
)
J
(U
mf
)
F
_ _
x
2
J
(195)
The equation was later found to be also applicable for
binary systems with particles of dierent densities
(Chiba et al., 1979). For systems with dierent particle
densities, the particle size ratio restriction of 3 still
applies. Its extrapolation to multicomponent systems
poses diculties, however (Rowe and Nienow, 1975).
Chiba et al. (1979) also proposed two equations to
estimate the (U
mf
)
M
for a completely mixed bed and
the (U
mf
)
S
for a totally segregated mixture. By utilizing
the Ergun equation and the constant voidage assump-
tion, they proposed to estimate the (U
mf
)
M
for a com-
pletely mixed binary system as
(U
mf
)
M
= (U
mf
)
F
r
p
r
F
d
p
(d
p
)
F
_ _
2
(196)
The average mixture density and particle size are cal-
culated from
r
p
= f
VF
r
F
(1 f
VF
)r
J
(197)
d
p
= [f
NF
(d
p
)
3
F
(1 f
NF
)(d
p
)
3
J
]
1=3
(198)
where f
VF
is the volume fraction of otsam and f
NF
is
the number fraction of the otsam particles and can be
evaluated as
f
NF
= 1
1
f
VF
1
_ _
(d
p
)
F
(d
P
)
J
_ _
3
_ _
1
(199)
For a completely segregated bed, the following equa-
tion should be employed.
(U
mf
)
M
=
(U
mf
)
F
1 (U
mf
)
F
=(U
mf
)
J
]x
F
(U
mf
)
F
=(U
mf
)
J
_
(200)
To make use of Eq. (196) requires prior knowledge of
(U
mf
)
F
and Eq. (200), (U
mf
)
F
and (U
mf
)
J
.
Uchida et al. (1983) modied the equation by
Cheung et al. (1974), substituting the volume fraction
for the weight fraction in the original equation and
introducing an additional empirical constant, m, and
suggested the following equations.
(U
mf
)
M
= (U
mf
)
F
(U
mf
)
J
(U
mf
)
M
_ _
(1f
F
)
m
(201)
where
m = 0:17
d
J
d
F
_ _
r
F
r
J
_ _ _ _
0:437
(202)
The volume-mean particle diameter rather than the
harmonic mean diameter should be used in Eq. (202).
Using a wide variety of particles of dierent densities
and sizes, these authors found that the minimum uid-
ization velocity depended more strongly on the volu-
metric fraction rather than on the mass fraction of the
particles. The minimum uidization velocity also is
closely related to the mixing state of the mixture,
conrming the observation by others in earlier
investigations.
12.2.4 Minimum Fluidization Velocity of a
Multicomponent Mixture
For multicomponent particles of equal density, Rowe
and Nienow (1975) proposed to calculate the minimum
uidization velocity of the mixture using the equation
(U
mf
)
M
= (U
mf
)
1
e
e
1
_ _
3
1 e
1
1 e
_ _
2n
_ _
1=n
x
1
d
p1
d
p2
x
2
d
p1
d
p3
x
3
. . .
_ _
1(3=n)
(203)
where n can be taken to be 1.053, an empirical value.
Kumar and Sen Gupta (1974) obtained the mini-
mum uidization velocity from the typical logarithmic
plot of bed pressure drop versus uidizing velocity for
4 single, 17 binary, 6 ternary, and 3 quaternary-
component mixtures and found that the following
simple empirical equation correlated the data well.
(Re)
mf
=
d
p
U
mf
r
f
m
= 0:0054(Ar)
0:78
(204)
where
Ar =
d
3
p
r
f
(r
p
r
f
)g
m
2
(205)
r
p
=
n
i=1
r
pi
arithmetic mean (206)
and
d
p
= 1=
n
i=1
x
i
d
pi
harmonic mean (207)
Most of the literature correlations for calculating the
minimum uidization velocity of a multicomponent
mixture are derived based on the assumption that the
bed is completely mixed and homogeneous, corre-
sponding to the (U
mf
)
M
shown in Fig. 29.
12.2.5 Degree of Particle SeparationMixing and
Segregating Indexes
To describe properly the state of particle distribution
in a segregating gas uidized bed, two indexes, the
mixing index and the segregation index, have been
dened. If the approach to the perfect mixedness is
the primary interest, the mixing index is more appro-
priate. The mixing index for a binary system was rst
dened by Rowe et al. (1972) to be
M =
(X
J
)
u
X
J
(208)
where (X
J
)
u
is the fraction of jetsam in the upper part
of the bed and X
J
is the fraction of jetsam at the state
of perfect mixing. Both (X
J
)
u
and X
J
are usually
expressed in weight fraction. By this denition, the
state of perfect mixing has a mixing index of M = 1,
while a state of complete segregation, M = 0. The de-
nition of mixing index was expanded by Naimer et al.
(1982) to include cases where a unique value of (X
J
)
u
cannot be obtained.
To describe the degree of particle separation in the
bed, the segregation index dened by Chiba et al.
(1982) may be more convenient. They dened the seg-
regation index for a binary system, S, as
S =
(X
F
)
u
X
F
(209)
where (X
F
)
u
is the otsam weight fraction in the upper
part of the bed and X
F
is the otsam weight fraction at
the state of perfect mixing. By this denition, S = 1
describes a state of perfect mixing and S = 1=X
F
indicates a state of complete segregation. Since X
F
=
1 X
J
for a binary system, it can be readily derived
that
M =
1 SX
F
1 X
F
= S
(1 S)
X
J
(210)
Because of the complexity of describing even a binary
system, no general and useful mixing index has been
suggested for multicomponent systems. Even the
experimental studies are conspicuously lacking in the
literature.
12.2.6 Effect of Particle Size, Density, Shape, and
Gas Velocity
The segregation patterns for practical binary systems
are shown in Fig. 30 for otsam-rich systems and in
Fig. 31 for jetsam-rich systems (Nienow et al., 1978a).
Increasing operation velocity will drive the equilibrium
toward a better mixed state. Similar segregation pat-
terns can be drawn for binary systems with dierent
density and size ratios. For systems with large density
and size ratios, the segregation patterns are similar to
those shown in Figs. 30a and 30b, and in Figs. 31a and
31b. The segregation patterns shown in Figs. 30c and
31c correspond to systems with smaller size and density
ratios.
The eect of particle density, particle size, and gas
velocity on the mixing of binary, ternary, and quatern-
ary particle systems was studied by Nienow et al.
(1978a, 1987) experimentally. They observed that sig-
nicant improvement of mixing was obtained with per-
forated plate and standpipe distributors, compared to
a porous distributor plate at similar supercial operat-
ing velocity. They also showed that the mixing index
Figure 30 Practical states of equilibrium for otsam-rich binary mixtures.
M could be correlated with the gas velocity by the
following equation:
M =
1
(1 e
Z
)
(211)
where
Z =
U U
To
U (U
mf
)
F
_ _
e
U=U
To
(212)
The takeover velocity U
To
, where dM=d(U (U
mf
)
F
) is
a maximum and M = 0:5, could be estimated from the
following empirical equations:
U
To
(U
mf
)
F
=
(U
mf
)
J
(U
mf
)
F
_ _
1:2
0:9(r
R
1)
1:1
d
0:7
ER
2:2(x
J
)
0:5
(H
+
)
1:4
(213)
The r
R
in Eq. (213) is the density ratio; the d
ER
, the
size ratio; and the H
+
, the reduced bed height. They are
dened as
r
R
=
r
J
r
F
(214)
d
ER
=
f
J
d
J
f
F
d
F
(215)
H
+
= 1 exp
H
D
_ _
(216)
Equation (213) gave reasonable prediction of U
To
for
binary systems of dierent particle densities
(r
J
=r
F
,= 1). For binary systems with particles diering
only in size, i.e., r
J
=r
F
= 1, Eq. (213) gave U
To
values
signicantly larger than those measured experimen-
tally. Equations (211) and (213) are only recommended
for binary mixtures with volumetric jetsam concentra-
tion less than about 50%, and Eq. (213) is also only
good for systems with a particle size ratio less than 3.
Equation (213) cannot be applied to a bed with a high
aspect ratio where slugging occurs.
The equations of Nienow et al. were modied later
by Rice and Brainnovich (1986) and Peeler and Huang
(1989). A more recent study on segregation by size
dierence was conducted by Wu and Baeyens (1998).
They found that the excess gas ow rate required to
prevent segregation in a uidized bed with a wide size
distribution of powder can be calculated from the mix-
ing index expression shown in the equation
M = 1 0:0067d
R
1:33
G
B
A
_ _
0:75
(217)
For d
R
2, good mixing (M _ 0:9) can be achieved
when the visible bubble ow (G
B
=A) is larger than
about 0.094 m/s. Since temperature has limited eect
on the visible bubble ow rate, it is expected that tem-
perature has very little eect on the particle separation
due to dierence in size alone.
Simple quantitative equation can only be formu-
lated for binary systems of near spherical particles.
For binary systems of granular particles, Rowe et al.
(1972) suggested the following dependence for segrega-
tion tendency.
Segregation tendency
o [U (U
mf
)
F
]
r
J
r
F
_ _
2:5
d
B
d
S
_ _
0:2
(218)
Figure 31 Practical states of equilibrium for jetsam-rich binary mixtures.
According to Eq. (218), the eect of density dierence
is considerably more important than that of size dier-
ence. The main eect of the particle size is to alter the
U
mf
of the mixture. The exact form of the velocity
eect is yet to be determined.
12.2.7 Effect of Pressure and Temperature
The eect of pressure on particle separation in a gas
uidized bed was studied by Chen and Keairns (1975)
for systems of particles of both similar density (dolo-
mite particle systems) and dierent densities (char-
dolomite systems) up to a pressure of 690 KPa. They
found that the tendency of particle segregation
decreased with increases in pressure.
The pressure eect on solid mixing and segregation
of binary mixtures was also investigated by Chiba et al.
(1982) at pressures up to 0.8 Mpa. The total bed pres-
sure drop and axial particle distribution for various
systems using silica sand and coal char of dierent
sizes were obtained. The U
mf
prediction by Chenug
et al. (1974), originally developed for binary mixtures
of similar density and dierent particle sizes, was
found to be applicable not only to particles of dierent
sizes and densities at 2 atmospheric pressure but also
to high pressures. The mixing between the binary com-
ponents increased with increasing pressure. Through
photographic measurement of bubble size, shape, and
wake angle in a two-dimensional gas uidized bed,
they concluded that the increasing particle mixing (or
the decreasing particle segregation tendency) at higher
pressures was due primarily to the increases in wake
fraction of the gas bubbles at higher pressures. The
volumetric wake fraction was observed to increase
from 0.02 at 0.1 Mpa to about 0.1 at 0.8 Mpa almost
linearly.
12.2.8 Effect of Particle Segregation on Other
Fluidization Phenomena
The eect of particle segregation on ne elutriation
from gas uidized beds was studied by Tanimoto et
al. (1983) employing binary particle mixtures. They
found that in segregated beds, where substantial
amounts of nes tended to concentrate at the bed
surface, the elutriation characteristics were very much
dierent from those of the well-mixed beds. The
elutriation rate decreased with increases in bed height
rather than the opposite tendency observed in a well-
mixed bed. The higher bed height increased the elutria-
tion rate because of larger bubbles and because of
more particles thrown into the freeboard by those
larger bubbles. More nes tended to concentrate at
the bed surface with a lower bed height because there
was less mixing provided by smaller bubbles existing in
a shallower bed. The net eect was an increase in elu-
triation rate at lower bed heights. The concentration of
nes at the bed surface, apparently, dominated the
elutriation phenomenon in a segregated bed.
12.2.9 Rate of Particle Separation in a Gas
Fluidized Bed
Past studies on particle separation have concentrated
primarily on the mixing aspect of the phenomenon,
notably those by Rowe, Nienow, Chiba, and their
associates using two separate layers of otsam and jet-
sam as a starting mixture. To design a uidized-bed
separator for a particle system, however, not only the
degree of separation but also the rate of separation
must be known. The rate of separation of dierent
particle systems has not been studied systematically.
Limited data were reported by Rowe et al. (1972) for
coarse and ne ballotini particles. Chen and Keairns
(1978) presented a limited amount of data on the rate
of separation in a chardolomite system with an
unmixed starting mixture. The rate of separation of a
wide size distribution of glass beads with the same
particle density was investigated by Yoshida et al.
(1980). A simple but eective apparatus was designed,
and a comprehensive experimental program was car-
ried out to study the rate of separation of dierent
particle systems of dierent size and density ratios,
by Yang and Keairns (1982b). These studies were con-
centrated, however, on batch separation. Continuous
separation studies were also conducted by Yang et al.
(1984b), Nienow and Naimer (1980), Iya and Geldart
(1978), and Hussein et al. (1981), though only on a
limited number of systems for certain specic applica-
tions. Those investigations are reviewed here.
Rate of Particle Separation in Batch Systems
The experiments conducted by Chen and Keairns
(1978) were carried out using two-component mixtures
of dierent sizes and/or dierent densities arranged
with the heavy or large particles on top of the light
or small particles. The particle separation information
was obtained by sampling with a rotating turntable
from two sampling ports located along the side of
the bed at a rate of 1.5 to 3 seconds per sample when
the bed was uidized.
The study conducted by Yoshida et al. (1980) was
also for binary systems but with particles of similar
density. The bed of a selected binary system of glass
beads was rst uidized at a velocity higher than the
minimum uidization velocity of the large compo-
nent to promote the particle mixing. The velocity
was then decreased to a preselected value between
(U
mf
)
F
and (U
mf
)
J
. The bed was slumped after a pre-
determined duration of operation, and the bed was
removed layer by layer for analysis. The same proce-
dure was repeated with longer operation duration until
the steady state concentration prole was reached. It
was found that a clean interface could usually be
obtained between the otsam-rich and the jetsam-rich
beds when the particle size ratio of the binary compo-
nents was larger than 3. For systems with size ratio less
than 3, a transition region with changing composition
existed between the two segregated beds.
A more comprehensive investigation was carried
out by Yang and Keairns (1982b) employing acrylic
particles (r
p
= 1110 kg/m
3
) as the otsam and dolo-
mite particles (r
p
= 2610 kg/m
3
) as the jetsam. Both
types of particles have a relatively wide particle size
distribution. Four mixtures of 20, 40, 60, and 80 weight
percent of dolomite were studied at velocities ranging
from the minimum uidization velocity of the acrylic
particles to slightly higher than that of the dolomite
particles. The experiments were performed in a spe-
cially constructed bed with a main gas line leading to
the bed and a gas bypass line, both controlled by indi-
vidual but electrically interlocked solenoid valves. A
known concentration of acrylicdolomite mixture
was rst passed through a Rie sampler at least four
times to mix the mixture, and the mixture was then
placed in the uidized bed. Air was turned on to a
desirable reading on the rotameter with the solenoid
valve in the bypass line open and that in the main line
closed. The needle valve in the bypass line was then
adjusted to give a pressure drop in the line equivalent
to that expected from the uidized bed. An electrical
switch interlocking both solenoid valves was then
turned on to open the solenoid valve in the main line
and to shut the solenoid valve in the bypass simulta-
neously. After a predetermined time, the switch was
again turned o to reverse the ow. The uidized-
bed content was then vacuumed o layer by layer
and the particle concentrations analyzed by screening.
The procedure was then repeated for a dierent
separation time duration, for a dierent separation
gas velocity, or for a dierent particle mixture.
The transient particle concentration proles for a
20 w/o and 60 w/o dolomite starting mixture are pre-
sented in Figs. 32 and 33. They are shown for 3, 5, 10,
and 20 seconds of operation. When the jetsam concen-
tration is low (e.g., 20 w/o dolomite mixture), both the
top and bottom layers have relatively uniform concen-
tration proles throughout the transient time period at
dierent operating velocities (see Fig. 32). For the
60 w/o dolomite mixture, however, the same is true
only at lower operating velocities (Figs. 33a and
33b). At higher operating velocities, especially those
higher than the minimum uidization velocity of the
dolomite (Fig. 33d), concentration gradients start to
develop in both layers with a fuzzy transition between
them. At equilibrium, the upper uidized-bed layer
and the bottom packed-bed layer usually have uniform
particle concentrations if the operating uidizing velo-
cities are lower than the minimum uidization velocity
of the jetsam. There is also a small transition zone
between the two layers. The jetsam in the bottom
packed bed increases with an increase in operating
velocity, while the upper uidized bed almost consists
of pure otsam.
Visual observation of the experiments in this study
indicated that the particle separation mechanisms were
dierent at high and low uidization velocities with the
minimum uidization velocity of dolomite, (U
mf
)
J
, as
the demarcation. At an operating velocity lower than
(U
mf
)
J
, there was a distinct packed bed at the bottom
in equilibrium with a uidized bed at the top with a
short transition zone in between. Mixing between these
two beds appeared to be minimum. When the operat-
ing velocity was increased to higher than (U
mf
)
J
, the
whole bed appeared to be uidized, although a fuzzy
particle interface was usually discernible. Both particle
mixing and particle separation were apparently occur-
ring along the bed height. At equilibrium the particle
concentration proles in both layers were essentially
uniform. This was approximately true even when the
operating velocity was higher than (U
mf
)
J
. During the
transient time period, this was still true except for the
case with high jetsam concentration and high operating
velocity (Fig. 33d). It appears, then, that considerable
simplication in the mathematical model may be
possible, at least for the highly segregating system of
acrylicdolomite.
A simple mathematical model was developed by
Yang and Keairns (1982b) by assuming that the parti-
cle segregation is a uidized bed could be simulated by
two perfectly mixed uidized beds in series with parti-
cle interchange between them. The short transition
region observed experimentally was ignored in the
model. The particle exchange was accomplished by
bubble wakes from the bottom to the top uidized
bed and by bulk solids ow in the reverse direction.
The resulting equation is
ln
V
1
(1 (C
F
)
W
) V
J1
V
1
(1 (C
F
)
W
) V
o
J1
_ _
=
(U U
mf2
) A f
w
(1 "
w
)
V
1
t
(219)
If we assume further that (C
F
)
w
= 1:0, i.e., the bubble
wake contains pure otsam (acrylic particles), Eq.
(219) can be written as
ln
V
J1
V
o
J1
_ _
= mt (220)
where
m =
(U U
mf2
) A f
w
(1 "
w
)
V
1
(221)
There is certainly some experimental evidence indicat-
ing that the assumption of (C
F
)
W
= 1:0 is reasonable
at least for highly segregating systems (Yang and
Keairns, 1982b).
Equation (220) was used to t the experimental data
with excellent results. The success of the correlation
does suggest possible physical signicance of the para-
Figure 32 Transient particle separation proles for acrylicdolomite systems20 w/o dolomite concentration.
meter, m, dened in Eq. (221). A constant m means a
constant volumetric exchange rate between the two
uid beds during the transient period. A key assump-
tion in Eq. (220) is that (C
F
)
W
= 1:0. This assumption
is reasonable for a highly segregating system like the
acrylicdolomite system. The rate of particle separa-
tion depends critically on (C
F
)
W
as shown in Eq.
(219). Unfortunately, experimental values of (C
F
)
W
for dierent systems are generally not available.
The mixing and segregation kinetics of mixtures of
iron and glass particles, a strongly segregating bed,
were also studied by Beeckmans and Stahl (1987)
employing both initially fully mixed and initially com-
pletely segregated conditions. The data were analyzed
based on a mixed bed of jetsam and otsam residing on
top of a pure jetsam bed. The interchange velocities of
jetsam between the two beds were found to depend
strongly on the excess uidization velocity.
Continuous Operation and Industrial Applications
A good example of continuous particle separation is
the industrial application in the ash-agglomerating ui-
dized bed gasier. The ash remained from gasication
of coal is agglomerated into bigger and denser agglom-
erates and is continuously removed from the uidized
Figure 33 Transient particle separation proles for acrylicdolomite systems60 w/o dolomite concentration.
bed gasier via a uidized bed separator. Cold ow
simulation was conducted in a large scale cold model
3 meters in diameter using crushed acrylic particles
(r
p
= 1100 kg/m
3
) to simulate the char and sand
particles (r
p
= 2650 kg/m
3
) to simulate the ash
agglomerates (Yang et al., 1986).
Continuous mixing experiments of two particulate
species of dierent densities were also performed by
Nienow and Naimer (1980) in a gas uidized bed.
They found that the continuous operation did not
have any eect on the segregation pattern, and that
the behavior of a continuously operated gas uidized
bed could generally be predicted from batch experi-
ments. Similar mixing indices obtained from the
batch experiments could be used under continuous
operating conditions.
One promising application of the continuous parti-
cle separation operation is to use it as a solid-to-solid
heat exchanger. The heat carriers, which are usually
large and dense particles, are rained through a uidized
bed of nes. The application can be for drying as
studied by Baskakov et al. (1975), for supplying the
necessary heat to the endothermic reactions, as in the
Union Carbide agglomerated-ash coal gasication pro-
cess (Corder, 1973), or some other similar applications
like the commercial Nitro-Top process (Drake, 1973).
The dynamics of coarse and dense particles raining
through a gas uidized bed of ne particles were stu-
died by Baskakov et al. (1975), Iya and Geldart (1978),
and Hussein et al. (1981) in uidized beds of dierent
congurations. Baskakov et al. (1975) found that the
minimum uidization velocity ratio between the jetsam
and the otsam had essentially no eect on the quality
of particle separation for (U
mf
)
J
=(U
mf
)
F
> 18. When
the ratio was less than 7, the degree of particle separa-
tion deteriorated noticeably. The bulk density and the
mean size of the nes constituting the uidized bed
were found by Iya and Geldart (1978) to be the most
important parameters aecting the rain-through velo-
city of the coarse particles. The size of the coarse par-
ticles did not appear to be important on rain-through
velocity but it did aect the maximum circulation ux.
Circulation ux of coarse particles up to 2:3 10
5
kg/
hm
2
was observed.
The later study by Hussein et al. (1981) attained
only about 10% of the maximum circulation ux
observed by Iya and Geldart (1978), probably because
of the dierent sizes and densities of the particles
employed. They also observed that the size of the des-
cending balls and the uidizing gas were important,
contrary to that concluded by Iya and Geldart
(1978). Obviously more studies are necessary in this
area of particle separation. Hussein et al. (1981) also
investigated the heat transfer between the descending
balls and the uidized bed and determined the heat
transfer coecients to be in the range of 100 to 200
W/m
2
8K. This lies at the lower end of the heat trans-
fer coecients expected between immersed surfaces
and a gas uidized bed. A patent was granted for
retorting and/or gasifying solid carbonaceous mater-
ials, such as coal, coke, shale, and tar sands, employing
this concept (Mitchell and Sageman, 1979).
Continuous operation was also carried out by
Beeckmans and Minh (1977) using the uidized
counter-current cascade principle with encouraging
results. Separation can also be enhanced by the pre-
sence of an electrostatic eld (Beeckmans et al., 1979)
or baes of various designs (Kan and Sen Gupta,
1978; Naveh and Resnick, 1974).
12.3 Mathematical Models for Prediction of
Equilibrium Concentration Proles
A gas uidized bed is a complex reactor even if it con-
tains only particles of similar size, shape and, density.
The physical phenomena occurring in a gas uidized
bed depend not only on the particle characteristics, the
operating pressure, and the temperature that will
change the properties of the gas uidizing medium
but also on the physical size of the bed. New ndings
and surprises are still continuously being reported.
When the bed material consists of particles of dierent
sizes and densities, the mixing and segregation phe-
nomena are much less understood. The equilibrium
particle concentration prole established inside a gas
uidized bed at steady state is actually a dynamic equi-
librium between the competing processes of mixing
and segregation. Several mathematical models have
been proposed to evaluate both the transient and the
equilibrium particle concentration proles. However,
they are all restricted to systems of binary mixtures
only.
The models proposed so far can be broadly classi-
ed into two categories. The rst group of models is
constructed on the basis of mechanistic phenomena
observed in a bubbling gas uidized bed. Thus they
are usually similar in concept and only dierent in
details. They all divide the bed into the wake (or bub-
ble) phase and the bulk (or emulsion) phase. The
assumptions made for the particle exchange between
the two phases and the excursion of certain phenom-
ena distinguish each individual model. The models in
this group are due to Gibilaro and Rowe (1974),
Burgess et al. (1977), and Yoshida et al. (1980). The
second type of model is that suggested by Gelperin et
al. (1977). He assumes that the particle separation pro-
cess in a gas uidized bed is a diusion process and
lumps all various eects into an eective particle
diusivity and a relative displacement rate. Readers
can consult Yang (1986) or original articles for details.
12.4 Particle Separation Based on the Principle of
Particle Terminal Velocity
All the discussion so far is on the particle separation in
a gas uidized bed based on velocities very close to the
minimum uidization velocity of the jetsam. Particle
separation can also be accomplished by employing
velocities close to that of the terminal velocity of the
particles. The U-Gas coal gasication process devel-
oped by the Institute of Gas Technology utilized the
principle of terminal velocity for separating ash
agglomerates from char in the coal gasier. A compre-
hensive study on this subject using a uidized bed simi-
lar to that of the U-Gas process has been published by
Leppin and Sahay (1980). Some experiments were also
carried out by Chen and Keairns (1978).
NOMENCLATURE
A = cross-sectional area of uidized bed
A
c
= catchment area of a bubble stream
A
o
= area of distributor plate per hole, A
o
=
0 for porous plate
Ar = Archimedes number
A
t
= total cross-section area of a single jet or
of concentric jets
b = jet half-thickness
C; C
1
; C
2
= constants
(C
F
)
w
= volumetric otsam concentration
(fraction) in the wake phase
d
p1
; d
p2
---- = particle diameter of component 1, 2, and
---, respectively
d
B
; d
S
= particle diameter of big and small
particles, respectively
d
F
; d
J
= particle diameter of otsam and jetsam,
respectively
d
o
= diameter of a jet nozzle
d
or
= orice diameter
d
p
= particle diameter of a single component
system
d
p
= average particle size of a mixture
(d
p
)
F
= particle diameter of otsam
(d
p
)
J
= particle diameter of jetsam
d
pi
= particle diameter of ith component in a
multicomponent mixture
d
R
= particle diameter ratio, d
B
=d
S
D = diameter of a uidized bed
D
B
= bubble diameter
D
Bo
= initial bubble diameter
(D
B
)
max
= maximum bubble diameter
D
e
= equivalent bubble diameter
D
sr
= lateral dispersion coecient of solids in
uidized beds
e = coecient of restitution
F = force on a single particle; or total
amount of gas leakage during bubble
formation
f
F
; f
J
= volumetric fraction of otsam and
jetsam in the bed, respectively
F
H
= adhesion force transmitted in a single
contact between two particles
f
n
= bubble frequency
f
NF
= number fraction of otsam
F
o
= solids feed rate
Fr = Froude number
(Fr)
c
= critical Froude number for bubbling and
jetting demarcation
f
S
= slug frequency
f
VF
= volume fraction of otsam
f
w
= wake fraction, the volume of wake
divided by the combined volume of
bubble and wake
G = gas ow rate
G
B
= average visible volumetric bubble ow
G
j
= gas ow rate through the jet nozzle
h = height above the distributor
H = bed height of the uidized bed
H
1
; H
2
= bed height at stage 1 and stage 2,
respectively, during bed collapsing tests
H
3
; H
D
= bed heights at stage 3 during bed
collapsing tests
H
c
= bed height at critical point in bed
collapsing tests
H
e
= bed height of emulsion phase
H
max
= maximum bed height
H
mf
= bed height at minimum uidization
H
min
= minimum bed height
H
o
= initial bed height
H
o
= bed height at t = o
I = the uniformity index
J = solids circulation ux in uidized beds
K; K
1
; K
2
;
K
3
; K
4
= constants
L = bed depth; or bed width
L
j
= jet penetration depth
L
m
= jet penetration depth for multiple-jet
systems
L
mf
= bed depth at minimum uidization
L
S
= slug length
l
s
= orice separation distance or pitch
M = the mixing index
M
g
= mass ow rate of gas
(M
g
)
i
= mass ow rate of gas in the inner jet
(M
g
)
o
= mass ow rate of gas in the outer jet
M
s
= mass ow rate of solids
(M
s
)
i
= mass ow rate of solids in the inner jet
n = a constant
N
o
= total number of orices in a multiorice
distributor plate
N
T
= number of bed diameters between the
rear of a leading slug and the nose of a
trailing one
P = pressure
P
j
= pitch, distance between the centerline of
two jet nozzles
P = pressure drop
q = gas exchange rate between bubble and
emulsion phase; or bulk/wake exchange
rate per unit height
r = radial coordinate; or radial distance
from the jet axis
r
1=2
= radial position where jet velocity is
1
2
of
the maximum at jet axis
R = ratio of terminal velocity and minimum
uidization velocity; or radius of sphere
of penetration (see Fig. 14)
R
B
= radius of bubble
R
cf
= ratio of minimum uidization velocity at
pressure P over that at atmospheric
pressure
(Re)
bf
; (Re)
mf
,
(Re)
tf
= Reynolds numbers based on the
beginning uidization velocity, U
bf
, the
minimum uidization velocity, U
mf
, and
the total uidization velocity, U
tf
,
respectively
(Re)
p
= Reynolds number based on the particle
diameter
(Re)
t
= Reynolds number based on the terminal
velocity
R(t) = particle resident time distribution in
uidized beds
S = the segregation index
t = time
t
b
= time at the end of stage 1 in bed
collapsing tests
t
c
= time at the critical point in bed
collapsing tests
t
i
= mean residence time of particles in ith
stage
t
R
= mean residence time of particles
t
T
= bed turnover time
U = supercial uidization velocity
U
A
= absolute bubble velocity
U
B
= bubble velocity
U
b
= gas velocity at jet boundary
U
bf
= beginning uidization velocity
U
c
= a transition velocity between bubbling
and turbulent uidization regimes
U
cf
= complete uidization velocity
(U
g
)
i
= jet velocity in the inner jet
(U
g
)
o
= jet velocity in the outer jet
U
j
= average jet velocity at the nozzle
(U
J
)
c
= jet velocity of a concentric jet
(U
j
)
s
= jet velocity of a simple two-phase jet
U
k
= a transition velocity between bubbling
and turbulent uidization regimes
U
m
= maximum velocity at the jet axis
U
mb
= minimum bubbling velocity
U
mf
= minimum uidization velocity
(U
mf
)
1
= minimum uidization velocity of
component 1 in a multicomponent
mixture
U
mf2
= minimum uidization velocity of
uidized bed 2
(U
mf
)
atm
= minimum uidization velocity at
atmospheric pressure
(U
mf
)
F
; (U
mf
)
J
= minimum uidization velocities of
otsam and jetsam, respectively
(U
mf
)
I
; (U
mf
)
S
= minimum uidization velocities of
binary systems
(U
mf
)
M
= minimum uidization velocity of a
mixture
(U
mf
)
p
= minimum uidization velocity at
pressure p
U
r
= jet velocity at radial position r
U
S
= slug velocity; solid particle velocity; or
relative displacement rate between
jetsam and otsam particles
U
SA
= absolute slug velocity
(U
s
)
i
= solid particle velocity in the inner jet
U
t
= terminal velocity of a single particle
U
tf
= total uidization velocity
U
To
= the takeover velocity
U
tr
= transport velocity, a transition velocity
from the turbulent regime to the fast
uidization regime
V
1
= volume of solids in uid bed 1
excluding volume of the voids between
particles
V
B
= bubble volume
V
d
= volume of bubble drift
V
j
= mean particle velocity in the jet
V
J1
= volume of jetsam in uid bed 1
excluding the voids between particles
V
0
J1
= volume of jetsam in uid bed 1 at t = 0
V
S
= slug volume; or volume of bubble shell
V
w
= volume of bubble wake
W = bed weight
x = x coordinate; or distance from the jet
nozzle
x
i
= weight fraction of the ith component
x
F
; x
J
= average weight fraction of otsam and
jetsam, respectively
X
F
; X
J
= weight fraction of otsam and jetsam,
respectively
X
F
; X
J
= fraction of otsam and jetsam,
respectively, at the state of perfect
mixing
(X
F
)
u
; (X
J
)
u
= weight fraction of otsam and jetsam in
the upper bed, respectively
y = y coordinate
Y = correction for deviation from the two-
phase theory
Y
h
= height of stagnant emulsion region
between adjacent jets
z = axial coordinate
r
B
; r
S
= particle density of big particles and
small particles, respectively
(r
B
)
S
= bulk density of the small particles
r
f
= uid density
r
F
; r
J
= particle density of otsam and jetsam,
respectively
r
p
= particle density in a single component
system
r
p
= average particle density of a mixture
" = voidage
"
1
= voidage of bed consists of component 1
"
b
= bed voidage
"
B
= bubble fraction in the bed
"
B
= average volume fraction of visible
bubble phase
"
c
= voidage at critical bed height
"
e
= voidage of emulsion phase
"
j
= voidage inside the jet
"
mf
= voidage at minimum uidization
"
s
= fraction of bed occupied by the slugs
"
w
= voidage in the wake
"
z
= voidage outside of jet in the emulsion
phase at z
c
f
= uid stream function
c
p
= particle stream function
f = particle shape factor
f
B
= bubble shape factor
f
F
; f
J
= shape factor of otsam and jetsam
particles, respectively
a
b
= angle of included angle in a bubble
b = distributor porosity
b
d
= drift fraction, V
d
=V
B
b
w
= wake fraction, V
W
=V
B
l = the rate of competing eect between
circulation and segregation,
=w=k
s = standard deviation
m = uid viscosity
y = polar coordinate; included angle of a
bubble; or jet half-angle
y
s
= dimensionless subsidence time of a
powder
REFERENCES
Abramovich GN. The Theory of Turbulent Jets. Cambridge,
MA: MIT Press, 1963.
Anagbo PE. Derivation of jet cone angle from bubble theory.
Chem Eng Sci 35:14941495, 1980.
Babu SP, Shah B, Talwalkar A. Fluidization correlations for
coal gasication materialsminimum uidization velo-
city and uidized bed expansion ratio. AIChE Symp Ser
74(176):176186, 1978.
Baeyens J, Geldart D. Fluidization et ses Applications.
Toulouse, 1973, p 182.
Baeyens J, Geldart D. An investigation into slugging ui-
dized beds. Chem Eng Sci 29:255265, 1974.
Baskakov AP, Malykh GA, Shihko II. Separation of mater-
ials in equipment with a uidized bed and with continuous
charging and discharging. Int Chem Eng 15(2):286289,
1975.
Basov VA, Markhevka VI, Melik-Akhnazanov TKh,
Orochke DI. Investigation of the structure of a non-
uniform uidized bed. Intern Chem Eng 9:263266, 1969.
Beeckmans JM, Minh T. Separation of mixed granular solids
using the uidized counter-current cascade principle. Can
J Chem Eng 55:493496, 1977.
Beeckmans JM, Stahl B. Mixing and segregation kinetics in a
strongly segregated gas-uidized bed. Powder Technol
53:3138, 1987.
Beeckmans JM, Inculet II, Dumas G. Enhancement in seg-
regation of a mixed powder in a uidized bed in the pre-
sence of an electrostatic eld. Powder Technol 24:267
269, 1979.
Behie LA, Bergougnou MA, Baker CGJ, Bulani W. Jet
momentum dissipation at a grid of a large gas uidized
bed. Can J Chem Eng 48:158161, 1970.
Behie LA, Bergougnou MA, Baker CGJ. Heat transfer from
a grid jet in a large uidized bed. Can J Chem Eng 53:25
30, 1975.
Behie LA, Bergougnou MA, Baker CGJ. Mass transfer from
a grid jet in a large gas uidized bed. In: Fluidization
Technology. (Keairns DL, ed.). New York: McGraw-
Hill, 1976, Vol. 1, pp 261278.
Bi HT, Fan LS. Existence of turbulent regime in gas-solid
uidization. AIChE J 38:297301, 1992.
Bi HT, Grace JR. Flow regime diagrams for gas-solid uid-
ization and upward transport. Int J Multiphase Flow
21:12291236, 1995a.
Bi HT, Grace JR. Eect of measurement method on veloci-
ties used to demarcate the onset of turbulent uidization.
Chem Eng J 57:261271, 1995b.
Bi HT, Grace JR, Zhu JX. Regime transitions aecting gas-
solids suspensions and uidized beds. Chem Eng Res Des
73:154161, 1995.
Blake TR, Wen CY, Ku CA. The correlation of jet penetra-
tion measurements in uidized beds using nondimensional
hydrodynamic parameters. AIChE Symp Series
80(234):4251, 1984.
Bloore PD, Botterill JSM. Similarity in behavior between gas
bubbles in liquid and uidized solid systems. Nature
190:250, 1961.
Bourgeois P, Grenier P. The ratio of terminal velocity to
minimum uidizing velocity for spherical particles. Can
J Chem Eng 46:325328, 1968.
Buevich YA, Minaev GA. Mechanics of jet ows in granular
layers, evolution of single jets and the nucleation mechan-
ism. Inzh Fiz Zh 30(5):825833, 1976.
Burgess JM, Fane AG, Fell CJD. Measurement and predic-
tion of the mixing and segregation of solids in gas uid-
ized beds. Pac Chem Eng Congr 2:14051412, 1977.
Cai P, Chen SP, Jin Y, Yu ZQ, Wang ZW. Eect of operat-
ing temperature and pressure on the transition from
bubbling to turbulent uidization. AIChE Symp Series
85(270):3743, 1989.
Canada GS, McLaughlin MH, Staub FW. Flow regimes and
void fraction distribution in gas uidization of large
particles in beds without tube banks. AIChE Symp Ser
176(74):1426, 1978.
Catipovic NM, Jovanovic GN, Fitzgerald TJ. Regimes of
uidization of large particles. AIChE J 24:543547, 1978.
Chandnani PP, Epstein N. Spoutability and spout destabili-
zation of ne particles with a gas. In: Fluidization
(Ostergaard K, Sorensen A, eds.) New York:
Engineering Foundation, 1986, pp 233240.
Chehbouni A, Caouki J, Guy C, Klvana D. Characterization
of the ow transition between bubbling and turbulent
uidization. Ind Eng Chem Res 33:18891896, 1994.
Chen JLP, Keairns DL. Particle segregation in a uidized
bed. Can J Chem Eng 53:395402, 1975.
Chen JLP, Keairns DL. Particle separation from a uidized
mixturesimulation of Westinghouse coal-gasication
combustor-gasier operation. Ind Eng Chem Process
Des Develop 17:135141, 1978.
Cheung L, Nienow AW, Rowe PN. Minimum uidization
velocity of a binary mixture of dierent sized paticles.
Chem Eng Sci 29:13011303, 1974.
Chiba T, Kobayashi H. Eect of bubbling on solid segrega-
tion in gas uidized beds. In: FluidizationScience and
Technology (Kwauk M, Kunii, D. eds.). New York:
Gordon and Breach, 1982, pp 7990.
Chiba S, Chiba T, Nienow AW, Kobayashi H. The minimum
uidization velocity, bed expansion and pressure-drop
prole of binary particle mixtures. Powder Technol
22:255269, 1979.
Chiba S, Nienow AW, Chiba T, Kobayashi H. Fluidized
binary mixtures in which the dense component may be
otsam. Powder Technol, 26:110, 1980.
Chiba S, Kawabata J, Yumiyama M, Tazaki Y, Honma S,
Kitano K, Kobayashi H, Chiba T. Pressure eects on
solid mixing and segregation in gas-uidized beds of
binary solid mixtures. In: FluidizationScience and
Technology (Kwauk M, Kunii, D. eds.). New York:
Gordon and Breach, 1982, pp 6978.
Chitester DC, Kornosky RM, Fan LS, Danko JP.
Characteristics of uidization at high pressure. Chem
Eng Sci 39;253261, 1984.
Choi JH, Son JE, Kim SD. Generalized model for bubble
size and frequency in gas-uidized beds. Ind Eng Chem
Res 37:25592564, 1998.
Clift R, Filla M, Massimilla L. Gas and particle motion in
jets in uidized beds. Intern J Multiphase Flow 2:549561,
1976.
Corder WC. Proc. 5th Synthetic Pipeline Gas Symp,
Chicago, IL. 1973.
Darton RC, LaNauze RD, Davidson JF, Harrison D.
Bubble growth due to coalescence in uidized beds.
Trans Inst Chem Eng 55:274280, 1977.
Davidson JF. Symposium on uidizationcontribution to
discussion. Trans Inst Chem Eng 39:230232, 1961.
Davidson JF, Harrison D. Fluidized Particles. Cambridge:
Cambridge University Press, 1963.
Davidson JF, Schuler BOG. Bubble formation at an orice
in an inviscid liquid. Trans Inst Chem Eng 38:335342,
1960.
Davidson JF, Harrison D, Guedes de Carvalho JRF. On the
liquidlike behavior of uidized beds. Ann Rev Fluid Mech
9:5586, 1977.
Davies RM, Taylor G. The mechanics of large bubbles rising
through extended liquids and through liquids in tubes.
Proc Roy Soc (London) A200:375390, 1950.
De Michele G, Elia A, Massimilla L. The interaction between
jets and uidized beds. Ing Chim Ital 12:155162, 1976.
Dickman R, Forsythe WL. Laboratory prediction of ow
properties of uidized solids. Ind Eng Chem 45:1174
1185, 1953.
Doichev K, Todorov S, Dimitrov V. Transition between par-
ticulate and aggregative uidization at dierent state-of-
ow of solids. Chen Eng Sci 30:419424, 1975.
Donadono A, Massimilla L. Mechanisms of momentum
and heat transfer between gas jets and uidized beds.
In: Fluidization. (Davidson JF, Keairns DL, eds.)
Cambridge: Cambridge University Press, 1978, pp 375
380.
Donadono S, Maresca A, Massimilla L. Gas injection in
shallow beds of uidized, coarse solids. Ing Chim Ital
16:110, 1980.
Donsi G, Massimilla L, Colantuoni L. The dispersion of axi-
symmetric jets. In: Fluidization (Grace JR, Matsen JM,
eds.). New York: Plenum Press, 1980, pp 297304
Drake G. Proc Fert Soc No. 130, 1973.
Dumitrescu DT. Stro mung an einer Luftblase im senkrechten
Rohr. Z Angew Math Mech 23:139, 1943.
Ettehadieh B, Yang WC, Hadipur GB. Motion of solids,
jetting and bubbling dynamics in a large jetting uidized
bed. Powder Technol 54:243254, 1988.
Fane AG, Nghiem NP. Bubble induced mixing and segrega-
tion in a gas-uidized bed. J Chinese Inst Chem Eng
24:215224, 1983.
Filla M, Massimilla L, Vaccaro S. Solids entrainment in jets
in uidized beds: the inuence of particle shape. Journe es
Europe ennes sur la Fluidisation, Toulouse, France,
September 1981.
Filla M, Massimilla L, Vaccaro S. Gas jets in uidized beds
and spouts: a comparison of experimental behavior and
models. Can J Chem Eng 61:370376, 1983a.
Filla M, Massimilla L, Vaccaro S. Gas jets in uidized beds:
the inuence of particle size, shape and density on gas and
solids entrainment. Int J Multiphase Flow 9:259267,
1983b.
Furukawa J, Ohmae T. Liquid-like properties of uidized
systems. Ind Eng Chem 50:821828, 1958.
Gbordzoe EAM, Freychet N, Bergougnou MA. Gas transfer
between a central jet and a large two-dimensional gas-
uidized bed. Powder Technol 55:207222, 1988.
Geldart D. The eect of particle size and size distribution on
the behavior of gas-uidized beds. Powder Technol 6:201
205, 1972.
Geldart D. Types of gas uidization. Powder Technol 7:285
292, 1973.
Geldart D. Predicting the expansion of gas uidized beds. In:
Fluidization Technology. (Kearins DL, ed.) Washington,
DC: Hemisphere, 1975, pp 237244.
Geldart D. Gas Fluidization Technology. Chichester: John
Wiley, 1986.
Geldart D, Abrahamsen AR. Homogeneous uidization of
ne powders using various gases and pressures. Powder
Technol 19:133136, 1978.
Gelperin NI, Einstein VG. Heat transfer in uidized beds. In:
Fluidization. (Davidson JF, Harrison D, eds.) New York:
Academic Press, 1971, pp 471450.
Gelperin NI, Einstein VG, Nosov GA, Mamoshkina VV,
Rebrova SK. Teor Osnovy Khim Tehnol 1:383, 1967.
Gelperin NI, Zakharenko VV, Ainshtein VG. Segregation of
solid particles in a uidized bed and the equilibrium dis-
tribution. Theor Foundations Chem Eng (Thor Osnovy
Khim Tekh) 11:475481, 1977.
Gibilaro LG, Rowe PN. A model for a segregating gas
uidized bed. Chem Eng Sci 29:14031412, 1974.
Goossens WRA. Classication of uidized particles by
Archimedes number. Powder Technol 98:4853, 1998.
Goossens WRA, Dumont GL, Spaepen GL. Fluidization of
binary mixtures in laminar ow region. Chem Eng Prog
Symp Ser 67(116):3845, 1971.
Grace JR. The viscosity of uidized beds. Can J Chem Eng
48:3033, 1970.
Grace JR. Fluidized-bed hydrodynamics. In: Handbook of
Multiphase Systems. (Hetsroni G, ed.) Washington:
Hemisphere, 1982, pp 8-58-64.
Grace JR. Contacting modes and behavior classication of
gassolid and other two-phase suspension. Can J Chem
Eng 64:353363, 1986.
Grace JR, Clift R. On the two-phase theory of uidization.
Chem Eng Sci 29:327334, 1974.
Grace JR, Lim CT. Permanent jet formation in beds of par-
ticulate solids. Can J Chem Eng 65:160162, 1987.
Grace JR, Sun G. Fines concentration in voids in uidized
beds. Powder Technol 62:203205, 1990.
Habermann A, Winter, F. Hofbauer H, Gogolek PEG.
Residence time distribution of particles in uidized bed
reactors for metallurgical processes. In; Fluidization
(Fan LS, Knowlton TM, eds.). New York: Engineering
Foundation, 1998, pp 117124.
Harrison D, Leung LS. Bubble formation at an orice in a
uidized bed. Trans Inst Chem Eng 39:409413, 1961.
Harrison D, Davidson JF, de Kock JW. On the nature of
aggregative and particulate uidization. Trans Inst Chem
Eng 39:202211, 1961.
Hirsan I, Sishtla C, Knowlton TM. Presented at the 73rd
AIChE Annual Meeting, Chicago, IL, November 1980.
Horio M. Hydrodynamics of circulating uidizationpre-
sent status and research needs. In: Circulating Fluidized
Bed Technology III (Basu P, Horio M, Hasatani M, eds.).
Oxford: Pergamon Press, 1990, pp 314.
Hovmand S, Davidson JF. Chemical conversion in a slugging
uidized bed. Trans Inst Chem Eng 46:190203, 1968.
Hovmand S, Davidson JF. Pilot plant and laboratory scale
uidized reactors at high gas velocities: the relevance of
slug ow. In: Fluidization. (Davidson JF, Harrison D,
eds.) New York: Academic Press, 1971, pp 193259.
Hussein FD, Paltra PP, Jackson R. Heat transfer to a ui-
dized bed using a particulate heat transfer medium. Ind
Eng Chem Process Des Dev 20:511519, 1981.
Iya SK, Geldart D. Flow of coarse particles through a ui-
dized bed of nes. Ind Eng Chem Process Des Dev
17:537543, 1978.
Jackson R. The mechanics of uidized beds: the motion of
fully developed bubbles. Trans Inst Chem Eng 41:2248,
1963.
Kan DD, Sen Gupta P. Relative performance of baed and
unbaed uidized bed classiers. Indian J Tech 16:507
512, 1978.
Kececioglu I, Yang WC, Keairns DL. Fate of solids fed
pneumatically through a jet into a uidized bed. AIChE
J 30:99110, 1984.
Kehoe PWK, Davidson JF. The uctuation of surface height
in freely slugging uidized beds. AIChE Symp Ser
69(128):4148, 1973a.
Kehoe PWK, Davidson JF. Pressure uctuations in slug-
ging uidized beds. AIChE Symp Ser 69(128):3440,
1973b.
Kimura T, Matsuo H, Uemiya S, Kojima T. Measurement of
jet shape and its dynamic change in three-dimensional
jetting uidized beds. J Chem Eng Japan 27:602609,
1994.
Kimura T, Horiuchi K, Watanabe T, Matsukata M, Kojima
T. Experimental study of gas and particle behavior in the
grid zone of a jetting uidized bed cold model. Powder
Technol 82:135143, 1995.
King DF, Mitchell RG, Harrison D. Dense phase viscosities
of uidized beds at elevated pressure. Powder Technol
28:5558, 1981.
Knowlton TM. High-pressure uidization characteristics of
several particulate solids. AIChE Symp Ser 73(61):2228,
1977.
Knowlton TM, Hirsan I. The eect of pressure on jet pene-
tration in semi-cylindrical gas-uidized beds. In:
Fluidization (Grace JR, Matsen JM, eds.). New York:
Plenum Press, 1980, pp 315324.
Kondukov NB, Sosna MH. Khim Prom 6:402, 1965.
Kozin BE, Baskakov AP. Studies on the jet range in a bed of
granular particles. Khim I Teckhnol Topliv I Masel 3:47,
1967.
Kramers H. On the viscosity of a bed of uidized solids.
Chem Eng Sci 1:3537, 1951.
Kumar A, Sen Gupta P. Prediction of minimum uidization
velocity for multicomponent mixtures. Indian J Technol
12:225227, May 1974.
Kunii D, Levenspiel, O. Fluidization Engineering. 1st ed.
New York: John Wiley, 1969.
Kunii D, Levenspiel, O. Fluidization Engineering. 2nd ed.
Boston: Butterworth-Heinemann, 1991.
Kwauk M. Fluidization: Idealized and Bubbleless, with
Applications. New York: Ellis Horwood, 1992.
Lefroy GA, Davidson JF. The mechanics of spouted beds.
Trans Inst Chem Eng 47:T120T128, 1969.
Leppin D, Sahay GN. Cold-model studies of agglomerating
gasier discharge behavior. In: Fluidization (Grace JR,
Matsen JM, eds.). New York: Plenum Press, 1980, pp
429436.
Leva M. Fluidization. New York: McGraw-Hill, 1959.
Levenspiel O. The Chemical Reactor Omnibook. Corvallis:
OSU Book Stores, 1979.
Lim KS, Zhu JX, Grace JR. Hydrodynamcis of gassolid
uidization. Int J Multiphase Flow 21(Suppl):141193,
1995.
Markhevka VI, Basov VA, Melik-Akhnazanov TKh,
Orochko DI. The ow of a gas jet into a uidized bed.
Theor Foundations Chem Eng 5:8085, 1971.
Massimilla L. Gas jets in uidized beds. In: Fluidization. 2nd
ed. (Davidson JF, Clift R, Harrison D., eds.). London:
Massimilla L, Donsi G, Migliaccio N. The dispersion of gas
jets in two-dimensional uidized beds of coarse solids.
AIChE Symp Ser 77(205):1727, 1981.
Matheson GL, Herbst WA, Holt PH. Characteristics of
uidsolid systems. Ind Eng Chem 41:10991104, 1949.
Matsen JM. Evidence of maximum stable bubble size in a
uidized bed. AIChE Symp Ser 69(128):3033, 1973.
Matsen JM, Hovmand S, Davidson JF. Expansion of uid-
ized beds in slug ow. Chem Eng Sci 24:17431754, 1969.
Merry JMD. Penetration of a horizontal gas jet into a uid-
ized bed. Trans Inst Chem Eng 49:189195, 1971.
Merry JMD. Penetration of vertical jets into uidized beds.
AIChE J 21:507510, 1975.
Merry JMD. Fluid and particle entrainment into vertical jets
in uidized beds. AIChE J 22:315323, 1976.
Mitchell DS, Sageman DR. Countercurrent plug-like ow of
two solids. U.S. Patent 4,157,245 (1979).
Molerus O. Interpretation of Geldarts type A, B, C and D
powders by taking into account interparticle cohesion
forces. Powder Technol 33:8187, 1982.
Mori S, Wen CY. Estimation of bubble diameter in gaseous
uidized beds. AIChE J 21:109115, 1975.
Morooka S, Nishinaka M, Kato Y. Sedimentation velocity
and expansion ratio of the emulsion phase in a gassolid
uidized bed. Kagaku Kogaku 37:485493, 1973.
Murray JD. On the mathematics of uidization: steady
motion of fully developed bubbles. J Fluid Mech
21:465493, 1965.
Naimer, NS, Chiba T, Nienow AW. Parameter estimation
for a solids mixing/segregation model for gas uidized
beds. Chem Eng Sci 37:10471057, 1982.
Naveh E, Resnick W. Particle size segregation in
baed uidized beds. Trans Inst Chem Eng
52:5866, 1974.
Nguyen XT, Leung LS. A note on bubble formation at an
orice in a uidized bed. Chem Eng Sci 27:17481750,
1972.
Nicklin DJ, Wilkes JO, Davidson JF. Two-phase ow in
vertical tubes. Trans Inst Chem Eng 40:6168, 1962.
Nienow AW, Cheesman, DJ. The eect of shape on the
mixing and segregation of large particles in gas-uidized
bed of small ones. In: Fluidization. (Grace JR, Matsen
JM, eds.) New York: Plenum Press, 1980, pp 373380.
Nienow AW, Naimer NS. Continuous mixing of two parti-
culate species of dierent density in a gas uidized bed.
Nienow AW, Rowe PN, Cheung LYL. A quantitative ana-
lysis of the mixing of two segregating powders of dierent
density in a gas-uidized bed. Powder Technol 20:8997,
1978a.
Nienow AW, Rowe, PN, Chiba T. Mixing and segregation of
a small proportion of large particles in gas uidized beds
of considerably smaller ones. AIChE Symp Ser
74(176):4553, 1978b.
Nienow AW, Naimer NS, Chiba T. Studies of segregation/
mixing in uidized beds of dierent size paticles. Chem
Eng commun 62:5366, 1987.
Peeler JPK, Huang JR. Segregation of wide size range par-
ticle mixtures in uidized beds. Chem Eng Sci 44:1113
1119, 1989.
Perales JF, Coll T, Llop MF, Puigjaner L, Arnaldos J, Cassal
J. On the transition from bubbling to fast uidization
regimes. In: Circulating Fluidized Bed Technology III
(Basu P, Horio M, Hasatani M, eds.). Oxford:
Pergamon Press, 1990, pp 7378.
Rhodes MJ, Wang XS. Dening fast uidization. In:
Fluidization. (Fan LS, Knowlton TM, eds.) New York:
Rice RW, Brainnovich Jr JF. Mixing/segregation in two- and
three-dimensional uidized beds: binary systems of equi-
density spherical particles. AIChE J 32:716, 1986.
Richardson JF. Incipient uidization and particulate sys-
tems. In: Fluidization. (Davidson JF, Harrison D, eds.)
New York: Academic Press, 1971, pp 2664.
Rietema K. The eect of interparticle forces on the expansion
of a Ginigebeiys gas-uidized bed. Proceedings Intern
Symp Fluidization, Toulouse, 1967, pp 2840.
Roach PE. Dierentiation between jetting and bubbling in
uidized beds. Int J Multiphase Flow 19:11591161, 1993.
Romero JB, Johanson LN. Factors aecting uidized bed
quality. Chem Eng Prog Symp Ser 58(38):2837, 1962.
Rowe PN. Drag forces in a hydraulic model of a uidized
bed, part II. Trans Inst Chem Eng London 39:175180,
1961.
Rowe PN. The eect of bubbles on gassolids contacting in
uidized beds. Chem Eng Prog Symp Ser 58(38):4256,
1962.
Rowe PN. Experimental properties of bubbles. In:
Fluidization. (Davidson JF, Harrison D, eds.) New
York: Academic Press, 1971, pp 121192.
Rowe PN. Prediction of bubble size in a gas uidized bed.
Chem Eng Sci 31:285288, 1976.
Rowe PN, Nienow AW. Minimum uidization velocity of
multicomponent particle mixtures. Chem Eng Sci
30:13651369, 1975.
Rowe PN, Nienow AW. Particle mixing and segregation in
gas uidized beds. Power Tech 15:141147, 1976.
Rowe PN, Partridge BA, Cheney AG, Henwood GA, Lyall
E. The mechanisms of solids mixing in uidized beds.
Trans Inst Chem Eng 43:T271T286, 1965.
Rowe PN, Nienow AW, Agbim AJ. A preliminary quantita-
tive study of particle segregation in gas uidized beds
binary systems of near spherical particles. Trans Inst
Chem Eng 50:324333, 1972.
Rowe PN, MacGillivray HJ, Cheesman DJ. Gas discharge
from an orice into a gas uidized bed. Trans Inst Chem
Eng 57:194199, 1979.
Saxena and Vogel. The measurement of incipient uidization
velocities in a bed of corase dolomite at temperature and
pressure. Trans Inst Chem Eng 55:184189, 1977.
Schugerl K. Rheological behavior of uidized systems. In:
Fluidization. (Davidson JF, Harrison D, eds.) New
York: Academic Press, 1971, pp. 261292.
Schugerl K, Merz M, Fetting F. Rheologische Eigenschaften
von gasdurchstro mten Fliessbettsystemen. Chem Eng Sci
15:138, 1961.
Shakhova NA. Discharge of turbulent jets into a uidized
bed. Inzh Fiz Zh 14(1):6169, 1968.
Shakhova NA, Minaev GA. Aerodynamics of jets discharged
into uidized beds. Heat Transfer Soviet Research
4(1):133142, 1972.
Shi YF, Fan LT. Lateral mixing of solids in batch gassolids
uidized beds. Ind Eng Chem Process Des Dev 23:337
341, 1984.
Sit SP, Grace JR. Hydrodynamics and mass transfer in the
distributor zone of uidized beds. Proc Second World
Cong Chem Eng, Montreal, Canada, 1981, pp 8184.
Stewart PSB, Davidson JF. Slug ow in uidized beds.
Powder Tech 1:6180, 1967.
Subbarao D, Moghaddam E, Bannard JE. Lateral mixing of
particles in uidized beds. Chem Eng Sci 40:19881990,
1985.
Tanimoto H, Chiba S, Chiba T, Kobayashi H. Mechanism of
solid segregation in gas uidized beds: In: Fluidization.
(Grace JR, Matsen JM, eds.) New York: Plenum Press,
1980, pp 381388.
Tanimoto H, Chiba T, Kobayashi H. Eects of segregation
of ne elutriation from gas-uidized beds of binary solid
mixture. J Chem Eng Japan 16:149152, 1983.
Toomey RO, Johnstone HF. Gaseous uidization of solid
particles. Chem Eng Prog 48:220226, 1952.
Trawinski H. Chem Ing Tech 25:201, 1953.
Tsukada M, Horio M. Gas motion and bubble formation at
the distributor of a uidized bed. Powder Technol 63:69
74, 1990.
Turner JS. Buoyancy Eects in Fluids. Cambridge:
Uchida S, Yamada H, Tada I. Minimum uidization velocity
of binary mixtures. J Chinese Inst Chem Eng 14:257264,
1983.
Uno S, Kintner RC. Eect of wall proximity on the rate of
rise of simple air bubbles in a quiescent liquid. AIChE J
2:420425, 1956.
Vaccaro S. Analysis of the variables controlling gas jet
expansion angles in uidized beds. Powder Technol
92:213222, 1997.
Vaccaro S, Musmarra D, Petrecca M. A technique for mea-
surement of the jet penetration height in uidized beds by
pressure signal analysis. Powder Technol 92:223231,
1997.
Vaid RP, Sen Gupta P. Minimum uidization velocities in
beds of mixed solids. Can J Chem Eng 56:292296, 1978.
Valenzuela JA, Glicksman LR. Gas ow distribution in a
bubbling uidized bed. Presented at AIChE 74th
Annual Mtg, 1981.
Verloop J, Heertjes PM. On the origin of bubbles in gas-
uidized beds. Chem Eng Sci 29:11011107, 1974.
Wen CY, Yu YH. A generalized method for predicting the
minimum uidization velocity. AIChE J 12:610612,
1966a.
Wen CY, Yu YH. Mechanics of uidization. Chem Eng Prog
Symp Ser 62(62):100111, 1966b.
Wen CY, Horio M, Krishnan R, Kosravi R, Rengarajan P.
Jetting phenomena and dead zone formation on uidized
bed distributors. Proc Second Pacic Chem Eng Cong,
Vol. 2, 1182, 1977, Denver, Colorado.
Wen CY, Deole NR, Chen LH. A study of jets in a three-
dimensional gas uidized bed. Powder Technol 31:175
184, 1982.
Werther J. Bubble growth in large diameter uidized beds.
In: Fluidization Technology (Keairns DL, ed.).
Washington: Hemisphere, 1976a, pp 215235.
Werther J. Convective solids transport in large diameter gas
uidized beds. Powder Technol 15:155167, 1976b.
Whitehead AB. Prediction of bubble size in a gas-uidized
bed. Chem Eng Sci 34:751751, 1979.
Whitehead AB, Young AD. Fluidization performance in
large scale equipment. In: Proc Intern Symp on
Fluidization. Amsterdam: Netherlands University Press,
1967, pp 284293.
Whitehead AB, Dent DC, McAdam JCH. Fluidization
studies in large gassolid systems part V. long and short
term pressure instabilities. Powder Technol 18:231237,
1977.
Wilhelm RH, Kwauk M. Fluidization of solid particles.
Chem Eng Prog 44:201217, 1948.
Wu CS, Whiting WB. Interacting jets in a uidized bed.
Chem Eng Comm 73:117, 1988.
Wu SY, Baeyens. J Segregation by size dierence in gas
uidized beds. Powder Technol 98:139150, 1998.
Xavier AM, Lewis DA, Davidson JF. The expansion of bub-
bling uidized bed. Trans Inst Chem Eng 56:274280,
1978.
Yagi S, Kunii D. Fluidized-solids reactors with continuous
solids feedI. Residence time of particles in uidized
beds. Chem Eng Sci 16:364371, 1961.
Yang WC. Jet penetration in a pressurized uidized bed.
I&EC Fundamentals 20:297300, 1981.
Yang WC. Particle segregation in gas-uidized beds. In:
Encyclopedia of Fluid Mechanics, Vol. 4. (Cheremisino
NP, ed.) Houston: Gulf, 1986, pp 817852.
Yang WC. 30 years of industrial research on uidization
bridging the gap between theory and practice. In:
Fluidization (Fan LS, Knowlton TM, eds.). New York:
Engineering Foundation, 1998a, pp 3143.
Yang WC. Comparison of jetting phenomena in 30-cm and
3-m diameter semicular uidized beds. Powder Technol
100:147160, 1998b.
Yang WC. Engineering and application of recirculating and
jetting uidized beds. In: Fluidization, Solids Handling,
and Processing. (Yang WC, ed.) Westwood: Noyes, 1999,
pp 236330.
Yang WC, Keairns DL. Design and operating parameters for
a uidized bed agglomerating combustor/gasier. In:
Fluidization (Davidson JR, Keairns DL, eds.).
Cambridge: Cambridge University Press, 1978, pp 208
214.
Yang WC, Keairns DL. Estimating the jet penetration depth
of multiple vertical grid jets. I&EC Fundamentals 18:317
320, 1979.
Yang WC, Keairns DL. Momentum dissipation of and gas
entrainment into a gassolid two-phase jet in a uidized
bed. In: Fluidization (Grace JR, Matsen JM, eds.). New
York: Plenum Press, 1980a, pp 305314.
Yang WC, Keairns DL. The eect of an expanded section on
slugging. AIChE J 26:144148, 1980b.
Yang WC, Keairns DL. Solid entrainment rate into gas and
gassolid two-phase jets in a uidized bed. Powder
Technol 33:8994, 1982a.
Yang WC, Keairns DL. Rate of particle separation in a uid-
ized bed. I&EC Fundamentals 21:228235, 1982b.
Yang WC, Keairns DL, McLain DK. Gas mixing in a jetting
uidized bed. AIChE Symposium Series 80(234):3241,
1984a.
Yang WC, Revay D, Anderson RG, Chelen EJ, Keairns DL,
Cicero DC. Fluidization phenomena in a large-scale, cold-
ow model. In: Fluidization. (Kunii D, Toei R, eds.) New
York: Engineering Foundation, 1984b, pp. 7784.
Yang WC, Ettehadieh B, Anestis TC, Gizzie RE, Haldipur
GB. Fluidization phenomena in a large jetting uidized
bed. In: Fluidization V. (Ostergaard K, Sorensen A, eds.).
New York: Engineering Foundation, 1986, pp 95102.
Yang Z, Tung Y, Kwauk M. Characterizing uidization by
the bed collapsing method. Chem Eng Commun 39:217
232, 1985.
Yasui G, Johanson LN. Characteristics of gas pockets in
Yates JG, Cheesman DJ, Segeev YA. Experimental observa-
tions of voidage distribution around bubbles in a uidized
bed. Chem Eng Sci 49:18851895, 1994.
Yates JG, Wu KT, Cheesman DJ. Bubble coalescence from
multiple entry nozzles. In: Fluidization (Large JF,
Laguerie C, eds.). New York: Engineering Foundation,
1995, pp 1119.
Yerushalmi H, Cankurt NT. Further studies of the regimes
of uidization. Powder Technol 24:187205, 1979.
Yerushalmi H, Cankurt NT, Geldart D, Liss B. Flow regimes
in vertical gassolid contact systems. AIChE Symp Ser
176(74):113, 1978.
Yoshida K, Kameyama H, Shimizu F. Mechanism of particle
mixing and segregation. In: Fluidization (Grace JR,
Matsen JM. eds.). New York: Plenum Press, 1980, pp
389396.
Zenz FA. Bubble formation and grid design. Inst Chem Eng
Symp Ser 30:136145, 1968.
Zenz FA. Regimes of uidized behavior. In: Fluidization.
(Davidson JF, Harrison D, eds.) New York: Academic
Press, 1971.
Zenz FA. How ow phenomana aect design of uidized
beds. Chem Eng 84(27):8191, December 1977, pp 123.
Zenz FA. The uid mechanics of bubbling beds. Fibonacci
Quarterly 16(2):171183, 1978.
4
Elutriation and Entrainment
Joachim Werther and Ernst-Ulrich Hartge
1 INTRODUCTION
For most applications of uidized beds, the process,
e.g., the chemical reaction or the drying process,
takes place in the dense uidized bed either completely
or at least in major part. Nevertheless, in many units
the freeboard above the uidized bed takes the larger
volume of the whole unit compared to the uidized
bed. The task of the freeboard is mainly to prevent a
too large amount of the bed material from being
carried out of the unit by the gas stream. In order to
design properly the freeboard, it is necessary to know
not only the height, which is necessary to separate the
solids as eectively as possible from the gas, but also
the inuences of the freeboard diameter, of particle
properties, and of operating conditions on the entrain-
ment ux. The entrainment ux has to be known for
the design of gas/solid separators like cyclones or
lters. On the other hand, the loss of bed material
that is related to the entrainment may be important
for the technical and also the economic success of a
uidized bed application.
While for most processes the transport of solids out
of the uidized bed reactor is a disadvantage, some
processes take advantage of the loss of nes, examples
being uidized bed classiers (Tasirin and Geldart,
2000) and the incineration of sewage sludge, where
the ne ash is removed as y ash entrained from the
combustor.
In the present chapter we deal with the mechanisms
underlying the entrainment, some experimental nd-
ings, and the correlations to estimate the entrainment
and the height necessary to minimize the entrainment.
2 DEFINITIONS
At the very beginning we dene the terms that will be
used throughout this chapter. Two terms are often
used for the ow of solid out of uidized beds. These
are the terms entrainment and elutration. Entrainment
means the ux of solids carried out of the uidized bed
by the gas in kg per unit cross-sectional area and
second. Elutriation means the classifying eect of the
uidized bed entrainmentit characterizes the selec-
tive removal of particles of individual size from the
uidized bed.
The uidized bed is usually divided into dierent
vertical zones (Fig. 1). At the bottom there is the
dense uidized bed, above which the freeboard is
located. The term freeboard denotes the space above
the dense uidized bed. The height between the dense
bed surface and the gas outlet will be named here free-
board height H
FB
. It could be argued whether it would
be better to denote by freeboard height the total
distance between the dense bed surface and the top
of the vessel. However, most authors prefer the gas
outlet as the upper limit of the freeboard, since this
latter location determines the carryover of solids.
At higher gas velocities, the bed surface will start to
uctuate and will become more and more fuzzy. The
sharp transition between the dense bed and the dilute
freeboard will disappear, and there will be a zone char-
acterized by a sharp decay of the solids concentration.
This latter zone will be called the splash zone.
Experimental observations show that above the
dense bed or the splash zone, respectively, the solids
holdup gradually decays until it becomes constant or
at least nearly constant. The distance between this
point, where the solids concentration becomes (almost)
constant, and the surface of the uidized bed is called
the Transport Disengaging Height (TDH).
3 EJECTION OF PARTICLES INTO THE
FREEBOARD
In order to understand the process of entrainment we
will have a look at the governing mechanisms rst. The
whole process may be divided into three subprocesses.
At rst the solids have to be transported from the bed
into the freeboard; then they will rise, owing to their
initial velocity, adjust their velocity to the gas velocity
in the freeboard, and partly disengage; andprovided
that the freeboard is suciently highnally they will
be transported by the upowing gas to the gas outlet.
In the following we will rst deal with the mechanisms
that cause the particles to be ejected from the bed into
the freeboard.
The most important mechanism for the projection
of particles from the uidized bed surface into the free-
board is related to the eruption of bubbles (Levy et al.,
1983). In Fig. 2 schematic drawings of bubbles burst-
ing at the surface of a uidized bed are shown.
Dierent mechanisms are considered to be responsible
for the particle ejection, namely the ejection of parti-
cles from the bubble roof or nose (Fig. 2a) and the
ejection of particles transported upwards in the wake
of the bubble (Fig. 2b). While there is a general agree-
ment on the important role of the bubbles, no such
agreement exists with regard to the suitable mechan-
ism. While Zenz and Weil (1958) and Chen and Saxena
(1978) consider the nose mechanism to be the most
important, George and Grace (1978) assumed the
wake mechanism to be the governing one. Pemberton
and Davidson (1986) postulated that particles might be
ejected by both mechanisms, depending on the particle
size, the uidizing velocity, and the bed geometry.
From their measurements they deduced that bubbles
bursting one by one eject the solids from the roof while
coalescing bubbles preferably throw the wake into the
freeboard (cf. Fig. 2c). These ndings are also con-
rmed by Levy et al. (1983) and by Peters et al.
(1983), who took sequences of photographs to study
the mechanism of particle projection. Pemberton and
Davidson (1983) postulated that for particles belong-
ing to the Geldart group A uidized at high velocities,
the ejection from the roof predominates, while for
group B particles uidized with lower velocities, ejec-
tion from the wake is governing the process. A dier-
ence between the two mechanisms lies in the amount of
solids thrown into the freeboard and their size distri-
bution. While the solids thrown into the freeboard
from the bubble wake have nearly the same size dis-
tribution as the bed material, the solids thrown into the
freeboard from the bubble roof are much ner. The
bubble-roof model gives a much lower solids ux at
the bed surface than the wake model, the dierence
between the uxes being about one order of magni-
tude. Another dierence between the two mechanisms
is the dispersion of the solids ejected to the freeboard:
while the solids from the wake are thrown into the
freeboard as quite closely packed bulks of solids, the
Figure 1 Zones in a uidized bed vessel.
Figure 2 Bubbles bursting at the bed surface, which are
ejecting solids into the freeboard (a) from the roof, (b)
from the wake of a single bubble, and (c) from the wake of
two coalescing bubbles.
particles from the roof are well dispersed (Tanimoto et
al., 1984, Hazlett and Bergougnou, 1992).
Starting velocities of the particles at the bed surface
have been determined, e.g., by George and Grace
(1978) and by Peters et al. (1983) and have been
found to be signicantly higher than the bubble rise
velocity. George and Grace (1978) found that the start-
ing velocity of the particles ejected into the freeboard is
roughly twice the bubble rise velocity. These results
have been conrmed by Hatano and Ishida (1983),
who measured particle velocities by ber-optical
probes.
Another mechanism for the transport of particles
into the freeboard, which may be particularly signi-
cant for coarse particle uidized beds with a wide par-
ticle size distribution, is the interstitial velocity in the
dense bed. Fine particles at the top of the dense bed
will be transported to the bed surface and it will be
blown out into the freeboard (Bachovchin et al.,
1981). This latter transport mechanism is strongly
classifying, but possible uxes are low compared to
the uxes induced by the bursting bubbles.
4 GASSOLIDS FLOW IN THE FREEBOARD
After the solids have been ejected into the freeboard
with a certain starting velocity, they will decelerate and
either fall back to the uidized bed or will be trans-
ported by the upowing gas to the gas outlet. The most
simple assumption for the modelling of this process is
to assume that all particles behave independently of
each other and to use a simple ballistic model (Zenz
and Weil, 1958; Do et al., 1972). A comparison of
trajectories calculated with this model and experimen-
tally observed particle trajectories (Peters et al., 1983)
is given in Fig. 3. There is a quite good agreement, but
the small maximum heights between 1 and 2 cm should
be noted.
The simple ballistic model implies that only particles
with a terminal settling velocity u
t
less than the super-
cial velocity U are entrained. But experimental obser-
vations show that also larger particles may be
entrained (Geldart and Pope, 1983). Furthermore
according to this model, there is no explanation for
the experimental ndings that entrainment decreases
with increasing height even for uidized beds of nes,
which all are entrainable. A third contradiction of this
model to experimental ndings is that due to the
decreasing solids velocities, the solids concentration
above such a bed of nes would increase with height,
while all experimental observations show a decrease of
the solids concentration with height.
In a bed with a wide solid size distribution, coarse
particles with a settling velocity higher than the gas
velocity are entrained, which can be explained by inter-
actions between ne and coarse particles (Geldart et
al., 1979; Geldart and Pope, 1983). The exchange of
momentum may lower the settling velocity of the
coarse particles, while the nes are decelerated. But
this mechanism still gives no explanation of the
decreasing entrainment of nes with increasing height.
Measurements of instantaneous solids concentra-
tions (Hatano and Ishida, 1983; Yorquez-Ramirez
and Duursma, 2001) show that particles are not
moving individually through the freeboard; they form
clusters, which allow slip velocities between gas and
particles that are much higher than the single particles
terminal velocity. Such clusters will form also with very
ne particles and thus provide an explanation of why
the entrainment is decreasing with increasing height
even for ne particle systems (Geldart and Pope,
1983; Kunii and Levenspiel, 1989).
The ow structure inside the freeboard becomes
even more complicated when velocity proles of the
gas are taken into account. Such proles are due to
the laminar ow that prevails in many of the quite
small test units used for entrainment experiments.
For such a laminar prole, quite large particles may
Figure 3 Measured and calculated particle trajectories.
(After Peters et al., 1983.)
be entrained when rising in the center, while even very
ne particles will move downward at the wall. For
laminar velocity proles this eect is much more pro-
nounced than for turbulent velocity proles, which
prevail in larger units. This might also give an explana-
tion for the observed dependence of the entrainment
ux on the diameter of the freeboard. According to
Tasirin and Geldart (1998a) there is a critical freeboard
diameter above which the entrainment ux is nearly
independent of the freeboard diameter. They explain
the existence of this critical diameter by the transition
from laminar ow to turbulent ow.
Besides the velocity proles due to laminar or tur-
bulent ow, there are also temporary uctuations in
the gas ow caused by the eruption of bubbles at the
bed surface. A widely accepted model to explain the
freeboard turbulence caused by the bubbles is the idea
of ghost bubbles, postulated by Pemberton and
Davidson (1984). They assume that a bubble keeps
its identity even after breaking through the bed surface
and while traveling up through the freeboard. The
ghost bubbles are characterized by a circulating ow,
which continues the gas ow pattern inside and around
the bubbles in the uidized bed. Yorquez-Ramirez and
Duursma (2000, 2001), however, state that vortex rings
are generated by the erupting bubbles, which cause the
freeboard turbulence. They identify the uctuations of
the level of the bed surface as another source of gas
ow uctuations.
5 EXPERIMENTAL INVESTIGATIONS OF
ENTRAINMENT
After the description of the governing mechanisms
responsible for the entrainment of particles from
uidized beds, a short overview about experimental
investigations on entrainment and elutriation will be
given. Since there is very large number of experimental
investigations in the literature, we do not attempt to
give a complete survey.
In most cases, experimental results are given either
in terms of the total entrainment ux E
h
in kg/(m
2
s) at
the height h above the distributor or in terms of an
elutriation rate constant K
+
ih
, dened as the ratio of
the instantaneous rate of removal of solids of size d
pi
based on the cross-sectional area A to the fraction of
the mass of the bed material with size d
pi
:
k
+
ih
=
E
ih
(t)
x
Bi
(t)
(1)
This can be rewritten as
K
+
ih
=
1
A
d
dt
[x
Bi
(t) M
B
(t)]
x
Bi
(t)
(2)
E
ih
denotes the carryover ux of a component of size
d
pi
at the height h above the distributor based on the
cross-sectional area of the bed. x
Bi
gives the mass frac-
tion of particles with size d
pi
in the bed, and M
B
is the
total mass of the bed material. For steady-state opera-
tion, i.e., when the carryover is continuously recycled
to the bed, the fraction of particles with size d
pi
is
constant with time, and Eq. (1) can be rewritten as
K
+
ih
=
E
ih
x
Bi
(3)
For batch experiments, K
+
ih
has to be calculated from
the entrained mass m
i;t
of size d
pi
accumulated during
the time t. Therefore Eq. (1) has to be integrated.
Provided the total mass of bed material does not
change signicantly, this gives
m
i;t
= x
BiO
M
B
1 exp
K
+
ih
A
M
B
_ _ _ _
(4)
where x
Bi0
is the initial mass fraction of particles with
size d
pi
in the bed mass. In order to obtain the total
entrainment E
h
, the individual entrained mass uxes
E
ih
of particle sizes d
pi
have to be accumulated:
E
h
=
i
E
ih
=
i
x
Bi
K
+
ih
( ) (5)
The mass fraction of the particles with size d
pi
in the
entrained ow can be calculated from
x
i;E
=
E
ih
E
h
=
x
Bi
K
+
ih
E
h
(6)
Above a certain freeboard height, the entrainment ux
and the elutriation rate constants become essentially
independent of height. This height is called the trans-
port disengaging height (TDH). The entrainment
uxes and elutriation rate constants above this height
are given the index o, i.e. E
o
, E
io
. In the following,
the inuence of operating and design parameters will
be discussed wherever possible in terms of the elutria-
tion rate constants K
+
ih
or K
+
io
.
5.1 Inuence of Particle Size
For ne particles with a terminal velocity u
t
below the
gas velocity in the freeboard, the elutriation rate con-
stant K
+
io
increases with decreasing particles size.
While there is a general consensus about this when
group A powders are concerned there are some indica-
tions that there is a critical particle size, below which
the elutriation rate constant K
+
io
levels o or even
decreases with decreasing size. Figure 4 shows experi-
mental results measured by Smolders and Baeyens
(1997) with limestone and a-alumina. The experimen-
tally determined values for K
+
io
are plotted against the
particle size. In spite of some scatter, the general
decrease in the elutriation rate constants below about
10 mm is obvious. Similar results were also presented
by Tasirin and Geldart (1998b), Santana et al. (1999),
and Tasirin et al. (2001).
The same trend was also observed for uidized beds
with binary particle mixturesnes in a coarse particle
uidized bedby Ma and Kato (1998). The reason for
this eect is probably the formation of particle agglom-
erates in the case of very ne particles for which adhe-
sion forces are large compared to the gravitational
forces. Baeyens et al. (1992) suggest based on their
measurements a correlation for the critical particle dia-
meter d
p;crit
, at which the inection occurs:
d
p;crit
r
0:725
s
= 10325 (7)
d
p;crit
is to be inserted in mm and r
s
in kg/m
3
,
respectively. Ma and Kato (1998) introduce a critical
cohesion number
N
+
coh
=
C
r
s
d
p;crit
g
= 4:5 (8)
as the criterion for the critical particle diameter d
p;crit
,
with C = 0:455 r
0:269
s
where r
s
is to be inserted in
kg/m
3
. Experiments by Briens et al. (1992) also show
the decrease of the elutriation rate constant with
decreasing size for very ne particles. These latter
authors showed furthermore that there is no inuence
of electrostatic eects on the critical particle size, even
though the absolute values of the entrainment uxes
were signicantly changed by electrostatic loading.
For the coarse particles with a terminal settling
velocity u
t
above the supercial gas velocity in the free-
board U, it is often assumed that they cannot be
entrained if the freeboard is suciently high, but mea-
surements by Geldart and Pope (1983) and Geldart et
al. (1979) have shown that coarse particles may be
entrained, too, if there is a sucient entrainment ux
Figure 4 Elutriation rate constant K
+
io
versus the particle size for nes and supernes for (a) limestone and (b) a-alumina. (Data
from Smolders and Baeyens, 1997.)
of nes. The amount of entrained coarses particles is
signicantly inuenced by the ux of nes.
While there is an inuence of the entrainment of
nes on the carryover of coarse material, no inuence
of the nes content on the elutriation rate constant of
the nes can be found (Taneda et al., 1998). This is an
assumption implicitly already made in Eq. (3).
5.2 Inuence of Supercial Gas Velocity
The inuence of the supercial gas velocity is quite
obvious. Both the ejection of particles from the dense
bed into the freeboard and the transport through the
freeboard are aected by the supercial velocity. In
Fig. 5, data measured by Choi et al. (1998) for sand
are given as an example. In general, the elutriation rate
increases proportionally with the gas velocity to a
power of 2 to 4.
5.3 Effect of Freeboard Geometry
5.3.1 Inuence of Freeboard Height
Close to the bed surface the carryover from the uid-
ized bed decreases signicantly with increasing height.
The decay in the entrainment becomes nearly constant,
when the transport disengaging height (TDH) is
exceeded (Fig. 6). Like the entrainment ux, the solids
volume concentration in the freeboard decreases with
increasing distance from the bed surface and becomes
nearly constant above the TDH.
According to measurements by Nakagawa et al.
(1994), the shape of the gas outlet has no signicant
inuence on the solids holdup in the freeboard. Figure
7 shows their measurements in a 0:15 0:15 m
2
unit
with two dierent outlet congurations. As outlets, a
608 pyramidal with gas exit at the top, and a cubic cap
with gas exit on one side, have been used.
5.3.2 Effect of Diameter
Only few studies of the inuence of scale on the
entrainment are available in the literature. Lewis et
al., (1962) made a study on entrainment with units of
0.019 to 0.146 m diameter. As a result, they stated that
entrainment is independent of size for units larger than
0.1 m in diameter. This result is supported by ndings
of Colakyan and Levenspiel (1984) and Tasirin and
Geldart (1998c). The latter authors measured with
units of 0.076 m and 0.152 m, respectively, and
found the entrainment in the larger unit to be higher
than in the smaller unit. Colakyan and Levenspiel
(1984) found no signicant inuence of the diameter
on the entrainment in beds of 0.1, 0.3, and 0.9 m
diameter.
Figure 5 Inuence of the supercial gas velocity on the elu-
triation rate, D
t
= 0:1 m, air at ambient conditions. (Data
measured by Choi et al., 1998.)
Figure 6 Inuence of height above the dense bed surface on
the entrainment ux for dierent gas velocities.
(D
t
= 0:45 m; d
p;50
= 260 mm, quartz sand, data measured
by Demmich, 1984.
Figure 7 Axial particle holdup distribution obtained with
dierent gas outlet congurations. (Nakagawa et al., 1994.)
While the above cited ndings are all for units with
constant diameter, Briens et al. (1990), Smolders and
Baeyens (1997), and Tasirin and Geldart (1998a)
studied the eect of an enlarged freeboard, i.e., of a
freeboard diameter larger than the uidized bed dia-
meter. In this case the gas velocity is reduced by the
square of the ratio of bed diameter D
bed
to freeboard
diameter D
FB
, which leads to a signicant reduction of
the entrainment ux by increasing the freeboard dia-
meter. Smolders and Baeyens (1997) found
E
o
o (D
bed
=D
FB
)
4
. Tasirin and Geldart (1998a)
found that the elutriation rate constant K
+
io
is only
determined by the gas velocity of the expanded section,
provided the expanded section is suciently high. If
there is not sucient length in the extended diameter
section to form a uniform gas velocity prole across
the whole cross-sectional area, the entrainment ux
will be higher than expected for the averaged velocity
in the expanded diameter section.
5.4 Inuence of Bed Height and Internals in the Bed
Taking into account the role of bubbles for the trans-
port of particles into the freeboard, an inuence of bed
height and also of internals in the uidized bed could
be expected. Baron et al. (1990) studied the inuence of
the bed height, using a uidized bed column of 0.61 m
diameter and silica sand with a surface mean diameter
d
p
of 65 mm. Their results for the entrainment ux E
o
above the TDH and the entrainment ux E
0
just above
the bed surface are shown in Figs. 8a and 8b, respec-
tively. The results show a slight increase of the entrain-
ment uxes with increasing bed height, the inuence
being more pronounced for the entrainment ux E
0
at the bed surface and higher gas velocities. The reason
for the increase of the entrainment ux is the increase
of bubble sizes with height, which obviously dominates
the decrease in bubble frequency and therefore in the
probability of bubbles to coalesce near the bed surface.
This indicates, that at least for the system studied by
Baron et al., the bubble nose mechanism is the dom-
inating mechanism for particle ejection into the free-
board. Choi et al. (1989), however, found no inuence
of bed height on entrainment during their study of
entrainment from uidized bed combustors. In this
study they used two uidized bed combustors with
cross-sectional areas of 0:3 0:3 m
2
and 1:01
0:83 m
2
, respectively. The smaller one was equipped
with an extended freeboard of 0:45 0:45 m
2
. In
fact, the bed height eect in Fig. 8 is small compared,
e.g., to the eect of the gas velocity.
Tweedle et al. (1970) showed by inserting screen
packings into the uidized bed that the entrainment
was reduced due to the reduced bubble size.
5.5 Inuence of Internals in the Freeboard
The picture of the inuence of internals in the free-
board is not very clear. On the one hand, they may
act as a kind of gassolid separators, thus helping to
reduce the TDH (Martini et al., 1976) and the entrain-
ment ux (Baron et al., 1988). On the other hand, they
may increase the entrainment due to the increased gas
velocity in the freeboard (Tweedle et al., 1970), or they
may not aect the entrainment at all (George and
Grace, 1981). Specially designed baes and inserts in
the freeboard may help to reduce the entrainment ow
signicantly (Harrison et al., 1974).
5.6 Inuence of Temperature and Pressure
Studies of the inuence of pressure have been carried
out by Chan and Knowlton (1984a, 1984b). They
Figure 8 Eect of the bed height on the entrainment ux (a)
above the TDH and (b) just above the bed surface.
(D
t
= 0:61 m, silica sand, d
p
= 61 mm, data measured by
Baron et al., 1990.)
showed that the TDH increases linearly with pressure
and that also the entrainment ux increases with pres-
sure (Fig. 9). These eects may directly be related to
the increasing density of the gas and thus decrease in
single particle terminal velocity. Eects of the changing
viscosity should not play any dominant role, since the
inuence of the pressure on viscosity is small in the
pressure range investigated.
Another important operating parameter is the tem-
perature, since most uidized bed applications are
operated at elevated temperatures, while most investi-
gations on entrainment are performed at ambient con-
ditions. Studies on the inuence of temperature are
reported by George and Grace (1981), Choi et al.
(1997), and Wouters and Geldart (1998). George and
Grace varied the temperatures in a quite narrow range
using silica sand as bed material and did not nd an
inuence of temperature on the entrainment. Choi et
al. investigated a much wider range of temperatures
and found a minimum of the entrainment for tempera-
tures of about 450 to 700K in the case of sand (Fig. 10).
Similar results were also obtained for emery
(r
s
= 3891 kg/m
3
) and for cast iron (r
s
= 6158 kg/m
3
)
particles as bed material. Wouters and Geldart mea-
sured a decrease of the elutriation rate constant with
increasing temperature in the range between 270 and
670K for the ne FCC used. They did not nd a mini-
mum in their plot of the entrainment rate vs. the tem-
perature.
6 MODELING
Considering the complex uid mechanics in the free-
board and at the bed surface of the bubbling bed, it is
not astonishing that up to now no rigorous model that
applies the methods of computational uid dynamics
(CFD) has been published in the literature. This would
be quite a dicult task, particularly because a very
wide range of solids volume concentrationsranging
from e
mf
in the suspension of the dense bed to nearly
zero above the TDHhas to be covered. Models of
the gassolid ow in the freeboard are therefore based
on either simplied physical descriptions or empirical
correlations.
6.1 Entrainment for Freeboard Heights Exceeding
TDH
As early as 1958 Zenz and Weil (1958) introduced the
idea that the freeboard above the TDH behaves like a
pneumatic transport line at choking conditions. This is
the assumption on which most of the models up to now
are based. The dierences are mainly in the calculation
of the choking load and of the particle size distribution
of the entrained particles. Zenz and Weil (1958) calcu-
lated for each particle size class contained in the bed
the choking load G
si
(kg/m
2
s) for monosize particles
of diameter d
pi
with a correlation originally developed
for pneumatic transport. The elutriation ux for these
particles is then assumed to be the product of the mass
fraction x
Bi
of these particles in the bed and the chok-
ing load G
si
:
E
io
= x
Bi
G
si
(9)
This implies that, for all particle size classes, at least a
mass ux equal to the elutriation ux E
io
is ejected
from the uidized bed into the freeboard.
This choking load model has been modied by
Gugnoni and Zenz (1980). They proceeded in two
stages. First, they estimated the particle size distribu-
tion by using the Zenz and Weil model with a some-
Figure 9 Inuence of pressure on the total entrainment
rates. (D
t
= 0:3 m, sand, d
p
= 70 mm, data measured by
Chan and Knowlton, 1984a.)
Figure 10 Inuence of temperatures on the elutriation rate
constant. (Sand, data measured by Choi et al., 1997.)
what updated choking correlation. In a second step,
the total ux of solids entrained above the TDH was
then calculated with a correlation based on the calcu-
lated mean diameter of the entrained solids and the
dierence between the gas velocity and the minimum
uidizing velocity.
Briens and Bergougnou (1986) also used a two-step
approach. They again assume that for each particle
size the ow above the TDH is limited by choking,
i.e., it cannot be larger than the choking load attribu-
ted to that specic particle size. Furthermore, the ux
of each particle size class above the TDH cannot be
larger than the ux of solids of this size ejected from
the bed to the freeboard. While these assumptions are
sucient to calculate the entrainment from a monosize
uidized bed, there is one more assumption necessary
to calculate the fraction of the choking load attributed
to each specic particle size; for this purpose they
assumed that the size distribution of the entrained
material above the TDH will adjust itself so that the
total entrainment ux reaches a maximum. To solve
this model, correlations for the choking load and for
the entrainment of particles from the bed into the free-
board are needed. The model then has to be solved
iteratively.
Models that focus more on the entrainment of the
particles from the dense bed to the freeboard are given
by George and Grace (1978) and by Smolders and
Baeyens (1997). Both models rst calculate the entrain-
ment ux E
i0
at the bed surface based on such bubble
properties as size, frequency, and velocity. For the ux
above the TDH, it is simply assumed that only parti-
cles with a terminal velocity less than the gas velocity
in the freeboard are entrained, the mass ux of these
particles being the same as at the bed surface:
E
io
=
E
i0
for U
ti
< U
FB
0 for U
ti
_ U
FB
_
(10)
These two models will give the same entrainment ux
at the bed surface as above the TDH for uidized beds
operated with solids having all terminal velocities less
than the supercial velocity. This latter nding is in
contradiction with empirical ndings which show a
decay in the entrainment with height even for such
uidized beds of nes.
Besides the models described above there is quite a
large number of empirical correlations for the elutria-
tion rate constant K
+
io
above the transport disengaging
height. A selection of the most frequently cited and
most recent correlations is given in Table 1.
Comments with regard to special areas of application
are also given in this table. All the correlations have
been converted so that the parameters are to be
inserted in SI units, although the original correlations
may have required the insertion of data in other units.
In Fig. 11, numerical values for the elutriation constant
K
+
io
predicted by the dierent correlations are plotted
vs. the gas velocity and compared with experimental
data for sand of about 90 mm measured by Colakyan
and Levenspiel (1984), Choi et al. (1998), and George
and Grace (1981). It can be noticed that there are
signicant dierences in magnitude and predicted ten-
dencies, but also the experimental data show a consid-
erable scattering. In Fig. 12, in addition, the elutriation
rate constant is plotted vs. the particle diameter. Also
in this gure, signicant deviations between the corre-
lations can be noticed even in the tendencies.
Both Fig. 11 and Fig. 12 could lead to the conclu-
sion that it is rather hopeless to predict the entrain-
ment from uidized beds with a reasonable degree of
accuracy. However, we see also that a group of corre-
lations is not too far away from the experimental
results. In a given situation one should always carefully
select one or more correlations, which have been devel-
oped on a data basis, which includes the situation
under consideration. For example, one should cer-
tainly not expect a correlation developed for ne par-
ticle systems to hold for a coarse particle system. The
gas velocity, the type of solids, and particularly the
particle size distribution should be for a given applica-
tion within the range of tested parameters of the
selected correlation. In any case, the results of such
correlations should be treated with care: an uncertainty
of plus/minus 100% should be taken into account.
6.2 Estimation of TDH
An important parameter for the design of a uidized
bed vessel is the TDH. To reduce the carryover from a
uidized bed, the freeboard should have, wherever pos-
sible, a height of at least the TDH. On the other hand,
with respect to entrainment there is no additional
advantage of increasing the height of the vessel beyond
the TDH. As already mentioned, the denition of the
TDH is not very sharp, since there is only a gradual
decay of the entrainment ux vs. the height at the level
of TDH. Again, there is no commonly accepted
method for the calculation of the TDH, but just several
empirical correlations. Some of these correlations are
listed in Table 2. As can be seen from this table, the
inuencing parameters dier between the correlations.
While some of them take only the supercial velocity
into account, without any regard to the solids in the
bed, other correlations depend on one or more of the
Table 1 Correlations for the Elutriation Constant K
+
io
U, m/s D, m
d
p;bed
mm Reference Comments
1 K
+
io
g d
2
pi
m(U u
ti
)
2
= 0:0015 Re
0:6
t
0:01 Re
1:2
t
0.31.0 0.071.0 0.11.6 Yagi and Aochi
(1955) as cited
by Wen and
Chen (1982)
Cited by dierent authors
in dierent ways,
original publication
inaccessible
2
K
+
io
r
g
U
=
1:26 10
7
U
2
g d
pi
r
2
p
_ _
1:88
for
U
2
g d
pi
r
2
p
< 3:10
1:31 10
4
U
2
g d
pi
r
2
p
_ _
1:18
for
U
2
g d
pi
r
2
p
> 3:10
_
_
0.30.7 0:05 0:53 0.040.2 Zenz and Weil
(1958)
Correlation aiming at FCC
uidized beds
3
K
+
io
r
g
(U U
ti
)
= 1:52 10
5
(U U
ti
)
2
g d
pi
_ _
0:5
Re
0:725
i
0.61.0 0.102 0.7 Wen and
Hashinger (1960)
4
K
+
io
r
g
(U U
ti
)
= 4:6 10
2
(U U
ti
)
2
g d
pi
_ _
0:5
Re
0:3
t

r
s
r
g
r
g
_ _
0:15
0.92.8 0.0310.067 0.71.9 Tanaka et al.
(1972)
5
K
+
io
r
g
U
= A 130 exp 10:4
u
ti
U
_ _
0:5
U
mf
U U
mf
_ _
0:25
_ _
with A = 10
3
. . . 10
4
0.62.4 0:91 0:91 0.061.0 Merrick and
Highley (1974)
Correlation derived for
bubbling uidized bed
combustors
6
K
+
io
r
g
U
= 23:7 exp 5:4
u
ti
U
_ _
0.63.0 0.076
0.3
0.060.35
1.5
Geldart et al.
(1979)
7
K
+
io
r
g
U
= 9:43 10
4
U
2
g d
p
_ _
1:65
for 58 _
U
2
g d
pi
_ _
_ 1000; 0:1m=s _ U _ 0:3 m=s;
0 < d
p
_ 74 mm
0.10.3 0:61 0:61 00.125 Lin et al. (1980)
8
K
+
io
kg
m
2
s
_ _
= 0:011 r
s
1
u
ti
U
_ _
2
0.93.7
0:92 0:92
0:3 0:3
0.31.0 Colakyan et al.
(1981), Colakyan
and Levenspiel
(1984)
Focus on Geldart Group B
and D particles
9
K
+
io
kg
m
2
s
_ _
= 2:8 10
2
U u
ti
U
_ _
1:6
r
s
r
g
r
s
_ _
0:54U
2:1
t
D
h
Kato et al. (1985)
10
K
+
io
r
g
U
= 1:6
U
u
ti
_ _
1
u
ti
U
_ _
Sciazko et al
(1991)
11
K
+
io
kg
m
2
s
_ _
= 5:4 10
5
r
s
U
0:2
_ _
3:4
1
u
ti
U
_ _
2
for d
pi
_
10325
r
0:725
s
0.20.7 0.030.78 Baeyens et al
(1992)
Correlation takes cohesive
forces into account and
is focused on supernes
in group A and C
systems
12
K
+
io
kg
m
2
s
_ _
= 0:35r
s
U(1 e)
H
with
(1 e)
H
= 7:41 10
3
R
1:87
A
0:55
t
H
0:64
FB
and R =
x
i
U U
ti
U
ti
_ _
for u
ti
< U
0.10.6 0.071
0:08 0:08
0:15 0:15
0.030.2 Nakagawa et al.
(1994)
13
K
+
io
kg
m
2
s
_ _
=
23:7 r
g
U
2:5
exp 5:4
u
ti
U
_ _
for Re < 3000
14:5 r
g
U
2:5
exp
_
5:4
u
ti
U
_
for Re > 3000
_
_
_
with Re =
D
t
U
n
0.20.8 0.076, 0.152 0.0170.077 Tasirin and
Geldart (1998c)
14
K
+
io
d
p
m
= Ar
0:5
exp 6:92 2:11F
0:303
g

13:1
F
0:902
d
_ _
with F
g
= g d
p
(r
s
r
g
) (gravity force per projection area)
F
d
= C
d
r
g
U
2
2
(drag force per projection area)
0.37.0 0.061.0 0.051.0 Choi et al. (1999) Correlation based on a
wide range of dierent
units, materials, and
operating conditions,
e.g., temperature and
pressure
All parameters to be inserted in SI units.
following parameters: the bubble diameter, the column
diameter, and solids and gas properties.
For uidized beds with FCC or comparable solids,
Zenz and Othmer (1960) have given a graphical repre-
sentation of the TDH vs. the uidizing velocity U with
the bed diameter as a parameter (Fig. 13). Here a
signicant inuence of the bed diameter can be
recognized. The ZenzOthmer diagram is based on
much industrial expertise and it is certainly helpful as
a rst guess.
6.3 Entrainment for Freeboard Heights Below TDH
To describe exactly the processes below the TDH, it is
necessary to describe the formation and disintegration
of clusters or strands, the acceleration and deceleration
of clusters, the turbulent structures of the gas ow, and
the interaction between the solids and the gas turbu-
lence. All these eects are inuenced by the processes
inside or at the surface of the dense bed. The
approaches published up to now reduce the complexity
by reducing the ow either to three phases (Kunii and
Levenspiel, 1989) or to two entrainment ows in par-
Figure 11 Comparison of predicted elutriation rate con-
stants with measurements by Colakyan and Levenspiel
(1984), Choi et al. (1998), and George and Grace (1981).
Sand, d
p
= 90 mm, for correlation 13 D = 0.6 m was
assumed, and the terminal velocity was calculated with the
correlation for the drag coecient by Kaskas and Brauer as
cited by Brauer (1971). The numbers denote the various cor-
relations in Table 1.
Figure 12 Comparison of dierent correlations for the elu-
triation rate constantinuence of particle size. Sand, U =
1 m/s, for correlation 13 D=0.6 mwas assumed, the terminal
velocity was calculated with the correlation for the drag coef-
cient by Kaskas and Brauer as cited by Brauer (1971). The
numbers denote the various correlations in Table 1.
Table 2 Correlations for the Calculation of the Transport Disengaging
Height (TDH)
Equation Reference
1 TDH = 0:85U
1:2
(7:33 1:2 log U) Chan and Knowlton (1984b)
2 TDH = 1000
U
2
g
Fournol et al. (1973)
3 TDH = 18:2d
b
George and Grace (1978)
4 TDH = 13:8d
b
Fung and Hamdullahpur (1993)
5
TDH =
1500H
b
Re
p
Ar
Sciazko et al. (1991)
allel (Large et al., 1976). Since both models leadat
least from the practical point of viewto similar
results, we will focus here on the latter one. Large et
al. assume that the entrainment ux E
ih
for a given
component i consists of two uxes in parallel, one
ux E
io
owing continuously from the bed to the out-
let and a second ux of clusters or strands ejected from
the bed, which decreases exponentially with increasing
height h above the bed surface:
E
ih
= E
io
E
i0
exp(a
i
h) (11)
or
E
ih
= x
Bi
K
+
io
E
i0
exp(a
i
h) (12)
with E
i0
being the ux of component i ejected from the
bed to the surface.
To calculate the entrainment for freeboard heights
below the TDH we need now in addition to the elutria-
tion rate constant K
+
io
the entrainment ux at the bed
surface and the decay constant a
i
. For the decay con-
stant the value varies according to Wen and Chen
(1982) between 3.5 and 6.4 m
1
. They recommend a
constant value of a = 4:0 m
1
for systems for which
no detailed information is available.
The entrainment at the bed surface E
i0
can be
recalculated from entrainment measurements at two
dierent heights, if such data are available. If no
experimental data are available a correlation for the
entrainment at the bed surface has to be used. Some
of the correlations published in the literature are given
in Table 3.
NOMENCLATURE
a = Decay constant dened by Eq. (11), m
1
A = Cross-sectional area of the bed, m
2
d
b
= Diameter of a bubble, m
d
p
= Particle size, m
D = Diameter of the uidized bed column, m
D
FB
= Diameter of the column in the freeboard region, m
D
h
= Hydraulic diameter of the uidized bed column
(4A/circumference), m
E
o
= Total entrainment ux above TDH, kg/(m
2
s)
Table 3 Correlations for the Calculation of the Entrainment Flux E
0
at the Bed Surface
Equation Reference
1
E
0
= 3d
p
(1 "
mf
)(U U
mf
)
d
b
bubble nose model Pemberton and Davidson (1986)
E
0
= 0:1 r
s
(1 "
mf
)(U U
mf
) bubble wake model
2
E
0
= 3:07 10
9
A d
b
(U U
mf
)
2:5
r
3:5
g
g
0:5
m
2:5
Wen and Chen (1982)
3 E
0
= 9:6 A (U U
mf
)
2:5
A d
b
298
T
_ _
3:5
(T in kelvins) Choi et al. (1989)
Figure 13 Transport disengaging height vs. excess gas velo-
city U U
mf
for particles of the FCC catalyst type. The
parameter is the bed diameter. The freeboard is assumed to
have the same diameter as the bed. (After Zenz and Othmer,
1960.
E
0
= Total entrainment ux at the surface of the
uidized bed, kg/(m
2
s)
E
h
= Total entrainment ux at height h, kg/(m
2
s)
E
io
= Entrainment ux of component i above TDH, kg/
(m
2
s)
E
i0
= Entrainment ux of component i at the surface of
the uidized bed, kg/(m
2
s)
E
ih
= Entrainment ux of component i at height h, kg/
(m
2
s)
h = Height above the distributor, m
K
+
io
= Elutriation rate constant E
io
x
Bi
above TDH, kg/
(m
2
s)
K
+
ih
= Elutriation rate constant E
ih
X
Bi
at height h, kg/
(m
2
s)
M
b
= Mass of solids in the uidized bed, kg
Re = Reynolds number Re =
UDr
g
m
, -
Re
t
= Particle Reynolds number dened by Re
t
=
m
t
d
p
r
g
m
,-
t = Time, s
T = Temperature, K
U = Supercial velocity, m/s
U
mf
= Minimum uidizing velocity, m/s
u
t
= Terminal velocity of a single particle, m/s
x
Bi
= Mass fraction of particles of component i in the
uidized bed, -
Greek symbols
r
g
= Gas density, kg/m
3
r
s
= Solids density, kg/m
3
m = Gas viscosity, Pa s
REFERENCES
Bachovchin DM, Beer JM, Sarom AF. An investigation
into the steady-state elutriation of nes from a uidized
bed. AIChE Symp. Ser. 77:7685, 1981.
Baeyens J, Geldart D, Wu SY. Elutriation of nes from gas
uidized bed of Geldart A-type powderseect of adding
supernes. Powder Technology 71:7180, 1992.
Baron T, Briens CL, Bergougnou MA. Reduction of particle
entrainment from gas-uidized beds with a screen of oat-
ing balls. AIChE Symp. Ser. 84:5057, 1988.
Baron T, Briens CL, Galtier P, Bergougnou MA. Eect of
bed height on particle entrainment from gas-uidized
beds. Powder Technology 63:149156, 1990.
Brauer H. Grundlagen der Einphasen- und Mehrphasen-
stro mungen. Frankfurt a. Main: Verlag Sauerla nder,
1971, pp 196200.
Briens CL, Bergougnou MA. A new model for entrainment
from uidized beds. AIChE Journal 32:233238, 1986.
Briens CL, Bergougnou MA, Barton T. Reduction of particle
entrainment from gas-uidized beds. Prediction of the
eect of disengagement zones. Powder Technology 62:
135138, 1990.
Briens CL, Bergougnou MA, Inculet II, Baron T, Hazlett
JD. Size distribution of particles entrained from uidized
beds: electrostatic eects. Powder Technology 70:5762,
1992.
Chan IH, Knowlton TM. The eect of pressure on entrain-
ment from bubbling gas uidized beds. In: Kunii D, Toei
R, eds. Fluidization. New York: Engineering Foundation,
1984a pp 283290.
Chan IH, Knowlton TM. The eect of system pressure on
the transport disengaging height above bubbling uidized
beds. AIChE Symp. Ser. 80:2433, 1984b.
Chen TP, Saxena SC. A theory of solids projection from a
uidized bed surface as a rst step in the analysis of
entrainment process. In: Davidson JF, Keairns DL, eds.
Fluidization II. London: Cambridge University Press,
1978, pp 151159.
Choi J-H, Son JE, Kim S-D. Solid entrainment in uidized
bed combustors. Journal of Chemical Engineering of
Japan 22:597606, 1989.
Choi J-H, Choi K-B, Kim P, Shun D-W, Kim S-D. The eect
of temperature on the particle entrainment rate in a gas
uidized bed. Powder Technology 92:127, 1997.
Choi J-H, Ryu HJ, Shun D-W, Son JE, Kim S-D.
Temperature eect on the particle entrainment rate in a
gas uidized bed. Ind. Eng. Chem. Res. 37:1130, 1998.
Choi J-H, Chang I-Y, Shun D-W, Yi C-K, Son J-E, Kim
S-D. Correlation on the particle entrainment rate in gas
uidized beds. Ind. Eng. Chem. Res 38:24912496, 1999.
Colakyan M, Levenspiel O. Elutriation from uidized beds.
Powder Technology 38:223232, 1984.
Colakyan M, Catipovic, N Jovanovic G, Fitzgerald T.
AIChE Symp. Ser. 77:6675, 1981.
Demmich J. Mechanisms of solids entrainment from ui-
dized beds. Ger. Chem. Eng. 7:386394, 1984.
Do HT, Grace JR, Clift R. Particle ejection and entrainment
from uidised beds. Powder Technology 6:195200, 1972.
Fournol AB, Bergougnou MA, Baker CGJ. Solids entrain-
ment in a large gas uidized beds. Canadian Journal of
Chemical Engineering 51:401404, 1973.
Fung AS, Hamdullahpur F. Eect of bubble coalescence on
entrainment in gas uidized beds. Powder Technology
77:251265, 1993.
Geldart D, Pope D. Interaction of ne and coarse particles in
the freeboard of a uidized bed. Powder Technology
34:9597, 1983.
Geldart D, Cullinan J, Georghiades S, Gilvray S, Pope DJ.
The eect of nes on entrainment from gas uidized beds.
Trans. Inst. Chem. Eng. 57:269275, 1979.
George SE, Grace JR. Entrainment of particles from aggre-
gative uidized beds. AIChE Symp. Ser. 74:6774, 1978.
George SE, Grace JR. Entrainment of particles from a pilot
scale uidized bed. Canadian Journal of Chemical
Engineering 59:279284, 1981.
Gugnoni RJ, Zenz Fa. In: Grace JR, Matsen JN, eds.
Fluidization. New York: Plenum Press. 1980, pp 501
508.
Harrison D, Aspinall PN, Elder J. Suppression of particle
elutriation from a uidised bed. Trans. Inst. Chem.
Engrs. 52:213216, 1974.
Hatano H, Ishida M. Study of the entrainment of FCC par-
ticles from a uidized bed. Powder Technology 35:201
209, 1983.
Hazlett JD, Bergougnou MA. Inuence of bubble size dis-
tribution at the bed surface on entrainment prole.
Kato K, Tajima T, Mao M, Iwamoto H. In: Kwauk M,
Kunii D, Zheng J, Hasatani M, eds. Fluidization 85
Science and Technology Amsterdam: Elsevier. 1985, pp
134147.
Kunii D, Levenspiel O. Entrainment of solids from uidized
beds I. Hold-up of solids in the freeboard; II. Operation
of Fast Fluidized Beds. Powder Technology 61:193206,
1989.
Large JF, Martinie Y, Bergougnou MA. Interpretative
model for entrainment in a large gas uidized bed.
International Powder Bulk Solids Handling and
Processing Conference, Chicago, 1976.
Levy EK, Caram HS, Dille JC, Edelstein S. Mechanisms for
solids-ejection from gas-uidized beds. AIChE Journal
29:383388, 1983.
Lewis WK, Gilliland ER, Lang PM. Entrainment from ui-
dized beds. Chem. Eng. Prog. Symp. Ser. 58:6572, 1962.
Lin L, Sears JT, Wen CY. Elutriation and attrition of char
from a large uidized bed. Powder Technology 27:105
115, 1980.
Ma X, Kato K. Eect of interparticle adhesion forces on
elutration of ne powders from a uidized bed of binary
particle mixture. Powder Technology 95:93101, 1998.
Martini Y, Bergougnou Ma, Baker CGJ. Entrainment reduc-
tion by louvers in the dilute phase of a large gas uidized
bed. In: Keairns DL, ed. Fluidization Technology, Vol. II.
1976, pp 2940.
Merrick D, Highley J. Particle size reduction and elutriation
in a uidized bed process. AIChE Symp. Ser. 70:366378,
1974.
Nakagawa N, Arita S, Uchida S, Takamura H, Takarada T,
Kato K. Particle hold-up and elutriation rate in the free-
board of uid beds. J. Chem. Eng. Japan. 27:7984, 1994.
Pemberton ST, Davidson JF. Elutration of ne particles
from bubbling uidized beds. In: Kunii D, Toei R, eds.
Fluidization IV. New York: Engineering Foundation,
1983, pp 275282.
Pemberton ST, Davidson JF. Turbulence in the freeboard of
a gas uidized bed, the signicance of ghost bubbles.
Chem. Eng. Sci 39:829840, 1984.
Pemberton ST, Davidson JF. Elutriation from uidized
bedsI. Particle ejection from the dense phase into the
freeboard. Chem. Eng. Sci. 41:243251, 1986.
Peters MH, Fran L-S, Sweeney TL. Study of particle ejec-
tions in the freeboard region of a uidized bed with an
image carrying probe. Chem. Eng. Sci 38:485487, 1983.
Santana D, Rodr guez JM, Mac as-Mach n. A. Modelling
uidized bed elutriation of ne particles. Powder
Technology 106:110118, 1999.
Sciazko M, Bandrowski J, Raczek J. On the entrainment of
solid particles from a uidized bed. Powder Technology
66:3339, 1991.
Smolders K, Baeyens J. Elutriation of lines from gas uidized
beds: mechanisms of elutriation and eect of bed geo-
metry. Powder Technology 92:3546, 1997.
Tanaka I, Shinohara H, Hirose H, Tanaka Y. Elutriation of
nes from uidized bed. J. Chem. Eng. Japan 5:5762,
1972.
Taneda D, Takahagi H, Aoshika S, Nakagawa N, Kato K.
Eects of the powder properties upon the hold-up and the
elutriation rate of ne particles in a powderparticle ui-
dized bed. In: Fan L-S, Knowlton TM, eds. Fluidization
IX. New York: Engineering Foundation, 1998, pp 317
324.
Tanimoto H, Chiba T, Kobayashi H. The mechanism of
particle ejection from the surface of a gas-uidized bed.
Int. Chem. Engng. 24:679685, 1984.
Tasirin SM, Geldart D. Entrainment of nes from uidized
beds with expanded freeboard. Chem. Eng. Commun.
166:217230, 1998.
Tasirin SM, Geldart D. The entrainment of nes and super-
nes from uidized beds. Powder Handling and
Processing 10:263268, 1998b.
Tasirin SM, Geldart D. Entrainment of FCC from uidized
bedsa new correlation for the elutriation rate constants
K
+
io
. Powder Technology 95:240247, 1998c.
Tasirin SM, Geldart D. Separation performance of batch
uidized bed air classication. Powder Handling and
Processing 12:3946, 2000.
Tasirin SM, Anuar N, Rodzi F. Entrainment of nes (group
C particles) from uidized beds. In: Kwauk M, Li J, Yang
WC, eds. Fluidization X. New York: Engineering
Tweedle TA, Capes CE, Osberg GL. Eect of screen packing
on entrainment from uidized beds. Ind. Eng. Chem.
Process Des. Development 9:8588, 1970.
Wen CY, Chen LH. Fluidized bed freeboard phenomena:
entrainment and elutriation. AIChE Journal 28:117128,
1982.
Wen CY, Hashinger RF. Elutriation of solid particles from a
dense phase uidized bed. AIChE Journal 6:220226,
1960.
Wouters IM., Geldart D. Entrainment at high temperatures.
In: Fan L-S, Knowlton TM, eds. Fluidization IX. New
York: Engineering Foundation, 1998, pp 341348.
Yagi S, Aochi T. Paper presented to the Society of Chemical
Engineers of Japan, Spring Meeting, 1955.
Yorquez-Ramirez MJ, Duursma GR. Study of the ow pat-
tern above an erupting bubble in an incipiently uidized
bed using image shifting. Chem. Eng. Sci 55:20552064,
2000.
Yorquez-Ramirez MJ, Duursma GR. Insights into the
instantaneous freeboard ow above a bubbling uidised
bed. Powder Technology 116:7684, 2001.
Zenz FA, Othmer DF. Fluidization and Fluid-Particle
Systems. New York: Reinhold, 1960.
Zenz PA, Weil NA. A theoreticalempirical approach to the
mechanism of particle entrainment from uidized beds.
AIChE Journal 4:472479, 1958.
5
Effect of Temperature and Pressure
J. G. Yates
University College London, London, United Kingdom
1 INTRODUCTION
Practically all industrial gassolid uidized bed reac-
tors operate at temperatures well above ambient, and
some, such as those used in the production of polyole-
ns, also operate at elevated pressures. It is therefore
important to know how uidized beds behave under
high temperatures and/or pressures and if possible to
predict this behavior from observations made under
ambient conditions. The emphasis here will be on
those aspects of the subject that are of direct relevance
to the design and operation of uidized bed plant; a
comprehensive review of the more academic aspects
can be found in the review by Yates (1996).
As is made clear in other parts of this publication,
solid particles in uidized beds are held in suspension
by the upward ow of gas. The velocity at which the
particles rst become suspended is the minimum ui-
dization velocity, u
mf
, and as the gas velocity through
the bed is increased, the particles pass through a num-
ber of ow regimes characterized as bubbling, turbu-
lent, and fast. The transition velocities between one
regime and another, as well as the terminal fall velocity
of single particles, u
t
, are all inuenced by the physical
properties of the gas, and these in turn are inuenced
by the temperature and pressure under which the bed is
operated. Changes in the surface properties of the ui-
dized solids, particularly with increasing temperature,
can also have an important inuence on bed behavior.
These and other factors will be considered in detail
below.
The review will start by considering the two veloci-
ties at the extremes of the ow regimes, i.e. u
mf
and u
t
.
2 MINIMUM FLUIDIZATION VELOCITY
This may be found by measuring the pressure drop
through a bed of particles as a function of the gas
velocity. The frictional pressure drop, p, through a
packed bed of monodispersed spheres, is given by the
Ergun equation:
p
L
=
150(1 e)
2
e
3
mu
(fd
p
)
2
1:75(1 e)
e
3
r
f
u
2
fd
p
(1)
where u is the uid supercial velocity, L is the bed
height, d
p
is particle diameter, f is the particle spheri-
city, m is the uid viscosity, r
f
is the uid density, and e
is the voidage of the bed. The rst term on the right
hand side of Eq. (1) represents the pressure loss
through viscous eects and is the dominant term in
the laminar ow region; the second term is the loss
due to inertial forces and will be dominant at high
Reynolds numbers.
At u
mf
the weight of the bed is fully supported by
the ow of gas, and the pressure drop through the bed
is then equal to the bed weight per unit area:
p
L
= 1 e
mf
( ) r
p
r
f
_ _
g (2)
where e
mf
is the minimum uidization voidage and
r
p
is the particle density. Combining (1) and (2) and
multiplying through by d
3
p
r
f
=(1 e
mf
)m
2
:
Ga = 150
(1 e
mf
)
f
2
e
3
mf
Re
mf

1:75
fe
3
mf
Re
2
mf
(3)
where
Ga =
d
3
p
r
f
(r
p
r
f
)g
m
2
; Re
mf
=
d
p
u
mf
r
f
m
(4)
Wen and Yu (1966) showed that the voidage and shape
factor terms in Eq. (3) could be correlated with much
experimental data on the basis of the two approxima-
tions
(1 e
mf
)
f
2
e
3
mf
- 11
1
fe
3
mf
- 14 (5)
They lead to a modied form of Eq. (3):
Ga = 1650 Re
mf
24:5 Re
2
mf
(6)
which may be rearranged to
Re
mf
=
(C
1
)
2
C
2
Ga
_
C
1
(7)
The two constants C
1
and C
2
in the original Wen and
Yu correlation are 33.7 and 0.0408, respectively, so
that, knowing the eects of temperature and pressure
on the gas density and viscosity terms contained in Eq.
(4), u
mf
can be calculated over any range of these para-
meters. However, Yang et al. (1985) pointed out that
several other sets of constants have been proposed by
other workers, making, Eq. (6) unreliable for predict-
ing u
mf
at elevated temperatures and pressures. Yang et
al. (1985) also pointed out that the Wen and Yu cor-
relation is nothing more than a relationship between
Reynolds number, Re, and drag coecient, C
D
, since
for a single spherical particle
C
D
pd
2
p
4
r
f
u
2
2
=
pd
3
p
6
r
p
r
f
_ _
g (8)
or
C
D
=
4
3
d
p
r
p
r
f
_ _
g
r
f
u
2
(9)
and
Ga =
3
4
Re
2
C
D
(10)
The Wen and Yu correlation [Eq. (7)] can be re-
arranged to give
C
D
Re C
1
( )
2
C
2
1
0:75C
2
Re
2
(11)
showing the relationship between Re and C
D
.
Yang et al. (1985) proposed using the methodology
developed by Barnea and Mizrahi (1973) to obtain
values for u
mf
at elevated temperatures and pressures.
These two workers proposed a general correlation for
pressure drop through xed beds of spherical particles
based on a discrete particle model corrected for parti-
cle interaction. They incorporated a voidage term into
a standard C
D
vs. Re plot for a single sphere to give the
following modied values for a multiparticle system:
(Re)
e
= Re
1
e exp 5(1 e)=3e [ ]
_ _
(12)
(C
D
)
e
= C
D
e
3
1 (1 e)
1=3
(13)
These correlations were extended by Barnea and
Mednick (1975) to calculate u
mf
. They suggested plot-
ting the dimensionless diameter (Re
2
C
D
)
1=3
mf
against
the dimensionless velocity (Re=C
D
)
1=3
mf
which at the
point of minimum uidization can be expressed as
Re)
e
(C
D
)
e
_ _
1=3
mf
=
Re
C
D
_ _
1=3
mf
1 1 e
1=3
mf
_ _ _ _
1=3
e
4=3
mf
exp 5 1 e
mf
( )=9e
mf
( )
_
_
_
_
(14)
Yang et al. (1985) combined Eq. (14) with the standard
drag correlations recommended by Clift et al. (1978)
(Table 1) and found good agreement with experimental
values obtained at pressures of up to 63 bar and tem-
peratures of up to 900
C.
A qualitative appreciation of the way u
mf
is aected
by elevated pressures was given by Rowe (1984), who
rearranged Eq. (3) to express u
mf
in terms of the
operating variables as:
u
mf
=
m
r
f
d
p
42:9 1 e
mf
( )
1 3:0 10
4
e
3
mf
1 e
mf
( )
2
Ga
_ _
1=2
1
_
_
_
_
_
_
(15)
Rowe (1984) used Eq. (15) to show how pressure
would be expected to aect u
mf
for particles with a
density of 1250 kg=m
3
and a range of sizes (Fig. 1).
In these calculations the minimum uidization voidage
was set at 0.5 and the shape factor at unity; it is clear
that for particles below about 100 mm in diameter
[Group A in the Geldart (1973) classication], pressure
is predicted to have little if any eect on u
mf
. The
reason for this is that the ow of gas around these
small particles is laminar and so the uidparticle
interaction force is dominated by the gas viscosity
(which is essentially independent of pressure in the
range considered); the Galileo number, Ga, is a linear
function of pressure. As particle size increases, how-
ever, inertial forces become more important and at d
p
> 500 mm (Geldart Group B) they begin to dominate
over the viscous forces; this causes u
mf
to decrease
sharply with pressure up to about 20 bar and more
gradually thereafter. Similar conclusions had been
reached by King and Harrison (1982) who showed
that on the basis of Eq. (3) u
mf
is independent of pres-
sure for laminar ow (Re
mf
< 0:5) while for turbulent
ow (Re
mf
> 500) u
mf
is inversely proportional to the
square root of gas density and hence pressure; pressure
dependence would be expected to be weaker in the
intermediate regime.
Olowson and Almstedt (1991) measured the u
mf
of a
range of particles in Geldart Groups B and D at pres-
sures from 0.1 to 0.6 MPa and found a general
decrease with increasing pressure. They found the
eect of pressure to be well described by the Ergun
equation and a number of simplied correlations
derived from it, although the accuracy of these correla-
tions diered signicantly.
The eect of temperature on u
mf
may also be seen
by examination of Eq. (15). Since the density of a gas is
inversely proportional to its absolute temperature, gas
density will decrease with increasing temperature; the
viscosity of a gas on the other hand increases with
increasing temperature m
g
being proportional to T
n
where n is usually between 0.5 and 1.0. The combined
eect of changes in density and viscosity results in the
Galileo number decreasing steadily with increasing
temperature, but it is not immediately obvious from
Eq. (15) how this will aect u
mf
. Figure 2 shows Eq.
(15) plotted with a square root dependence of gas
Table 1 Recommended Drag Correlations. w = log
10
Re
Range Correlation
(A) Re < 0:01 C
D
= 3=16 24=Re
(B) 0:01 < Re _ 20
log
10
C
D
Re
24
1
_ _
= 0:881 0:82w 0:05w
2
i.e., C
D
=
24
Re
1 0:1315 Re
(0:820:05w)
_ _
(C) 20 < Re _ 260
log
10
C
D
Re
24
1
_ _
= 0:7133 0:6305w
i.e., C
D
=
24
Re
1 0:1935 Re
0:6305
_ _
(D) 260 _ Re _ 1500 log
10
C
D
= 1:6435 1:1242w 0:1558w
2
(E) 1:5 10
3
_ Re _ 1:2 10
4
log
10
C
D
= 2:4571 2:5558w 0:9295w
2
0:1049w
2
(F) 1:2 10
4
< Re < 4:4 10
4
log
10
C
D
= 1:9181 0:6370w 0:0636w
2
Source: Clift et al., 1978.
Figure 1 Eect of pressure on u
mf
, based on Eq. (15). (From
Rowe, 1984.)
viscosity on temperature, and it is clear that a rise in
temperature causes the u
mf
of Group A powders to
decrease, although the eect is relatively modest. For
Group B materials, the eect is rst to raise the value
through the gas density term and then to cause a
decrease as the viscous term starts to dominate.
Experimental verication of these trends comes from
the work of Botterill and Teoman (1980), Svoboda and
Hartman (1981), Botterill et al. (1982), Wu and
Baeyens (1991) and Raso et al. (1992); all authors,
however, stress the importance of choosing the correct
value of the emulsion phase voidage, e
mf
, in using the
Ergun equation for these predictions.
Recent work by Formisani et al. (1998) has shown
the eect of increasing temperature on the voidage of
xed and uidized beds, e
0
and e
mf
respectively. They
studied beds of silica sand, glass ballotini, salt, and
uidized cracking catalyst (FCC) and found for all
an increase in voidage with increasing temperature,
the dependence being expressed in the linear form by
e
T
= e
amb
k T T
amb
( ) (16)
where the subscripts refer to temperature and ambient
respectively and the parameter k is a function of par-
ticle properties. The authors conclude that the eect is
due to the variation with temperature of interparticle
forces, which runs parallel to the thermal variation of
gas properties; the authors demonstrate that the pre-
dictive ability of the Ergun equation is only restored if
the dependence of e
mf
is correctly accounted for.
Yamazaki et al. (1995) studied the eect of chemi-
sorbed water on the minimum uidization voidage of
silica particles at high temperatures and concluded that
e
mf
decreases as humidity increases, the cause being the
reduction in van der Waals forces of wet surfaces due
to the lower value of their Hamaker constant.
The eects of temperature and pressure on e
mf
have
also been considered from a theoretical standpoint by
Rowe (1989).
3 TERMINAL FALL VELOCITY OF SINGLE
PARTICLES
The force acting on a single particle falling through a
quiescent uid is
F = M
du
dt
(17)
= (particle weight) (drag force)
=
pd
3
p
6
(r
p
r
f
)g C
D
pd
2
p
4
r
f
u
2
2
(18)
At u
t
du
dt
= 0
Hence
u
t
=
4d
p
r
p
r
f
_ _
g
3r
f
C
D
_ _
1=2
(19)
where C
D
is, again, an empirical drag coecient and a
function of the Reynolds number (see Table 1). The
form of this function has been a frequent subject of
investigation, and work in this area has been reviewed
by Khan and Richardson (1987) and by Hartman and
Yates (1993). The relationship is complicated by such
factors as the ow regime (laminar, turbulent, or
intermediate) and the size, density, and shape of the
particles involved. Haider and Levenspiel (1989)
devised a method by which u
t
can be calculated directly
for nonspherical particles from their size, sphericity,
etc. and from the physical properties of the system.
The method uses a dimensionless particle size and a
dimensionless gas velocity dened respectively by:
d
+
p
= d
p
r
f
(r
p
r
f
)g
m
2
_ _
1=3
u
+
t
= u
t
r
2
f
m(r
p
r
f
)g
_ _
1=3
(20)
Then
u
+
t
=
18
(d
+
p
)
2
2:335 1:744f
s
(d
+
p
)
0:5
_ _
1
(21)
for 0:5 < f
S
< 1 where f
S
is the particle sphericity.
Figure 2 Eect of temperature on u
mf
, based on Eq. (15).
(From Rowe, 1984.)
Terminal velocities can then be found directly from
eqs (20) and (21). Based on this method, the terminal
fall velocities in air of nonspherical particles in Groups
A, B, and D of the Geldart classication have been
calculated as functions of temperature and pressure
(Table 2 and Figs. 3 and 4).
It can be shown that in the Stokes ow region
(Re
t
< 0:1):
u
t
=
d
2
p
(r
p
r
f
)g
18m
f
(22)
so that apart from its negligibly small inuence on the
density dierence term in Eq. (21), pressure will have
no eect on u
t
; temperature will of course have an
inuence, through its eect on gas viscosity. Figure 3
shows the eect of temperature and pressure on the
terminal velocity of a Group A material (for which
Re
t
< 1), and it is clear that both cause a monotonic
decrease under all superambient conditions. The pres-
sure eect is transmitted through the gas density term,
and the temperature eect through the gas viscosity
term in Eq. (20). In the case of powders in Groups B
and D (Fig. 4), increasing pressure has a stronger eect
on u
t
than in the case of Group A, but the trend is in
the same direction.
Increasing the temperature on the other hand leads
to an increase in u
t
owing to the dominance of inertial
forces over viscous forces in the case of these materials
and to the eect of temperature in decreasing gas
density.
4 BUBBLING BEDS
4.1 Pressure
It has long been recognized that pressure exerts a
strong inuence on the bubbling behavior of gas-
uidized beds. The literature up to 1993 was reviewed
by Yates (1996), and this is reproduced and extended
to 1998 in Table 3. The three principal groups of pow-
ders in Geldarts classication will be considered in
turn.
4.1.1 Group A Powders
A general observation is that with these powders, while
u
mf
is unaected (for the reasons given above), the
region of bubble-free expansion between u
mf
and the
minimum bubbling velocity, u
mb
, increases with
increasing pressure; this is in accord with the theory
of Foscolo and Gibilaro (1984). Furthermore, at the
same values of volumetric ow rate, bubbles in beds of
Group A materials generally become smaller as pres-
sure increases. This could be due to (1) a greater pro-
portion of gas owing through the emulsion phase
following an increase in emulsion-phase voidage or
(2) a decrease in the stability of bubbles leading to
their breakup into smaller voids. The question of
there being a maximum stable bubble size was rst
explored by Harrison et al. (1961) and later developed
Table 2 Particle Properties Used
in u
t
Calculations.
d
p
=m r
p
=kg m
3
f Group
0.0001 1500 0.8 A
0.0005 1500 0.8 B
0.001 2500 0.8 D
Figure 3 Eect of temperature and pressure on terminal fall velocity of Group A particles in air.
by Davidson and Harrison (1963). The theory is based
on the assumption that as a bubble rises in a uid bed
the shear force exerted by the particles moving down-
wards relative to the bubble sets up a circulation of gas
within the bubble and that the velocity of this circula-
tion, u
c
, approximates to the bubble rise velocity, u
b
.
When through coalescence the bubble diameter
increases to the point where u
b
> u
t
, the terminal fall
velocity of the bed particles, the solids in the wake will
be drawn up into the bubble, causing it to break up
into smaller bubbles with lower rise velocities. The
bubble would thus not be expected to exceed a certain
size determined by the value of u
t
, and beds of small,
light particles should show smoother uidization
characteristics than those of coarser materials. Now
u
t
decreases as gas pressure increases (see above), and
so if the foregoing theory is correct, Group A powders
should show smoother behavior at elevated pressures;
this eect has been widely observed (Homann and
Yates, 1986).
An alternative theory developed by Upson and Pyle
(1973) and Clift et al. (1974) suggests that bubble
breakup is caused by an instability in the bubble roof
allowing particles to rain down through the void and
so divide it into two. On the basis of hydrodynamic
theory due to Taylor (1950), Clift et al. (1974) showed
that the only factor determining the stability of the
bubble roof is the apparent kinematic viscosity of the
emulsion phase and that bubbles will become less
stable as this viscosity decreases. A decrease in dense-
phase viscosity would result from an increase in the
voidage of that phase, and as we have already seen
this is an increasing feature of beds of Group A pow-
ders as operating pressure is increased. King and
Harrison (1980) reported results of a study of particles
of Groups A and B at pressures of up to 25 bar. They
used x-rays to observe both slugs and bubbles and
found that in all cases of Group A materials, breakup
occurred by ngers of particles falling in from the roof
and that the eect was more pronounced the higher the
pressure.
In a separate study King et al. (1981) determined the
viscosity of the emulsion phase of uidized beds of
glass ballotini of a range of sizes by measuring the
rate of fall under gravity through the bed of a small
metal sphere. The beds were uidized by both carbon
dioxide and nitrogen at pressures of up to 20 bar. They
found that an increase in gas pressure led to a substan-
tial decrease in the viscosity of the nest powder but
that the viscosities of beds of powders larger than
about 100 mm were almost independent of pressure.
4.1.2 Group B Powders
In the case of these materials King and Harrison (1980)
reported no eect of pressure on bubble size up to 25
bar, but Homann and Yates (1986) found that mean
bubble diameters increased slightly up to 16 bar and
decreased thereafter up to 60 bar; the extent of these
eects was dependent on the vertical position in the
bed at which the observations were made.
Homann and Yates also measured changes in the
value of the bubble velocity coecient, K, with increas-
ing pressure:
u
b
= K
gd
b
2
_ _
1=2
(23)
The work gave values of K for a freely bubbling bed of
alumina powder at a point 40 cm above the distributor,
decreasing from 0.9 at 1 bar to 0.7 at 20 bar; this was
followed by a gradual increase to 1.2 at 60 bar. The
observed decrease up to 20 bar is consistent with the
Figure 4 Eect of temperature and pressure on terminal fall velocity in air of particles in Groups A, B, and D.
Table 3 Experimental Observation of Effects of Pressure on Gas-Fluidized Beds
Reference
Bed cross
Section, cm
Particle
properties
material r
p
; gcm
3
d
p
; mm Gas
Max.
pressure,
kPa
Observations with increasing
pressure (P, kPa)
Godard and Richardson diam: = 10:0 (i) diakon 1.18 125 air 1414 U
mb
-U
mf
increases.
(1968) (ii) phenolic
resin
0.24 186
Botterill and Desai
(1972)
diam: = 11:4 (i)
(ii)
(iii)
copper
shot
sand
coal
150, 625
160, 800,
2740
1430
air CO
2
1000 Quality of uidization increases for
large particles; heat transfer
coecients double from 1 to 10
bar for larger particles but not
smaller ones.
Knowlton
(1977)
diam: = 29:2 (i)
(ii)
coals
cokes
and chars
siderite
(FeCO
3
)
1.12
1.57
3.91
230780
290
N
2
6900 (i) Fluidization becomes smoother
at P > 1035.
(ii) Bed entrainment increases
signicantly at P > 1035.
(iii) No change in bed expansion or
density.
Saxena and Vogel diam: = 15:2 (i) dolomite 2.46 765 air 834 U
mf
decreases and follows Ergun
(1977) (ii) sulphated
dolomite
3.19 717 correlation.
Canada and MacLaughlin
(1978)
30.5 square glass
beads
2.48
2.92
650
2600
air
R-12
9000 Slugging less pronounced at high
pressure; heat transfer coecients
increase.
Crowther and Whitehead diam: = 2:7 (i) synclyst 0.9 63 Ar 6900 U
mb
-U
mf
increases; fully particulate
(1978) (ii) coals 1.3 1963 CF
4
uidization occurs with nest
particles.
Denloye and Botterill
(1978)
diam: = 11:4 (i) copper
shot
8750 160, 340,
620
air,
argon
1000 Bed-to-surface heat transfer
coecient increases.
(ii) sand 2600 160, 570,
1020, 2370
Co
2
and
freon
(iii) soda
glass
2450 415
Guedes de Carvalho et al.
(1978)
diam: = 10 (i)
(ii)
ballotini
sand
2.9
2.69
64
211
74
N
2
CO
2
2800 No signicant eects with 74 mm
sand and 21 mm ballotini; bubbles
become smaller and break up
from rear with 64 mm ballotini.
Subzwari et al.
(1978)
2-D bed
46 15
FCC
Powder
0.95 60 air 700 e
i
increases; number and size of
bubbles decrease: bubbles split by
division from roof: maximum
stable bubble size decreases.
Varadi and Grace
(1978)
2-D bed
31 1:6
sand 2.65 250295 air 2200 U
mb
-U
mf
increases; no increase in
bubble splitting; bubbles split by
division from roof.
Borodulya et al.
(1980)
diam. 10 cm sand 2.48
2.60
1261220 air 8000 Bed-to-surface heat transfer
coecient increases.
King and Harrison
(1980)
diam. 10 cm ballotini 2.9 64
101
475
air
N
2
CO
2
2500 For 64 mm particles, bubbles become
smaller and less stable; no eect
of pressure on larger particles.
Knowlton and Hirsan
(1980)
semicircle
diam.
30.5 cm
(i)
(ii)
(iii)
sand
char
siderite
(FeCO
3
)
2.63
1.16
3.99
438
419
421
Nz 5171 Jet penetration length increases;
existing correlations underestimate
jet lengths.
Rowe and MacGillivray
(1980)
diam.=
20 cm
silicon
carbide
3.19 58 air 400 Bubbles become smaller; bubble
velocity increases; visible bubble
ow diminishes.
Xavier et al. diam. =
10 cm
(i)
(ii)
(iii)
(iv)
ballotini
ballotini
ballotini
polymer
2.82
2.97
2.91
0.55
61
475
615
688
CO
2
N
2
N
2
N
2
2500 Bed-to-surface heat transfer for ne
powders only slightly aected; for
larger particles, heat transfer
coecient increases with pressure.
Kawabata et al.
(1981)
2-D bed
30 1
sand 2.63
2.58
2.59
300
430
600
air 800 U
mf
decreases; bubble sizes
unchanged, but they become
atter; bubble velocity decreases.
Guedes de Carvallo et al.
(1982)
diam.=5 ballotini 2.9 64
101
N
2
500 Mass transfer between slug and the
dense phase does not decrease
with pressure as much as expected
if gas diusivity alone is the
dominant process.
Table 3 Continued
Reference
Bed cross
Section, cm
Particle
properties
material r
p
; gcm
3
d
p
; mm Gas
Max.
pressure,
kPa
pressure (P, kPa)
King and Harrison
(1982)
diam.=10 (i)
(ii)
(iii)
ballotini
sand
polymer
2.82
2.97
2.66
0.55
61475
81, 288
688
N
2
CO
2
2500 For ne particles, U
mf
and e
mf
are
unaected, but U
mb
and e
mb
increase; for large particles,
U
mf
= U
mb
and decrease with
increasing pressure. But e
mf
= e
mf
and is unaected.
Rowe et al.
(1982)
diam.=7.6 FCC powder 0.82 70, 82 Ar
air
Co
2
2000 U
mb
-U
mf
increases
Sobreiro and Monteiro
(1982)
diam.=4.5 (i)
(ii)
(iii)
ballotini
alumina
pyrrhotite
2.3
2.8
4.0
125, 177
250
88, 125
177
125
N
2
3500 U
mf
is pressure independent in
lamina ow regime but decreases
with increasing pressure at higher
Re values; e
mf
is pressure-
independent; U
mb
-U
mf
increases
for ne powders.
Barreto et al.
(1983)
diam.=10 FCC powders 1.26
0.88
5898 N
2
2000 Bubble size and velocity decrease;
bubble frequency increases with
pressure; more gas ows in bubble
wakes and less in bubble voids.
Borodulya et al.
(1983)
diam.=10.5 (i)
(ii)
glass
beads
sand
2630
2630
2580
2700
3100
1250
1225
794
air 8100 Bed-to-surface heat transfer
coecient increases.
Rowe et al.
(1984)
17:5 12:5 (i)
(ii)
alumina
silicon
carbide
1.42
3.19
450
262
N
2
8000 Geldart Group B behavior at low
pressures changes to Group A
behavior at high pressures; bubble
velocity increases after a small
initial decrease; bubble volume
decreases after slight initial
increase; bubbles atten at high
pressures.
Chan and Knowlton
(1984)
diam.=11.5 sand 2.60 20300 N
2
3100 Solids entrainment increases sharply
with increasing pressure.
Verkooijen et al.
(1984)
diam.=10 FCC powder 61 butene
in N
2
1000 Number of interphase transfer units
decreases from 1 to 2 bar and
increases sharply above 5 bar.
Chitester et al.
(1984)
diam.=10.2
10:2 1:9
(i)
(ii)
(iii)
(i)
(ii)
coals
chars
ballotini
coal
char
1.25
1.12
2.47
1.25
1.12
88361
157376
96374
195
203
N
2
6485 e
i
increases; for coarse particles, bed
height increases at a given gas
ow rate; for ne particles it does
not always do so. No evidence for
maximum stable bubble size up to
810 kPa. Bubble splitting occurs
from roof. Bed appears
homogeneous at highest pressures.
Weimer and Quaderer
(1985)
13 carbon
carbon
carbon
0.850
0.850
0.850
66
108
171
COH
2
8500 For 66 mm powder, e
i
increased
from 0.53 to 0.74, u
i
increases
sevenfold, d
bmax
decreases sixfold;
pressure eects were similar but
of lower magnitude.
Homann and Yates
(1986)
17:8 12:7 alumina
alumina
silicon
carbide
1.417
1.488
3.186
450
695
184
N
2
8000 d
max
decreases but bubble velocity
increases; bubble stability
decreases; bubble ow in center of
bed increases.
Chan et al.
(1987)
38.1 sand
sand
sand
coke
breeze
char
2.565
2.565
2.565
1.507
1.251
100
200
400
400
400
N
2
3200 Bubble size decreases, bubble
frequency increases.
Jacob and Weimer
(1987)
9.7 carbon 0.85 44
122
COH
2
12,420 Particulate bed expansion is
adequately described by Foscolo
Gibilaro theory.
Cai et al.
(1989)
28.4 silica gel
silica gel
silica gel
silica gel
FCC
FCC
resin
sand
0.834
0.706
0.711
0.844
1.172
1.667
1.330
2.580
476
280
165
1057
65
53
566
98
air 800 Higher pressure causes smaller
bubbles and lower transition
velocity from bubbling to
turbulent regimes.
Table 3 Continued
Reference
Bed cross
Section, cm
Particle
properties
material r
p
; gcm
3
d
p
; mm Gas
Max.
pressure,
kPa
pressure (P, kPa)
Carsky et al.
(1990)
13.5 sand
re clay
2.65
2.00
600
600
air 1300 Bubble size decreases up to 400 kPa
and remains constant thereafter.
Olowson and Almstedt
(1990)
20 30 sand 2.60 700 air 1600 Pierced bubble length decreases
withincreasing pressure and with
decreasing u-u
mf
.
Olsson et al.
(1995)
20 30 sand 2.60 700 air 1600 Signicant dierences between beds
operated with and without tubes.
Initial increase with pressure in
mean pierced bubble length, rise
velocity, bubble volume fraction
and visible bubble ow rate.
These level o at pressures above
about 0.5 MPa. As pressure
increases, gas through-ow
velocity decreases signicantly.
Wiman et al.
(1995)
20 30 sand 2.60 700 air 1600 Tube erosion decreases at high
pressures. Bed-to-tube heat
transfer coecient increases
signicantly with increasing
pressure.
Llop et al.
(1996)
5 sand 2.65 213, 450
728, 1460
air 101 At reduced pressure, uidization
behavior is similar to that at
atmospheric pressure. A modied
Wen and Yu correlation is
developed to calculate u
mf
.
Wiman et al.
(1997)
20 30 sand 2.60 450, 700 air 1600 Tube erosion in a sparse tube bank
higher than in densely packed
bank. Larger sand more erosive
than smaller; strong coupling
between tube erosion and bubble
rise velocity.
Wiman et al.
(1998)
20 30 sand 2.60 450 air 1600 A dimensionless drag force is a
suitable scaling parameter in
many cases. Increased gasparticle
interaction at high pressures
combined with turbulent
uctuations in the gas phase can
account for increased instability
of bubbles.
observations of Kawabata et al. (1981) on similar
Group B powders. The increase in K above 20 bar
coupled with the decrease in bubble diameter at these
pressures means that although bubbles are becoming
smaller, their rise velocities are increasing, the reverse
of the tendency observed at ambient pressure.
Homann and Yates (1986) also reported an increase
in bubble interaction leading to coalescence as pressure
was increased, evidence that was further supported by
the observed changes in the lateral distribution of bub-
bles; at atmospheric pressure bubbles were more or less
evenly distributed across the bed, but at higher pres-
sures the bubble ow was increasingly concentrated
toward the vertical axis of the bed. This would seem
to indicate an increase in bubble coalescence leading to
the establishment of a central channel in which the
resistance to gas ow is relatively low. Increased
bubble coalescence should of course result in the for-
mation of larger bubbles, but as mentioned above,
bubble diameters tend to decrease as pressure rises
above 20 bar, and it would seem likely that the expla-
nation is that the stability of the coalesced bubbles is
lower at the higher pressures, and that once formed
they break up into ever smaller units. Indeed at the
highest pressures studied, individual bubbles were
hard to identify on the x-ray lms, the whole bed hav-
ing taken on the appearance of an ill-dened foaming
mass of uidized material.
The results of Homann and Yates (1986) were
corroborated by Olowson and Almstedt (1990) who
used a combined capacitance and pressure probe to
measure bubble characteristics in a bed of powder
that was close to the Geldart B/D boundary
(d
p
= 0:7 mm, r
p
= 2600 kg=m
3
). They operated at
pressures of up to 16 bar and at excess gas velocities
from 0.1 to 0.6 m/s and found that the mean bubble
frequency, the mean bubble rise velocity, the mean
bubble volume fraction, and the visible bubble ow
rate rose with both increasing pressure and excess gas
velocity. The mean pierced length of bubbles was
found to decrease, after an initial increase, with
increasing pressure. An interesting observation was
made with the pressure probe, which showed that
the throughow velocity of gas inside the bubbles
decreased with increasing pressure, a fact that could
also contribute to the growing instability of bubbles,
since it is the drag force exerted by the throughow-
ing gas on particles in its roof that is responsible for
bubble stability (Campos and Guedes de Carvalho,
1992); the drag force is proportional to the square
of the uid velocity [Eq. (8)]. This question was also
considered by Olowson and Almstedt (1992).
Almstedts group at Chalmers University has
carried out a comprehensive series of studies on pres-
surized uidized beds both with and without internal
tube banks of dierent geometries (Olsson et al., 1995;
Wiman et al., 1995; Wiman and Almstedt, 1997, 1998).
The conclusions of these studies are summarised in
Table 3. There has been some interest in operating
uidized beds at subambient pressures, and published
work has been reviewed by Kusakabe et al. (1998) and
Fletcher et al. (1993). Llop et al. (1996) studied the
uidization of coarse sand with a shape factor of 0.6
and more spherical particles of millet (0.9) over a range
of pressures up to atmospheric. They found similar
behavior at low pressures to that observed at atmo-
spheric pressure and developed a modied Wen and
Yu correlation to predict u
mf
over the low-pressure
range.
4.1.3 Group D Powders
These large, dense materials have been relatively little
studied compared with those in Groups A and B. They
are often dicult to uidize smoothly, and there is a
strong tendency towards spouting. Little is known
about how pressure aects their bubbling behavior
although Rowe (1984) has drawn attention to the
inherent instability of uidized Group D particles
through the strong eect of pressure on their minimum
uidization velocity (Fig. 1). It is apparent from this
that any small pressure changes that occur within the
bed as a result of bubbles forming and passing through
it will have a signicant eect on u
mf
that could lead to
local deuidization and generally unstable behavior.
Some indication of the way high pressure operation
could inuence Group D uidization comes from the
study by King and Harrison (1980) of spouted beds.
They measured the minimum spouting velocity of 1.1
mm diameter glass ballotini at pressures of up to 20
bar and found a marked decrease with increasing pres-
sure. It would thus seem likely that Group D materials
should follow the same trends as those shown by
Group B powders but at higher pressures.
4.2 Temperature
Compared with the eects of high pressures referred to
above, there has been relatively little work devoted to
bubble dynamics at high temperatures. Furthermore,
some of what has been published is contradictory. In
an early study Mii et al. (1973) examined uid beds of
graphite particles (Group B) at temperatures of up to
800
C and found that both the frequency of bubble

formation and the uidity of the bed increased with
increasing temperature; similar conclusions were
reached by Yoshida et al. (1974) with beds of micro-
spherical catalyst particles between 500 and 1000
C
and by Otake et al. (1975) with beds of Group A
cracking catalyst. Geldart and Kapoor (1976) studied
bubble sizes in beds of spherical steel shot
(d
p
= 0:118 mm) at up to 300
C and found a reduction

of up to 25% under equivalent conditions of bed
height and excess gas velocity. This would be in keep-
ing with the change in interstitial gas ow with rising
temperature predicted by the theory of Foscolo and
Gibilaro (1984). Kai and Furasaki (1985) also reported
an improvement in the quality of uidization in beds
of FCC at up to 370
C:
Sittiphong et al. (1981) studied the eruption dia-
meter of bubbles in beds of large particles
(d
p
= 3 mm) of a refractory material (Group D) and
found a signicant increase with temperature under
equivalent conditions. This is contrary to what has
been observed in small particle systems such as those
mentioned above, but it is probably a result of the
reduction in the value of minimum uidization velocity
for these large particles at high temperatures referred
to earlier (Fig. 2); thus at the same value of the uidi-
zation velocity more gas will ow as bubbles as u
mf
decreases.
An important quantity in the design of bubbling-
bed reactors containing internal heat transfer coils
is the bed expansion ratio, d. A model for d in terms
of a dimensionless drag force, F
+
, was developed by
Lofstrand et al. (1995) and compared with experimen-
tal work carried out with the pressurized unit at
Chalmers University and an atmospheric pressure
uidized bed boiler tted with bed internals. The bed
expansion ratio is dened as
d =
H
fl
H
mf
H
fl
(24)
and for a freely bubbling bed without internals the
data correlated with
d = 0:11(F
+
1)
0:34
(25)
where
F
+
=
Re
e
mf
f
2
Ar
150
1 e
mf
e
mf
1:75fRe
_ _
(26)
For beds with staggered tube banks, the correlation
was
d = 0:11 F
+
1 ( )
0:34ST
(27)
where
S =
W
W N
h
d
_ _
0:89
(28)
and
T =
H
fl
H
fl
N
v
d
_ _
0:27
(29)
In these expressions N
h
and N
v
are the average number
of tubes in the horizontal and vertical directions,
respectively, d is the tube diameter, and W is the bed
width. The correlations apply adequately to a wide
range of operating conditions with Group B and D
powders and temperatures of up to 850
C.
5 JET PENETRATION
When gas rst enters a uidized bed from an orice in
the supporting grid it does so either in the form of
discrete bubbles or as a amelike jet that decays at
some point above the grid into a bubble stream (Fig.
5). Whether one or the other forms seems to depend on
the properties of the bed material. Thus Rowe et al.
(1979) observed the point of entry of a gas using a
nonintrusive x-ray technique and saw only bubbles
forming with a well dened frequency. Observations
in two-dimensional or semicylindrical beds frequently
show the presence of jets that tend to be stabilized by
the wall.
Grace and Lim (1987) proposed on the basis of
much experimental evidence that the criterion for the
formation of jets should be
Figure 5 A jet entering a uidized bed from an immersed
nozzle. (From Knowlton and Hirsan, 1980.)
d
or
d
p
_ 25:4 (30)
where d
or
and d
p
are the diameters of orice and par-
ticle, respectively. Under all other conditions, bubbles
rather than jets would be expected to form.
A considerable number of studies have been
reported on jet penetration, and several correlations
have been developed giving the penetration length as
a function of the physical properties of gas and parti-
cles and of operating conditions (Knowlton and
Hirsan, 1980). However, most of these correlations
were produced from experimental data obtained at
ambient conditions, and they fail when applied at ele-
vated temperatures and pressures. Hirsan et al. (1980)
measured jet penetrations in beds of Group B powders
at pressures of up to 50 bar and produced a correlation
for L
max
(Fig. 5) in terms of the Froude number and
the ratio of uid to particle density:
L
max
d
0
= 26:6
r
f
r
p
_ _
0:67
u
2
o
gd
p
_ _
0:34
u
u
cf
_ _
0:24
(31)
where u
o
is the orice gas velocity and u
cf
is the
supercial velocity necessary to completely uidize
the polydispersed powder. The correlation shows that
the penetration length increases with pressure but
decreases as the velocity of the uidizing gas increases,
results also obtained by Yates et al. (1986). Yang
(1981) used the same data as Hirsan et al. (1980) and
produced a slightly dierent correlation:
L
max
d
0
= 7:65
1
R
cf
_ _
r
f
r
p
r
f
_ _
u
2
o
gd
0
_ _ _ _
0:472
(32)
where
R
cf
=
(u
cf
)
pressure
(u
cf
)
atmosphere
(33)
Yates and Cheesman (1987) measured jet penetrations
in three-dimensional beds of two coarse powders at
pressures of up to 20 bar at ambient temperature and
at temperatures of up to 800
C at ambient pressure.
The pressure results gave a correlation identical to Eq.
(31) but the high temperature measurements had
slightly dierent values of the coecient and exponent;
a correlation obtained from the combined results at
temperature and pressure was
L
max
d
0
= 9:77
1
R
cf
_ _
r
f
r
p
r
f
_ _
u
2
o
gd
0
_ _ _ _
0:38
(34)
This would appear to be the only correlation cur-
rently available which draws together results from
measurements at both elevated temperatures and
pressures.
6 HIGH VELOCITY OPERATION
6.1 General Characteristics
The overall structure of a uidized bed changes as the
uidizing gas velocity is increased. At low velocities,
the particles thrown by bursting bubbles into the free-
board region above the bed surface fall back after a
short time, but as the gas velocity inceases and bub-
bling becomes more vigorous, the concentration of
particles in the freeboard at any one time increases.
The eruption of bubbles gives rise to pressure uctua-
tions within the bed, and these uctuations increase
with increasing velocity. The velocity at which the pres-
sure uctuations peak, U
c
, has been assumed by some
workers to mark the onset of a transition from bub-
bling to turbulent uidization that is complete at the
higher velocity, U
k
, where the uctuations level o. In
this turbulent regime, at velocities in excess of U
k
, the
bed still has an upper surface, although it is much more
diuse than that present in the bubbling state, and
considerable carryover of particles occurs.
As the gas velocity is increased beyond U
k
a point is
reached at which the particles are transported out of
the bed altogether either in dense-phase or in dilute-
phase pneumatic transport; the velocity at which this
occurs is called the transport velocity, U
tr
. In order
to maintain a constant inventory of particles in the bed
at velocities in excess of U
tr
it is necessary to recycle
them via external cyclones and a standpipe, a geometry
known as a circulating uidized bed. A comprehen-
sive review of the literature on circulating systems has
been given by Grace et al. (1997), and the subject is
also dealt with in Chapter 19 of this handbook.
6.2 Temperature and Pressure Effects
The main interest here has been their eect on solids
carryover and heat transfer (see below). Some work on
transition velocities between ow regimes has been
reported, however. Thus Cai et al. (1989) studied the
eect of operating temperature (50 to 500
C) and pres-
sure (0.1 to 0.8 MPa) on the transition from bubbling
to turbulent uidization of eight powders in Groups A
and B of the Geldart classication uidized in a
column 150 mm in diameter and 3.8 m in height.
They found that the transition velocity, U
c
, increased
with increasing temperature but decreased with
increasing pressure. They correlated their results with:
U
c
gd
p
_ =
m
f20
m
f
_ _
0:2
K
r
f20
r
f
_ _
r
p
r
f
r
f
_ _
D
f
d
p
_ _ _ _
0:27
(35)
where D
f
is the diameter of the uidized bed, the sub-
script 20 refers to physical properties measured at 20
C
and 1 bar, and
K =
0:211
D
0:27
f
2:42 10
3
D
1:27
f
_ _
1=0:27
(36)
A study of a circulating bed at pressures of up to 50
bar was reported by Wirth (1992), but the type of ow
investigated (i.e., transitional, turbulent, or fast) is not
clear from the report.
Tsukada et al. (1993) determined the eect of pres-
sures of up to 0.7 MPa on the three velocities U
c
, U
k
,
and U
tr
of an FCC powder. They found each velocity
to decrease with increasing pressure raised to the
power 0:3, a result not very dierent from that of
Cai et al. (1989). Other conclusions were that the dilute
phase volume fraction at U
tr
was 0.8 at ambient pres-
sure and decreased slightly with pressure, and that the
dimensionless core diameter at U
tr
was about 0.8 and
was insensitive to pressure.
6.3 Entrainment and Elutriation
As has been demonstrated by a number of studies (e.g.,
Rhodes, 1989) there are strong links between the
phenomena of turbulent and fast uidization and
the entrainment and elutriation of bed particles that
occur at lower gas velocities. Entrainment occurs
when gas bubbles burst at the bed surface and throw
particles up into the freeboard region. At low gas velo-
cities these particles quickly fall back into the bed and
are retained, but as the uidizing velocity is increased,
more particles are transported to ever greater heights
above the bed surface, and there exists a particle
density gradient extending some distance above the
surface. For suciently tall freeboards there will be a
certain height at which the density gradient eventually
falls to zero and above this height the entrainment ux
will be constant. This height is called the transport
disengaging height or TDH. If the bed solids have a
wide size distribution and the gas velocity in the free-
board exceeds the terminal fall velocity of the smaller
ones, then these will be carried out of the system or
elutriated.
There has been a large number of experimental
studies of entrainment and elutriation, and as in turb-
ulent and fast beds there are considerable areas of dis-
agreement amongst them. The main reason for this
is the often wide disparity between equipment scale
(particularly bed diameter) and the size and size distri-
bution of the particles investigated. Comprehensive
reviews have been given by Kunii and Levenspiel
(1991) and Tasirin and Geldart (1998).
It is conventional following Leva (1951) and Yagi
and Kunii (1955) to consider elutriation to be a rst-
order process such that the rate of elutriation of
particles within a particular size range, d
pi
, is directly
proportional to the mass fraction of that size range, x
i
,
in the bed. Thus
1
A
t
d
dt
(x
i
M) =
+
i
x
i
(37)
where A
t
is the bed cross-sectional area, M is the mass
of particles in the bed and
+
i
is the elutriation rate
coecient with units of kgm
2
s
1
.
From Eq. (37) it may be seen that
x
i
= x
i0
exp
+
i
A
t
t
M
_ _
(38)
where x
i0
is the initial mass fraction of the particles at
time zero, so that the higher the value of
+
i
the greater
will be the rate of removal of particles from the bed.
Many empirical correlations exist giving
+
i
in terms of
the physical properties of gas and particles (Kunii
and Levenspiel, 1991), but all should be applied with
caution outside the range of conditions under which
they were established.
It is clear from an examination of these empirical
correlations that the terminal fall velocity is an impor-
tant factor in determining the value of the elutriation
rate coecient for any given size of particle. It is also
apparent that the rate coecient will increase as u
t
decreases, so that increasing pressure would be
expected to increase the rate of elutriation. Chan and
Knowlton (1984) studied elutriation from a bed of
Ottawa sand with a wide size distribution at pressures
of up to 31 bar. They found the solids entrainment rate
to increase signicantly with increasing pressure and
uidizing gas velocity (Fig. 6).
The elutriation rate coecient was found to be
linearly proportional to the gas density up to an oper-
ating pressure of 20.7 bar but to increase rapidly and in
a nonlinear manner at higher pressures. At pressures of
up to 3.5 bar, the total entrainment rate was propor-
tional to gas velocity to the power 4.1 but increased
sharply to the 8.4 power of velocity at a pressure of 31
bar. Chan and Knowlton also observed that particles
were entrained at uidizing gas velocities lower than
their terminal fall velocity and attributed this to
momentum transfer from smaller particles to bigger
ones, causing them to be carried over at velocities
lower than the corresponding u
t
. A similar observation
had been made previously at ambient pressure by
Geldart et al. (1979). Pemberton and Davidson
(1983) also studied the eect of increased pressure on
solids entrainment from bubbling beds and found a
similar trend in the entrainment rate coecient to
that reported by Chan and Knowlton (1984). They
attributed the increase in entrainment to the fact that
at high pressures gas bubbles within a bed are generally
smaller than for the same volumetric owrate at ambi-
ent pressure; and that the entrainment ux produced
by particle ejection from the roof of a bursting bubble
is given by
E =
3d
p
r
p
1 e
mf
( ) U U
mf
( )
d
e
(39)
where d
e
is the diameter of the sphere with the same
volume as that of the bubble. Thus as d
e
decreases, E
increases. Alternatively, ejected particles could origi-
nate from the wakes of two bubbles that coalesce
close to the surface. In this case the entrainment ux
is independent of bubble diameter and so would be
expected to be independent of pressure.
The eect on entrainment of increasing temperature
was studied by Findlay and Knowlton (1985). They
increased gas temperature while maintaining the
system pressure constant, so that the dominant eect
was an increase in gas viscosity. This would also have
the eect of decreasing terminal fall velocity and so
increasing entrainment at a given uidizing velocity;
this is just what was observed, as may be seen from
Fig. 7.
7 HEAT TRANSFER
7.1 Bubbling Beds
Their excellent heat transfer properties constitute one
of the more attractive features of uidized beds from a
design point of view. Transfer between gas and parti-
Figure 6 Eect of pressure and uidizing gas velocity on solids entrainment based on the data of Chan and Knowlton (1984).
(From Kunii and Levenspiel, 1991.)
Figure 7 Eect of gas viscosity on entrainment. (From
Findlay and Knowlton, 1985 and Knowlton, 1992.)
cles is normally very ecient, largely as a result of the
high surface area of the particulate phase; a cubic
metre of particles of diameter 100 mm has a surface
area of the order of 30,000 m
2
(Botterill, 1986). Gas-
to-particle transfer is thus rarely a limiting factor and
will not be discussed further.
More interesting from a practical point of view is
heat transfer between the bed material and immersed
surfaces, and a number of studies of the eects of tem-
perature and pressure on the mechanism of the process
have been published. It is generally accepted that the
overall heat transfer coecient, h, between an
immersed surface and a gas-uidized bed can be
expressed as the sum of three components:
h = h
pc
h
gc
h
r
(40)
where h
pc
, h
gc
, and h
r
are the particle convective, gas
convective, and radiative transfer coecients, respec-
tively. The radiative component is of signicance only
above about 600
C in bubbling beds but becomes

important at lower temperatures in the lean-ow risers
of circulating systems (see below). The gas convective
term is of importance only for beds of large Group B
and Group D materials.
For Group A and small Group B powders h
gc
is
insignicant under normal operating conditions, so in
the absence of radiation eects it is h
pc
that dominates
the heat transfer process. Mickley and Fairbanks
(1955) rst pointed out that bed-to-surface heat trans-
fer is an unsteady state process in which packets of
emulsion-phase material carry heat to or from the sur-
face residing there for a short period of time before
moving back into the bulk of the bed and being
replaced by fresh material. On the basis of such a
model the rate of heat transfer will be a maximum at
the instant of contact but will decrease as the residence
time of the packet at the surface increases and the local
temperature gradient is reduced; when the surface is
surrounded by a gas bubble, however, the transfer
rate will fall rapidly to a very low value. The motion
of the packets is of course directly related to the ow of
gas bubbles through the bed, and their surface scour-
ing action will increase as bubble ow increases up to a
point where bubbles are the dominant phase; the rate
of heat transfer by this mechanism would thus be
expected to rise to a maximum value before falling
away. The MickleyFairbanks model gives a value
for the instantaneous heat transfer coecient as
h
i
=
k
mf
r
mf
C
mf
_ _
pt
_ _1=2
(41)
where k
mf
, r
mf
, and C
mf
are the thermal conductivity,
density, and heat capacity of the emulsion phase,
respectively, and t is the residence time of the packet
at the wall. A time averaged heat transfer coecient,
h
p
, may be obtained from Eq. (41) if the residence time
distribution of the packets is known (Xavier and
Davidson, 1985):
h
p
=
_
t
0
h
i
dt
t
(42)
= 2
k
mf
r
mf
C
mf
(u u
mf
)
pL
_ _
1=2
(43)
where for a bubbling bed L is half the equivalent
bubble diameter and r
mf
C
mf
may be assumed equal
to r
s
(1 e
mf
)C
s
. In a bubbling bed, the presence of
the bubbles must be allowed for in calculating h
pc
:
h
pc
= h
p
(1 e
b
) (44)
where e
b
is the volume fraction of bubbles in the bed.
The MickleyFairbanks model was found to over-
predict heat transfer coecients, and Baskakov (1964)
suggested that this was due to an additional resistance,
1=h
f
, caused by the presence of a thin gas layer between
the packet and the wall.
If the two resistances act independently and in
series, then
h
pc
=
1
(1=h
p
) (1=h
f
)
_ _ (1 e
b
) (45)
where, assuming the two-phase theory to apply
(1 e
b
) =
u
b
(u u
mf
u
b
)
(46)
and
h
f
=
mk
g
d
p
(47)
k
g
being the gas thermal conductivity and u
b
being the
rise velocity of a single isolated bubble. Xavier and
Davidson (1985) showed that literature values for m
ranged from 4 to 10; they recommended a value of 6
for design purposes.
The eective conductivity of the emulsion phase
may be expressed as
k
mf
= k
0
e
0:1r
g
C
g
d
p
u
mf
(48)
k
0
e
being the conductivity of a xed bed containing a
stagnant gas.
The particle convective heat transfer coecient may
then be found from the physical properties of the
system using Eqs. (48, 5052), and (53) (Xavier and
Davidson, 1985).
Examination of these relationships will show the
expected eects of pressure and temperature.
7.1.1 Effect of Pressure
The main eect of increasing pressure will be to raise
the gas density r
g
, but this will have only a slight eect
on the particle convective term through its inuence on
k
mf
[Eq. (48)]. In the case of small particles, however,
the suppression of bubbling caused by increasing pres-
sure (see above) would be expected to increase the heat
transfer coecient by improving the quality of ui-
dization near the transfer surface. This was observed
experimentally by Borodulya et al. (1980), who found
an increase of 30% in the maximum heat transfer
coecient for 0.126 mm sand between 6 and 81 bar
pressure.
For larger particles, the eect of pressure is to
increase the gas convective component of the transfer
coecient, since in the case of these materials
h
gc
o Re
mf
(49)
Experimental conrmation of this trend comes from
the work of Botterill and Desai (1972), Botterill and
Denloye (1978a,b), Staub and Canada (1978), Canada
and McLaughlin (1978), Borodulya et al. (1980), and
Xavier et al. (1980).
The eect of pressure on bed-to-immersed-tube heat
transfer was investigated by the Chalmers University
group (Olssen and Almstedt, 1995), who found a
signicant increase in the bed-to-tube heat transfer
coecient with increasing pressure. Results from a pre-
vious study with the same bed showed a decrease in
tube erosion with increasing pressure (Wiman et al.,
1995). The work showed that while the convective
heat transfer is most strongly coupled to the local
bubble frequency, the erosion is most strongly coupled
to the local bubble rise velocity and the local bubble
ow rate.
7.1.2 Effect of Temperature
Increasing temperature has two eects (1) by decreas-
ing gas density, the gas convective component of heat
transfer is decreased slightly, and (2) by increasing the
thermal conductivity of the gas, the eectivness of
packets of emulsion phase in contact with the transfer
surface is increased. The overall eect for Group A and
B powders is to increase the convective transfer coe-
cient as was shown by Botterill and Teoman (1980). In
the case of Group D powders where the gas convective
component is dominant, an increase in temperature
actually causes a decrease in the heat transfer coe-
cient. Botterill et al. (1981) showed that the radiant
component of Eq. (40) begins to be important above
600
C, this was supported by the subsequent study of

Ozkaynak et al. (1983).
7.2 Circulating Beds
In circulating uidized bed (CFB) combustors, heat is
transferred from the gassolid mixture owing through
the riser to water-cooled surfaces in the wall. To model
the convective heat transfer process, therefore, it is
necessary to know, among other factors, the structure
of the owing suspension and its residence time at the
wall. In a combustor operating at anything up to
1000
C, radiation heat transfer is also an important

consideration and can in fact dominate over particle
convection. The subject is thus complex, and we are
still a long way from being able to describe the process
in any detail. Reviews have been written by Grace
(1986), Glicksman (1988), Leckner (1991), and
Glicksman (1997).
The ow structure in CFB risers has been described
above; it was shown to consist of a lean region in the
center with a higher density annulus at the wall. The
solidgas suspension in the annulus is believed to move
down the wall and to be responsible for transferring
heat by convection. In CFB boilers, the risers are nor-
mally square in cross section with membranetube
walls through which the heat transfer uid circulates.
In large units (> 50 MWth) the walls do not provide
sucient cooling surface, and it is necessary to
immerse tube bundles in the riser (Leckner, 1991). In
bubbling pressurized uid bed combustors, heat is
transferred to densely packed tubes that signicantly
alter the ow pattern compared to freely bubbling
beds. There are few reports on this in the literature,
but work by Olsson et al. (1995) adddresses the
problem. Given the core-annular ow pattern in risers,
it is to be expected that models developed for bed-to-
surface heat transfer in bubbling beds can be applied in
a suitably modied form to CFBs. Subarao and Basu
(1986) used a cluster model similar to that of Mickley
and Fairbanks (1955) and assumed that at any given
time the fraction of surface covered by particle clusters
is f and the fraction uncovered is (1 f ). The average
heat transfer coecient can then be written as
h = h
c
f h
d
(1 f ) (50)
where h
c
and h
d
are the time-averaged heat transfer
coecients for clusters at the wall and in the dilute
phase, respectively. Lints and Glicksman (1993) used
the Baskakov model of heat transfer in bubbling beds
[Eq. (45)] to derive an expression for h
c
in a CFB as
h
c
=
1
h
f
1
h
e
_ _
1
=
dd
p
k
g
pt
k
c
c
ps
r
s
(1 e
c
)
_ _ _
1
(51)
where h
e
is the transfer coecient due to conduction
within a cluster, d is the ratio of lm thickness to par-
ticle diameter, c
ps
and r
s
are, respectively, the heat
capacity and density of the solid particles, e
c
is the
cluster voidage, and k
c
is the thermal conductivity of
the cluster. If the thermal conductivity of the cluster is
estimated using Eq. (48) then Eqs. (50) and (51) have
ve parameters that must be determined in order to
predict the overall heat transfer coecient: d, t, e
c
, f ,
and h
d
. Lints and Glicksman (1993) discuss studies
carried out to determine these parameters and show
that experiment and model predictions agree quite
well for laboratory scale CFBs. They sound a note of
caution, however:
Although the same basic model of heat transfer
should apply and the same hydrodynamic para-
meters should govern the process in larger beds
the relation of those parameters within the wall
layer to the average conditions across the entire
bed can be expected to change substantially
between beds of dierent size. This remains a
critical area for further study.
The inuence of the radiative component of the heat
transfer coecient was studied by Wu et al. (1989),
using a refractory lined reactor column 7.32 m high
and 152 152 mm in cross section. The circulating
bed material was coarse sand, and the temperature
range covered was 340880
C. Average heat transfer

coecients to both a vertical tube and a membrane
wall were found to increase almost linearly with
suspension density and with temperature.
Radiation was found to play a signicant role,
especially at high temperatures and low suspension
densities. The radiative component was estimated by
treating the gassolids suspension as a gray body such
that
h
rad
=
s T
4
susp
T
4
surf
_ _
1=e
susp
1=e
surf
1
_ _
T
susp
T
surf
_ _ (52)
where T
susp
and T
surf
are the temperatures of the sus-
pension and tube surface, respectively. The emissivity
values, e
susp
and e
surf
, were set at 0.91and the estimated
and experimental heat transfer coecients were found
to be in good agreement (Table 4).
A cross plot of the overall transfer coecient
against suspension temperature for the two transfer
surfaces shows that the relative magnitude of increase
of the radiation component is dierent for the two
geometries owing, the authors conclude, to the better
view factor of the vertical tube.
7.3 Application of Dimensional Analysis
The foregoing discussion of uid bed heat transfer was
based on theories of the hydrodynamic mechanism by
which bed solids and gas come into contact with an
immersed surface. An alternative approach to the
problem is to arrange the relevant variables into
dimensionless groups, thereby enabling behavior to
be predicted over a wide range of operating conditions.
An example of this is provided by the work of Molerus
(1992a,b, 1993), who applied dimensional analysis to
identify seven relevant groups based on the physical
properties of the particulate material and the uidizing
gas:
p
1
=
d
3
p
g(r
p
r
g
)
2
m
2
p
2
=
r
p
r
g
p
3
=
C
g
m
k
g
Pr
p
4
=
C
p
m
k
g
p
5
=
k
p
k
g
p
6
=
hd
p
k
g
Nu
p
7
=
r
p
c
p
k
g
g
3
_
(u u
mf
)
(53)
where in addition to the normal symbols k represents
thermal conductivity (Wm
1
K
1
). The maximum heat
transfer coecient in the laminar ow regime is then
shown to be
h
max
l
l
k
g
1
k
g
2C
p
m
_ _
= 0:09 (54)
where l
l
is a laminar ow length scale dened by
l
l
=
m
g
_
(r
p
r
g
)
_ _
2=3
(55)
For Archimedes numbers in the range 10
5
< Ar < 10
8
the method gives
Nu
max
= 0:146Pr
1=3
(56)
and for Ar > 10
8
,
Nu
max
= 0:02469Ar
0:4304
Pr
1=3
(57)
The predictions agree well with observations for both
bubbling and circulating beds in the absence of radia-
tion eects. For a more detailed account, the reader is
referred to the paper by Molerus (1993).
8 SINTERING AND AGGLOMERATION
With certain uidized bed materials, as the tempera-
ture of the bed is increased, a point is reached at which
the particles begin to sinter by a softening of the sur-
face and the formation of interparticle bonds. This
temperature, which is often lower than the fusion tem-
perature of the bulk material, is called the minimum
sintering temperature, T
s
, and beds operated at tem-
peratures higher than this can suer catastrophic
deuidization through large-scale agglomeration of
particles. On the other hand, increasing the uidizing
velocity at bed temperatures in excess of T
s
can prevent
deuidization, as can increasing the initial size of the
particles. Whether beds deuidize would seem to
depend on a balance between the cohesiveness or
stickiness of the particle surfaces and the kinetic
energy of the particles due to the uid forces acting
on them. Tardos et al. (1985a) stated that basic
knowledge in the area is so limited that industry relies
mostly on empiricism to avoid deuidization of beds
containing sticky particles, and despite the increased
number of publications on the subject since that date,
there is still no mechanistic model available that can
reliably predict under what conditions deuidization
will occur. A number of attempts at modeling the
process have been made, however, and these will be
reviewed following a brief description of sintering
and ways of measuring it.
Sintering occurs as a result of the migration of lat-
tice vacancies or the movement of atoms in the surface
of particles (Siegell, 1984). If the rate of sintering is
suciently high, then when particles of the same mate-
rial come into contact, bonds will be formed between
them leading to the formation of agglomerates. The
temperature at which this occurs, T
s
, can be measured
with a dilatometer in which the relative expansion
contraction of a sample of granular material under
load is measured as its temperature is increased
(Compo et al., 1984). The change in length or dilation,
L, of a sample of glass beads divided by its initial
length, L
0
, is recorded and plotted as a function of
temperature. An initial increase in the length ratio
due to thermal expansion of the bulk material is
observed, but at a higher temperature, as the surface
begins to soften, and contraction counteracts
expansion, the dilation ceases, and the curve shows a
plateau. The temperature at which the curve turns
sharply downward and at which the sample contracts
at a rapid rate is the sintering temperature T
s
.
Compo et al. (1987) combined dilatometer measure-
ments with deuidization experiments for a variety of
materials and concluded that the materials can be
divided into two categories, (1) those such as glass
beads, coal powder, and polyolen granules that
agglomerate quickly after T
s
has been reached and
the increase in gas velocity needed to keep the bed
uidized increases exponentially, and (2) the ionic
solids such as sodium chloride and calcium chloride
that form relatively weak agglomerates and for which
only a slight increase in uidizing velocity is necessary
to restore uidization once agglomeration has begun.
None of the materials studied was found to deuidize
below its minimum sintering temperature.
A system in which deuidization can present a par-
ticular problem is the combustion of low-rank coals in
a circulating uid bed (CFB) combustor. Some low-
rank coals have a high content of sodium and sulphur,
and at operating temperatures in the region of 850
C
Table 4 Comparison of Experimental Data with
Estimated Heat Transfer Coefcient for a Vertical
Tube
kg=m
3
C W=m
2
K
Susp
Dens
Avg
susp
temp
Tube
surf
temp
Gas
conv
comp
Particle
conv
comp
Rad
comp
Total
est
h
Exp
data
h
15 701 83 13 19 68 100 106
587 56 13 19 48 80 89
343 45 14 19 22 55 62
60 701 83 13 76 68 157 166
587 56 13 76 48 137 144
343 45 14 76 22 112 110
Source: Wu et al., 1989
these are converted to molten alkali sulphates, which
are deposited along with coal ash on particles of the
bed material, usually sand or something similar, caus-
ing them to agglomerate and deuidize. The processes
taking place are complex and poorly understood,
although experimental studies by Manzoori and
Agarwal (1993, 1994) have claried some of the details.
Mechanistic models of the deuidization process
have been explored by a number of authors and prin-
cipally by Tardos and Pfeer and their coworkers at
City College New York. Thus Tardos et al. (1985a) set
out to predict the limiting gas velocity U
s
, necessary to
break the largest agglomerate in the bed and thereby to
keep a bed of sticky particles uidized at temperatures
above the minimum sintering temperature. They
looked for conditions in which the dynamics of the
bonding forces holding the agglomerates together
and the forces leading to breakup (due to the motion
of gas bubbles in the bed) are in equilibrium. For
simplicity they assumed that the largest agglomerate
occupied the entire cross section of the bed but con-
ceded that in large beds deuidization would be likely
to occur before this size was reached; in these cases the
predicted limiting velocity, U
s
, should be considered a
lower bound for the uidizing gas velocity. Further
simplifying assumptions were that the agglomerates
were cylindrical in shape and were xed in the bed
and not freely buoyant, and that the forces acting on
them due to the ow of interstitial gas were negligible,
the main force being shear due to the motion of
bubbles. Based on an equation of Livshits et al.
(1978) the maximum force, G
v
, acting on the agglom-
erate was found in terms of the excess velocity
(u u
mf
) and related to the pressure, q, acting on the
bottom face of the cylinder by
q =
4G
v
pd
2
ag
(58)
where d
ag
is the diameter of the agglomerate. This
pressure will cause failure of the structure if its value
exceeds q
max
where
q
max
= s
y
2h
d
ag
_ _
2
A
1
(59)
in which s
y
is the yield strength of the agglomerate and
A
1
is a coecient approximately equal to 2. From
these equations a relationship is found between the
excess velocity and the yield strength of the agglomer-
ate, a quantity that was calculated on the basis of a
technique devised by Rumpf (1977). The result is a
rather complicated expressions, the evaluation of
which requires knowledge of two fundamental physical
quantities, the surface viscosity Z
s
and the yield
strength of a sinter neck s
s
, both as functions of tem-
perature. The surface viscosity can be found from
dilatometer data, and the yield strength can be either
estimated using adhesion theory or calculated from
experimental data (Tardos et al., 1985b).
An alternative approach based on the concept of the
dissipation of the kinetic energy of particle collisions
via surface viscosity has been studied by Ennis et al.
(1991). It is particularly applicable to uid bed granu-
lation through the introduction of liquid binding
agents but does appear to have some relevance for
high temperature deuidization. The mechanism of
coalescence is considered to be a function of a binder
Stokes number St
v
dened as
St
v
=
2Mu
0
3pma
2
(60)
where M is the mass of a particle, u
0
is the initial
relative velocity of two particles, a is the particle
radius, and m is the surface viscosity. For two colliding
particles to rebound, the Stokes number must exceed a
critical value given by
St
+
v
= 1
1
e
/
_ _
ln
h
/
h
a
_ _
(61)
where e
/
is the coecient of restitution of the particles,
h
/
is the thickness of the binder on the surface, and h
a
is the characteristic length of surface asperities.
Particles colliding with Stokes numbers less than this
critical value are considered to have coalesced. The
theory compared favorably with results of the de-
uidization experiments reported by Gluckman et al.
(1975).
Seville et al. (1998) also identied two types of ad-
hesion leading to agglomeration: viscoplastic which
occurs with glassy materials, and a second type caused
by the formation of an excessive amount of liquid on a
surface resulting from melting or chemical reaction.
Both types can lead to deuidization. The authors
assumed that particles in a uidized bed remain in
close proximity in quiescent zones with little relative
motion until they are disturbed by bubbles; the
residence time, t
b
, of particles in these zones may be
suciently prolonged to allow sintering to occur. They
modeled the behavior of viscoplastic materials on the
basis of t
b
and a second characteristic time, t
s
, which is
that needed to form a strong sintered bond. Then
equating the sintering time with the time required to
turn the bed material over once, which is derived from
two-phase theory as approximately
t
b0
=
H
mf
(u u
mf
)
(62)
which gives a relationship between the critical excess
velocity required just to break up the agglomerates and
the surface viscosity of the sintering particles:
u u
mf
=
K
1
K
2
m
0
exp(E=RT)
(63)
Here K
1
is approximately equal to H
mf
and K
2
is the
critical size of the sinter neck. The denominator in
Eq. (63) represents the temperature dependence of
the surface viscosity. Favorable comparisons were
drawn between the theory and deuidization
experiments with polyethylene granules over a range
of temperatures.
As was stated at the beginning of this chapter,
practically all uidized bed reactors operate at
temperatures above the ambient, and yet the eect of
temperature on the surface properties of uidized
solids is far from well understood. The work reported
by the above authors has made a valuable contribution
to our understanding, but the whole question of inter-
particle forces in uidized beds and the eect on them
of operating temperature and pressure are areas much
in need of further study.
NOTATION
A; A
t
= cross-sectional area, m
2
Ar = Archimedes number
C = heat capacity, J kg
1
K
1
C
D
= drag coecient
D
f
= bed diameter, m
D
g
= gas diusivity, m
2
Pa s
1
d = diameter, m
d
e
= diameter of equivalent sphere, m
E = entrainment ux, kg m
2
s
1
E
mb
= modulus of elasticity, Nm
2
e = emissivity
e
/
= coecient of restitution
Ga = Galileo number
G
v
= force [Eq. (58)], N
g = acceleration due to gravity, ms
2
h = heat transfer coecient, Wm
2
K
1
h
/
= binder thickness, m
h
a
= length of surface asperities, m
K = bubble velocity coecient
k = thermal conductivity, Wm
1
K
1
k
c
= chemical rate coecient, ms
1
L = height, m
M = mass, kg
m = constant in Eq. (47)
Nu = Nusselt number
Pr = Prandtl number
p = pressure, Pa
Re = Reynolds number
Sh = Sherwood number
St = Stokes number
T = temperature, K
t
c
= particle burnout time, s
U; u
0
= supercial velocity, ms
1
U
c
= transition velocity, ms
1
U
k
1
U
s
= slip velocity, ms
1
U
tr
1
u = velocity, ms
1
u
t
= terminal fall velocity, ms
1
W = width, m
X = crossow factor
x = mass fraction
z = vertical coordinate, m
z
i
= location of point of inection [Eq. (35)], m
Greek Symbols
a; g = stoichiometric coecients
d = bed expansion ratio
e = voidage
f = sphericity
i
= elutriation rate coecient, kg m
2
s
1
m = viscocity, Ns m
2
r = density, kg m
3
s = yield strength, Nm
2
t = residence time, s
Subscripts
b = bubble
c = cluster
cf = complete uidization
D = dense phase
f = uid
= uidized
g = gas
gc = gas convective
max = maximum
mb = minimum bubbling
m = minimum uidization
o, or = orice
p = particle
pc = particle convective
r = radiant
s = solids
t = terminal fall
REFERENCES
Barnea E, Mizrahi J. A generalised approach to the uid
dynamics of particulate systems: general correlation of
uidization and sedimentation in solid multiparticle sys-
tems. Chem Eng J 5:171, 1973.
Barnea E, Mednick RL. Correlation of minimum uidization
velocity. Trans I Chem E 53:278, 1975.
Barreto GF, Yates JG, Rowe PN. The eect of pressure on
the ow of gas in uidized beds of ne particles. Chem
Eng Sci 38:19351945, 1983.
Baskakov AP. The mechanism of heat transfer between a
uidized bed and a surface. Int Chem Eng 4:320, 1964.
Borodulya VA, Ganzha VG, Podberezsky AI. Heat transfer
in a uidized bed at high pressure. In: Fluidization (JR
Grace, JM Matsen, eds.). Plenum Press, New York, 1980,
201207.
Borodulya VA, Ganzha VL, Podberezsky AI, Upadhyay SN,
Saxena SE. High pressure heat transfer investigations for
uidized beds of large particles and immersed vertical
tube bundles, Int J of Heat and Mass Transfer 26:1577
584, 1983.
Botterill JSM. Fluid bed heat transfer. In: Gas Fluidization
Technology (D Geldart, ed.). John Wiley, Chichester,
1986.
Botterill JSM, Denloye AOO. Bed to surface heat transfer in
a uidized bed of large particles. Powder Tech 19:197
203, 1978a.
Botterill JSM, Denloye AOO. A theoretical model of heat
transfer to a packed or quiescent uidized bed. Chem Eng
Sci 33:509515, 1978b.
Botterill JSM, Desai M. Limiting factors in gas-uidized bed
heat transfer. Powder Tech 6:231238, 1972.
Botterill JSM, Teoman Y. Fluid-bed behaviour at elevated
temperatures. Fluidization 5 (JR Grace, JM Matsen,
eds.). Plenum Press, New York, 1980, 93100.
Botterill JSM, Teoman Y, Yuregir KR. Temperature eects
on the heat transfer behaviour of uidized beds. AIChE
Symp Ser 77:(208)330340, 1981.
Botterill JSM, Teoman Y, Yuregir KR. The eect of operat-
ing temperature on the velocity of minimum uidization
bed voidage and general behaviour. Powder Technology
31:101110, 1982.
Cai P, Chen SP, Jin Y, Yu ZQ, Wang ZW. Eect of tempera-
ture and pressure on the transition from bubbling to tur-
bulent uidization. AIChE Symp Ser 85:(270)3743, 1989.
Campos JBLM, Guedes de Carvalho JRF. Drag force on the
particles at the upstream end of a packed bed and the
stability of the roof of bubbles in uidized beds. Chem
Eng Sci 47:40574062, 1992.
Canada GS, MacLaughlin MH. Large particle uidization
and heat transfer at high pressure. AIChE Symp Ser
74:(176)2737, 1978.
Carsky M, Hartman M, Ilyenko BK, Makhorin KE. The
bubble frequency in a uidized bed at elevated pressure.
Powder Tech 61:251254, 1990.
Chan IH, Knowlton TM. The eect of pressure on entrain-
ment from bubbling gas uidized beds. In: Fluidization
(D Kunii, R Toei, eds.). Engineering Foundation, New
York, 1984, pp 283290.
Chan IH, Shishtla C, Knowlton TM. The eect of pressure
on bubble parameters in gas uidized beds. Powder Tech
53:217235, 1987.
Chitester DC, Kornosky RM, Fan LS, Danko JP.
Eng Sci 39:253261, 1984.
Clift R, Grace JR, Weber ME. Stability of bubbles in ui-
dized beds. I & E C Funds 13:4551, 1974.
Clift R, Grace JR, Weber ME. Bubbles, drops and particles,
Academic Press, New York, 1978.
Compo P, Pfeer R, Tardos GI. Minimum sintering tem-
peratures and deuidization characteristics of uidizable
particles. Powder Tech 51:85101, 1987.
Compo P, Tardos GI, Mazzone D, Pfeer R. Minimum sin-
tering temperatures of uidizable particles. Part Charact
1:171177, 1984.
Crowther ME, Whitehead JC. Fluidization of ne particles at
elevated pressure. In: Fluidization (JF Davidson, DL
Kearins, eds.). Cambridge University Press, 1978, pp
6570.
Davidson JF, Harrison D. Fluidised Particles, Cambridge
University Press, 1963.
Denloye AOO, Botterill JSM. Bed to surface heat transfer in
a uidized bed of large particles. Powder Tech 19:197
203, 1978.
Ennis BJ, Tardos GI, Pfeer R. A microlevel-based charac-
terization of grannulation phenomena. Powder Tech
65:257272, 1991.
Formisani B, Girimonte R, Mancuso L. Analysis of the ui-
dization process of particle beds at high temperature.
Chem Eng Sci 53:951962, 1998.
Findlay JG, Knowlton TM. Final Report for US Dept. of
Energy, Project DE-AC21-83MC20314, 1985.
Fletcher JV, Deo MD, Hanson FV. Fluidization of a multi-
sized Group B sand at reduced pressure. Powder Tech
76:141147, 1993.
Foscolo PU, Gibilaro LG. A fully predictive criterion for the
transition between particulate and aggregate uidization.
Chem Eng Sci 39:16671675, 1984.
Geldart D. Types of gas uidization. Powder Tech 7:285
292, 1973.
Geldart D, Cullinan J, Georghiades S, Gilvray D, Pope DJ.
The eect of nes on entrainment from gas uidized beds.
Trans I Chem E 57:269275, 1979.
Geldart D, Kapoor DS. Bubble sizes in a uidized bed at
elevated temperatures. Chem Eng Sci 31:842843, 1976.
Glicksman LR. Circulating uidized bed heat transfer. In:
Circulating Fluidized Bed Technology 2 (P Basu, JF
Large eds.). Pergamon Press, Oxford, 1988.
Glicksman LR. In: JR Grace, AA Avidan, TW Knowlton,
eds. Circulating Fluidized Beds. Blackie, London, 1997,
pp 261311.
Gluckman MJ, Yerushalmi J, Squires AM. Deuidization
characteristics of sticky or agglomerating beds. In:
Fluidization Technology (DL Kearins, ed.) Vol 2.
Hemisphere, New York, 1975, pp 395422,.
Goddard K, Richardson JF. The behaviour of bubble-free
uidized beds. I Chem E Symp Ser No. 30:126135, 1968.
Grace JR, Avidan AA, Knowlton TM, (eds.). Circulating
Fluidized Beds. Blackie, London, 1997.
Grace JR, Lim CT. Permanent jet formation in beds of par-
ticulate solids. Can J Chem Eng 65:160162, 1987.
Grace JR. Heat transfer in circulating uidized beds. In:
Circulating Fluidized Bed Techonolgy (P Basu, ed.).
Pergamon Press, Oxford, 1986.
Guedes de Carvalho JRF, King DF, Harrison D.
Fluidization of ne particles under pressure. In:
Fluidization (JF Davidson, DL Kearins, eds.).
Cambridge University Press, 1978, pp 5964.
Guedes de Carvalho JRF, King DF, Harrison D. Mass
transfer from a slug in a uidized bed at elevated pres-
sures. Chem Eng Sci 37:10871094, 1982.
Haider A, Levenspiel O. Drag coecient and terminal velo-
city of spherical and non-spherical particles. Powder Tech
58:6370, 1989.
Harrison D, Davidson J, De Kock JW. On the nature of
aggregative and particulate uidization. Trans Inst Chem
Engrs 39:202211, 1961.
Hartman M, Yates JG. Free-fall of solid particles through
uids. Collect Czech Chem Commun 58:961982, 1993.
Hirsan I, Shishtla C, Knowlton TM. The eect of bed and jet
parameters on vertical jet penetration length in gas ui-
dized beds. Paper presented at 73rd Annual AIChE
Meeting, Chicago, Illinois, Nov 1620, 1980.
Homann AC, Yates JG. Experimental observations of ui-
dized beds at elevated pressures. Chem Eng Commun
41:133149, 1986.
Jacob KV, Weimer AW. High pressure particulate expansion
and minimum bubbling of ne carbon powders. AIChE J
33:16981706, 1987
Kai T, Furusaki S. Behaviour of uidized beds of small par-
ticles at elevated temperatures. J Chem Eng Japan 18:113
118, 1985.
Kawabata J, Yumiyama M, Tazaki Y, Honma S, Chiba T,
Sumiya T, Ehdo K. Characteristics of gas uidized beds
under pressure. J Chem Eng Japan 14:8589, 1981.
Khan AR, Richardson JF. Resistance to motion of a solid
sphere in a uid. Chem Eng Commun 62:135150, 1987.
King DF, Harrison D. The bubble phase in high pressure
uidized beds. In: Fluidization (JR Grace, J M Matsen,
eds.). Plenum Press, New York, 1980, 101107.
King DF, Harrison D. The dense phase of a uidized bed at
elevated pressures. Trans I Chem E 60:2630, 1982.
King DF, Mitchell FRG, Harrison D. Dense phase viscos-
ities of uidized beds at elevated pressures. Powder Tech
28:5558, 1981.
Knowlton TM. High pressure uidization characteristics
of several particulate solids: primarily coal and coal-
derived materials. AIChE Symp Ser 73:(161)2228,
1977.
tration in semi-cylindrical gas-uidized beds. In:
Fluidization (JR Grace, JM Matesen, eds.). Plenum
Press, New York, 1980, pp 315324.
Kunii D, Levenspiel O. Fluidization Engineering.
Butterworths, Boston, 1991.
Kusafabe K, Kuriyama T, Morooka S. Fluidization of ne
particles at reduced pressure. Powder Tech 58:125130,
1998.
Leckner B. Heat transfer in circulating uidized bed boilers.
Circulating Fluidised Technology 3 (P Basu, M Horio, M
Hasatani, eds.). Pergamon Press, Oxford, 1991.
Leva M. Elutriation of nes from uidized systems. Chem
Eng Prog 47:(1)3945, 1951.
Lints MC, Glicksman LR. Parameters governing particle-to-
wall heat transfer in a circulating uidized bed. Preprints
4th Int. Conf. on Circulating Fluidized Beds. AIChE J
350355, 1993.
Livshits YY, Tamarin AI, Zabrodsky SS. Maximum forces
acting on a body immersed in a uidized bed. Fluid. Mech
Sov Res 7:30, 1978.
Llop MF, Madrid F, Arnaldos J, Casal J. Fluidization at
vacuum conditions. A generalized equation for the predic-
tion of minimum uidization velocity. Chem Eng Sci
51:51495158, 1996.
Lofstrand H, Almstedt AE. Andersson S. Dimensionless
model for bubbling uidized beds. Chem Eng Sci
50:245253, 1995.
Manzoori AR, Agarwal PK. The role of inorganic matter in
coal in the formation of agglomerates in circulating uid
bed combustors. Fuel 72:10691075, 1993.
Manzoori AR, Agarwal PK. Agglomeration and deuidiza-
tion under simulated circulating uidized-bed combustion
conditions. Fuel 73(4):563568, 1994.
Mickley HS, Fairbanks DF. Mechanism of heat transfer to
Mii T, Yoshida K, Kunii D. Temperature eects on the
characteristics of uidized beds. J Chem Eng Japan
6:100102, 1973.
Molerus O. Heat transfer in gas uidized beds, Part 1.
Powder Tech 70:114, 1992a.
Molerus O. Heat transfer in gas uidized beds Part 2.
Dependence of heat transfer on gas velocity. Powder
Tech 70:1520, 1992b.
Molerus O. Arguments on heat transfer in gas uidized beds.
Chem Eng Sci 48:761770, 1993.
Olowson PA, Almstedt AE. Inuence of pressure and uidi-
zation velocity on the bubble behaviour and gas ow dis-
tribution in a uidized bed. Chem Eng Sci 45:17331741,
1990.
Olowson PA, Almstedt AE. Inuence of pressure on the
minimum uidization velocity. Chem Eng Sci 46:637
640, 1991.
Olowson PA, Almstedt AE. Hydrodynamics of a bubbling
uidized bed: inuence of pressure and uidization velo-
city in terms of drag force. Chem Eng Sci 47:357-366,
1992.
Olssen SE, Almstedt AE. Local instantaneous and time-aver-
aged heat transfer in a pressurized uidized bed with hor-
izontal tubes: inuence of pressure, uidization velocity
and tube-bank geometry. Chem Eng Sci 50:32313245,
1995.
Olssen SE, Wiman J, Almstedt AE. Hydrodynamics of a
pressurized uidized bed with horizontal tubes: inuence
of pressure uidization velocity and tube bank geometry.
Chem Eng Sci 50:581592, 1995.
Otake T, Tone S, Kawashima M, Shibata T. Behaviour of
rising bubbles in a gas-uidized bed at elevated tempera-
ture. J Chem Eng Japan 8:388392, 1975
Ozkaynak T, Chen JC, Frankeneld TR. An experimental
investigation of radiation heat transfer in a high tempera-
ture uidized bed. In: Fluidization (D Kunii, R Toei,
eds.). Engineering Foundation, New York, 1983, pp
371378.
Pemberton ST, Davidson JF. Elutriation of ne particles
from bubbling uidized beds. In: Fluidization (D Kunii,
R Toei, eds.). Engineering Foundation, New York, 1983
pp 275282.
Raso G, DAmore M, Formisani B, Lignola PG. The inu-
ence of temperature on the properties of the particulate
phase at incipient uidization. Powder Tech 72:7176,
1992.
Rhodes MJ. Upward ow of gas/solid suspensions, Part 2. A
practical quantitative ow regime diagram for the upward
ow of gas/solid suspensions. Chem Eng Res Des 67:30
37, 1989.
Rowe PN. The eect of pressure on minimum uidization
velocity. Chem Eng Sci 39:173174, 1984.
Rowe PN. The dense phase voidage of ne powders uidized
by gas and its variation with temperature, pressure and
particle size. In: Fluidization (JR Grace, LW Shemilt, MA
Bergougnou, eds.). Engineering Foundation, New York,
1989, pp 195202.
Rowe PN, Foscolo PU, Homann AC, Yates JG. X-ray
observation of gas uidized beds under pressure. In:
Fluidization (D Kunii, R Toei, eds.). Engineering
Foundation, New York, 1983, pp 5360.
Rowe PN, Foscolo PU, Homann AC, Yates JG. Fine pow-
ders uidized at low velocity at pressures up to 20 bar with
gases of dierent viscosity. Chem Eng Sci 37:11151117,
1982.
Rowe PN, MacGillivray HJ. A preliminary x-ray study of the
eect of pressure on a bubbling gas uidized bed. Inst
Energy Symp Ser 4:19, 1980.
Rowe PN, MacGillivray HJ, Cheesman DJ. Gas discharge
from an orice into a gas uidized bed. Trans I Chem E
57:194199, 1979.
Rumpf H. Particle Adhesion, in Agglomeration 1977 (KVS
Shastry, ed.). AIME, 1997.
Saxena SC, Vogel GJ. The measurement of incipient uidiza-
tion velocities in a bed of coarse dolomite at temperature
and pressure. Trans I Chem E 55:184189, 1997.
Seville JPK, Siloman-Pug H, Knight PC. Modelling of sin-
tering in high temperature uidized beds. Powder Tech
97:160169, 1998.
Siegell JA. High temperature deuidization. Powder Tech
38:1322, 1984.
Sittiphong N, George AH, Bushnell D. Bubble eruption dia-
meter in a uidized bed of large particles at elevated tem-
peratures. Chem Eng Sci 36:12591260, 1981.
Sobreiro LEC, Monteiro JLF. The eect of pressure on ui-
dized bed behaviour. Powder Tech 33:95100, 1982.
Staub FW, Canada GS. Eect of tube bank and gas density
on ow behaviour and heat transfer in a uidized bed. In:
Fluidization (JF Davidson, DL Kearins, eds.). Cambridge
University Press, 1978, pp 339344.
Subarao D, Basu P. A model for heat transfer in circulating
uidized beds. Int J Heat Mass Transfer 29:487489,
1986.
Subzwari MP, Clift R, Pyle DL. Bubbling behaviour of ui-
dized beds at elevated pressures. In: Fluidization (JF
Davidson, DL Kearins, eds.). Cambridge University
Press, 1978, pp 5054.
Svoboda K, Hartman M. Inuence of temperature on inci-
pient uidization of limestone, lime, coal ash and corun-
dum. I and EC Proc Des Dev 20:319326, 1981.
Tardos G, Mazzone D, Pfeer R. Destabilization of uidized
beds due to agglomeration. Part 1: Theoretical model.
Can J Chem Eng 63:377383, 1985a.
Tardos G, Mazzone D, Pfeer R. Destabilization of uidized
beds due to agglomeration. Part 2: Experimental verica-
tion. Can J Chem Eng 63:384389, 1985b.
bedsa new correlation for elutriation rate constants.
Powder Tech 95:240247, 1998.
Taylor GI. The instability of liquid surfaces when accelerated
in a direction perpendicular to their planes. Proc Roy Soc
A201:192196, 1950.
Tsukada M, Nakanishi D, Horio M. The eect of pressure
on phase transition from bubbling to turbulent uidiza-
tion. Int J Multiphase Flow 19:2734, 1993.
Upson PC, Pyle DL. The stability of bubbles in uidized
beds. In: Fluidization and Its Applications. Ce padue` s-
E
ditions, Toulouse, 1973, pp 207222.

Varadi T, Grace JR. High pressure uidization in a two-
dimensional bed. In: Fluidization (JF Davidson, DL
Keairns, eds.). Cambridge University Press, 1978, pp.
5558.
Verkooijen AHM, Rietema K, Thoenes D. The inuence of
pressure on the conversion in a uidized bed. In:
Fluidization (D Kunii, R Toei, eds.). Engineering
Foundation, New York, 1983, pp 541546.
Weimer AW, Quarderer GJ. On dense phase voidage and
bubble size in high pressure uidized beds of ne powders.
AIChE J 31:10191028, 1985.
Wen CY, Yu YH. Mechanics of uidization. Chem Eng Prog
Symp. Ser 62:100111, 1966.
Wiman J, Almstedt AE. Inuence of pressure, uidization
velocity and particle size on the hydrodynamics of a freely
bubbling uidized bed. Chem Eng Sci 53:21672176,
1998.
Wiman J, Almstedt AE. Aerodynamics and heat transfer in a
pressurized uidized bed: Inuence of pressure, uidiza-
tion velocity, particle size and tube geometry. Chem Eng
Sci 52:26772696, 1997.
Wiman J, Mahpour B, Almstedt AE. Erosion of horizontal
tubes in a pressurized uidized bed: inuence of pressure,
uidization velocity and tube bank geometry. Chem Eng
Sci 50:33453356, 1995.
Wirth KE. Fluid mechanics of pressurized circulating ui-
dized beds. In: Fluidization (OE Potter, DJ Nicklin,
eds.). Engineering Foundation, New York, 1992, pp
113120.
Wu RL, Grace JR, Lim CJ, Brereton CMH. Suspension-to-
surface heat transfer in a circulating uidized bed com-
bustor. AIChE J 35:16851691, 1989.
Wu SY, Baeyens J. Eect of the operating temperature on
the minimum uidization velocity. Powder Tech 67:217
220, 1991.
Xavier AM, Davidson JF. Heat transfer in uidized beds. In:
Fluidization (JF Davidson, R Clift, D Harrison, eds.).
Academic Press, London, 1985.
Xavier AM, King DF, Davidson JF, Harrison D. Surface-
bed heat transfer in a uidized bed at high pressure. In:
Fluidization (JR Grace, JM Matsen, eds.). Plenum Press,
1980, pp 209216.
Yagi S, Kunii D. Studies on combustion of carbon particles
in ames and uidized beds. Proc. 5th Int. Symp.,
Combustion. Van Nostrand Reinhold, New York, 1955,
pp 231244.
Yamazaki R., Han NS, Sun ZF, Jimbo G. Eect of chemi-
sorbed water on bed voidage of high temperature uidized
beds. Powder Tech 84:1522, 1995.
Yang WC. Jet penetration in a pressurized uidized bed. I
and EC Fundams, 20:297300, 1981.
Yang WC, Chitester DC, Kornosky RM, Keairns DL. A
generalised methodology for estimating minimum uidi-
zation velocity at elevated pressure and temperature.
AIChE J 31:10861092, 1985.
Yates JG, Bejcek V, Cheesman DJ. Jet penetration into ui-
dized beds at elevated pressures. In: Fluidization (K
Ostergaard, S Sorensen, eds.). Engineering Foundation,
New York, 1986, pp 7986.
Yates JG, Cheesman DJ. Unpublished results, 1987.
Yates JG. Eects of temperature and pressure on gassolid
uidization. Chem Eng Sci 51:167205, 1996.
Yoshida K, Ueno T, Kunii D. Mechanism of bed to wall heat
transfer in a uidized bed at high temperatures. Chem
Eng Sci 29:7782, 1974.
6
Gas Distributor and Plenum Design in Fluidized Beds
S. B. Reddy Karri
Particulate Solid Research, Inc., Chicago, Illinois, U.S.A.
Joachim Werther
1 INTRODUCTION
The gas distributor (also called a grid) in a uidized
bed reactor is intended to induce a uniform and stable
uidization across the entire bed cross section, prevent
nonuidized regions on the grid, operate for long per-
iods (years) without plugging or breaking, minimize
weepage of solids into the plenum beneath the grid,
minimize attrition of the bed material, and support
the weight of the bed material during startup and shut-
down. In practice, grids have taken a variety of forms,
a few of which are discussed in subsequent pages.
Whatever the physical form, all are fundamentally
classiable in terms of the direction of gas entry:
upward, laterally, or downward. The choice depends
on prevailing process conditions, mechanical feasibil-
ity, and cost. In the past, grid design has been more of
an art than a science. However, more recent studies
now allow grid design based on scientic principles.
2 TYPES OF GRIDS
2.1 Perforated Plates (Upwardly Directed Flow)
Main Advantages
Simple fabrication; most common; inexpensive; easy to
modify hole size; easy to scale up or down; easy to
clean can be at, concave, convex, or double dished;
ports are easily shrouded.
Possible Disadvantages
Bed weepage to plenum; can be subject to buckling or
thermal distortion; requires peripheral seal to vessel
shell; requires support over long spans; high pressure
drop required if weepage during operation is to be
minimized
2.2 Bubble Caps and Nozzles (Laterally Directed
Flow)
Main Advantages
Depending on the design, weeping is reduced or totally
avoided; good turndown ratio; can incorporate caps as
stiening members; can support internals.
Expensive; dicult to avoid stagnant regions; more
subject to immediate bubble merger; dicult to
clean; dicult to modify; not advisable for sticky
solids; requires peripheral seal; ports not easily
shrouded.
Details of some nozzles that are currently used in
circulating uidized beds (CFB) combustors are shown
in Fig. 1 (VGB, 1994). There are signicant dierences
between bubble caps (No. 7 in Fig. 1) and nozzles (No.
1 in Fig. 1) with respect to the prevention of solids
back ow: in the case of nozzles, the high velocity of
the gas jet prevents the solids from owing back into
the wind box. On the other hand, in the case of the
bubble cap design, the gas owing out of the bubble
cap into the bed has a rather low velocity. In this case,
the backow of solids is avoided by letting the gas ow
downward from the holes in the inner tube to the lower
edge of the cap. The separation distance s
bc
is respon-
sible for the sealing eect of the bubble cap.
2.3 Sparger (Laterally or Downwardly Directed
Flow)
Main Advantages
Can minimize weeping; good turndown ratio; low pres-
sure drop; can support internals; can undergo thermal
expansion without damage; ports are easily shrouded;
well suited to multilevel uid injection; solids can ow
from above the grid to below.
Deuidized solids beneath the grid; can be a less
forgiving mechanical design.
Figure 1 Distributors and nozzles used in large circulating uidized bed combustors. (After VGB, 1994.)
2.4 Conical Grids (Laterally Directed Flow)
Main Advantages
Promotes solid mixing; prevents stagnant solids
buildup; minimizes solids segregation. Facilitates the
easy discharge of solids.
Dicult to construct; requires careful design to ensure
good gas distribution; requires high pressure drop for
good gas distribution.
2.5 Pierced Sheet Grids (Laterally Directed Flow)
Produced by punching holes in a relatively thin plate.
Holes are of a semielliptical shape with slanting,
strongly conical openings in the direction of entry. It
is primarily used in uid bed drying applications.
Holes can be oriented in such a way to promote certain
mixing patterns or drive the solids toward discharge
nozzle.
Main Advantages
Promotes solid mixing; prevents stagnant solids
buildup. Facilitates discharge of most of the solids.
The holes are angled so that the grids can be non-
weeping for coarse solids.
Dicult to construct, facilitates only small hole sizes,
requires reinforcement underneath the sheet to support
the bed.
Among the foregoing advantages and limitations,
the designer must select those most pertinent or critical
to the process application. There are, for example,
instances in which solids below the grid level are toler-
able, where grid thermal expansion is signicant,
where bed solids are very friable, where pressure
drop, and therefore the cost of compressive horse-
power, is critical, where solids are sticky and must
be kept in motion throughout, where internal impellers
or stirrers must be provided, or where grids are
expected to have a short life due to corrosion. These
and many other specics have dictated a host of design
variations, some of which are illustrated below. It
should be emphasized that each application requires
thoughtful engineering consideration before nal
design selection.
3 GRID DESIGN CRITERIA
3.1 Jet Penetration
Gas owing from the grid holes can take the form of
either a series of bubbles or a permanent jet, depending
on system parameters and operating conditions.
However, a permanent jet prevails for most industrial
conditions. Jet penetration is one of the most impor-
tant design parameters since it helps in
1. Determining how far to keep the bed internals,
such as feed nozzles, heat exchanger tubes, etc.,
away from the grid to minimize erosion of
internals.
2. Deciding on grid design parameters such as
hole size and the gas jet velocity required to
achieve a certain jetting region.
3. Minimizing or maximizing particle attrition at
grids.
Knowlton and Hirsan (1980) reported that the jet
penetration for upwardly directed jets uctuated
greatly. Karri (1990) noted that jet penetration can
vary as much as 30% for upwardly directed jets.
However, the jet emanating from a downwardly direc-
ted grid hole is stable, and its penetration length does
not signicantly uctuate with time. Figure 2 indicates
jet penetration congurations for jets oriented
upwardly, horizontally, and downwardly. According
to Karri, the jet penetrations for various orientations
can be approximately related by:
L
up
- 2L
hor
- 3L
down
(1)
There are numerous jet penetration correlations (Zenz,
1969; Shakhova, 1968; Merry, 1971; Yang and
Keairns, 1979; Knowlton and Hirsan, 1980; Yates et
al., 1986; Blake et al., 1990; Roach, 1993) in the litera-
ture. Massimilla (1985) and Karri (1990) have shown
that the jet penetrations predicted by these correlations
can vary by a factor of 100 or more. Among them,
Merrys correlation for horizontal jets was shown
(Karri, 1990; Chen and Weinstein, 1993; Roach,
1993) to give reliable predictions, although this corre-
lation was derived for horizontal jets issuing into an
incipiently uidized bed, which is not exactly the same
situation as for a grid jet. Merrys correlation to calcu-
late the penetration of horizontal jets is
L
hor
d
h
= 5:25
r
g;h
U
2
h
r
p
(1 e
mf
)gd
p
!
0:4
r
g;b
r
p
!
0:2
d
p
d
h

0:2
(2)
The jet penetration lengths for upwardly and down-
wardly directed jets can be calculated from Eq. (1).
These equations take into account the eects of pres-
sure and temperature on jet penetration. Knowlton
and Hirsan (1980) and Yates et al. (1986) found that
the jet penetration increases signicantly with system
pressure. In addition, Findlay and Knowlton (1985)
found that the jet penetration decreases with increasing
system temperature. Bed internals should not be placed
in the jetting zone near the grid, otherwise the internals
could be severely eroded.
3.2 Grid Pressure Drop Criteria
For a grid, achieving equal distribution of gas ow
through many parallel paths requires equal resistances
and sucient resistance to equal or exceed the maxi-
mum value of any unsteady state pressure uctuation.
It has been determined experimentally that the head
of solids in some uidized beds above an upwardly
directed grid port can vary momentarily by as much
as 30%. This is due to large uctuations in the jet
penetration for an upwardly directed jet, as discussed
in the previous section. The equivalent variation down-
stream of a downwardly directed port is less than 10%.
Thus as a rule of thumb, the criteria for good gas
distribution based on the direction of gas entry are
(Karri, 1990):
1. For upwardly and laterally directed ow:
P
grid
_ 0:3P
bed
(3)
2. For downwardly directed ow
P
grid
_ 0:1P
bed
(4)
and
3. Under no circumstances should the pressure
drop across a large-scale commercial grid be
less than 2500 Pa, i.e.,
P
grid
_ 2; 500 Pa (5)
Figure 2 Jet penetrations at grid holes for dierent orientations.
Several investigators (Hiby, 1964; Zuiderweg, 1967;
Whitehead, 1971; Siegel, 1976; Mori and Moriyama,
1978) have found the ratio of distributor pressure drop
to bed pressure drop to be in the range of 0.015 to 0.4.
If turndown is desired, the grid pressure drop
criteria (Eqs. 3 and 4) should apply at the minimum
gas ow rate. This can be a problem for circulating
uidized bed combustors, since this means that under
full load the grid pressure drop will be unacceptably
high. Also, if the grid is curved, i.e., concave, convex,
or conical, the criterion must apply with respect to the
lowest hole on the grid. Take an example of a uid bed
with curved grid, as shown in Fig. 3.
A pressure balance across the curved grid can be
written as
P
h
(Highest hole) = P
h
(Lowest hole)
r
B
g (H
high
H
low
)
(6)
i.e.,
P
h
(Highest hole) = P
h
(Lowest hole)
480 9:8 0:9
= P
h
(Lowest hole) 4235 Pa
(7)
Therefore the lowest grid hole has the lowest pressure
drop, and hence the pressure drop criterion must apply
with respect to the lowest hole on the grid.
3.3 Design Equations
The following equations can be used to design perfo-
rated plates, spargers, and bubble cap types of grids:
Pressure drop across the grid:
P
grid
= Kgr
B
L
B
P
grid
_ 2; 500 Pa (8)
where K = 0:3 for upward and lateral gas entry and 0.1
for downward gas entry.
The gas velocity through the grid hole (orice equa-
tion):
U
h
= C
d
2P
grid
r
g;h
s
(9)
The orice discharge coecient, C
d
, is typically about
0.6 for gas owing through an orice in a pipe (for a
ratio of orice diameter to pipe diameter in the range
of 0 to 0.2). This value of the orice coecient is for a
sharp-edged orice. However, grids are not sharp
edged, and the orice coecient is then greater than
0.6. A typical value of C
d
for a grid hole is about 0.8.
Actually, the value of C
d
depends on the grid plate
thickness and the hole pitch. It can be calculated
from Fig. 4 (Karri, 1991).
Volumetric ow rate of gas:
Q = N
pd
2
h
4
U
h
(10)
Figure 3 A typical uid bed showing a curved perforated plate.
3.3.1 Hole Size
To increase the gas residence time in the bed, it is
desirable to introduce the greatest number of small
gas bubbles as possible into the bed. This can be
achieved by maximizing N at the expense of d
h
in
Eq. (10) (within the limits of mechanical, cost, and
scaleup constraints). To minimize stagnant zones, the
number of grid holes per m
2
should be _ 10. In prac-
tice, the number of grid holes per square meter should
be greater than 20.
3.3.2 Hole Layout
To increase the uniformity of uidization, it is com-
mon to lay out the holes in triangular or square pitch,
as shown in Fig. 5. All the holes in a grid with trian-
gular pitch are equidistant. This is not the case for a
grid with square pitch. Triangular pitch will also result
in more holes per unit area.
The relationship between the grid hole pitch, L
h
,
and the number hole density (holes per unit area of
the bed), N
d
, depends on whether the holes are laid
out in triangular or square pitch.
3.4 Additional Criteria for Sparger Grids
Additional distribution criteria are used for sparger
grids. To keep the pipe header pressure drop down
to acceptable levels and to ensure good gas distribu-
tion, the following criteria (Karri, 1990) should be met:
1. The manifold should be sized based on the
following equation:
D
2
m
N
h
d
2
h
!
2
> 5 (11)
Figure 4 Grid hole discharge coecient design chart.
Figure 5 The relationship between hole density and grid hole pitch for both triangular and square pitch.
The parameters in Eq. (11) are dened in Fig.
6. Eq. (11) ensures that the pressure drop in the
manifold is negligible compared to the pressure
drop in the holes, which are determining the
grid pressure drop.
Similarly, the main header pipe should be
sized based on the equation
D
2
head
N
m
D
2
m
!
2
> 5 (12)
2. In some instances, two to three dierent hole
size are used on a given manifold to get better
gas distribution.
3. The gas velocity in the header/manifold pipe
should be < 25 m=s for best distribution.
4. Holes should not be located closer than one
D
m
from any sharp bend or tee in the
header/manifold to prevent solids from being
sucked into the manifolds due to the vena
contracta eect.
3.5 Port Shrouding or Nozzle Sizing
Shrouds are generally placed around grid holes to
reduce the velocity at the gassolid interface and
reduce particle attrition (de Vries et al., 1972).
Shrouds simply consist of short pipes centered over
the smaller grid holes that have been selected in size
and number to operate at a hole velocity dened by
Eq. (9).
To be eective, shrouds must be long enough to
contain the expanding (11
included angle) gas jet

leaving the grid orice (Karri, 1991).
As can be seen from Fig. 7, the minimum shroud
length should be:
L
min
=
D
s
d
h
2 tan 5:5
(13)
In practice, it is prudent to increase L
min
by a factor of
50 to 100%. A shroud length less than L
min
causes
signicantly more erosion and attrition than no shroud
at all. Signicant attrition can also occur if the shroud
is not centered over the smaller hole.
The nozzle or shroud details inside a sparger pipe
grid are illustrated in Fig. 8.
If properly sized and installed, particle attrition is
reduced by a factor (Karri, 1990) calculated from
particle attrition without shrouds
particle attrition with shrouds
=
D
s
d
h

1:6
(14)
4 PARTICLE ATTRITION AT GRIDS
Solids immediately surrounding the gas jets issuing
from the grid are ingested into the jets. These particles
are accelerated and collide with the particles near the
tip of the jet. Figure 9 depicts how the particles
are picked up and slammed into a uidized, yielding
bed for an upwardly directed jet. However, downward-
pointing jets generally issue into a nonuidized area of
particles. Therefore particles picked up by downwardly
directed jets, issuing into a nonyielding unaerated bed,
result in a greater degree of particle attrition than those
for upwardly directed jets. Karri (1990) reported that
downwardly directed jets have approximately twice the
steady-state attrition rate as that of upwardly directed
jets. The attrition rates for upwardly and laterally
directed jets are essentially the same. Grid jet attrition
is discussed in greater detail in Chapter 8 of this book.
5 EROSION
5.1 Erosion at Bed Walls and Internals
Erosion in the grid region is primarily due to high-
velocity submerged jets impinging on distributor
parts, bed walls, or bed internals. Therefore one should
estimate the jet penetration heights for a given grid
design and check for the following:
1. Bed internals should not be placed in the jet-
ting zone near the grid, otherwise the internals
could be severely eroded.
Figure 6 Manifold sparger grid showing the denitions of
various parameters.
2. Nozzles should not be located any closer than
half the jet penetration height from the bed
wall.
The basic equation for erosion rate is of the form
(Karri, 1990)
Erosion o
K
e
r
2
g;h
U
3
h
d
2
h
d
3
p
r
2
p
(15)
5.2 Erosion at Distributor Nozzles
Erosion in the nozzle or orices is often associated with
weepage of solids. This can be avoided by carefully
designing a grid with the proper pressure drop criteria,
as presented in Sec. 3.2. Poorly designed bubble caps
tend to have erosion problems due to the secondary
circulation of solids. Therefore bubble caps should be
designed to minimize secondary circulation of solids.
Erosion has often been experienced at the nozzles
used in CFB combustors (Fig. 1). A dominant mechan-
ism leading to erosion is the pressure-induced gas ow
reversal that will be discussed below in Sec. 6. Solids
which have entered into the nozzle during a period of
ow reversal are entrained out once the gas ows at
high velocity in the outward direction again. The
entrained solids in the high-velocity ow in the nozzle
hole may cause severe erosion of the wall of the hole. A
Figure 7 (a) Diverging free jet; (b) shroud too short to contain the jet; (c) minimum shroud length required to contain jet.
Figure 8 Shroud design for a sparger grid.
second mechanism of erosion was observed by high-
speed video in cold models (Hartge and Werther,
1998): even when the gas was owing out of the hole
into the bed, a region near the mouth of the orice
could be observed where the gas jet entrained particles
into the hole. These entrained particles caused erosion
at the outer edge of the hole. Figure 10 shows the
photography of a nozzle that had been painted in
black before the experiment. After 60 hours of opera-
tion, the erosion marks were clearly visible. They were
particularly obvious at the lower edges of the holes,
which is due to the fact that the jet issuing from a
horizontal bore tends to bend into the upward direc-
tion (see Fig. 2), which gives more surface area to
entrain solids at the lower edge of the hole.
6 WEEPAGE OF SOLIDS
Solids weepage has been a major problem during the
development of circulating uidized bed (CFB)
combustors in the past two decades. Seemingly well-
designed nozzle grids experienced weepage to such an
extent that CFB boilers had to be shut down after
several days or weeks of operation, because most of
the bed inventory had wept through the grid. More
recent investigations (Hartge and Werther, 1998;
Karri, 1991) have revealed that pressure uctuations
in the dense bottom bed of the CFB riser may cause
the backow of solids through the grid hole. Figure 11
shows measurements of the pressure drop between the
wind box and at a height of 0.3 m above the distributor
in a circulating uidized bed. The average pressure
drop of gas distributor and bed was about 40 mbar
during these measurements. As can be seen, sometimes
negative pressure drops occurred, i.e., the pressure at a
height of 0.3 m above the distributor was higher than
Figure 9 The mechanism of particle attrition at a sub-
merged jet.
Figure 10 Erosion marks around the gas outlet holes of
the nozzle with 5 mm diameter holes. (From Hartge and
Werther, 1998.)
Figure 11 Pressure ucutations measured in a circulating
uidized bed between the windbox and at a height of 0.3 m
above the distributor (riser diameter, 0.4 m; supercial gas
velocity, 3 m/s; solids mass ux, 22 kg/m
2
s).
the pressure below the distributor. In such cases, gas
ow reversal occurs, which results in weepage of solids
into the nozzles (Fig. 12).
In order to prevent such a ow reversal, the design
pressure drop of the nozzle in the example of Fig. 11
should be roughly 20 mbar larger, i.e., the pressure
drop of the grid has to take into account the largest
possible pressure uctuations. These pressure drop
uctuations are signicantly higher for Group B par-
ticles with a wide particle size distribution which are
typical of CFB combustion. Another reason that
makes it dicult to keep the grid pressure drop always
higher than the largest pressure uctuations in the bed
is the necessity of frequent turndown operations for
CFB combustors. Reducing the load to 50% means a
reduction of the grid pressure drop to 25% of its value
at full load. In CFB combustion, the bubble cap design
of Fig. 1 (case 7) has shown to be more eective. Here
the solids have to be transported upwards in the annu-
lus between the bubble cap and central tube against
gravity during the gas ow reversal. The solids are
prevented from entering the central tube if the separa-
tion length s
bc
is kept between 70 and 100 mm. Figure
13 shows solids inow during the period of gas ow
reversal.
7 EFFECTS OF TEMPERATURE AND
PRESSURE
System temperature and pressure aect the momentum
of grid jets via the gas density. The momentum of the
gas jets is proportional to r
g;h
U
h
. When the tempera-
ture is increased, the gas density decreases. For the
same gas jet velocity, this decreases the momentum
of the jets and therefore decreases the jet penetration
and the attrition at the grid. Similarly, when system
pressure is increased, gas density increases, gas jet
momentum increases, and therefore the jet penetration
and the attrition at the grid are increased.
8 PLENUM DESIGN
The plenum, or windbox, is the chamber immediately
below the grid. If the bed-pressure-drop-to-grid-
pressure-drop ratio is high enough, the plenum design
will probably not be too important. However, for the
case where this ratio is low, the plenum design may
determine whether the bed will operate satisfactorily.
The typical plenum designs showing various cong-
urations for introducing gas into the plenum are
illustrated in Fig. 14. Common sense dictates that cer-
tain plenum designs be preferred over others. If the gas
enters the plenum from the bottom it is preferable that
the plenum have a large enough distance between the
outlet of the supply pipe and the grid to prevent the gas
from preferentially passing through the middle of the
grid. When gas enters a plenum from the side, it is
preferable to rout the gas to the middle of the plenum
(Fig. 14c) rather than have the supply pipe end at the
wall of the plenum. In addition, horizontal-to-vertical
down gas entry (Fig. 14c) is preferable over the
horizontal-to-vertical up gas entry (Fig. 14b).
If the gassolid or gasliquid suspension needs to be
introduced into the plenum, as for example in a poly-
ethylene reactor and some FCC regenerators, it is
Figure 12 Solids owing into the nozzle shaft during gas
ow reversal.
Figure 13 Solids owing into the cap during gas ow rever-
sal.
preferable to introduce the suspension at the lowest
point of the plenum (Fig. 14a,d,e) to minimize the
accumulation of solids or liquids in the regions in-
accessible to reentrainment For two-phase systems, it
is preferable to have some sort of deection device
(Fig. 14d,e,f) between the outlet of the supply pipe
and the grid to prevent the solids from preferentially
passing through the middle of the grid due to their high
momentum. This preferential bypassing of solids
causes maldistribution of gas. In addition, the cong-
uration of Figs. 14e and 14f are preferable over the
congurations of Figs. 14a,d.
9 POWER CONSUMPTION
Since the grid contributed a considerable fraction of
total pressure drop across a given uid bed system, it
is always important to estimate the power consump-
tion of the blower that drives the gas through this
system. Suppose a stream of gas is to be compressed
from an initial pressure of P
1
to a higher pressure of P
2
to pump it through the entire uid-bed system. Using
thermodynamics for adiabatic reversible compression
with negligible kinetic and potential energy eects, the
ideal shaft work to compress each kilogram of gas is
given by
W
s;ideal
=
P
2
P
1
dP
r
g
(16)
If an ideal gas behavior is assumed, then Eq. (16) trans-
forms into
W
s;ideal
=
g
g 1
P
1
Q
1
P
2
P
1

(g1)=g
1
" #
(17)
or
W
s;ideal
=
g
g 1
P
2
Q
2
1
P
1
P
2

(g1)=g
" #
(18)
Due to heat of compression, the raise in temperature
can be calculated from
T
2
= T
1
P
2
P
1

(g1)=g
(19)
where g = ratio of specic heats of gas 1:67, 1.4, and
1.33 for monatomic, diatomic, and triatomic gases,
respectively.
Figure 14 Dierent plenum congurations.
However, for real operations with its frictional
losses, the actual work required is always greater
than the ideal and is given by
W
s;actual
=
W
s;actual
Z
(20)
where Z is the blower eciency, approximately given
by
Z = 0:550.75 for a turboblower
= 0:60.8 for a roots blower
= 0:80.9 for an axial blower or a two-
stage reciprocating compressor
Equation (20) can be used not only for power con-
sumption but also to size the correct horse power
motor to drive the blower.
The actual temperature of gas leaving a well-insu-
lated (adiabatic) but not 100% ecient compressor is
then calculated from
T
2
= T
1
T
1
Z
1
P
2
P
1

(g1)=g
1
" #
(21)
10 DESIGN EXAMPLES
10.1 FCC Grid Design
Example 1. A 13-m-ID bed of FCC catalyst
(d
p
= 60 mm) 3 m deep is to operate at a supercial
gas velocity of 0.6 m/s. The bed density is 480 kg=m
3
.
the density of the gas entering the bed is 0:64 kg=m
3
.
Design the following grid types: (1) a at perforated
plate, and (2) concentric-ring type downow sparger.
Assume the grid thickness to be 0.025 m.
Solution. Perforated Plate Design
Determine P
bed
and P
grid
:
P
bed
= gr
B
L
B
= 9:8 480 3 = 14,112 Pa
Choose P
grid
to be 30% of P
bed
P
grid
= 0:3 P
bed
= 4,234 Pa
Determine the gas velocity through the grid holes
(assume a typical value for c
d
0:77):
U
h
= C
d
2P
grid
r
g;h
s
= 0:77
2 4234
0:64
r
= 88:6 m=s
Determine the volumetric gas ow rate at the con-
ditions below the grid. For this example, assume
that the temperature of the gas below the grid is
the same as in the bed. This may not be the case
in an actual plant.
Q = U
sup
pD
2
4
= 0:6
p(13)
2
4
= 79:6 m
3
=s
Determine the number of grid holes required:
Since Q = N
d
2
h
4
U
h
Therefore
N =
Q
U
h
1
pd
2
h
=4
=
79:6
88:6
1
pd
2
h
=4
=
1:14
d
2
h
The hole density is
N
d
=
N
(p=4)D
2
=
1:14
d
2
h
1
(p=4)(13)
2
=
0:0086
d
2
h
Determine the hole pitch for a triangular arrange-
ment:
L
h
=
1
N
d
sin 60
p = 11:59 d
h
Downwardly Directed Gas Sparger Design
Choose P
grid
to be 10% of P
bed
P
grid
= 0:1P
bed
= 1,411 Pa = 14:4 cmH
2
O
1411 Pa is less than the minimum of 2500 Pa P
required for a grid. Therefore use P
grid
=
2500 Pa.
U
h
= 0:77
2 2500
0:64
r
= 68 m=s
N =
79:6
68
1
pd
2
h
=4
=
1:5
d
2
h
Various combinations of N and d
h
satisfy the pres-
sure drop requirements for the two grid type as
shown in the table:
To proceed with the design, it is necessary to select a
hole size (judgment call). For the purpose of this
d
h
Number of holes (N)
m perforated plate downow sparger
0.005 45,600 60,000
0.01 11,400 15,000
0.025 1,824 2,400
0.05 456 600
example, a hole size of 0.025 m will be chosen to
compare the dierent grid types. This hole dia-
meter does not result in an excessive number of
holes for both types of grids.
Check the value for C
d
for the perforated plate:
N
d
= 13:8 holes=m
2
and L
h
= 0:29 m
t
d
h
=
0:025
0:025
= 1
From Fig. 3
C
d
L
h
d
h

0:1
= 0:96
C
d
= 0:96
0:025
0:29

0:1
= 0:75 vs. 0.77 (initial guess)
There is a fairly good agreement between the initial
and calculated values for C
d
. If not, one must
repeat the calculations using the calculated C
d
until both values agree.
Therefore, for a 0.025 m hole diameter, the
perforated plate has 1,824 holes arranged in a
triangular pitch of 0.29 m. The hole density is
13:8 hole=m
2
.
For the sparger grids, it remains to determine the
sparger conguration and pipe-header size. Pipe
headers can be laid out in various congurations.
The design calculations will depend on the con-
guration one chooses.
Concentric-Ring Sparger. Consider for example,
a conguration of four concentric rings of 0.4 m dia-
meter supplied by a number of gas entry points.
This design results in 2,401 holes.
Determine the hole pitch:
L
h
=
97:13
2401
= 0:04 m
To determine the header-pipe size, rst determine
the maximum number of holes in ring section
supplied by a single eective entry of gas. If out-
ermost ring is supplied by four gas entry points,
then the number of eective gas entry points is 8,
and the number of holes in each section of ring
No. 4 would be N
h
= 978=8 = 122. Then Eq. 11
gives
D
2
head
N
h
d
2
h
!
2
> 5
D
2
head
122 0:025
2
!
2
> 5 or D
head
> 0:41 m
Summary. For an orice diameter of 0.025 m, the
downwardly directed concentric-ring sparger has
2,401 nozzles placed on four concentric rings.
The pitch is 0.04 m. Sometimes the holes are
staggered on the sparger pipe. Also it is a com-
mon practice to place two nozzles at a given cross
section as shown in Fig. 8.
Example 2. For the conditions of Example 1 of
perforated plate design, estimate the submerged jet
height in the uidized bed.
Solution. Perforated Plate
U
h
= 88:6 m=s r
g;h
= 0:64 kg=m
3
r
g;b
= 0:5 kg=m
3
d
h
= 0:025 m N = 1; 824
d
p
= 60 mm r
p
= 1440 kg=m
3
e
mf
= 0:42
Gas jet penetration depth using Merrys correlation
(Eq. 3.4.2) for horizontal jets
L
hor
d
h
= 5:25
r
g;h
U
2
h
r
p
(1 e
mf
)gd
p
!
0:4
r
g;b
r
p
!
0:2
d
p
d
h

0:2
Sparger Grid, Concentric Ring Type
Ring no.
(i)
Radius of each ring
(r
i
), m
Length of each ring
(L
i
) 2r
i
, m % of total length
Number of holes on each ring
(N
i
)
1 1.43 8.98 9.24 222
2 3.05 19.16 19.73 474
3 4.68 29.41 30.28 727
4 6.30 39.58 40.75 978
Total = 97.13 2,401
L
hor
= 5:25
0:64 88:6
2
1440(1 0:42)9:8 65 10
6
!
0:4
0:5
1440

0:2
65 10
6
0:025
!
0:2
0:025 = 0:32 m
From Eq. (1),
L
up
- 2L
hor
- 2 0:32 - 0:64 m
Example 3. For the conditions and the perfo-
rated plate dened in Example 1, design a shroud
having an ID twice that of the grid hole, i.e.,
D
S
= 2d
h
= 0:05 m
Solution. Perforated Plate
The minimum length of the shroud should be
L
min
=
0:05 0:025
2 tan 5:5
= 0:13 m
The gas jet velocity emanating from the shroud is
U
h;s
= U
h
d
h
D
s

2
= 88:6
0:025
0:05

2
= 22:2 m=s
Particle attritions rate will be reduced by a factor
calculated from Eq. (14):
particle attrition without shrouds
particle attrition with shrouds
=
D
s
D
h

1:6
=
0:05
0:025

1:6
= 3:0
Thus adding a shroud to the grid reduces the attri-
tion rate to 67% of the rate without a shroud.
10.2 Polyethylene Reactor Grid Design
Example 4. Design a at perforated-plate grid
for the polyethylene reactor schematically shown in
Fig. 15 and calculate the gas jet penetration depth.
Use a triangular pitch. System parameters are
U
sup
= 0:5 m=s r
g;h
= 19:2 kg=m
3
r
g;b
= 17 kg=m
3
r
p
= 641 kg=m
3
r
B
= 272 kg=m
3
P
grid
= 0:4P
bed
d
h
= 0:01 m d
p
= 508 mm e
mf
= 0:45
t = 0:019 m
Solution
Determine P
bed
and P
grid
P
bed
= gr
B
L
B
= 9:8 272 12:2 = 32,520 Pa
P
grid
= 0:4P
bed
= 13,008 Pa
Determine the gas velocity through the grid hole
(trial and error). Assume C
d
= 0:8:
U
h
= C
d
2P
grid
r
g;h
s
= 0:8
2 13008
19:2
r
= 29:5 m=s
Determine the volumetric ow rate of gas
Q = U
sup
pD
2
4
= 0:5
p(4:6)
2
4
= 8:3 m
3
=s
Determine the number of grid holes required:
N =
Q
U
h
1
pd
2
h
=4
=
8:3
29:5
1
(p=4)(0:01)
2
= 3582
Hole density:
N
d
=
3582
(p=4)(4:6)
2
= 215 holes/m
2
Figure 15 Schematic of polyethylene reactor.
Determine the hole pitch:
L
h
=
1
N
d
sin 60
p =
1
215 sin 60
_ = 0:073 m
Check the value for C
d
:
t
d
h
=
0:019
0:01
= 1:9
From Figure 3.4.4,
C
d
L
h
d
h

0:1
= 0:98
therefore
C
d
= 0:98
0:01
0:073

0:1
= 0:803 - 0:80 (great guess)
Gas jet penetration depth using Merrys correlation
[Eq. (2)] for horizontal jets:
L
hor
d
h
= 5:25
r
g;h
U
2
h
r
p
(1 e
mf
)gd
p
!
0:4
r
g;b
r
p
!
0:2
d
p
d
h

0:2
L
hor
= 5:25
19:2 29:5
2
641(1 0:45)9:8 508 10
6
!
0:4
17
641

0:2
508 10
6
0:1
!
0:2
0:01 = 0:55 m
From Eq. (1),
L
up
- 2L
hor
- 2 0:55 - 1:1 m
Coalescence factor:
l =
L
h
L
up
=2
=
0:073
1:1=2
= 0:13 < 1
Therefore Jets coalesce. The low value of l indicates
that the bed of solids is probably suspended
above the coalesced jets. Therefore the solids
rarely come into contact with the grid. This
type of design reduces the chances of grid plug-
gage due to sticky polyethylene solids.
Summary: The perforated plate has 3,582 holes, each
of 0.01 m diameter, arranged in a triangular pitch
of 0.073 m. The hole density is 215 holes=m
2
.
10.3 Power Consumption
Example 5. Determine the compressor power to
pass reactant gas into the plenum of the uid bed
system. Also calculate the temperature rise due to
heat of compression. The system parameters are
P
grid
= 6 kPa; P
bed
= 15 kPa; P
cyclonesfilters
= 12 kPa; pressure at the exit of the filters = 350 kPa.
Gas entering the compressor: T
1
= 20
C; P
1
=
101 kPaq; Q
1
= 10 m
3
=s.
Use Z = 0:85; g = 1:4
Solution
Determine compressor discharge pressure, P
2
:
P
2
= P
exit
P
cyclonesfilters
P
bed
P
grid
= 350 12 15 6 = 388 kPa
Determine ideal power consumption, W
s;ideal
W
s;ideal
=
g
g 1
P
1
Q
1
P
2
P
1

(g1)=g
1
" #
W
s;ideal
=
1:4
1:4 1
101 10
383
101

(1:41)=1:4
1
" #
= 1638 kW
Determine actual power consumption, W
s;actual
:
W
s;actual
=
W
s;ideal
Z
=
1638
0:85
= 1927 kW (or 2587 hp)
Determine the temperature rise, T
2
:
T
2
= T
1
T
1
Z
1
P
2
P
1

(g1)=g
1
" #
T
2
= 293
293
0:85
383
101

(1:41)=1:4
1
" #
= 453 (or 180
C)
NOMENCLATURE
C
d
= discharge coecient; see Fig. 4
d
h
= grid hole diameter, m
d
p
= Sauter mean particle size, m
D = diameter of uid bed, m
D
head
= diameter of the main header pipe, m
D
m
= diameter of the manifold pipe, m
D
s
= shroud or nozzle diameter, m
g = gravitational acceleration 9:8 m=s
2
H
high
= elevation of highest grid hole for curved grid,
m
H
low
= elevation of lowest grid hole for curved grid,
m
K = grid pressure-drop coecient; see Eq. (8)
0.3 for upward gas entry;
0.1 for lateral and downward gas entry
K
e
= erosion constant, Eq. (15)
L
B
= operating bed depth, m
L
down
= jet penetration for downwardly directed jet, m
L
h
= grid hole pitch, cm
L
hor
= jet penetration for horizontally directed jet, m
L
min
= minimum shroud or nozzle length, m
L
s
= shroud or nozzle length, m
L
up
= jet penetration for upwardly directed jet, m
N = number of grid holes
N
d
= number of hole density (holes per unit area of
the bed), holes=m
2
N
h
= maximum number of holes per manifold pipe
section supplied by gas entry
N
m
= number of manifolds on the main header
supplied by single gas entry point
P
1
= pressure of gas entering the blower, Pa
P
2
= pressure of gas leaving the blower, Pa
Q = total volumetric gas ow entering the grid,
m
3
=s
Q
1
= total volumetric gas ow entering the blower,
m
3
=s
Q
2
= total volumetric gas ow leaving the blower,
m
3
=s
t = grid thickness, m
T
1
= temperature of gas entering the blower,

K
T
2
= temperature of gas leaving the blower,

K
U
h
= velocity of gas through the grid hole, m/s
U
sup
= supercial gas velocity, m/s
W
s;actual
= actual power consumption due to shaft work,
W
W
s;ideal
= ideal power consumption due to shaft work,
W
a = energy eciency factor
g = ratio of specic heats of gas
r
B
= operating bed density, kg=m
3
r
g;b
= density of gas at bed operating conditions,
kg=m
3
r
g;h
= density of gas entering the grid hole (plenum
conditions), kg=m
3
r
p
= particle density, kg=m
3
e
mf
= voidage at minimum uidizing conditions
y = included angle of gas jet, degrees
P
bed
= pressure drop across the dense bed, Pa
P
grid
= pressure drop across the grid, Pa
P
h
= pressure drop across the grid hole, Pa
= particle shape factor
Z = compressor eciency
REFERENCES
Blake TR, Webb H, Sunderland PB. Chem Eng Sci 45:365,
1990
Chen L, Weinstein H. AIChE J 39(12):1901, 1993.
Findlay J, Knowlton TM. Final Report for US Dept of
Energy, Project, DE-AC21-83MC20314, 1985.
Hartge E-U, Werther J. In: Fan LS, Knowlton TM, eds.
Fluidization IX. Engineering Foundation, 1998, p 213.
Hiby JW. Chem-Ing Techn 36:228, 1964.
Karri SBR. PSRI Research Report No. 60, 1990.
Karri SBR. Grid Design Chapter, PSRI Design Manual,
1991
Knowlton TM, Hirsan I. In: Grace JR, Matsen JM, eds.
Fluidization. Plenum Press, 1980, p 315.
Massimilla L. In: Davidson JF et al., eds. Fluidization.
Academic Press, 1985, p 133.
Merry JMD. Trans Instn Chem Engrs 49:189, 1971.
Mori S, Moriyama A. Inst Chem Eng 18:245, 1978.
Roach PT. Fluid Dyn Res 11:197, 1993.
Shakhova NA, Inzh Fiz Zh 14(1):61, 1968.
Siegel R. AIChE J 25:590, 1976.
Sishtla C, Findlay J, Chan I, Knowlton TM. In: Grace JR,
Shemilt LW, Bergougnou MA, eds. Fluidization VI.
Engineering Foundation, 1989, p 581.
VGB Powertech, Design Report M218H, VGB
Kraftwerkstechnik, 1984.
de Vries RJ, van Swaaij WPM, Mantovani C, Heijkoop A.
Proc 5th Europ Symp Chem Reaction Engng,
Amsterdam, 1972, p 3959.
Whitehead AB. In Davidson JF, Harrison D, eds.
Fluidization. Academic Press, 1971, p 781.
Yang WC, Keairns DL. Ind Eng Chem Fundam 18:317,
1979.
Yates JG, Bejcek V, Cheesman DJ. In: Ostergaard K,
Sorensen A, eds. Fluidization V. Engineering
Foundation, 1986, p 79.
Systems. Reinhold, 1960, p 171.
Zenz FA. Inst Chem Eng Symp 30:136, 1968.
Zuiderweg. In: Drinkenburg, ed. Proc Int Symp on
Fluidization. Netherlands University Press, 1967, p 739.
7
Effect of Internal Tubes and Bafes
Yong Jin, Fei Wei, and Yao Wang
Tsinghua University, Beijing, Peoples Republic of China
In the past sixty years, worldwide research has led to
an essential understanding of the most important
properties of uidized beds. In particular, it has been
found that the ow structures of GS, LS, and GL
S uidization mainly vary with increasing uid velo-
city. For example, several ow patterns or regimes
have been identied in gassolid uidized beds
(Grace, 1986), such as particulate uidization, bub-
bling/slugging uidization, turbulent uidization, fast
uidization, and pneumatic conveying regimes (Fig. 1).
The complexity of gassolid ow structure of the
above regimes is as follows:
1. There are quite dierent ow behaviors on an
equipment scale, such as the intensity of gas
solid overall movement and backmixing, gas
solid dispersion, and the residence time distri-
bution (RTD) of both gas and solid in beds.
2. The bed parameters on a micro scale vary with
the radial/axial bed locations and time.
3. There are many substructures in beds, such as
bubbles, clouds, wakes, channels, and clusters;
their size distribution in time and space, as well
as the frequency of their presence and disap-
pearance, are normally quite complicated.
4. The eect of particle size and size distribution
on ow regime is complex.
The inuence of complex ow structures on reactor
performance is complicated, even in scaled up reactors.
Internals are usually introduced to modify the above
gassolid ow structures, in an eort to form a more
uniform and active gassolid ow to enhance heat and
mass transfer so as to improve the overall performance
of uidized bed reactors, especially to make the scale-
up easier. Therefore the study and improvement of
internals are very important for raising the productiv-
ity of uidized bed reactors and extending the useful-
ness of uidization processes.
In this chapter the mechanism of the eect of inter-
nals to improve the performance of uidized beds is
discussed, and highly ecient internal tubes and baes
used in gassolid uidized beds are presented. First,
the inherent structures of two-phase ow without
internals are introduced in Sec. 1 so that the impor-
tance and necessity of internals for gassolid uidized
beds can be explored. Second, various internal tubes
and baes are collected from the literature; they are
classied and discussed in Sec. 2. Recent progress and
some simple criteria for designing in this area are also
covered. Section 3 focuses on explaining the eects of
internals on two-phase ow.
1 INHERENT FLOW STRUCTURES OF GAS
SOLID FLOW
Bubbles, clusters and nonuniform ow structures are
inherent characteristics of the gassolid uidization
systems, which greatly inuence the performance of
uidized beds in many applications. Thus an under-
standing of the inherent ow structures is essential
for introducing internal tubes and baes with clear
purpose, and judging the improvement eect of inter-
nals objectively.
1.1 Flow Structure of Bubbling Fluidization
When the gas velocity approaches and goes over the
mf
, all ne bed parti-
cles that belong to Group A in the Geldart particle
classication scheme are suspended by the gas. The
whole bed expands uniformly, and the gas passes the
bed through the particles. This is called the particulate
uidized regime. As the gas velocity increases to bub-
bling velocity, U
b
, bubbles begin to appear in the bed.
But for the Group B particles, the U
mf
and U
b
are the
same value.
One of the most characteristic phenomena of gas
solid uidized beds is the formation of gas bubbles,
which dominate the behavior of uidized beds in rela-
tively low-velocity regimes. In analyzing the behavior
of bubbling uidized beds, it is essential to distinguish
between the bubble phase and the emulsion phase, the
latter consisting of particles uidized by interstitial gas.
A bubbling bed can conveniently be dened as a bed in
which the bubble phase is dispersed and the emulsion
phase is continuous.
For a gassolid uidized bed, in which most gas
passes through the bed as bubbles, the operational
region of gas velocity is greatly extended by the pre-
sence of bubbles. This is one of the main reasons that
GS uidized beds are more widely used in industries
than LS uidized beds.
In addition, the rising bubbles cause the motion of
particles, which obviously intensies the solids mixing
on the macro scale and leads to the temperature uni-
formity and high bed/surface heat transfer character-
istic. This is another attractive advantage of GS
uidization.
Consider the case of gas forming a bubble within a
mass of particles. The only way in which the solids can
be moved out of the way to form a bubble is that there
must be a force being exerted by the gas owing past
the solids. Most of the gas then nds its way back into
the bubble to complete its passage through the bed.
The resulting ow into and out of the emulsion
phase enhances the mass transfer. In the case where
the bubble eruption and collapse occur at the bed sur-
face, there is also intensely enhanced mixing due to
Figure 1 Flow regime of gas-solid uidization. (From Grace, 1986.)
convection eects compared with normal diusion
eects.
However, when a bed does not contain internals of
any sort, the movement of the bubbles in the bed is
unrestricted. As bubbles rise, they gradually increase in
size and tend to move horizontally toward the center of
the bed (Fig. 2). Much of the gas is short-circuited
through the bubbles, which greatly limits the conver-
sion and selectivity of a chemical reaction, especially
for the Group B particles at high supercial gas velo-
cities.
Bubble size is very important for a GS uidized
bed. The gas in a bubble can directly contact the bed
particles only in the clouds around the bubbles. The
performance of a uidized bed reactor can be
improved by decreasing the bubble size and renewing
the bubble surface for interchanging the gas between
bubbles and the interstitial gas in the emulsion phase.
The equilibrium bubble is one that is constantly split-
ting and coalescing. In a small pilot plant or lab facil-
ity, bubbles will not grow to very large size. However,
as the system is scaled up, the bubble size increases and
with it the bubble velocity. At higher supercial gas
velocities, especially for the group B particles, the max-
imum bubble diameter appears to be as wide as the bed
diameter. Such large bubbles would violently shake the
unit as tons of catalyst is splashed when they come out
of the bed surface. In addition, mass transfer from
such an enormous bubble would be so poor that it
would limit the ability of reaction. In order to prevent
either of these phenomena from taking place, we must
nd out a mechanism that can limit the maximum
stable bubble size.
Therefore the main eect of internals in a bubbling
uidized bed is to break up and renew bubbles, thus
enhancing the interchange of gas between the bubbles
and the emulsion phase. Chemical reactions and mass
transfer can be improved only when bubbles are small
and evenly distributed throughout the bed volume.
1.2 Flow Structure of Turbulent Fluidization
As the gas velocity of a bubbling uidized bed is slowly
raised, the heterogeneous two-phase character of the
bed rst peaks and then gradually gives way to a con-
dition of increasing uniformity in a turbulent state in
which large discrete bubbles or voids are wholly
absent. A turbulent uidized bed is considerably
more diuse than a bubbling uidized bed because of
the larger renewal frequency of bubbles and the greater
freeboard activity at higher gas velocities. On the
whole, a turbulent uidization has found a more
wide application eld than a bubbling uidization.
The most attractive feature of a turbulent uidized
bed is its uniform temperature eld. Thus the conver-
sion can be obviously improved at fully developed
turbulent uidization for most isothermal catalytic
uidized bed reactors compared with bubbling
uidization. At the same time, the solids backmixing
is also violent in turbulent uidization, which in turn
intensies the gas backmixing. A wide gas RTD is very
harmful to the selectivity of a chemical process with
side reactions.
As the renewal frequency of bubbles is already fairly
large in a turbulent uidized bed, introducing internals
into such a bed is mainly not for breaking bubbles,
Figure 2 Bubble behavior in a free uidized bed.
eliminating gas short-circuits or enhancing gas inter-
change between the bubble phase and the emulsion
phase, but for limiting the gas axial backmixing and
promoting the radial movement of the gas and solids.
The transition to a turbulent behavior is related to
particle size and density. The larger the particle size
and density, the higher the transition gas velocity,
U
c
. It has been found that the transition from a bub-
bling to a turbulent regime occurs at a smaller U
c
when
suitable internals are applied, which is especially
important for Group B particles. Detailed information
is given in Sec. 3.4.
1.3 Flow Structure of Fast Fluidization
Transition from turbulent to fast uidization occurs at
the transport velocity, U
tr
, where signicant numbers
of particles are carried out from the top of the column.
At the same time, continuous and smooth feeding of
solids into the bottom of the riser should be
maintained in order to keep the stability of a fast ui-
dization. Thus a fast-uidization regime is connected
with a circulating uidized bed (CFB).
Nonuniform ow structure in space is the character-
istic of a typical CFB coreannulus structure in a CFB
along the radial direction. In the center of the bed, gas
and particle velocities are higher and solids concentra-
tion lower, while in the wall region gas and particle
velocities are much lower and solids concentration is
much higher (Fig. 3). When at high solids circulating
rate, downow of gassolid will exist in the wall
region, which does not exist in bubbling and turbulent
uidized beds.
The axial solids distribution is obviously aected by
the end conguration of the bed, the total solids
inventory, and the gas velocity. A typical axial solids
distribution is dense in the riser bottom and dilute in
the top section (Fig. 4). When the inlet restriction is
strong, the axial solids concentration decreases gradu-
ally from the bottom to the top section, presenting an
exponential distribution. Reducing the solids inlet
restriction results in a higher solids circulation rate
and higher solids holdup in the riser bottom region,
and an axial solids prole like S comes into being.
This is attributed to pressure buildup in the solid
circulation system. The inexion point of the S,
where the solids density turns from high to low, rises
with the increase of solids inventory obviously.
This nonuniform ow structure in space leads to
two main problems: (1) inecient gassolid contact.
On the one hand, most gas only comes into contact
with dilute particles in the center of the bed, while in
the wall region, where there are large amounts of
clusters, only a little gas passes by. On the other
hand, the radial gas exchange and diusion rate
between the center and the wall region is rather low,
which results in a large dierence of chemical reaction
rate between them. (2) Signicant gas and solids back-
mixing is due to the nonuniform radial gassolid ow
structure, especially when there is downow in the wall
region.
Thus for a fast uidization, internals are used to
redistribute the axial and radial gassolid ow struc-
ture, that is, to improve the uniformity of gassolid
ow structure in space and to change the coreannulus
ow structure, so as to promote radial gassolid
exchange.
Figure 3 Radial distribution of gas/solids velocity and
solids concentration in a CFB.
1.4 Flow Structure of Pneumatic Conveying
When a continuous particle feeding of a CFB is cut o,
particles become dilute and well distributed in the bed.
Because of the low solids holdup and short residence
time, this bed can only be used in pneumatic
conveying, or in particular chemical reactors, where
the reaction rate is very high or extra active catalyst
is used. All internals used in risers can be applied in
this situation, so it will not be discussed here.
In general, nonuniform structures, in both time and
space, is widespread in bubbling, turbulent, and fast
uidization regimes. On the one hand, such nonunifor-
mity can enhance the mass and heat transfer of a bed.
On the other hand, it decreases the contact eciency of
gas and solids and makes the scale-up rather dicult.
Internals are usually introduced not to eliminate the
nonuniform ow structure completely but to control
its eect on chemical reactions. The function of inter-
nals varies in dierent uidization regimes, as do the
types and parameters of internals. Taking these pur-
poses into consideration, internals may be successfully
applied to catalytic reactors with high conversion and
selectivity, and some other physical processes.
If you have one of the following four requirements,
please do not hesitate to choose a suitable type of
internals:
1. To modify the nonuniform ow structure of
gassolids, that is, to avoid the overall non-
uniformity on equipment-scale, and to make
the nonuniformity in time much smaller than
the characteristic time of a chemical reaction;
for example, to reduce the average bubble size
in a bubbling uidized bed.
2. To enhance the gassolid contact and subse-
quently the mass and heat transfer, so as to
eliminate them as rate limiting factors to a che-
mical reaction.
3. To improve the gassolid movements and mix-
ing along the radial direction, and to decrease
the axial gassolid backmixing to a certain
extent.
4. To control the holdup of solids and to reduce
the carryover of particles.
2 CLASSIFICATION OF VARIOUS
INTERNALS
Internals include tubes, baes, and other obstacles
inside a uidized bed. In some cases, for example,
with a strongly exothermic or endothermic reaction
in which substantial amounts of heat need to be
removed or supplied from the bed, immersed tubes
may be a necessity as surfaces for heat exchange. In
other cases, baes or other obstacles can act as inter-
nals to improve the quality of uidization or to divide
a bed into a number of stages in parallel or in series to
promote smooth uidized bed operation. It should be
recognized that cyclone diplegs, downow standpipes,
feeding nozzles, and detective probes only come into
contact with a very small part of the bed, and their
eects on overall bed behavior are likely to be negligi-
ble. Although the end congurations of the inlet and
the outlet of a uidized bed (i.e., of a CFB) can also
aect the two-phase ow structure, they will not be
discussed here. Therefore, in this section, only baes,
tubes, and some novel geometric structures, which are
introduced to signicantly strengthen the contact e-
ciency of gassolids and to improve the uidization
performance, are treated. Internals have dierent
structural shapes and arrangements in a bed, hence
each has its own performance characteristics.
Internals reported in the literature are classied into
ve groups as baes, tubes, packings, inserted bodies,
and other congurations, which are listed in Table 1 in
detail.
2.1 Bafes
Baes can prevent bubbles from growing continu-
ously, redistribute bubbles across the cross section of
the bed, strengthen the heat and mass exchange
Figure 4 Axial distribution of solids concentration in a
CFB.
Table 1 Various Types of Internals
Category Type Conguration Main features References Aspect studied Application eld
Baes
Wire mesh 1. Increase bubble splitting
2. Low erosion
Dutta and Suciu
(1992)
Eectiveness of baes Chemical reaction process
(synthetic reaction of ethyl
acetate)
Perforated plate 1. Improve radial
solids distribution
2. Increase bubble
frequency
Gelperin (1971)
Kono (1984)
Zheng (1990)
Dutta and Suciu
(1992)
Zhao (1992)
Hartholt (1997)
Classication of a binary
mixture of coal particles
Bubble frequency, dense bed
height
Radial distribution of solids
Eectiveness of baes
Solids backmixing
Mixing and segregation of
solids
Turbulent uidized bed
Gassolid uidized bed
Circulating uidized bed
Single- or
multiple-turn
plate
1. Enhance gas and
solids exchange
2. Decrease elutriation
3. Improve radial bubble
distribution
Pilot chemical reactors
(syntheses of phthalic
anhydride and
butadiene)
Louver plate 1. Enhance gas and solids
exchange
2. Decrease elutriation
3. Improve radial bubble
distribution
Jin (1982) Bubble behavior Pilot chemical reactors
Ring 1. Improve radial voidage
distribution
2. Enhance radial gas and
solids mixing
3. Improve gas-solids
contact eciency
4. Suppress axial solids
mixing
5. Increase conversion of
ozone decomposition
reaction
Zhu et al. (1997)
Zheng et al.
(1991b,1992)
Jiang et al. (1991)
Radial and axial
voidage distributions
Chemical reaction
(ozone decomposition)
Fast uidized bed
Catalytic circulating
uidized bed area
Tubes
Horizontal
banks
1. Split bubbles
2. Improve the uniformity
of gas distribution
Olsson (1995)
Lo fstrand (1995)
Olowson (1994)
Bayat (1990)
Yates (1987)
Levy (1986)
Jodra (1983, 1979)
Glass and Harrison
(1964)
Tube bundles design for
pressure uidized bed
Bed expansion
Bubble behavior and gas ow
distribution
Tube bundles design with the
consideration of tube erosion
Bubble size, tube erosion, tube
bundle design
Bubble eruption, particle
elutriation process
Prediction of the bubble size
distribution
Bubble size modication, tube
bundles design
Visual observation
Two dimensional uidized
bed bubble uidized bed
Pressure uidized bed
Cylindrical reactor (butanol
and ammonia)
Vertical banks 1. High heat-exchange
coecient
2. Low erosion
phenomenon
3. Low scale-up eect
Volk (1962) Scale up of chemical reactor Turbulent uidized bed
Packings
Station, irregular 1. Keep bubbles small and
uniformly distributed
2. Reduce carryover of
particles
3. Increase bed expansion
4. Increase chemical
conversion
5. Impede solids motion
Huang (1997) Pressure drop
Solids holdup
Gassolidsolid uidized bed
reactor
Station, regular 1. Increase pressure drop
2. Increase solids holdup
3. Improve gassolid
contact eciency
van der Ham et al.
(1991, 1993,
1994)
Hydrodynamics and
mass transfer
(CFB)
Small scale CFB
Pilot-plant scale CFB
Floating 1. Binary particles with
dierent size and density
2. Larger particles oat
freely in the bed
Goikhman (1969)
Inserted
bodies
Pagoda-shaped
bodies
(See Fig. 8) 1. Break up bubbles
2. Enhance gassolid
contact
Jin et al. (1982) Bubble behavior
Fluid mechanics
Pilot-plant trials
reactor
Inserted
bodies
(cont.)
Ridge-shaped
bodies
(See Fig. 9) 1. Control bubble size
2. Improve the quality of
uidization
3. Extend the range of
turbulent uidization
Jin et al. (1986) Bed expansion
Bed homogeneity
Change of ow region
Emulsion phase
behavior
Bubble uidized bed
Inverse cone 1. Improve radial solids
distribution
2. Decrease solids holdup
Zheng et al. (1990) Radial solids
distribution
Blu bodies 1. Increase pressure drop
2. Decrease solids holdup
3. Increase gas and particle
velocities
4. Improve radial voidage
distribution
5. Enhance gassolid
contact eciency
6. Reverse radial ow
pattern
Gan et al. (1990) Concentration proles Fast uidized bed
Spiral ow
pates
1. Enhance gassolid
contact
2. Lengthen solids residence
time
Li (1997)
Cui (1996)
Vortex strength
Cocurrent and countercurrent
ow patterns
Bed pressure drop
Average density
Fast uidized bed
Other
congur-
ations
Swages 1. Decrease pressure drop Davies and
Graham
(1988)
Pressure drop Vertical pneumatic
conveying tubes
Center
circulating
tube
(See Fig. 12) 1. Very high solid
circulation rate
2. Uniform residence time
distribution
3. Simple structure
Milne et al.
(1992, 1994)
Liu et al. (1993)
Wang et al. (1993)
Fusey et al. (1986)
Stable operation
Temperature
distribution
Axial solid
distribution
Table 1 Continued
Category Type Conguration Main features References Aspects studied Application eld
between gas and solids, and decrease the rate of solids
elutriation.
Moreover, baes are useful for continuous pro-
cesses since they divide a uidized bed into an arbitrary
number of stages in series. Unbaed beds are notable
for remarkable temperature uniformity, rapid solids
mixing (approximate a well-mixed stage), and substan-
tial amounts of gas backmixing, except for columns
with large height/diameter ratios. These characteristics
can be altered considerably by horizontal partitions.
The RTD of gas and solids can be eectively narrowed
in a baed uidized bed (Overcashier, 1959), though
not to the extent corresponding to an equivalent num-
ber of perfect-mixing stages in series, and it is possible
to establish temperature gradients between successive
stages of the bed.
At the same time, the vibration of the uidized bed
can be alleviated by eliminating large bubbles and rea-
lizing uniform distribution of bubble size. With the
decrease of the bubble size and velocity, a bed with
internals has a higher bed surface than a free bed at
the same gas velocity. However, the baes tend to
impede solids movement so that surface/bed heat
transfer coecients decrease; particle segregation can
also occur, and then it is dicult to have a full uidi-
zation in all compartments simultaneously. In addi-
tion, the total pressure drop across the bed will be
slightly increased by the horizontal baes.
As shown in Table 1, there are many kinds of
baes used in practice, such as wire meshes, perfo-
rated plates, single- or multiple-turn plates, ring baf-
es, and so on. Dutta and Suciu (1992) initiated an
experimental investigation to study systematically the
capacity of baes of various designs for breaking
bubbles in a medium-sized cold bed. The experimen-
tal conditions chosen and the bae designs are
believed to approach the hydrodynamic conditions
of many commercial bubbling beds more closely
than did earlier studies, such as those by Jin et al.
(1982), Jodra and Aragon (1983), and Yates and
Ruiz-Matinez (1987). Dutta and Sucius results indi-
cate that the eectiveness of a bae may be directly
correlated with the percentage of the opening area
and the number of openings per unit cross section
of the bae.
When compared with free uidized beds, beds con-
taining horizontal wire meshes or perforated plates
usually have smaller bubbles and uidize more
smoothly. In the 1970s, mesh baes were used in the
synthetic reaction process of ethyl acetate, where the
active carbon was used as carrier.
The single- and multiple-turn plates are very eec-
tive in breaking bubbles. Although this kind of bae is
much more complex in structure, it is not easily eroded
and can be used without maintenance for several years.
Single- and multiple-turn plates have been successfully
used to improve the synthetic yield and selectivity
of phthalic anhydride and butadiene in pilot-scale
uidized bed chemical reactors.
The function of louver plates is quite similar to that
of single- and multiple-turn plates. The only dierence
between them is that a louver plate cannot lead to gas
swirling. It is also reported that louver plates have been
used in pilot reactors, for example, for the synthesis of
aniline.
When gas velocity is high enough, for instance, the
reverse ow of solids through the bae slots is inhib-
ited by the upow gas. Thus the expansion ratio of the
bed increases very much, and a dilute phase of particles
forms under each bae at the same time, which is not
good for chemical reactions. The ow of solids
between the stages and the radial distribution of gas
are of special importance in the design of multistage
uidized beds. A type of bae plates with underow
downcomers has been developed that substantially
improves the performance of multistage uidized
beds (Kono and Hwang, 1983)
The inuence of ring baes on the reduction of
nonuniformity of a fast uidization was investigated
by Zheng et al. (1991). Jiang et al. (1991) also installed
ring baes in a riser 0.1 m in diameter to test their
eect on chemical conversion. Three ring baes with
opening area ratios of 70%, 90%, and 95% were
installed in a riser 7.6 cm in diameter and 3 m in height
by Zhu et al. (1997). They found that the ring baes
could reduce radial nonuniformity and evenly redistri-
bute solids in the radial direction. A detailed study on
the radial solids distribution around the ring bae
shows the formation of a more dense region above
the ring with a 70% opening ratio as compared with
that above those rings with opening ratios of 90% and
95%, and the formation of a more dilute region below
all three rings. Results of the axial voidage distribution
with the presence of rings show the formation of a
zigzag type axial prole instead of the regular S-shaped
prole.
Although baes are recommended to help reactor
performance, their eectiveness will depend on the
bae design and the hydrodynamic conditions in the
reactor, controlled by the following four factors:
2.1.1 Suitable Particle Systems
For group B particles for which bubble growth is
signicant, it is possible to keep the bubbles small
by using baes. However, for group A particles, the
bubbles are usually small enough and will not be radi-
cally aected by the baes unless the baes are spaced
suciently close to interfere with the overall
solids circulation. Therefore baes are usually applied
to processes using uidizing particles much dierent
(say larger and/or heavier) from the conventional
FCC-type catalysts, as has been demonstrated by
Dutta and Suciu (1992)
2.1.2 Reasonable Design
A careful design should consider the choice of a bae,
which is eective in narrowing the RTD of solids, and
the right conguration of it.
Perforated plates and ring baes are adaptable to
many solids processing operations which have a low
reaction rate and demand higher solid-phase conver-
sion.
The general criteria for the selection of optimum
ring openings are to obtain uniform radial solids dis-
tribution and to prevent solids accumulation (dead
zone).
For the selection of baes, besides the requirements
for good solids distribution, one should also reduce the
dead zone and erosion. For example, rings with an
inclined surface on top would encourage solids to
slide into the core region.
After a proper type of bae is selected, the bae
diameter and spacing should be decided. Usually the
bae diameter is smaller than that of the reactor. Thus
there is an annular gap near the reactor wall, which
contributes to axial solid backmixing. A certain degree
of solid backmixing can overcome the shortcomings of
solid segregation and the obvious axial temperature
gradient in a baed uidized bed. However, too big
a gap will lead to serious gas shortcuts that can sig-
nicantly decrease the bae eciency. So a suitable
annular gap should be selected in the range of 10 to
50 mm. To allow solids to be exchanged from stage to
stage, the openings in the baes should be made
suciently large that particles can pass through. Or
alternatively, downcomers that are similar to those
used in distillation columns may be provided.
The bae spacing is another important design
parameter. We know that there are few rules for select-
ing bae spacing. Many practices indicate that proper
bae spacing approximately equals the bed diameter.
However, most pilot plant beds in China of 1 to 3
meters in diameter have bae spacing of between
400 and 600 mm.
The scale-up of a uidized bed with baes is not
straightforward, since the eect of baes on the gas
and solid ow is complex and is also likely dependent
on the bed diameter. In Zhus study (1997), the 90%
ring can be considered to be the optimum ring opening
for the 76 mm riser. This may not be true for a riser
with a larger diameter.
2.1.3 Appropriate Operating Condition
The operating conditions are so extremely important
that baes may or may not improve the uidized bed
performance. For example, with a ring bae, the
improvement on the radial ow structure of a CFB is
achieved by properly coupling the ring size and the
operating conditions (Zhu et al., 1997). Determining
the optimum operating conditions requires a trial
and error approach.
2.1.4 Correct Installation
If transverse baes, of suitable types and correct
design are not installed quite horizontally, or if they
move while the bed is being operated, instability of
operation and other serious consequences like dead
spots and channeling may occur.
2.2 Tubes
Although the improvement of gassolid contact is
usually the main objective when various baes are
selected, some concomitant improvement, when a
bank of heat exchanger tubes must be used, is also
welcome. The uidized beds with tubes immersed in
the bed have a number of important applications in
industry.
To take advantage of the excellent heat transfer
characteristics of gas uidized beds, they are frequently
packed with heating or cooling tubes through which a
suitable heat transfer uid is circulated. Supply or
removal of heat in this way serves to control the tem-
perature of the bed, which is a major factor in the
design of chemical reactors whose product selectivity
is an important consideration.
Heat exchanging tubes have an important eect on
the performance of the bed even when they are located
relatively far below the free surface of the bed. A vast
majority of investigations on beds containing tubes has
been devoted to attempts to improve the heat transfer
between the bed and the immersed tubes. However, the
same important eects that tubes have on uidized bed
performance have not been stressed in detail.
Major problems associated with the design of a tube
bank are related to the orientation arrangement of
tubes, the optimal tube diameter, the tube spacing,
and the distance between the tube bank and the dis-
tributor. All of them control or are related to the ow
of solids in the tube bank, and they consequently
determine the heat transfer coecient between the
bed and the tube surface and the rate of mass transfer
in the bed; nally they determine the rate of chemical
reaction, if the bed is used as a reactor.
Heat transfer tubes may be positioned either verti-
cally or horizontally within the bed, but whatever
orientation they take; they will inevitably exert an
inuence on the pattern of gas ow through the bed.
Horizontal tubes are commonly used in uidized
bed combustors. The rst qualitative information on
the ow patterns near horizontal tubes was obtained
by visual observation of the phenomena in two-dimen-
sional beds (Glass and Harrison, 1964). Sitnai and
Whitehead (1985) summarized the results of dierent
experimental work on hydrodynamics in uidized beds
containing in-bed horizontal tubes, but most of the
investigations were made under atmospheric condi-
tions. Furthermore, Almstedt (1985), Almstedt and
Ljungstro m (1987), and Almstedt and Zakkay (1990)
presented measurements of in-bed hydrodynamics
from pressurized uidized bed combustors (PFBC)
burning coal. While these beds contain horizontal
cooling tubes, the possibility of varying operating con-
ditions during combustion is limited. More systematic
studies of the inuence of pressure, uidization velo-
city, and tube bank geometry on the hydrodynamics
have been carried out in model beds operating at room
temperature, i.e., by Staub et al. (1980) and recently by
Olowson (1994) and Olsson et al. (1995).
Provided particle transport conditions are not
approached, a thin cushion of air will form at the
underside of an object immersed in an air uidized
bed, while a deuidized region of solid particles will
rest on the topside. Especially when a tube diameter is
fairly large, because of these eects, horizontal tubes or
cylindrical obstacles do not have particularly good par-
ticlesurface contact except at the obstacle surface near
the ends of the horizontal diameter of the obstacle. The
relatively poor heat transfer coecients at the top and
bottom of horizontal tubes are partly compensated for
by good heat exchange at the tube sides where bubbles
may sometimes be formed. At uidizing ow rates
close to incipient uidization, horizontal tubes may
be at least as satisfactory for heat transfer purposes
as vertical tubes, but at higher ow rates the vertical
orientation is normally preferable. Horizontal tubes
may cause rising bubbles to split, but this is only really
eective when the tubes are much smaller than the
bubbles and when direct hits occur. In any case,
bubble coalescence usually takes place within a short
distance above the horizontal obstacle, so the overall
eect of horizontal tubes on the mean bubble size in
the bed is seldom large, unless, of course, the tubes
form an array, that lls the bed.
In industrial application of uidization, horizontal
heat transfer tubes are used to improve the uniformity
of gas distribution at the base of a bed equipped with a
low-pressure drop distributor. Except where horizontal
tubes are used to aid gas distribution in this manner, or
where a bed has a very small height/diameter ratio,
uidization considerations suggest that it is generally
preferable to arrange tubes vertically in uidized beds
rather than horizontally.
Optimum tube spacing of a horizontal tube tank
depends on the size of the solids used. The choice of
horizontal spacing determines the extent of the vertical
solids movement. The distance of separation between
horizontal tubes is a major factor controlling the bub-
ble breakage in uidized beds. Reducing the distance
of separation leads to increased bubble splitting to the
extent that the average volume of the daughter bubbles
is reduced to about 25% of that of the parent bubble
(Yates, 1987).
Various triangular arrangements are used; the most
typical, in order to increase tube packing density, are
(1) triangular with vertical separation between rows
equal to center-to-center horizontal tube spacing, (2)
equilateral, and (3) Rivesville arrangement, in which
face-to-face tube spacing between vertically staggered
tubes are half of the horizontal face-to-face tube
spacing (Cherrington and Golan, 1978). The limits of
horizontal spacing (center to center) generally used are
between two and four tube diameters, with three-tube-
diameter spacing being favorable in experimental
studies. For commercial design, where high surface
area of tubes per unit bed volume is required, close
spacing and small diameter tubes are preferable. The
larger the ratio of bubble diameter to tube diameter,
the more likely are the bubbles to split. The tube spa-
cing is not critical at high uidizing velocities; in that
case, the radial mixing of solids is intensive. Closer
tube spacing usually results in a slight decrease in
heat transfer coecient, but the penalty is compen-
sated by a higher heat exchange area per unit bed
volume (Sitnai, 1985). The horizontal arrangement
aects the heat transfer of a tube bank much more
than the vertical arrangement (Zabrodsky et al.,
1981). In some cases, with a triangular array of tubes
that occupied 8% of the bed volume, the bubble size is
roughly 1.5 times the tube spacing (Glicksman et al.,
1987). It seems possible, by matching the geometrical
arrangement of the tubes with the operating conditions
of the bed (i.e., gas velocity, particle size, U
mf
, etc.), to
exert control over the size of the bubbles.
A vertical tube bank is usually used for its high
heat-exchanging ability and low erosion wastage,
although it is not as eective as a horizontal tube
bank in breaking bubbles. However, it has been
observed that the heat transfer coecient will decrease
along vertical tubes in the upward direction in a circu-
lating uidized bed boiler, because many particles
move along the tubes, keeping close to the wall and
with little radial exchange. This can be prevented by
putting up cornices along the tube bundle 200 mm
apart (Fig. 5), which may eectively improve the heat
exchanging eciency.
Rising bubbles, which may be as large as several feet
in diameter, set up forces that can cause vigorous
transverse vibrations of immersed vertical tubes.
There is evidence that these vibrations improve heat
transfer rates and bed uniformity, so that it is not
usually desirable to damp out these vibrations comple-
tely, whereas the vibrations set a minimum on the
thickness of the vertical tubes and need to be consid-
ered when specifying the size, shape, and material of
construction of the tubes. Moreover, vertical tubes
divide the bed into a number of parallel longitudinal
compartments standing side by side. It has been
asserted that scale-up should be based on the equiva-
lent diameter (4 times the eective diameter of the ui-
dized bed over the eective saturation of the bed),
which goes a long way toward overcoming the pro-
blems of the scale-up eect. This is especially true for
beds in turbulent uidization regimes of Geldart A
particles. The reason is that for ne particles, the big-
gest stable bubble size is rather small and is sucient
for a vertical tube bank to prevent bubbles from cen-
tralizing while they are rising. But for coarse particles,
the narrow spaces between the vertical tubes make it
easy for channeling and gushing to occur because of
the inability to break bubbles. These phenomena tend
to have very adverse eects at higher gas velocities. For
this reason, vertical tube banks are widely used when
the weight fraction of nes in the catalyst is relatively
high. Volk et al. (1962) used rods 4.8 cm in diameter as
vertical inserts in a uidized bed, and scale-up was
made on that basis; favorable chemical conversions
in large-scale beds were obtained.
It is reported that a large pilot uidized bed reactor
with vertical tubes (shown as Fig. 6), whose equivalent
diameter is 1020 cm, can achieve the same conversion
and yield as a small reactor 1020 cm in diameter (Ye,
1987). This result has been veried in the iron reduc-
tion process and in the synthesis process of carbon
monoxide hydrogenation.
On the other hand, slugging and channeling are
easily present in parallel compartments between verti-
cal tubes, which are often regarded as harmful to the
performance of uidized beds. The gassolid exchange
and heat transfer rate may signicantly decrease when
slugging and channeling take place. This can be pre-
vented by choosing a proper tube diameter and spacing.
When the ratio of the tube diameter to the average
bubble size is larger than 1/3, gushing ow occurs in
the bed. When this ratio is smaller than 1/5, some bub-
bles adhere to the vertical tube, and the uidization
performance will be improved by decreasing the bubble
velocity and the bubble coalescence (Ye, 1987). The gap
width between any part of adjacent surfaces should be
at least 30 particle diameters. If smaller gaps exist,
channels tend to appear and lead to a general deteriora-
tion in the performance of the uidized bed.
2.3 Packings
Early at the beginning of the development of uidiza-
tion (in the 1950s), Rasching rings and Berl saddle
packings were packed in uidized beds to enhance
Figure 5 Structure of a vertical tube bank with cornices. 1
cornice; 2 vertical tube.
the eciency of gassolid contact. But they prevent the
particle movement so seriously that the bed would
easily be deuidized.
Up to the 1960s, packings in uidized beds received
considerable attention in North America, and beds
lled with packings have fallen into two groups
(Grace and Jarrison 1970):
1. Screen-packed beds in which the packings con-
sist of small open-ended wire mesh screen
cylinders
2. Packed uidized beds in which small particles
are uidized in the interstices of much larger
unuidized particles
Compared with other types of packings, screen cylin-
der packings occupy only 5% of the bed volume and
thus attract special attention. There were many inves-
tigations of it on gassolid contact, bed expansion,
bubble size, solids entrainment, uidization quality,
and eect on chemical reactions. It was found that
the screen cylinder type tends to keep bubbles small
and uniformly distributed so that slugging is
suppressed, carryover of particles is reduced, bed
expansion is increased, and chemical conversions are
increased. Although metal screens have many advan-
tages, they are easy to transgure in uidization beds
for poor strength, and cannot be operated for a long
period of time. Moreover, solids motion is impeded so
that heat transfer rates are less favorable than in
corresponding unpacked systems, and segregation of
particles may be signicant. Dead spots and channel-
ing also tend to occur, and it may be dicult to dump
the uidized particles from a bed containing xed
packings. But till now, chemical reactors packed with
metal screens are still applied in the acrylonitrile indus-
try, and they have eectively improved the synthesis
yield of acrylonitrile.
In the 1990s, a large research eort was put into the
second group and some equipment has been devel-
oped. For example, the gassolidsolid circulating
uidized bed reactor (GSSCFR) is such a novel reactor
that can carry out, in addition to a primary reaction, a
heat transfer, a mass transfer, or a momentum transfer
operation or even another reaction simultaneously. In
a GSSCFR, a packed bed and a uidized one are inter-
connected as shown in Fig. 7. Large catalyst particles
are xed in the packed bed. Another amount of ne
particles is uidized, which is often used as an adsor-
bent or a heat carrier, circulating between the packed
bed and the uidized bed. For reversible reactions, the
powder adsorbent can selectively adsorb and remove
the desired product while it ows through the packed
bed, so the reactor can shift the equilibrium in favor of
the desired product. For strongly exothermic or
endothermic reactions, the powder can remove or
supply the heat of reaction and make a uniform tem-
perature distribution in the reaction zone.
In a GSSCFR, the behavior of a gassolid two-
phase concurrent ow through the packed bed greatly
aects the mass and heat transfer. The gas and powder
ow concurrently through a packed bed to intensify
the gassolid contact (Yamaoka, 1986; Shibata et al.,
1991; Song et al., 1995; Van der Ham et al., 1996).
Huang et al. (1996a,b) extended the research started
by Song et al. (1995), and examined the axial distribu-
tion of powder holdup in the packed bed; they devel-
oped a mathematical model to predict the powder mass
ux and the mean dynamic powder holdup. They
found that the powder mass ux is inuenced by the
powder inventory, the gas velocities, and the void of
Figure 6 Vertical internals used in the iron reduction process. 1 vertical tube; 2 crosspiece.
the packed bed. More powder inventory, less gas velo-
cities, and larger voids of the packed beds are of benet
for increasing the powder mass ux and hence for
enhancing the heat and mass transfer.
A GSSCFR exhibits good operating stability and
simple construction and can be operated at a wider
range of supercial gas velocity than a gassolid
solid trickle ow reactor (Yamaoka, 1986; Shibata et
al., 1991; Song et al., 1995). The experimental results of
oxidationdehydrogenation of butene to butadiene
show that the C
4
H
8
conversion and C
4
H
6
selectivity
can be signicantly increased in a GSSCFR (Huang et
al., 1998).
Some early Soviet work (Goikhman et al., 1968)
suggests that packings (which are much larger than
the uidized particles) may be more eective if they
are light enough to oat freely in the bed instead of
being essentially xed in position as described above.
This appears to be a peculiar approach, but more work
is necessary before oating packings can be evaluated
with condence.
Despite the considerable amount of research work
that has been carried out in the laboratory on packings
in uidized beds, the technique has not yet found a
place in large-scale industrial applications because of
its fatal weakness of preventing solids movement.
It is well known that, compared with a packed bed
reactor, the most outstanding advantage of a uidized
bed reactor is its strong ability for heat exchange. That
is why it is especially suitable for processes with strong
heat eects. The favorable bed/surface heat transfer
rates and solids mixing characteristics of gas uidized
beds result from the motion of particles, motion caused
by rising bubbles; therefore it is important not to
impede the particle movement greatly. However,
most packings in a uidized bed seriously prevent the
movement of particles, and they are harmful to heat
transfer too. Though the eective thermal conductivity
of a packed uidized bed can be about 75 times that of
a normal packed bed (Ziegler, 1963), it is still lower
than that of a uidized bed. It seems that packed ui-
dized beds are only useful when it is not practicable to
divide the solid material into particles ne enough to
permit uidization at reasonable uid ow rates, or
when a highly exothermic reaction takes place on the
surface of the large xed particles.
2.4 Insert Bodies
It has been suggested that a combination of horizontal
and vertical internals has an important eect on
improving uidization quality. Thus various kinds of
insert bodies come into being, such as: pagoda-shaped
bodies by Jin et al. (1982), ridge-shaped bodies by Jin
et al. (1986b), inverse cones by Zheng et al. (1990), half
oval blu-bodies by Gan et al. (1990), and spiral ow
plates by Li (1997).
On the basis of an empirical investigation on the
eects of the behavior of the bubbles in uidized
beds and dierent types of internals on the motions
Figure 7 Structure of a gassolidsolid circulating uidized reactor (GSSCFR).
of the bubbles and the solid particles, Jin et al. (1982)
have designed a type of pagoda-shaped body (Fig. 8).
The pagoda-shaped bodies are made up of three sepa-
rate parts welded together: a perforated pipe with
square cross section, an inclined sieve plate inside the
pipe, and a square cornice. The vertical perforated pipe
is like the normal type of vertical tube, its purpose
being to restrict the bubble size and to guard against
any tendency of the bubbles to congregate in the center
of the bed as they move upwards. Most of the bubbles
are forced into the perforated pipes through the sieve
openings and the larger holes, only to be redispersed
through the holes further up the pipes in the form of
strings of small bubbles. However, owing to the pre-
sence of four larger holes and numerous sieve holes on
the wall of the pipe, radial movement of the solid par-
ticles is quite vigorous, which is advantageous as far as
heat and mass transfer are concerned.
The eect of the inclined sieve plates inside the pipe
and the cornice outside the pipe is to break up and
renew the bubbles. At higher gas velocities, especially,
the eect is to force the bubble phase to pass through
the sieve holes or around the pipe. During this process,
the bubbles are continuously broken up and renewed,
the long-distance backmixing of solids and gas in the
emulsion phase is cut o, and relatively high gassolid
exchange coecients between the bubble phase and the
emulsion phase are obtainable. In addition, as the
perforated pipe is empty, two-phase ow occurs
both inside and outside the pipe, and the eective
space inside the bed is not reduced. Also, the inclined
sieve plates inside and the cornices outside adjacent
pipes are arranged in an overlapping fashion, thereby
making the spatial distribution of solids more uniform
and avoiding the formation of air cushions in the
bed. Vertical and radial circulation of the particles is
thus improved.
From the viewpoint of mechanical design, the
pagoda-shaped bodies are simple to construct, easy
to fabricate, and mechanically strong. During startup
and shutdown, they do not experience any violent
impingement of the catalyst, and during operation,
they do not readily suer from thermal deformation,
and can remain in working condition for a long period
of time. Since the orientation of the tubes is the same
as that of heat exchanger tubes, their installation,
maintenance, and overhaul are quite easy. Through
experiments in a laboratory reactor (30 cm in I.D.), a
medium-sized reactor (1 m in I.D.), and a pilot reactor
(2 m in I.D.), pagoda-shaped bodies have proved to be
very eective at high gas velocity, catalyst loading rate,
and reaction yield.
The ridge-shaped bodies (Jin, 1986b), which are said
to be capable of controlling the size of the bubbles,
improving the quality of uidization, and extending
the range of turbulent uidization, are composed of
vertical heat exchange tubes and ridge baes (Fig.
9). Figure 10 shows the structure of an oxidizing reac-
tor with ridge-shaped bodies.
The pagoda- and ridge-shaped bodies are used in
turbulent uidized beds. However, for a CFB, a
more simple construction is adopted, such as inverse
Figure 8 Construction of the pagoda-shaped bodies.
cones and blu bodies. According to Gan et al. (1990),
by forcing gas and solids to ow upward in the narrow
ring section near the wall, half oval blu bodies can
reverse the radial ow distribution in a circulating ui-
dized bed and cause higher solids concentration and
lower gas and solids velocities in the core, and lower
solids concentration and higher gas and solids veloci-
ties in the wall region. Gas and solids are forced to ow
upward in the narrow annular region near the wall. At
the same time, the gassolid mixing along the radial
direction could be also greatly intensied.
Besides the single- or multiple-turn plate, the spiral
ow plate is another kind of internal that may lead to
gas and solid swirling. The main functions of a spiral
ow plate are lengthening the residence time of solids
and enhancing the contact between gas and solids.
There are many kinds of spiral ow plates, with dier-
ent structures and opening area ratios, and even with
dierent arrangements, which are applied in drying
processes (Fig. 11). The vortex strength is a common
index to evaluate them. Li (1997) found that a reversed
conic plate is better than a conic spiral ow panel in
PVC drying processes.
Generally, the introduction of solid obstacles into a
bed further complicates the system, and there is as yet
no relevant theoretical analysis available.
2.5 Other Congurations
There are also other elements in uidized beds, not
installed as internals, but necessary for some purposes.
One example is that by Davies and Graham (1988).
They swaged indentations 3.5 mm deep and 16 mm
wide onto the wall of a riser 0.152 m in diameter to
test their eect on pressure drop in the riser. They
found that swages evenly spaced along the circular
riser wall would scrape the dense downow solids
away from the wall and cause higher bed voidage
and lower pressure drop in the fully developed region
than those in a bare tube.
Another example is the strange nose in a circulat-
ing uidized boiler. As we know, the circulating coal
ash usually feeds into a boiler from one side of the bed
wall, which easily results in annular ow and decreases
the burning eciency. Thus a protruding object like a
nose is set at the other side in order to prevent the
nonuniform ow.
The third example, fast uidization, can also be
achieved in a concentric circulating uidized bed.
One option is to have solids ow up in the center
and down in the annular shell. This has been called
an internally circulating uidized bed. Alternatively,
one can have solids ow up in the annular shell and
Figure 9 Construction of the ridge-shaped bodies.
ow down in the center, leading to what has been
called the integral circulating uidized bed. These
two options are shown in Fig. 12. Since both have
been abbreviated to ICFB, the former will be identied
as i-ICFB and the latter as o-ICFB with the i and o
indicating the inside and outside location of the riser.
The i-ICFB originated from the spouted bed with a
draft tube. Preliminary hydrodynamic studies by Milne
et al. (1992a,b) and by earlier workers (Fusey et al.,
1986) show stable operation of this system. One advan-
tage is that a very high solids circulation rate (near
600 kg=m
2
:s) can be realized in this system (Milne et
al., 1994). Milne et al. (1994) have shown that a stable,
isothermal operation at high temperature can also be
achieved in an i-ICFB unit and have suggested that it
can be ideal for short contact time reactions where
Figure 10 Structure of an oxidizing reactor with ridge-shaped bodies.
uniform residence time distribution and stable high
temperature operation are needed, such as in pyrolysis.
o-ICFB has been developed as an alternative to the
regular CFB systems to address the concerns of non-
uniform solids feeding at the riser bottom and the
energy requirements of a large amount of external
solids separation and recirculation. There are several
advantages: (1) simple structure; (2) high solids circu-
Figure 11 Diagrams of multistage dryer with spiral ow plates. 1spiral ow bed; 2spiral ow panel; 3rst-stage
cyclone; 4measure tank; 5storage tank; 6buttery value; 7electric heater; 8owmeter; 9second-stage cyclone;
10inlet.
lation rate; (3) uniform solids feeding. As for i-ICFB,
the solids circulation rate is not independently con-
trolled but varies with riser and auxiliary gas velocities.
Liu et al. (1993a,b) and Wang et al. (1993) found that
axial solids distributions in the riser of o-ICFB are very
similar to those in regular CFB risers. The cross-
sectional average solids holdup is somewhat higher
than that in regular CFB risers operated under similar
conditions (Wang et al., 1993). Typical radial solids
concentration proles measured by Wang et al.
(1993) show that a core-annulus structure still exists
inside the annular riser, with the dense layer at the
outer wall slightly thicker than that on the inner wall
(Fig. 13). Measuring at two vertical planes in the annu-
lar riser on the opposite sides of the central standpipe,
Wang et al. (1993) found that the radial voidage
distributions are symmetrical.
It has already been stressed that it is impossible to
specify a single type of internal that is best for all
applications. This is because each successful applica-
tion of uidization depends upon dierent properties
of uidized beds to varying degrees, and a type of
internal that helps one process may hinder another.
For example, particle segregation is an adverse eect
in many applications, but not, of course, if a uidized
bed is used as a particle classier; then an internal
member that promotes segregation (i.e., xed pack-
ings, horizontal screens) will be appropriate. In many
continuous operations, horizontal baes may be of
considerable benet. Where heat transfer is of para-
mount importance or where an existing system does
not perform as well as expected, vertical tubes warrant
serious consideration. Clearly, combinations of various
kinds of internals are likely to be useful in certain
applications.
3 EFFECT OF INTERNAL TUBES AND
BAFFLES
Many references in the literature mention that the
improved uidization results from the use of tubes
and baes as internals. Harrison and Grace (1971)
and Sitnai and Whitehead (1985) have reviewed the
inuence of internals on the behavior of uidized
beds for reactor design purposes. For example, on
the one hand, most transverse baes in a uidized
bed eectively narrow the gassolid residence time dis-
tribution, prevent rushing at the surface of the bed,
and reduce the entrainment of solid particles. On the
other hand, a disadvantage of the use of baes in a
bed is the presence, underneath each bae, of a region
of lower density where the solid particles are more
dilute. Another disadvantage is the axial size-classica-
tion of particles. Furthermore, increasing the diameter
of the uidized bed magnies these problems.
In this section, the eect of internals on a uidized
bed is analyzed from the following four aspects.
3.1 Bubble Behavior
Bubble size is a very important parameter in the design
and simulation of uidization. We can accept that
Figure 12 Structure of integral circulating uidized bed.
Figure 13 Typical radial solids concentration proles in an
o-CFB.
most of the eects that internals produce in a uidized
bed are derived from the same primary eect: control
of the bubble size and its distribution. The movement
of bubbles in a bed containing no internals has already
been discussed in Sec. 1.1.
Jin et al. (1982) have investigated the ability of inter-
nals to break up bubbles in a two-dimensional uidized
bed by the photographical method (Fig. 14). For com-
parison, Fig. 14a shows the bubbles in a free uidized
bed. When an internal module is introduced, such as a
bae, a tube or a pagoda-shaped body, the situation is
modied to dierent degrees according to the module
type, as shown in Figs. 14b,c,d.
Figure 15 shows the eect of transverse baes
(louver plates) on the behavior of bubbles at dierent
gas velocities. The baes do not visibly inuence the
bubble behavior (coalescence or rise velocity) at low
gas velocities (Fig. 15a). Small bubbles push upward
past each layer of baes, only to collapse when they
reach the surface of the bed. With the increase of gas
velocity, a small, nonstationary, particle-free section is
observed underneath a bae with bubbles splitting
above it. At much higher gas velocities, the eects of
baes become apparentbreakup, regeneration, and
uniform distribution of bubbles. When the supercial
gas velocity further increases, stationary particle-free
sections are observed under each bae and smaller
bubbles are formed above them. The particles, which
follow the bubbles upwards, mainly circulate through
the bed in one of two ways: through the spaces around
the baes, or directly past the louvers in the baes.
Figure 15b illustrates the rain of particles, which is
observed coming o the louver plates. Under these
conditions, contact between the gas and solids is
excellent.
Figure 14 Bubble behavior in uidized beds with dierent internals.
Figure 15 Eect of transverse baes (louver plates) on the behavior of bubbles at dierent gas velocities.
As the supercial gas velocity is increased still
further, bubbles aggregate into large cushions under-
neath the baes. The fraction of the volume of the bed
which is occupied by the cushions of coalesced bubbles
keeps on increasing with the gas velocities (see Fig.
15c). At higher gas velocities (Fig. 15d), a very large
part of the bed is occupied by these air cushions, giving
a very large void ratio and expansion ratio. Since a
too-high gas velocity aects the vertical circulation of
the particles, there is a gradual lessening and eventual
disappearance of the rain of particles coming o the
louvers of the baes. The particles tend to become
classied according to their neness. A thick air cush-
ion is also detrimental to heat exchange.
Figure 16 shows the eect of vertical tube banks on
the behavior of the bubbles. The main eect of the
presence of pipes inside a bed is to even up the spatial
distribution of the bubbles within the bed and to
reduce their tendency to congregate near the center
of the bed. Such pipes also tend to have relatively little
eect on bubble breakup.
To sum up, within a denite range of gas velocities,
transverse baes as well as vertical tubes are able to
control, break up and renew the bubbles and thereby
substantially improve the gassolid contact within the
bed. At higher gas velocities, however, such baes and
tubes will have a detrimental eect on uidization per-
formance.
The presence of horizontal tubes decreases the
dense-phase uidity, which prevents bubbles from
accelerating and consequent coalescence. Horizontal
tubes tend to cause the bubbles to split, resulting in
smaller bubbles in the region just above the tubes.
Processes of bubble splitting are followed by coales-
cence. This reaches an equilibrium with a relatively
small bubble, similar in size to the tube spacing.
Under some circumstances, horizontal tubes will
cause splitted bubbles to recombine, thus nullifying
the benecial eects of splitting (Yates, 1987). Large
bubbles can rise over horizontal rods without breaking
up (Xavier et al., 1978), and the bed expansion is not
much inuenced by the rods (Newby and Keairns,
1978).
Figure 17 shows the eect of pagoda-shaped bodies
on the behavior of the bubbles in a two-dimensional
bed at dierent gas velocities. When the gas velocity is
very low, a large number of small bubbles adhere to
the surface of the perforated pipe, passing back and
forth through the holes during the course of the
upward movement, with the inserts exerting a denite
eect on the bubble velocities (Fig. 17a). As the gas
velocity is raised, there is a slight increase in the bubble
diameter, but the presence of the inserts prevents the
bubbles from growing too much; some of the bubbles
move away from the wall of the pipe (Fig. 17b). At
higher supercial gas velocity, it is possible to observe
each individual bubble being constrained and broken
up by the pagoda-shaped bodies (Fig. 17c). At still
higher supercial gas velocities, it becomes more and
more dicult to discern the behavior of individual
bubbles (Fig. 17d). Under these conditions, the bed
could be classied into two regimes: a bubble phase,
in which groups of bubbles are of no denite shape,
and in which the solid particles are sparsely scattered
and move upward more or less in train with the
bubbles; and an emulsion phase, in which the solid
particles are present in greater concentration and mov-
ing downward. The equivalent diameter of the bubbles
is basically constant over a rather wide range of super-
cial gas velocities, irrespective of the height of the
bed. Raising the supercial gas velocity under these
conditions increases the volume of the bubble phase
Figure 16 Eect of vertical tube banks on the behavior of the bubbles.
and intensies turbulence. Also, there is an increas-
ingly rapid exchange between the bubble and emulsion
phases throughout the bed.
In general, by catching and breaking up the bubbles,
the internals have the eect of reducing their size. On
one hand, when a bubble is split by an internal, the
sum of the volumes of the daughter bubbles is less than
the volume of the parent one, indicating that some of
the original bubble gas leaks into the emulsion phase
of the bed during the splitting process. Furthermore,
the amount of leakage increases as the number of
daughter bubbles increases. These facts could be of
importance in the design of uidized bed reactors.
With the reduction of the mean bubble size, the
expansion ratio of a bed changes. Figure 18 shows
quite clearly the eects that dierent types of internals
have on the expansion ratio in a two-dimensional ui-
dized bed. On the other hand, small bubbles rise rela-
tively slowly in a uidized bed, and so the result is to
increase the total concentration of bubbles in the bed.
A thick mist of small bubbles has a benecial eect on
the conversion of a chemical reaction.
Two dierent methods have been developed for pre-
diction of the bubble size in a bed with internals
formed by horizontal rows of tubes (Jodra and
Aragon, 1983). One, based on a new correlation,
makes it possible to nd the size of a bubble above
the tube as a function of the operating conditions
and characteristics of the tube arrangement. The
other is based on the general theory of bubble forma-
tion in submerged orices. Both methods can be used
to estimate bubble sizes, which are acceptably close to
the experimental values, and can be applied to other
types of internals, such as perforated plates, meshes
and arrangements of horizontal bars of noncircular
cross sections.
Besides splitting bubbles, internals alter the eruption
mechanism for single bubbles at the bed surface,
causing a signicant fraction of the single bubbles to
generate large wake eruptions. Levy et al. (1986) have
found that without any internals in the bed, when sin-
gle bubbles erupt, the bulge material of a bubble
always reached a greater height above the bed free sur-
face than the wake material. With horizontal tubes
present, up to 60 percent of the single bubble eruptions
are characterized by the upward projection of wake
material, reaching a height above the free surface
which far exceeds those attained by the bulge material.
The wake-dominated eruption of the single bubble
appears to be due to the formation of elongated ver-
tical cavities in the region between the internals and the
free surface. A quantitative understanding of the
eects of internals on bubble eruptions may lead to
Figure 17 Eect of pagoda-shaped bodies on the behavior of the bubbles.
Figure 18 Eects of internals on the expansion ratio in a
two-dimensional uidized bed.
techniques for designing the internals to minimize
solids elutriation.
Recently, it has been observed that the bubble beha-
vior is aected by pressure. Olsson (1995) has found
that there is an initial increase in the mean pierced
length, the mean bubble rise velocity, the mean bubble
volume fraction, and the mean visible bubble ow rate
as the pressure is increased. With a further increase in
pressure, these visible bubble ow parameters start to
level o or even decrease. Generally speaking, this
decrease is more pronounced for the high excess gas
velocity than for low velocity. This behavior is gov-
erned by a balance between bubble coalescence and
bubble splitting. At low pressures, bubble coalescence
dominates over splitting, while, as the pressure is
increased, the splitting process becomes predominant.
Moreover, the distribution of bubbles is also related
to pressure (Olsson, 1995). Experiments show that at
low excess gas velocity, there is a redistribution of the
visible bubble ow toward the center of the bed as the
pressure is increased. At high excess gas velocity, how-
ever, the inuence of pressure is not the same. For
lower pressures, there is still a redistribution of the
bubbles toward the center of the bed with increasing
pressure, but for higher pressures the distribution
becomes somewhat more uniform again as the pressure
is increased further. Olssons explanation for the bub-
ble distribution becoming more uniform at combina-
tions of high excess gas velocity and pressure was that
the bed was developing a more turbulent behavior,
with increased bubble splitting and consequently a
more dispersed bubble ow.
3.2 Flow Distribution
As discussed in Sec. 1, nonuniform ow is an inherent
characteristic of gassolid uidization systems. A typi-
cal radial solids distribution in a riser without any
inserts is not uniform: the solids concentration is low
in the center and high in the wall region. In many
circumstances, internals are aimed at redistributing
gas and solids ow, in an eort to form a more
uniform ow structure to improve interphase contact
eciency so as to increase the overall performance of
reactors.
Ring baes and swages scrape the downowing
solids away from the wall towards the center, leading
to a more uniform radial solids distribution. Using an
optical ber concentration probe, Zheng et al. (1991,
1992) have showed that the radial voidage distribution
is much more uniform with ring baes present (Fig.
19). Jiang et al. (1991) have conrmed these ndings by
showing that the conversion of the ozone decomposi-
tion reaction is improved with the addition of ring
baes. Arrangement of a narrow seam near the wall
further improves the performance. The gas preferably
ows upward in the annular region so that the radial
distribution becomes more uniform. Similarly, a half-
oval blu body forces the gas and solids to ow
upwards in the narrow annular region (Fig. 20).
Hence the dilute core and dense annulus ow pattern
are disrupted (Gan et al., 1990). As a consequence,
gassolid contacting is greatly improved.
However, not all internals provide favorable gas
and solids redistribution. Salah et al. (1996) have
found that ring baes with smaller opening areas
can make the radial voidage distribution less uniform
under low gas velocity. With a ring of 44% open area,
Zheng et al. (1990) reported increased nonuniformity
under certain operating conditions. For risers operated
at high gas velocity and high solids ux, e.g., the riser
in FCC units, particles mainly ow upwards at the wall
so that the installation of ring type baes may actually
induce particle downow, leading to more radial ow
segregation. Fortunately, Zheng et al. (1990) have
found that a plate with larger orices and more open-
ing area near the wall is very eective in ow redistri-
bution. Therefore, properly designed internals can help
to redistribute the gas and solids ow.
At the same time, internals can also change the
average voidage. The results of Zheng et al. (1992)
indicate that ring baes signicantly reduce the aver-
age solids concentration. Gan et al. (1990) also found
that their blu body bae reduced the average solids
concentration by scraping o superuous particles
Figure 19 Radial voidage distribution in a ring-baed ui-
dized bed.
near the wall. However, Jiang et al. (1991) reported a
higher local solids concentration in the presence of ring
baes. With the packed bar internals, which take up a
large cross-sectional area of the bed (50%), van der
Ham et al. (1993) measured 1.6 to 5 times higher solids
holdup compared with an empty column. Although
part of this increase was due to particles settling on
top of the bars, at least some was due to the bars. It
would appear that baes with less ow restriction
(e.g., less reduction in cross-sectional area) mainly
increase the actual gas velocity, scraping the dense
downow solids away from the wall, leading to higher
bed voidage and lower pressure drop as observed by
Davies and Graham (1988) and Zheng et al. (1991,
1992), while baes with a signicant reduction of
cross-sectional area constrict the solids ow, leading
to lower bed voidage (e.g., Jiang et al., 1991; Van der
Ham et al., 1993). The blu body seems to be an excep-
tion. It occupies 50% of the riser cross section, signi-
cantly increases the gas and particle velocities, and
causes a higher pressure drop; it does not increase
the solids holdup. This may be attributed to the
smooth parabolic shape facing the gas and solids ow.
Besides the gas and solids distributions, the installa-
tion of internal baes can signicantly aect the
pressure distribution in a uidized bed.
Pressure uctuations in the uidized bed have been
used successfully as a measure of bubbling intensity,
and the method is found particularly useful for the
study of behavior of beds with internals (Kang et al.,
1967; Newby and Keairns, 1978; Staub and Canada,
1978). Jiang et al. (1991) measured an increase in
pressure drop when four ring baes with an opening
area of 56% were installed in their riser 0.1 m in I.D.
Gan et al. (1990) experienced a localized high-pressure
reduction across their blu body, accompanied by
signicant solids acceleration.
3.3 Gas and Solids Mixing
The degree of axial mixing of gas and particles in ui-
dized beds is important for many continuous as well as
batch processes, and control thereof is desirable. It is
very important to limit backmixing and maintain good
gassolid contact while using uidization for solid or
gassolid processing, as these operations need high
solid-phase conversion and gas-phase utilization.
First, gas backmixing is substantially reduced by
immersed obstacles. The higher the gas velocity the
smaller the gas backmixing (Ye, 1987). There is little
gas backmixing between the top and the bottom of
each horizontal bae. The axial gas-mixing coecient
is lowered only in beds of small particles or at low gas
velocities. The larger the particles and the higher the
gas velocity, the less eective is the horizontal baes.
Horizontal tubes completely suppress gas back-
mixing in a large range of velocities, but vertical
tubes are less eective at higher velocities. A note-
worthy result with vertical tubes is the development
of marked nonuniformity of radial gas mixing.
Secondly, most internals are likely to suppress axial
solids backmixing. Subdividing the bed with horizontal
baes is a common and successful practice to improve
gassolid contact and to limit solids backmixing. Many
congurations of baes not only can constrain bubble
growth but also can narrow the RTD of gas and solids.
For example, a 1 m I.D. bed with bae spaces of 100
to 200 mm, with a perforation ratio smaller than 30%
and a small space between the plate and the wall, can
reduce axial solids backmixing and lead to a large tem-
perature gradient along the axis. Zheng et al. (1992)
have found that the axial solids dispersion of a ring
bae is only half of that in an empty column.
The baed uidized bed, when used to carry out a
reaction, is a kind of multistage uidized reactor,
which can be modeled by treating the overall mixing
of solids as a series of perfectly mixed stages with solids
backmixing between them. Consequently, the determi-
Figure 20 Flow pattern in a riser with a blu body.
nation of the solids backmixing rate between stages is
important.
Experimental results show that, for a given bed
material, the main factors inuencing the solids back-
mixing rate are gas velocity, bae free area, and bae
spacing. Based on a theory of particle entrainment, a
mathematical model has been developed to predict the
solids backmixing rate in smooth bubbling uidization
(Zhao et al., 1992).
Modeling of solids mixing in larger scale beds with
horizontal tubes has been reported by several work-
ers. Solids tracers having similar properties to the bed
material are usually used in mixing experiments. One
of the simpler models (Sitnai et al., 1981) postulates
one upow and two downow regions. From the
interpretation of mixing data, a high exchange rate
of solids between the main downowing dense phase
and the upward moving drift phase is inferred. At
uidizing velocities of up to 1 m/s with particles
about 700 m, the axial solids dispersion coecient
has little eect on the overall solids transport, and
the solids exchange coecient between regions
remains constant.
Solids mixing in beds with an assembly of vertical
tubes has not been widely investigated. Ramamoorthy
and Subramanian (1981), applying a one-dimensional
diusion model, determined an empirical correlation
to predict the axial diusion coecient at given uidiz-
ing velocity and packing density of vertical rods. The
diusivity is substantially decreased in the presence of
internals and further decreased with a reduction in the
spacing of internals.
Most internals appear to improve the radial gas and
solids mixing. Using FCC particles with a mean
diameter of 89 m impregnated with ferric oxide as
catalysts for ozone decomposition, Jiang et al. (1991)
observed a more uniform radial distribution of ozone
concentration in the riser with ring baes (Fig. 21) and
attributed this to higher radial gas and solids mixing.
Owing to the enhanced gassolid contact eciency,
increased gas and solids radial mixing and suppressed
axial mixing, CFB reactor performance is expected to
improve when internal baes are present. Jiang et al.
(1991) showed that ozone conversion was signicantly
higher when ring baes were installed in a CFB reac-
tor.
In summary, the addition of baes, tubes, and
other obstacles restricts large-scale solids movement,
reduces the gas and solids backmixing, improves the
radial gas and solids mixing, and causes segregation of
particles of dierent sizes.
3.4 Transition to Turbulent Fluidization
Flow patterns in uidized bed chemical reactors have a
strong inuence on the performance of reactors. Most
commercial catalytic reactors are run successfully in
the turbulent uidized region. Numerous investigators
have used the method of analyzing the pressure uc-
tuations in the bed to characterize the transition from
bubbling to turbulent uidization; see, e.g., Canada et
al. (1978), Yerushalmi and Cankurt (1979), Sadasivan
et al. (1980), Satija and Fan (1985), Rhodes and
Geldart (1986), Noordergraaf et al. (1987),
Andersson et al. (1989), Cai et al. (1989), and
Olowson (1991). These uctuations are caused mainly
by the activity of the bubbles in the bed, e.g., by the
bursting of bubbles at the surface and by the coales-
cence and splitting of bubbles in the bed.
The transition to a more turbulent behavior is asso-
ciated with many factors, such as uidization pressure,
particle properties, and the geometric conditions of the
bed. It is also inuenced by the internals in the bed.
The eect of internals on the transition velocity U
c
has
been systematically studied by Jin et al. (1986a). It can
be seen from Fig. 22 that with internals in beds, the
transition from a bubbling to a turbulent regime occurs
at smaller U
c
. Johnsson and Andersson (1990) have
also suggested that tubes in a bed can cause a transi-
tion to a turbulent behavior. Similar result has been
obtained by Olsson et al. (1995). They found that the
tube banks caused the bubbles to break up at lower
pressure and gas velocity, resulting in a more turbulent
behavior than for the bed without tubes.
Figure 21 Radial distribution of ozone concentration in a
riser with ring baes.
Figure 22 Eect of internals on the transition from bubbling to turbulent regime. (a) Free bed, (b) bed with vertical tubes, (c)
bed with pagoda-shaped bodies.
Experiment results indicate that dierent types of
internals have eects on U
c
to dierent extents. Jin et
al. (1986a) have proposed a correlation to predict U
c
as
follows:
U
c
gd
p
p
kD
f
d
p
f

n
where n 0:27 and D
f
is a parameter called perfor-
mance diameter which has length dimension. The
KD
f
is found to characterize the geometric structure
of the beds, and its value is obtained as follows:
KD
f
3:67 10
3
for free bed (280 280 mm)
KD
f
2:32 10
3
for bed with vertical tubes
KD
f
3:42 10
4
for bed with pagoda-shaped
bodies
The average error of above formula is about 6.58%.
REFERENCE
Almsedt AE. Measurements of bubble behavior in a pressur-
ized uidized bed burning coal, using capacitance
probesPart II. Proceedings of the 8th International
Conference on Fluidized Bed Combustion. Houston,
1985, pp 865877.
Almstedt AE, Ljungstro m EB. Measurements of the bubble
behavior and the oxygen distribution in a pilot scale pres-
surized uidized bed burning coal. Proceedings of the 9th
International Conference on Fluidized Bed Combustion,
Boston, 1987, pp 575585.
Almstedt AE, Zakkay V. An investigation of uidized-bed
scaling-capacitance probe measurements in a pressurized
uidized bed combustor and a cold model bed. Chem Eng
Sci 45:10711078, 1990.
Andersson S, Johnsson F, Leckner B. Fluidization regimes in
non-slugging uidized beds. Proceedings of the 10th
International Conference on Fluidized Bed Combustion.
San Francisco, 1989, pp 239247.
Cai P, Chen SP, Yu ZQ, Jin Y. Eect of operating tempera-
ture, pressure and internal baes on the average voidage
of gassolid uidized beds. Petro Chem Eng 18(9):610
615, 1989.
Canada GS, McLaughlin MH, Staub FW. Flow regimes and
void fraction distribution in gas uidization of large par-
ticles in beds without tube banks. AIChE Symp Ser No
176, 74:1426, 1978.
Cherrington DC, Golan LP. Am. Pet. Inst. Proc. Div.
Rening. 43rd. 57:164182, 1978.
Davies CE, Graham KH. Pressure drop reduction by wall
baes in vertical pneumatic conveying tubes. In
CHEMECA 88, Australias Bicentennial International
Conference for the Process Industries, Institute of
Engineers, Australia, Sydney, Australia, August 1988,
Vol. 2, pp 644651.
Dutta S, Suciu GD. An experimental study of the eective-
ness of baes and internals in breaking bubbles in uid
beds. J Chem Eng (Japan.) 25(3):345348, 1992.
Gan N, Jiang DZ, Bai DR, Jin Y, Yu ZQ. Concentration
proles in fast-uidized bed with blu-body (in Chinese).
J Chem Eng Chinese Univ 3(4):273277, 1990.
Glass DH, Harrison D. Chem Eng Sci 19:10011002, 1964.
Glicksman LR, Lord WK, Sakagami M. Bubble properties
in large-particle uidized beds. Chem Eng Sci 42(3):479
491, 1987.
Goikhman ID, Oigenblik AA, Genin LS, Filippova LA.
Eect of oating spherical packing on the residence-time
distribution of gas in a uidized bed (in Russian). Khim
Technol Topl Masel 13(10):3638, 1968.
Grace JR. Contacting modes and behaviour classication of
gas-solid and other two-phase suspensions. Can J Chem
Eng 64:353363, 1986.
Grace JR, Jarrison D. Design of uidised beds with internal
baes. Chem Proc Eng 6:127130, 1970.
Harrison D, Grace JR. In: Davidson JF, Jarrison D, eds.
Fluidization. New York: Academic Press, 1971, pp 599
626.
Huang SY, Wang ZW, Jin Y, Yu ZQ. Solids mass ux and
mean dynamic solids hold-up in the packed section of the
gas-solid-solid circulating uidized bed (in Chinese).
Chem Reaction Eng Tech 12:369376, 1996a.
Huang SY, Wang ZW, Jin Y, Yu Z. Axial pressure drop and
powders hold-up distribution in packed section of gas-
solid-solid circulating uidized bed (in Chinese). J Chem
Eng Chinese University 10:377383, 1996b.
Huang SY, Wang ZW, Jin Y. Studies on gas-solid-solid
circulating uidized bed reactors, Chem Eng Sci 54(13
14):20672075, 1999.
Jiang P, Bi HT, Jean RH, Fan LS. Bae eects on perfor-
mance of catalytic circulating uidized bed reactor.
AIChE J 37:13921340, 1991.
Jin Y, Yu ZQ, Zhang L, Shen JZ, Wang ZW. Pagoda-shaped
internal baes for uidized bed reactors. Intern Chem
Eng 22(2):269279, 1982.
Jin Y, Yu ZQ, Wang ZW, Cai P. A criterion for transition
from bubbling to turbulent uidization. In: O
stergaard K,
So rensen A, eds. Fluidization V. Elsinore, Deamarh, May
1823, 1986a, pp 289296.
Jin Y, Yu ZQ, Zhang L, Yao WH, Cai P. Ridge type internal
bae for uidized bed reactor. Petro Chem Eng (China)
15(5):269277, 1986b.
Jodra LG, Aragon JM. Prediction of the bubble-size distri-
bution in uidized beds with internal baes. Inter Chem
Eng 23(11):1830, 1983.
Johnsson F, Andersson S. Expansion of a bubbling uidized
bed with internals. Report A. Department of Energy
Conversion, Chalmers University of Technology,
Go teborg, Sweden, 1990, pp 90184.
Kang WK, Sutherland JP, Osberg GL. Pressure uctuations
in a uidized bed with and without screen cylindrical
packings. Ind Eng Chem Fundam 6(4):499504, 1967.
Kono HO, Hwang JT. The staged underow uidized bed.
AIChE Symposium Series 222, Vol. 78, pp 3746, 1983.
Levy EK, Freeman RA, Caram HS. Eect of horizontal
tubes on bubble eruption behavior in a gas uidized
bed. In: K O
stergaard, A So rensen, eds. Fluidization V.

Elsinore, Denmark, May 1823, 1986, pp 135142.
Li LX. Experimental study on spiral ow and design and
application of a new combined drying system. MS thesis,
Tsinghua University. 1997.
Liu DJ, Zheng CG, Li H, Kwauk M. A preliminary investi-
gation on integral circulating uidized bed (in Chinese).
Eng Chem Metal 6467, 1993a.
Liu DJ, Zheng CG, Li H, Kwauk M. Preliminary studies on
integral circulating uidized bed (in Chinese). In:
Proceedings 6th National Conf. on Fluidization.
Wuhan, China, October 1993b, pp 6366.
Milne BJ, Berruti F, Behie LA. Solids circulation in an
internally circulating uidized bed (ICFB) reactor. In:
Fluidization VII (OE Potter, DJ Nicklin, eds.).
Engineering Foundation, New York, 1992a, pp 235242.
Milne BJ, Berruti F, Behie LA, de Bruijn TJW. The intern-
ally circulating uidized bed (ICFB): a novel solution to
gas bypassing in spouted beds. Can J Chem Eng 70:910
915, 1992b.
Milne BJ, Berruti F, Behie LA, de Bruijn, TJW. The hydro-
dynamics of the internally circulating uidized bed at high
temperature. In: Avidan AA (ed.), Circulating Fluidized
Bed Technology IV. AIChE, New York, 1994, pp 2831.
Newby RA, Keairns DL. Fluidized bed heat transport
between parallel, horizontal tube-bundles. In: Davidson
JF, Keairns DL, eds. Fluidization. London: Cambridge
Noordergraaf IW, van Kijk A, van der Bleek CM.
Fluidization and slugging in large-particle systems.
Olowson PA. Hydrodynamics of a uidized bed-inuence of
pressure and uidization velocity. PhD thesis, Chalmers
University of Technolgy, Go teborg, 1991.
Olowson PA. Inuence of pressure and uidization velocity
on the hydrodynamics of a uidized bed containing hor-
izontal tubes. Chem Eng Sci 49(15):24372446, 1994.
Olsson SE, Wiman J, Almstedt AE. Hydrodynamics of a
pressurized uidized bed with horizontal tubes: inuence
of pressure, uidization velocity and tube-bank geometry.
Chem Eng Sci 50(4):581592, 1995.
Overcashier RH. Some eects of baes on a uidized system.
AIChE J 5(1):5460, 1959.
Ramamoorthy S, Subramanian N. Axial solids mixing and
bubble characteristics in gas-uidized beds with vertical
internals. Chem Eng J (Lausanne) 22(3):237242, 1981.
Rhodes MJ, Geldart D. Transition to turbulence? In:
O
stergaard K, So rensen A, eds. Fluidization V. New

York: Elsinore, Deamarh, 1986, pp 281288.
Sadasivan N, Barreteau K, Laguerie C. Studies on frequency
and magnitude of uctuations pressure drop in gas-solid
Salah H, Zhu JX, Zhou YM, Wei F. Eect of internals on
the hydrodynamic circulating uidized beds. Chinese
Society of Particuology, Beijing, China, pp Eq 1.11.6,
May 1996.
Satifa S, Fan LS. Characteristics of slugging regime and
transition to turbulent regime for uidized beds of large
coarse particles. AIChE. J, 31:15541562, 1985.
Shibata K, Shimizu M, Inaba S, Takahashi R, Yagi J. One-
dimensional ow characteristics of gas-powder two phase
ow in packed beds (in Japanese). Iron Steel 77:236243,
1991.
Sitnai O, Dent DC, Whitehead AB. Bubble measurement in
gas-solid uidized beds. Chem Eng Sci 36(9):1583, 1981.
Sitnai O, Whitehead AB. Immersed tubes and other inter-
nals. In: Davidson JF, Clift R, Harrison D, eds.
Fluidization, 2nd ed. Academic Press, London, 1985, pp
473493.
Song X, Wang Z, Jin Y, Tanaka Z. Powder Technology
83:127131, 1995.
Staub FW, Canada GS. Eect of tube bank and gas density
on ow behavior and heat transfer in uidized beds. Fluid
Proc Eng Found Conf 2nd, London: Cambridge Univ
Press: 1978, pp 339344.
Staub FW, Wood RT, Canada GS, Mclaughlin MH. Two-
phase ow and heat transfer in uidized beds. EPRI
Report CS-1456, RP525-1, 1980.
Van der Ham AGJ, Prins W, Van Swaaij WPM.
Hydrodynamics of a pilot-plant scale regularly packed
circulating uidized bed. AIChE Symp Ser 89(296):53
72, 1993.
Van der Ham AGJ, Heesink ABM, Prins W, Van Swaaij
WPM. Proposal for a regenerative high-temperature pro-
cess for coal gas cleanup with calcined limestone. Ind Eng
Chem Res 35(5):14871495, 1996.
Volk W, Johnson CA, Stotler HH. Eect of reactor internals
on quality of uidization. Chem Eng Progr 58(3):4447,
1962.
Wang ZL, Yao JZ, Liu SJ, Li HZ, Kwauk M. Studies on the
voidage distribution in an internally circulating uidized
bed (in Chinese). In: Proceedings 6th National Conf. on
Fluidization, Wuhan, China, October 1993, pp 150154.
Xavier AM, Lewis DA, Davidson JF. Trans Inst Chem Eng
56:274280, 1978.
Yamaoka H. Iron Steel (in Japanese) 72:403410, 1986.
Yates JG, Ruiz-Martinez RS. Interaction between horizontal
tubes and gas bubbles in a uidized bed. Chem Eng
Comm 62:6778,1987.
Ye YH. Internals for uidized beds. In: Wang ZX, Ye YH.
eds. Chemical Engineering Handbook (section 20,
Fluidization). Beijing: Chemical Industry Press, 1987, pp
138177.
Yerushalmi J, Cankurt NT. Further studies of the regimes of
uidization. Powder Technol 24:187205, 1979.
Zabrodsky SS, Epanov YG, Galershtein DM, Saxena SC,
Kolar AK. Int. J. Heat Mass Transfer 24(4):571579,
1981.
Zhao J, Zhong X, Xu HQ. A model of solid backmixing
between stages in a gas-uidized bed with perforated baf-
es. Powder Technol 73:3741, 1992.
Zheng CG, Tung YG, Zhang WN, Zhang JG. Impact of
internals on radial distribution of solids in a circulating
uidized bed (in Chinese). Eng Chem Metallurgy
11(4):296302, 1990.
Zheng CG, Tung YK, Li HZ, Kwauk M. Characteristics of
fast uidized beds with internals. In: Potter OE, Nicklin
DJ, eds. Fluidization VII. Engineering Foundation, New
York, 1992, pp 275284.
Zheng CG, Tung YK, Xia YS, Hun B, Kwauk M. Voidage
redistribution by ring internals in fast uidization. In:
Dwauk M, Hasatani M, eds. Fluidization 91. Science
and Technology. Beijing: Science Press, 1991, pp 168177.
Zhu JX, Salah M, Zhou YM. Radial and axial voidage dis-
tributions in circulating uidized bed with ring-type inter-
nals. J Chem Eng Jpn 30(5):928937, 1997.
Ziegler EN, Brazalton WT. Ind Eng Chem (Proc Des Dev),
2:276, 1963.
8
Attrition
Joachim Werther and Jens Reppenhagen

1 INTRODUCTION
In any uidized bed processbe it a combustor, a
dryer, or a chemical reactorthe bed material is in
vigorous motion and thus inevitably subjected to
mechanical stress due to interparticle collisions and
bed-to-wall impacts. This mechanical stress leads to a
gradual degradation of the individual bed particles,
which is quite often an unwanted phenomenon that
is therefore termed attrition.
The main consequence of attrition in uidized bed
processes is the generation of nes that cannot be kept
inside the system. Hence attrition creates an additional
burden on the ltration systems, i.e., the collection
systems must be larger (Merrick and Highley, 1974;
Vaux, 1978), and most importantly there is a loss of
valuable material. In a combustion process, for exam-
ple, the attrition-induced production of nes will lead
to a loss of both unburned fuel and sorbent, causing a
decrease in the eciency of combustion and sulfur
retention (Pis et al., 1991; Ray et al., 1987b). Other
examples are the heterogeneously catalyzed chemical
reactors, where the compensation for the lost material
by the addition of fresh makeup catalyst can amount
to very high operating expenses due to which a given
process may even become uneconomical (Contractor et
al., 1989; Patience and Mills, 1994).
Simultaneously with that unwanted production of
nes, there is a further signicant eect of attrition,
namely its inuence on the bed particle size distribu-
tion: it causes in any case a gradual shrinking of the so-
called mother particles from which the nes are drawn
until they are nally lost from the system as well (Kunii
and Levenspiel, 1969; Ray et al., 1987b). In addition,
there might be even a fragmentation of bed particles,
resulting in a signicant increase in the number of
particles and a corresponding decrease in the mean
particle size (Blinichev et al., 1968, Pis et al., 1991).
Whether with a batch process or a continuous one,
attrition will thus always lead to dierences in the
bed particle size distribution in comparison to that of
the feed material. Some authors (such as Kraft et al.,
1956; de Vries et al., 1972) report, for example, on the
eect that the bed material in a considered FCC pro-
cess got too coarse by the addition of makeup catalyst
in compensation for the lost mass of nes. Braca and
Fried (1956), however, observed the opposite eect,
where the bed material of the FCC process got too
ne since the attrition-produced nes were mostly
kept inside the system.
Since the particle population determines almost all
relevant mechanisms in a uidized bed system, attri-
tion may thus strongly aect the performance of a
uidized bed process. For instance, the elutriation
and entrainment eects (cf. Chapter 4 of this book),
the heat transfer from bed to inserts (Molerus, 1992) or
the conversion and selectivity of reactions (Werther,
1992) are aected either directly or through the bed
hydrodynamics by the particle size distribution.
Therefore some authors (such as Ray et al., 1987a)
even claim that a uidized bed with attritable materials
Current aliation: BMHClaudius Peters, Buxtehude,

Germany.
cannot be designed and simulated appropriately unless
the attrition activity is described.
Naturally, the primary interest in the eld of attri-
tion is to reduce its extent to a minimum. An obvious
approach is to use an optimum attrition resistant bed
material. Hence the majority of work deals with the
particles degradation in standardized bulk attrition
tests to determine the relative tendency of materials
to attrit (Davuluri and Knowlton, 1998). These so-
called friability tests are quite useful in the quick
and rst assessment of a candidate bed material.
But they solely focus on one factor aecting attrition,
namely the materials properties, whereas the prevail-
ing stress mechanisms may dier considerably from
those in the process. For that reason, the pure tests
cannot predict the quantitative extent of attrition that
will occur with the tested materials in a technical
uidized bed process. They cannot even guarantee
that the ratio of the attrition rates of two materials
will be the same in the test and in the process
(Werther and Reppenhagen, 1999a; Boerejn et al.,
2000).
Hence the pure attrition tests are not sucient
when a quantitative prediction of the attrition-
induced material loss or a prediction of the eect
on the bed particle size distribution is required. This
is for instance the case in the design procedure of a
process, where the capacity of the dust collection
system and the lifetime of the bed material must be
evaluated (Vaux and Fellers, 1981; Zenz, 1974) and
whereabove allthe hydrodynamics and thus the
bed particle size distribution must be clear (Ray et
al., 1987a; Zenz, 1971). Another example is when a
new generation of catalyst is to be developed for an
existing uidized bed reactor: here it might not be
sucient to know the relative hardness in comparison
to the previous catalyst generation. For a comprehen-
sive costbenet analysis, it is rather necessary to pre-
dict the attrition-induced loss rate and to be sure
about the hydrodynamics and thus the particle popu-
lation of the bed material. The same is valid when a
fuel or a sorbent is to be changed in a combustion
process.
In order to achieve such a comprehensive insight
into the attrition phenomena of a given process, it is
necessary to consider meticulously each individual
eect of the various inuencing factors related to the
material properties, the system design, and the
operating conditions. The present chapter attempts to
summarize systematically the ndings available in the
open literature and thus to draw up a possible
approach in the attrition modeling.
The rst key point in dealing with attrition seems to
be a distinction between the dierent ways of the par-
ticles degradation, such as fragmentation or abrasion
(e.g., Blinichev et al., 1968; Pell 1990), which are based
on quite dierent mechanisms. For this reason Sec. 2
will deal with the various modes of attrition and the
factors aecting them. Furthermore, there should be
an unambiguous denition of the way in which the
attrition extent and/or its impact on the particle size
distribution is measured and quantied. This denition
must suit both the prevailing attrition modes and the
attrition-induced eect that is to be considered.
Section 3 will therefore deal with the respective assess-
ment procedures. This provides a basis on which the
inuences of both the material properties and the
process conditions, i.e., system design and operating
conditions, can be described.
Considering the inuence of the material properties,
it must be acknowledged that attrition is a statistical
eect, i.e., there are dierences in the attrition suscept-
ibility of the individual bed particles, and the attrition
stress is acting randomly on them. As a consequence,
there is usually not the inuence of one particular
property evaluated; rather there is an evaluation of
the materials statistical attritability as a whole.
Section 4 summarizes the relevant attrition test proce-
dures. The results of these attrition tests must then be
transferred to the actual process by means of a physi-
cally sound description of the process conditions
(Werther and Reppenhagen, 1999; Boerejn et al.,
2000). This, however, requires a distinction between
dierent regions of the system that apply dierent
types of stress to the solids. For this reason, Sec. 5
rst summarizes the attrition mechanisms prevailing
in the relevant sources of a uidized bed system, and
Sec. 6 then nally deals with a description of attrition
in the entire process.
2 THE MODES OF ATTRITION AND THE
FACTORS AFFECTING THEM
Both the mechanism and the extent of attrition depend
not only on the properties of the solids but also on the
process design and to a large extent on the process
conditions. Clift (1996) has stated that attrition is a
triple-level problem, i.e., one must deal with phenom-
ena on three dierent length and time scales: of the
processing equipment, of the individual particles, and
of the subparticle phenomena such as fracture, which
leads to the formation of nes. The appearance of
attrition, therefore, can dier between the various pro-
cesses and even between individual regions in a given
process (cf. Sec. 5).
In general, the mode of attrition may vary from
pure abrasion to total fragmentation of the particles
(Blinichev et al., 1968; Pell, 1990). Abrasion implies
that exclusively asperities are removed from the par-
ticles surfaces. It thus produces a lot of elutriable nes,
whereas the particle size distribution of the mother
particles is hardly changed. In contrast to this,
fragmentation is a process of particle breakage into
similarly sized pieces. As a result, the number of par-
ticles increases, and the particle size distribution
becomes signicantly broader with a distinctly smaller
mean diameter compared to the original one. The two
modes and their dierent eects on the particle size
distribution are sketched in Fig. 1.
There are various attrition modes that can be
regarded as lying somewhere in between these two
extremes, as for example the initial breakage of fresh
catalyst particles from which irregularities and corners
are knocked o. All these modes are based on dierent
mechanisms and must therefore be described sepa-
rately (Pis et al., 1991).
The necessity of distinguishing dierent modes of
attrition is illustrated by the work of Reppenhagen
and Werther (1999a), for example, dealing with cata-
lyst attrition in cyclones. Focusing on the mode of pure
abrasion, they derived a physically sound equation for
the description of the attrition-induced production of
nes and thus for the attrition-induced material loss
through the cyclone overow (cf. Sec. 5.3). This
model approach is conrmed for various types of cat-
alysts in a wide range of operating conditions. But
when a specic threshold velocity at the cyclone inlet
is exceeded, a so-called surface fragmentation adds to
the production of nes and leads to a much higher loss
rate as predicted from the model approach (Fig. 19).
Reppenhagen and Werther (1999a) identied the sur-
face fragmentation by scanning electron microscope
(SEM) images (cf. Fig. 5) and could thus determine
the boundaries for pure abrasion conditions and thus
for their cyclone attrition model, which deals solely
with pure abrasion.
As it is obvious from the above example, the indi-
vidual modes do usually not occur separately. They are
rather combined in varying proportions, which makes
a description of the entire attrition process rather com-
plicated. The only attrition mode that can occur sepa-
rately is abrasion, since it has the lowest threshold
energy. The other attrition modes, the threshold energy
of which is much higher, are at least combined with
abrasion. In general, the extent of abrasion in relation
to fragmentation depends on various factors that will
be discussed below.
Figure 1 Attrition modes and their eects on the particle size distribution (q
3
mass density distribution of particle sizes d
p
).
Several rules have been suggested to identify the
dominant mode in a given case. They all use the change
in the particle size distribution as a criterion. For
instance, Dessalces et al. (1994) assessed the degrada-
tion behavior of various industrial FCC catalysts by
the change in the ratio of the 90% and 10% values of
the cumulative mass distribution of the particle sizes.
They got high values of this ratio (> 100) if abrasion
was dominant and smaller values (< 10) in the case of
fragmentation.
In the following, the various factors are summarized
that have been identied to aect attrition in its extent
or in its mechanisms. Basically, they can be classied
into two major groups, namely the various factors
related to material properties on the one hand and
the factors related to the process conditions on the
other.
2.1 Material Properties
2.1.1 Particle Structure
On the part of the material properties, the particle
structure has the most fundamental inuence on the
catalysts attrition behavior. The extent of degradation
as well as its mode will strongly depend on whether a
particle has a crystallized or an amorphous structure or
is an agglomerate (British Materials Handling Board,
1987).
As an example, catalysts that are used in uidized
bed reactors are usually produced by spray drying.
This type of catalyst is on the one hand rather suscep-
tible to attrition, but on the other hand it tends to
abrasion rather than to fragmentation. Hence it is
often assumed that abrasion is the only or at least
the governing attrition mode in uidized bed reactors
(e.g., Werther and Xi, 1993).
Arena et al. (1983) and Pis et al. (1991) studied the
attrition of amorphous materials such as coal or lime-
stone and found that the size distributions of the
attrited materials were independent of the initial
particle size and of most operating parameters. Ray
et al. (1987a) assumed that unlike crystalline materials,
amorphous materials may have some kind of natural
grain size to which the degradation nally leads.
2.1.2 Particle Size
The initial particle size is of primary interest with
respect to particle breakage, because smaller particles
tend to contain fewer faults in form of microcracks or
imperfections and are thus more dicult to break than
larger ones. However, the mechanisms of breakage will
not be further discussed in the present chapter. instead
we refer to a comprehensive survey given by the British
Materials Handling Board (1987).
2.1.3 Particle Size Distribution
Besides the breakage probability of the particles, the
size distribution determines another important
parameter with respect to attrition, namely the volume
specic surface area of the bed material. This para-
meter is particularly relevant with respect to abrasion,
where it leads to two contrary eects.
On the one hand, it determines the total surface area
that is exposed to abrasion. On the other hand, at
given process conditions, the specic surface area
determines the specic surface energy available for
attrition or abrasion. In other words, with a ner bed
material there may be an increase in total surface
exposed, but the energy that is acting on a unit surface
is decreasing. This is particularly relevant with respect
to mixtures. As an example, Arena et al. (1983) inves-
tigated coal attrition in a mixture with sand under hot
but inert conditions. As they increased the sand par-
ticle size while keeping its mass in the bed constant,
they observed an increase in the coal attrition rate.
They interpreted their results by assuming that the
abrasion energy is shared out on the entire material
surface. On the same basis Ray et al. (1987a) developed
their attrition rate distribution model for abrasion in
a uidized bed.
These ndings may also explain the observed inu-
ence of the nes content on the attrition propensity of
a given material: Forsythe and Hertwig (1949) already
noticed a reduction of the degradation of FCC cata-
lysts in jet attrition tests due to the presence of nes.
They themselves supposed some kind of cushioning
eect that limits the force of collision impact and
thus limits the degradation of the coarse particles.
The eect of ne particles is of strong interest because
they are produced by attrition, so attrition inhibits
itself if the ne particles remain in the system.
More recently, Reppenhagen and Werther (1999a)
and Werther and Reppenhagen (1999) found in their
catalyst attrition tests that the abrasion-induced pro-
duction of nes increases linearly with the surface
mean diameter of the tested material.
2.1.4 Particle Shape and Surface Structure
The particle shape is a relevant parameter, because
irregular and angular particles are inclined to have
their corners knocked o in collisions and thus become
rounder and naturally smaller with time. A macrosco-
pically smooth surface is therefore less prone to break-
age, but it may still undergo abrasion. With respect to
the latter, the microscopic surface structure is of inter-
est. Surface asperities may be chipped o and lead to
abrasion.
In their investigation on cyclone attrition,
Reppenhagen and Werther (1999a) compared, for
example, the attrition resistance of two dierent cata-
lysts, the chemical composition of which is identical
but their shape and surface structure is dierent (cf.
Fig. 2). Whereas the catalyst termed 97-G(fresh) is
irregularly shaped and has a rough surface, the catalyst
termed 97-R(fresh) is almost spherical, and its surface
shows only few asperities. Reppenhagen and Werther
(1999a) attributed these morphological dierences to
their nding that the 97-G(fresh) catalyst was almost
twice more attritable than the 97-R(fresh) material.
Both the particle shape and the surface structure
will inevitably change with time inside a process. This
will in any case happen as a gradual eect, which is
discussed in Sec. 2.2.3. But it may also happen instan-
taneously due to excessive stress, for example the sur-
face fragmentation of catalyst particles that has been
observed by Reppenhagen and Werther (1999a) (cf.
Figs. 19, 4, and 5).
2.1.5 Pretreatment and Preparation or Processing
History
The unsteady-state attrition rate of a given material
will normally depend on the pretreatment to which it
was subjected. The higher the intensity and the longer
the duration of the foregoing stress, the more probable
it is that the weakest points of the particles have
already been attrited.
The pretreatment may already occur in the pro-
cessing route of a given material. If we consider for
example the spray-drying process of catalysts, the
weakest agglomerates will immediately break down
during the production and handling of the material.
Otherwise, there is a pretreatment at earlier times in
the process itself. A quite often quoted example is
the dierence between a fresh catalyst material and
the corresponding equilibrium catalyst taken from
the process (Boerejn et al., 1998). In their cyclone
attrition tests, Reppenhagen and Werther (1999a)
compared the attrition resistance of an equilibrium
FCC catalyst obtained from a renery with that of
the fresh material. They found the fresh material to
be almost ve times more attritable than the equili-
brium catalyst and concluded that some of the mate-
rial properties must have been changed inside the
FCC process. These properties may be chemical
ones with resulting changes in the material strength,
but in any case Reppenhagen and Werther (1999a)
observed morphological changes in the SEM images
shown in Fig. 3. The surfaces of the fresh particles
are rougher than the ones of the equilibrium particles.
Consequently, there are more asperities that can be
chipped o and thus contribute to the attrition
extent. Furthermore, the equilibrium particles are a
bit more spherical than the fresh ones, which means
according to Sec. 2.1.4 an increase in the attrition
resistance.
Another example might be the sulfation of lime-
stone sorbents in combustion processes (Anthony
Figure 2 SEM images of two catalyst materials. The surface mean diameter was 112 microns in the case of the 97-R(fresh)
and 103 microns in the case of the 97-G(fresh). (From Reppenhagen and Werther, 1999a.)
and Granatstein, 2001). In any case, these are gradu-
ally acting phenomena, that are strongly related to the
inuence of the solids residence time, the eect of
which is discussed in Sec. 2.2. But there may be also
singular phenomena that aect the materials attritabil-
ity. A broken steam pipe in the stripping unit of an
FCC plant may for example cause a steam jet that
generates extra stress on the catalyst particles. Such a
singular event would not only cause a higher extent of
attrition during its occurrence but also it would most
probably have a long-term eect. The reason for this is
a serious damage of the morphology of the particles.
Reppenhagen and Werther (1999a) observed such a
phenomenon during their cyclone attrition tests as
they exceeded, for several passes through the cyclone,
a certain threshold energy. The result of this experi-
ment is shown in Fig. 4.
A batch of catalyst particles was passed in a rst
phase through a cyclone under conditions of pure
abrasion, then in a second phase under conditions
where additional surface fragmentation takes place,
and nally in a third phase again under conditions of
the rst phase of pure abrasion. It is quite obvious that
the surface fragmentation leads to an increased sus-
ceptibility to abrasion, i.e., at the end of the experiment
the abrasion rate is distinctly higher than at the begin-
ning, even though the operating conditions are identi-
cal. This can be explained by comparing SEM images
taken from the catalyst particles before and after the
subjection to the excessive stress (cf. Fig. 5). It can be
seen that parts of the particles surfaces have been
chipped o, leading to a new surface that is distinctly
rougher than the old one and therefore more prone to
abrasion. However, with an increasing number of
Figure 3 SEM images of an equilibrium catalyst and the corresponding fresh material. (Reppenhagen and Werther, 1999a.)
Figure 4 Inuence of surface fragmentation on the cyclone loss rate at conditions of pure abrasion (u
c;in
20 m=s,
c
0:1).
(From Reppenhagen and Werther, 1999a.)
passes under conditions of pure abrasion, the rough
surface, which is responsible for the increased loss
rate, would become smoother again and the cyclone
loss rate would then decrease to its initial value.
2.2 Process Conditions
The process conditions, which are resulting from the
system design and the operating conditions, will pri-
marily inuence the attrition of the bed material by
generating the stress on the individual particles. In
general, the stress leading to attrition of a given bulk
material may be a mechanical one due to compression,
impact, or shear, a thermal one owing to evaporation
of moisture or temperature shock, or a chemical one by
molecular volume change or partial conversion of the
solid into the gas phase.
2.2.1 Gas and Solids Velocity
The gas velocity is usually directly related to the par-
ticle velocity, which is the most important factor in
generating the mechanical stress by interparticle colli-
sions or by particlewall impacts (British Materials
Handling Board, 1987). The forces involved in the
degradation process may be generated by high-speed
collisions resulting preferably in breakage.
Alternatively, the energy may be transmitted through
a matrix of comparatively slowly moving particles
resulting mostly in abrasion. Particularly in the distri-
butor region of a uidized bed, where grid jets are
issuing into the bed of particles, the impact velocities
between particles can be extremely high and result in
signicant particle breakage.
2.2.2 Wall-Hardness
One can assume that the particle degradation increases
with the hardness of the vessel wall. This eect will
increase with increasing ratio of particle-to-vessel dia-
meter and may thus be only relevant inside a cyclone,
or in the vicinity of bed inserts.
2.2.3 Solids Residence Time
The relationship between the solids residence time and
the amount of material that is produced by attrition is
generally nonlinear. As an example, Fig. 6 shows the
typical time dependence of the attrition of a fresh cat-
alyst that is subjected to attrition in one of the test
devices described in Sec. 4.
The elutriated mass is dened to be the attrition
product, and consequently the attrition rate is dened
as elutriated mass per unit time. It is clearly seen that
the rate of attrition is decreasing with time. The reason
is that at the beginning the fresh catalyst particles have
very irregular shapes and contain many faults. This
results in a high rate of initial particle degradation,
during which the particles break and their edges and
asperities are knocked o. With progressing time, the
particles become smaller, rounder, and smoother, and
the number of their weak points decreases. The elutria-
tion rate therefore decreases continuously with time
and tends to a more or less constant value, which
can be interpreted as a kind of steady-state level
where only abrasion takes place.
Figure 5 SEM images of a given catalyst material (a) after subjecting it to pure abrasion, and (b) after subjecting it to abrasion
with additional surface fragmentation. (Note: The bridges between the particles in image (b) are not due to sintering but to an
excess gold layer resulting from the SEM preparation step.) (Reppenhagen and Werther, 1999a.)
The shape of the curve in Fig. 6 is typical for all
particle degradation processes. However, one has to
distinguish between batch and continuous processes.
In batch processes the whole bed material is always
at the same state of attrition, which changes during
the operating time according to Fig. 6. On the other
hand, there is a particle residence time distribution in
continuous processes. The attrition rate of the entire
material is therefore constant, although the state of
attrition of each particle again changes according to
Fig. 6.
Many attempts have been made to describe the time
dependence of the attrition rate in batch uidized bed
processes. Gwyn (1969) studied the degradation of cat-
alysts in a small-scale test apparatus and dened the
elutriated particles as the only attrition product. He
described the increase of the elutriated mass m
loss
with time t based on the initial solid bed mass m
b;0
by the now widely known Gwyn equation,
m
loss
m
b;0
K
a
t
b
1
For several batches of the same catalyst material with
quite dierent mass median diameters, Gwyn found
the exponent b to be constant, whereas the attrition
constant K
a
was found to decrease with mean particle
size. Equation (1) is therefore valid for a particular size
distribution only. Other empirical correlations for the
time dependence of the mass ratio m
loss
=m
b;0
have been
suggested by e.g. Pis et al. (1991) and Dessalces et al.
(1994).
2.2.4 Temperature
There are three conceivable temperature eects that
may inuence the particle degradation in either a direct
or an indirect way, i.e., thermal shock, changes in
particle properties, changes in the gas density.
The heating of fresh cold particles fed into a hot
process can cause various phenomena that may lead
to particle degradation. These are thermal stress,
decrepitation, evaporation of moisture, hydrate
decomposition, and impurity transformation. On the
other hand, particle properties such as strength,
hardness, and elasticity may also be aected by the
temperature. With respect to the resistance to degrada-
tion, there is an optimum temperature range for any
specic type of material. At lower temperatures, parti-
cles become brittle, while at higher temperatures they
may soften, agglomerate, or melt and lose discrete par-
ticulate properties. Consequently, it is important that
the particle friability, which is the major factor for the
attrition propensity of a given material, is assessed
under conditions that are similar to those found in
the process where the attrition is occurring.
Moreover, the temperature can have a strong eect
on the gas density, which aects the uidization state
and with it the particle motion and the stress the par-
ticles are subjected to.
2.2.5 Pressure
The absolute pressure is unlikely to have a direct eect
on attrition unless it aects the amount of adsorbed
surface layers. But there is again an eect on the gas
density that is similar to the eect of temperature men-
tioned above. Moreover, the rate of pressure change
may have more inuence.
2.2.6 Chemical Reaction
On the one hand, a chemical reaction of the particulate
material may generate stress within the particles lead-
Figure 6 Time dependence of catalyst attrition in a submerged jet test. D
t
0:05 m, u
or
100 ms
1
, d
or
2 mm, and HA-HPV
is a fresh catalyst, whereas FCC is a spent catalyst from a commercial unit). (From Werther and Xi, 1993.)
ing to fracture. In the case of gassolid reactions, the
particle degradation is also desired, because it acceler-
ates the reaction by extending the reactive surface. A
commercially relevant example is the particle degrada-
tion of solid fuels in combustion processes. This latter
topic has been extensively studied by Massimilla and
coworkers. The reader is referred for further details to
a comprehensive review given by Chirone et al. (1991).
On the other hand, a chemical reaction of the par-
ticulate material may aect the mechanical attrition
resistance of the particles, as for example the sulfation
of limestone sorbents in combustion processes
(Anthony and Granatstein, 2001) or the dierent
degrees of oxidation of a catalyst (Xi, 1993).
3 ASSESSMENT OF ATTRITION
As attrition basically aects everything from a single
particle to the entire process, there are various criteria
on the basis of which attrition may be qualitatively and
quantitatively characterized. One may base this assess-
ment on the observation of an individual particle.
Alternatively, the fate of a group of particles may
be examined, or the eect of attrition on the bulk
properties of the given material may be taken for this
assessment. The British Materials Handling Board
(1987) and Bemrose and Bridgwater (1987) give exam-
ples for the dierent methods.
The chosen type of assessment procedure basically
depends on whether the attrition propensity of a given
material, i.e., its material strength, or attrition in a
given process is to be considered. Furthermore, the
choice of the assessment procedure depends on the
attrition eect that is of interest. When the entire pro-
cess is considered, it is usually either the assessment of
the attrition-induced material loss or the assessment of
the attrition-induced changes in the particle size distri-
bution of the bed material.
3.1 Assessment of the Materials Strength
The assessment of the materials strength is commonly
based on so-called friability tests or attrition tests (cf.
Sec. 4), in which a single particle or a bulk sample of
the material is for a certain extent subjected to a spe-
cic stress. Data from these tests are usually presented
as single numbers called friability or attrition indices.
Most of these numbers are derived from a comparison
of the materials content of a previously dened frac-
tion of nes before and after the test (British Materials
Handling Board, 1987; Davuluri and Knowlton, 1998).
A frequently quoted example is the Hardgrove
Grindability Index, which was originally developed
to assess the grindability of coal. It measures the gen-
eration of material less than 74 microns produced in a
ball-ring pulverizer during 60 revolutions on a sample
with a size range between 590 and 1190 microns (cf.
ASTM D409-71).
Such indices may give a ranking of the materials
friability but cannot be directly related to the process
attrition, which would be the most appropriate assess-
ment. This means, if at given process conditions a
material A yields compared to a material B twice as
high attrition, the ratio of the respective attrition
indices of the materials A and B should be 2.
Since the attrition propensity of a given material
strongly depends on the type of stress to which it is
subjected (cf. Sec. 4), the requirement of a direct rela-
tion to the process will be met only by attrition tests
that simulate the process stress. The results from those
tests are usually expressed by a so-called attrition rate
[cf. Eqs. (2) and (3)], which is also taken for the assess-
ment of process attrition. Hence for each considered
process there is an empirical or even physically sound
equation available that describes the attrition rate in
dependence on the chosen operating conditions. The
materials strength is then simply expressed as a
proportionality constant that is independent of the
operating conditions, the so-called attrition rate con-
stant [e.g., K
j
in Eq. (7)].
3.2 Assessment of the Attrition-Induced Material
Loss
In uidized bed experiments, most authors consider
the attrition-induced material loss only, i.e., they
assume that all attrition products are elutriated.
Consequently, they measure either the decrease in
bed mass (Kono, 1981; Kokkoris and Turton, 1991,
1995) or the elutriated mass (Seville et al., 1992;
Werther and Xi, 1993). The results are commonly
expressed by a so-called total or overall attrition rate
r
tot
, which is accordingly dened by the relative change
of bed weight m
b
with time,
r
tot

1
m
b
dm
b
dt
2
or, alternatively, by the relative change in elutriated
mass m
loss
with time,
r
tot

1
m
b
dm
loss
dt
3
Whether Eq. (2) or Eq. (3) is chosen will normally
depend on the type of equipment used. When compar-
ing dierent authors works, attention should be paid
to the denition of m
b
. For batch processes it may be
either the initial bed mass or the instantaneous inven-
tory of the bed. Special attention has to be paid to a
denition of attrition rates in the case of continuous
processes where fresh material must be added to com-
pensate for reacted or withdrawn material and for the
attrition losses. The fresh material may contain elutri-
able nes which add to the measurable elutriation rate
thus leading to an apparently higher attrition rate.
Furthermore, it should be noted that a systems
material loss is also aected by the eciency and the
cut size of the gas/solids separation unit. Dierences in
this type of system must be taken into account when
dierent authors works are compared, or when a test
result is scaled-up to a full-scale process. Section 4.3,
which contains a description of the various test equip-
ment and procedures, will deal with this matter in more
detail.
Some authors (Vaux and Keairns, 1980, Sishtla et
al., 1989; Pis et al., 1991) dene the time integral of the
attrition rate as the attrition extent E
tot
t;
E
tot
t
_
t
0
r
tot
t dt 4
This denition, however, includes eects of initial
breakage and of initial nes content and will therefore
normally not give an unambiguous assessment of attri-
tion.
3.3 Assessment of Changes in the Particle Size
Distribution
The above denition of the attrition rate considers the
bed material as a whole and quanties solely the
production of elutriable material without taking all
breakage events or the shrinking of the so-called
mother particles into account. More insights into the
attrition mechanisms can be obtained from the obser-
vation of the change in the particle size distribution as
demonstrated by Zenz and Kelleher (1980) and by Lin
et al. (1980). An example of one of the results obtained
by Zenz and Kelleher (1980) is shown in Fig. 7.
Such a description of the eects of attrition gives
good information about changes in the content of par-
ticles in a particular size class. But it can only describe
the fate of a batch of particles where the extent of the
changes depends on the length of the considered time
interval during which the catalyst is subject to attrition.
Furthermore, this type of assessment includes eects of
initial breakage and the sifting of initial nes. As a
consequence of the above limitations, this type of
assessment is able to describe only the results of attri-
tion and not the dynamics of the attrition mechanisms
themselves. This can be best explained on the basis of
the mass balance of a given size interval i,
m
i
m
feed;i
m
loss;i

n
hi1
m
h;i

i1
k1
m
i;k
5
The mass content in the interval i may thus be
increased by a mass m
feed;i
that is fed to the system
Figure 7 Changes in the particle size distribution of a FCC catalyst material that was subjected to cyclone attrition. (Zenz and
Kelleher, 1980.) The respective experimental procedure is described in Sec. 5.3.
and may be decreased by a mass m
lossi
that is lost from
the system in a considered time interval. Furthermore,
there is an internal mass transfer between the indivi-
dual size intervals. The interval i receives attrited par-
ticles from coarser size intervals h and loses attrited
particles to size intervals with ner material, k, which
is in Eq. (5) expressed by the transfer masses m
h;i
and
m
i;k
, respectively.
The assessment on the basis of changes in the par-
ticle size distribution considers the left-hand side of
Eq. (5), i.e., the absolute change in mass m
i
is mea-
sured. But the individual contributions that led to these
changes, i.e., the terms on the right-hand side of
Eq. (5), are not considered by this type of assessment.
It may be possible to take the contributions of feed and
system loss by means of measurements into account,
but it is not possible at all to unravel the internal
transfer masses m
h;i
and m
i;k
between the size intervals
just by observing the changes in the particle size
distribution. For this reason, it is not possible to extra-
polate the description to a dierent duration of the
attrition process or even to transfer it to other initial
particle size distributions. For this purpose a descrip-
tion via particle population balances is needed.
A description of the attrition process that is based
on a particle population balance requires a detailed
knowledge about the kinetics of the attrition
mechanisms of the individual particles. In particular,
the inuence of the individual particle size on the attri-
tion extent, and the size distribution of the correspond-
ing attrition products, must be known. With regard to
comminution processes it is therefore very common to
divide the bulk material in several size ranges to which
both a probability of degradation and an individual
size distribution of attrition products are assigned.
The entire description is then realized by a concept
of two separate functions, namely the so-called
selection function or probability function Sx and
the breakage function Bx; y (Austin, 1972,
Schubert, 1975). The selection function describes the
mass fraction of a dierential size interval (x, x dx)
which is subjected to the particle size reduction and
therefore lost to smaller size intervals. The assignment
of this lost mass fraction to the particular size intervals
is realized by the breakage function. It describes the
mass fraction of breakage products of size y originat-
ing from a particle of size x.
Sx and Bx; y can be dened as vectors and
matrices where the elements are applied to discrete
particle size ranges. In detail, the vector elements s
i
describe the rate of material loss of a particular size
interval of mean diameter x
i
and the various matrix
elements b
i;k
describe the distribution of attrition pro-
ducts from the intervals i into the intervals k of smaller
particle sizes. With a vector of the feed size distribution
they can be combined in a matrix equation that yields a
size distribution vector of the product. Since these vec-
tor and matrix elements must be usually obtained
experimentally, the use of selection and breakage func-
tions is much more cumbersome than working with a
single index number. This concept has therefore been
used almost exclusively for the description of commer-
cially interesting comminution processes but very
seldom for the description of catalyst attrition
(Wei et al., 1977).
However, if attrition exclusively occurs by the mode
of abrasion, the respective particle balance appears to
be less complex. There are several mathematical
models of particle population balances suggested in
the literature that deal with that particular mode
(Levenspiel et al., 1969; Overturf and Reklaitis, 1983;
Ray et al., 1987b; Werner et al., 1995; Reppenhagen,
1999; Reppenhagen and Werther, 2001) for uidized
bed systems, or Hansen and Ottino (1997) for a gen-
eral description of the abrasion process. A comparison
with the above-discussed concept of selection and
breakage functions shows that all these approaches
assume that the particles of a given size fraction are
either entirely subjected to abrasion or left undamaged.
This means that the selection function is either com-
pletely omitted, since all particle size ranges are
assumed to be attritable (Levenspiel et al., 1969;
Overturf and Reklaitis, 1983; Werner et al., 1995), or
is reduced to a step function from zero where no abra-
sion is assumed to one at a particle size above which all
particles are assumed to be attritable (Ray et al.,
1987b; Hansen and Ottino, 1997). The respective
breakage functions are also comparatively simply
organized in the above approaches. The particle-size
dependence of the attrition extent and thus the mass
of attrition-produced nes per size interval and unit
time is either described by a rst-order kinetics
(Levenspiel et al., 1969) or by a potential dependence
(Hansen and Ottino, 1997). The entire mass of nes is
then either simply assumed to be lost (Levenspiel et al.,
1969) or is assigned to a particular size interval with a
given constant size distribution (Ray et al., 1987b;
Hansen and Ottino, 1997).
In most of the above approaches the balances are
solved analytically and the abrasion-induced shrinking
of the mother particles is thus taken into account by
simply shifting the size distribution in accordance with
the respective material losses. In contrast to this,
Reppenhagen and Werther (2001) recently developed
a discretized particle population balance, which
describes the mass transfer between neighboring size
intervals on the basis of the natural coupling with the
fractional nes production, i.e., the mass of produced
nes originating per unit time from a size interval K
i
corresponds to the material loss of the initial parent
particles of this fraction.
4 ATTRITION TESTS
The term attrition test generally covers all experi-
ments in which the eects of attrition are considered
and assessed. This might vary from the observation of
the degradation of only a single particle (Boerejn et
al., 2000) to the consideration of an entire process
(Werther and Reppenhagen, 1999). However, analo-
gously to the division of the assessment procedures,
the large number of experiments can be divided into
two major elds of application, namely the tests of
material friability and the experiments to study attri-
tion phenomena. They will be separately discussed in
the following sections. The relevant test devices will be
discussed afterwards.
4.1 Friability Tests
Friability tests are often used for a comparison of
dierent types of materials to select the most attri-
tion-resistant one (Vaux and Fellers, 1981; Davuluri
and Knowlton, 1998). A eld where friability tests
are of particular importance is catalyst development
(Dart, 1974). As an example, Contractor et al. (1989)
used a submerged-jet attrition test (described below) in
their development of a new generation of uidized bed
VPO-catalyst.
Several attempts have been made to develop a
standardized attrition test procedure to determine the
relative tendency of materials to attrit. A candidate
procedure had to meet at least the following two cri-
teria: the amount of sample material required should
be small and the test time should be relatively short
(Davuluri and Knowlton, 1998). Unfortunately, the
determination of a materials friability is not as simple.
The crucial point is the interaction of the material
properties with the type of stress to which a material
is subjected. In this context Pell (1990) gave a fre-
quently quoted thought experiment to illustrate the
diculties: If we took a batch of rubbers and a
batch of diamonds, and rubbed them on abrasive
paper, we would conclude that the diamonds were
more attrition resistant. If we instead struck the parti-
cles with a hammer we would conclude that the rubber
were more attrition resistant.
In this context Contractor et al. (1989) conclude
that the relative attrition rate depends on the attrition
test method used. Knight and Bridgwater (1985) sub-
jected spray-dried powders to a compression test, a
shear test, and a test in a spiral classier. They found
that each test gave a dierent ranking of the materials.
Werther and Reppenhagen (1999a) observed this phe-
nomenon as they subjected various types of uidized
bed catalysts to both a cyclone attrition and a
jet attrition test, each simulating one of the three
major attrition sources in uidized bed systems (cf.
Sec. 5).
Obviously, there can be no universal procedure for
the measurement of a materials propensity to attri-
tion. The attrition resistance is relative and depends
on both the material and the stress. An appropriate
attrition test should therefore duplicate at least the
dominant stress occurring in the considered uidized
bed process, in order to ensure that the ranking of the
materials in the test will be identical to that in the
process.
However, besides the simple ranking there is quite
often even a quantitative prediction of the process
attrition requested. This requires both an attrition
model with a precise description of the process stress
andas an input parameter to the modelprecise
information on the materials attritability under this
specic type of stress. This calls for attrition/friability
tests that duplicate the process stress entirely. As will
be elucidated in Sec. 5, the stress in a given uidized
bed system will be generated from at least three
sources, i.e., the grid jets, the bubbling bed, and the
cyclones. For each there is a corresponding friability
test procedure.
4.2 Experiments to Study Attrition Mechanisms
Attrition can normally not be investigated directly in a
large-scale process. It is, for example, impossible to
analyze the entire bulk of material, and it is nearly
impossible or at least very expensive to perform a para-
meter analysis in a running industrial process. For this
reason, attrition has to be investigated in small-scale
experiments. The results of these experiments require a
model or at least an idea of the governing attrition
mechanisms to be applied to the large-scale process.
In principle, there are two dierent philosophies of
attrition modeling:
The most commonly used philosophy is to design a
bulk test facility that simulates the process stress.
Examples of such test facilities are the various jet attri-
tion test devices (such as Forsythe and Hertwig, 1949;
Gwyn, 1969; Werther and Xi, 1993), which simulate
the stress generated in the vicinity of a gas distributor.
The results can either be directly extrapolated to a
large-scale process or reveal relationships that may
serve as guidelines in the design of this process. It
should be noted that most of these tests are batch
tests, but most full-scale processes are continuous.
Hence the transfer of the results can be dicult
owing to the time dependence of the degradation.
Furthermore, it must be taken into account that
such a bulk test can only give results that are based
on statistics, i.e., they reveal the average eect on all
particles but cannot indicate a specic mechanism of a
single particle breakage. In any case attention must be
paid when the properties of the bulk sample tested
dier from those in the real process. As an example,
Werther and coworkers (Werther and Xi, 1993;
Reppenhagen and Werther, 1999a) used previously
screened catalysts in their investigation of the mechan-
isms of jet and cyclone attrition but also evaluated the
eect of the particle size distribution to take the shift in
the particle size into account (Werther and
Reppenhagen, 1999).
In the second attrition-modeling philosophy
(Yuregir et al., 1987; Boerejn et al., 1998, 2000), it
is believed that the mechanics of particle interaction
in process test devices and in the large-scale processes
are not suciently understood to maintain the neces-
sary dynamic similarity and that the analysis of the
experimental data for relating the trend to the large-
scale operation is therefore not straightforward
(Boerejn et al., 2000). Instead, test devices have to
be used where the material is subjected to a well-
dened stress like compression, impact, or shear. The
chosen test should preferably apply the type of stress
that is considered as the main source of attrition in the
large-scale process. The complete process attrition can
than be described by coupling the test results with a
model that describes the motion of uid and particles
and with it the frequency and extent of the stress.
Unfortunately, the practical use of this latter approach
is at the limit of our current state of knowledge. First
of all, it is not easy to dene the relevant stress of a
process, i.e., the stress is usually composed of more
than one of the above mentioned pure types of stress,
and it is usually changing in dependence on various
parameters. As an example, Boerejn et al. (2000)
reported a gradual change of the grid-jet-induced stress
from shear to impact stress with increasing orice-to-
particle size ratio. Moreover, the necessary models of
the hydrodynamics in the various regions are not yet
fully developed (Boerejn et al., 2000). Nevertheless,
an attempt has been made in one case: Ghadiri and
coworkers (Ghadiri et al., 1992a, 1994, 1995;
Boerejn et al., 1998, 2000) have applied this concept
to grid-jet-induced attrition using a single particle
impact test.
4.3 Test Equipment and Procedures
Detailed reviews of the various test procedures are for
example given by Bemrose and Bridgwater (1987) and
the British Material Handling Board (1987). The pre-
sent section is therefore conned to the tests that are
most relevant to the subject of uidized bed attrition.
4.3.1 Tests Applying Well-Dened Stress
As mentioned earlier, one can distinguish three pure
and well-dened mechanical stresses on bulk solids
material, namely compression, impact, and shear.
There are numerous tests that are based on compres-
sion and shear, e.g., Paramanathan and Bridgwater
(1983), Neil and Bridgwater et al. (1994), Shipway
and Hutchings (1993), but they are not further dis-
cussed in this chapter because these stresses are usually
not relevant to uidized beds. On the other hand,
impact stress occurs whenever particles hit walls or
other particles. Attrition caused by impacts can thus
be observed, e.g., in grid jets, in the wake of bubbles, in
cyclones, or due to free fall. Consequently, there is a
great variety of impact tests that try to simulate these
particular stresses.
There are, for example, various drop shatter tests in
which the material falls under gravity onto a hard sur-
face or a xed bed. Such a test was carried out by Zenz
and Kelleher (1980), who considered catalyst attrition
due to free fall in a CFB downcomer. However, the
probably most relevant impact tests are those where
pneumatically accelerated particles are impacted onto
a target. Yuregir et al. (1986, 1987) pioneered this type
of test in their work on NaCl salt. In the meantime,
such test devices have found broad industrial applica-
tion as friability tests. For example, Fig. 8 shows the
setup used by Davuluri and Knowlton (1998). It
requires approximately 100 grams of material to con-
duct a test. The velocity at which the solids strike the
impact plate was varied from 46 to approximately
144 m/s, but in the upper range a material-specic
threshold velocity exists, above which the particles
completely shatter. This is for most materials a velocity
greater than 76 m/s. If velocities above the threshold
velocity are used, then no relative attritability informa-
tion can be obtained.
In addition to its use as a pure friability test, this
type of test has also been used in the above-mentioned
grid-jet modeling approach of Ghadiri and coworkers
(Ghadiri et al., 1992a, 1994, 1995; Boerejn et al.,
1998, 2000). In this approach the particles were fed
in a single array to the eductor.
4.3.2 GraceDavison Jet-Cup Attrition Test
The GraceDavison jet-cup attrition test is often used
to test the friability of catalysts (Weeks and Dumbill,
1990; Dessalces et al., 1994). The jet-cup apparatus is
sketched in Fig. 9.
The catalyst sample is conned to a small cup, into
which gas is tangentially added at a high velocity
(about 150 m/s). After a test run over a period of an
hour, the so-called Davison Index (DI) is determined
by measuring the increase in the weight fraction of
particles below 20 microns. The increase is determined
from both the fraction of elutriated ne material and a
particle size analysis of the remaining fraction in the
cup.
Some authors (Dessalces et al., 1994) assume that
the stress in the jet-cup is similar to that prevailing in
gas cyclones. With respect to ne catalysts, this type of
test works as good as the impact test described above,
but its applicability is limited to smaller sizes because
larger particles tend to slug in the small cylinder.
However, in the catalyst development, where at rst
only a little batch of catalyst is available, this appara-
tus is an important friability test, because it requires
only a small amount of material (approximately 5 to
10 g).
4.3.3 Fluidized Bed Tests
Fluidized bed tests may be used for both purposes, the
determination of the catalysts friability (Forsythe and
Hertwig, 1949; Gwyn, 1969) and the investigation of
attrition mechanisms (Werther and Xi, 1993). Most
uidized bed tests are currently carried out as so-called
submerged-jet tests, where high-velocity gas jets sub-
merged in a uidized bed produce high attrition rates
in a well-dened short period of time. The majority of
these tests are based on the device suggested by
Forsythe and Hertwig (1949) (Fig. 10a).
The setup consists of a 0.0254 m ID and 1.52 m long
glass pipe, which bears a canvas lter at its upper end,
and which is sealed by an orice plate at the bottom.
This latter plate contains a single 0.4 mm ID orice in
its center. The apparatus is operated in such a way that
the jet gas velocity approaches the speed of sound in
the orice. The lter keeps all material inside the sys-
tem. To assess the degradation extent, one should
screen the material by wet sieving through 325 mesh
(44 mm). The attrition rate is dened as the ratio of the
Figure 8 Schematic drawing of impact test facility used by
Davuluri and Knowlton (1998).
Figure 9 Schematic drawing of the GraceDavison jet-cup
attrition test. (After Weeks and Dumbill, 1990.)
increase in weight percent of 325 mesh material and
the weight percent age of 325 mesh particles in the
initial material. Using this test procedure, one has to
take into account that the attrition-produced nes that
are kept in the system may aect the materials attrit-
ability. Moreover, there are diculties in using this test
as a friability test, because the results are time inte-
grated and have to be assessed with the help of the
particle size distribution. The materials that are to be
compared should therefore have a similar initial parti-
cle size distribution and an identical pretreatment.
These diculties are avoided in the design suggested
by Gwyn (1969), shown in Fig. 10b. Here, the attrition
products are not kept inside the system; it is rather
assumed that they are elutriated. In the enlarged
diameter top section gravity separation denes the
limiting diameter of the elutriable particles. The attri-
tion rate is assumed to be given by the elutriation rate.
The steady-state elutriation rate can therefore be used
as a friability index.
It should be noted here that the quantitative results
obtained in a Gwyn-type attrition apparatus will in
general depend not only on the cut size of the gravity
separator but also on the entrainment and elutriation
conditions in the main column. Werther and Xi (1993)
compared, for example, attrition test results of the
same catalyst obtained from three dierently sized
Gwyn-type units, one with column A having 50 mm
ID and 500 mm height, one with column B having
50 m ID and 1000 mm height, and one with
column C having 100 mm ID and 1110 mm height.
Although in all columns the same stress was applied
to the material and the same gas velocity in the
enlarged top sections was adjusted, column A led to
signicantly higher results in the measured elutriated
mass. From a comparison of the particle size distribu-
tion of the elutriated material, it could afterwards be
concluded that the height of column A had not
exceeded the transport disengaging height (TDH),
and that coarser particles were allowed to reach the
gas outlet. From this observation, Werther and Xi
(1993) suggested that as a design rule for this type of
attrition test, the height between the bed surface and
the gas outlet should at least exceed the TDH.
Moreover, it should be noted that, when one com-
pares dierent materials test results, the density of the
particulate materials must be taken into account. If the
gas mass ow and the temperature are kept constant,
then a variation in the solids density will result in a
shift of the cut size and thus in the amount of material
collected as attrition product. Another point con-
cerns the particle size distribution of the bed material.
If the original solid sample is prepared by sieving e.g.
or sifting so that the smallest size is signicantly larger
than the cut size of the gravity separator (Werther and
Xi, 1993), one can be fairly sure that the elutriated
material is indeed due to attrition. On the other
hand, if the cut size of the gravity separator is located
Figure 10 Comparison of two dierent designs of a submerged jet test (a) after Forsythe and Hertwig (1949); (b) after Gwyn
(1969).
somewhere inside the original particle size distribution,
at steady state the elutriated material will inevitably
consist of both attrited debris and mother particles,
which have shrunk due to attrition to a size below
the cut size. It is clear that in this latter case the attri-
tion rate will dier from the one obtained with the
previous test. However, it must be generally noted
that the Gwyn-type test procedure is particularly suited
to measuring and investigating the attrition mode of
pure abrasion. It solely measures the produced and
elutriated nes but not the rest of the size reduction,
i.e., neither the shrinking of most of the mother
particles from which the nes are withdrawn nor the
breakage into debris larger than the cut size.
Both devices described above were developed in
order to test the friability of FCC catalysts.
Nowadays the application of these or similar tests is
a common procedure in the development of uidized
bed catalysts. Contractor et al. (1989), for example,
used a submerged-jet test to compare the attrition
resistance of newly developed VPO catalysts. In fact,
such tests can be applied to any type of uidized bed
processes. Sometimes they have to be slightly modied
to adapt them to the process under consideration. The
drilled plate may, for example, be substituted by a
porous plate if only attrition in the bed is of interest.
Even temperature and pressure can be adapted. Vaux
and Fellers (1981) investigated for example the friabil-
ity of limestone sorbent that is used for uidized bed
combustion. By surrounding a Gwyn-type test facility
with a heating system, they took thermal shock and
reaction into account.
4.3.4 Cyclone Tests
Even though cyclones are often regarded as the main
attrition source in a uidized bed system (Pell, 1990),
the number of publications on cyclone attrition test
experiments in the open literature is very small. It
may be that experiences with the performance of
cyclones are often considered proprietary. However,
this lack of information can also be explained by the
diculty of making a distinction between collection
eciency and attrition. The cyclone is a separator
designed to keep as much material as possible inside
the uidized bed system. The material that cannot be
kept inside the system, i.e., the loss or elutriation, is
usually dened as attrition. But in fact, the elutriation
rate originates from both the sifted original nes enter-
ing the cyclone (from the original feed particle size
distribution or produced by attrition in other parts of
the uidized bed system) and the nes that are actually
produced by attrition in the cyclone. Moreover, not all
attrition products will be directly elutriated from the
cyclone. Instead, a part will be collected by strands of
the material and will be transported via the solids
return line into the uidized bed. In subsequent passes
through the cyclone the accumulated attrition-pro-
duced nes will be elutriated owing to the sifting eect
of the cyclone. The same is valid when instead of the
attrition-induced elutriation rate the shift in the parti-
cle size distribution is considered: there is always a
combined eect of attrition and gas solids separation,
and above all there is the inuence of the other sources
of attrition in the solids circulation loop that feeds the
cyclone. However, the inuence of other attrition
sources can be eliminated by a test setup in which a
cyclone is operated in isolation and the material hand-
ling system does not create additional attrition stress
on the material.
Zenz (1974) and Zenz and Kelleher (1980)
considered attrition-induced changes in the particle
size distribution of FCC catalysts in isolated cyclones
of various sizes (0.1 m to 0.76 m ID). The experimental
setup simply consisted of the cyclones themselves, a
gassolids separation unit attached to the cyclone
overow consisting of gravity separator and lter
and a subsequent suction fan. In the experiments a
given batch of catalysts was sucked from a bucket
through the respective cyclone. The underow of the
cyclone was directly collected in another bucket, and
the particles leaving the cyclone through the overow
were collected in the gassolids separation unit. After
each pass the cyclones overow and underow were
mixed and once again sucked through the cyclone. The
reason for the recombination of over- and underow
after each pass was to eliminate the interaction with
the collection eciency of the cyclone and thus to
assign the measured eects to attrition only.
However, this procedure does not entirely correspond
to the operating conditions of a cyclone in a real pro-
cess, where the cyclone overow is usually lost and
thus is not able to reenter the cyclone. Therefore this
procedure might lead to results that do not entirely t
with process conditions, since attrition is also aected
by the particle size distribution and particularly by the
nes content of the material, i.e., there is some kind of
cushioning eect of the nes that reduces the entire
attrition of the material (cf. Sec. 2.1.3).
Reppenhagen and Werther (1999a) developed an
alternative method of studying the attrition mechan-
isms inside the cyclones independently of the collection
eciency. The basic idea is also to pass a batch of
particles several times through an isolated cyclone
but to consider the adjusting steady-state loss rate
instead of the particle size distribution of the material,
i.e., cyclone catch and cyclone loss are not mixed. The
experimental setup is shown in Fig. 11.
The cyclone is operated in a suction mode that
makes it possible to introduce the solids via a vibrat-
ing feeder into a tube attached to the cyclone inlet.
The separate feeding of the solids allows an indepen-
dent variation of gas inlet velocity u
c;in
and solids
loading m
c
. After each pass through the cyclone, the
collected material can be used as a feed material for
the next run by exchanging the underow and feeder
hoppers. The cyclone overow is connected to a sieve
and a lter in series. The appropriately sized sieve is
intended to collect accidentally elutriated mother par-
ticles, whereas the lter is intended to collect the
abrasion-produced ne particles. Since a part of
these abrasion-produced nes will not reach the lter
but stick to the sieve, after each pass the sieve is
cleaned. For this purpose the connection between
sieve and lter is detached and a nozzle is mounted
onto the top of the lter housing. With this vacuum-
cleaner-like arrangement the sieve is then cleaned
from its bottom side, by moving the sieve relative to
the nozzle. All particles smaller than its mesh size are
then sucked through the sieve and are collected on the
lter. Hence, the entire amount of elutriated abrasion-
produced nes can be measured by the increase in
weight of the lter.
Figure 12 gives an example of the experimental
results obtained with a previously screened spent
FCC catalyst, i.e., the smallest particles were
suciently larger than the particle size at which the
separation eciency of the cyclones becomes unity.
The cyclone loss rate per single pass is plotted against
the number of passes for given operating conditions
(gas velocity at the cyclone inlet u
c;in
, solids-to-gas
loading ratio m
c
). After each pass the cyclone loss
was determined from the increase of the lter weight,
and the loss rate was evaluated by relating the solids
mass in the loss to the respective solids input mass.
In analogy to Fig. 6, a large initial cyclone loss rate
is observed, which rapidly decreases and after 10 to 20
passes reaches a steady-state value. The high value of
the loss rate during the rst couple of passes is due to
the added eects of attrition and the sifting of nes that
could not be separated by the previous sieving. They
may have stuck to the surface of coarser particles and
are only removed inside the cyclone due to interparticle
and particlewall collisions. After 34 passes, the oper-
ating conditions were changed to a higher inlet velocity
of 20 m/s, which due to a constant solids feeding rate
led to a corresponding reduction in the solids loading.
The measurements show a slow increase of the cyclone
loss rate, which after another 15 passes leads to a steady
state again. This slow transition is probably caused by
abrasion-produced nes sticking to the larger particles
that are sent into the catch of the cyclone.
Figure 11 Experimental setup for the investigation of cyclone attrition. (After Reppenhagen and Werther, 1999a.)
Hence it can be concluded that for each operating
condition the nes concentration in the material tends
to a characteristic value of which the accumulation of
nes is balanced by the release of nes. When this
characteristic concentration is reached, the loss rate is
at steady state, i.e., it is equal to the production rate of
nes. Reppenhagen and Werther (1999a) suggested
that we take this steady-state value as a characteristic
value for both the assessment of a materials attritabil-
ity and the study of cyclone attrition mechanisms in
dependence on the various inuencing parameters. For
this purpose, they dened the cyclone attrition rate as
r
c

_ mm
loss;c;steady-state
_ mm
c;in
_ mm
a;c;fines
_ mm
c;in
6
where _ mm
loss;c;steady-state
is the cyclone loss ow at
steady state, _ mm
a;cfines
is the production rate of nes
due to attrition inside the cyclone, and _ mm
c;in
is the
mass ow into the cyclone.
It should be noted that Reppenhagen and Werther
(1999a) have exclusively concentrated on the attrition
of FCC and uidized bed catalysts, the abrasion-
produced nes of which were throughout all the
experiments entirely elutriable. In other words, inde-
pendently of the cyclone, the operating conditions, and
the catalysts material density, the produced nes were
suciently small to be in a particle size range in which
the grade eciency of the cyclone is smaller than unity.
However, for a transfer of the described test concept to
other investigations, the same attention must be paid
as in the case of the Gwyn-type tests, i.e., when com-
paring results with dierent materials the density of the
particulate material and its particle size distribution
must be taken into account. As for the latter para-
meter, Reppenhagen and Werther (1999a) have given
a simple example. They compared the test results of a
previously screened fraction with those of an
unscreened fraction of the same material. By taking
the dierences in the mean particle size into account
(cf. Sec. 5.3), they observed a distinctly higher loss rate
(approximately 25% higher) in the case of the
unscreened fraction owing to the additional loss of
suciently shrunk mother particles.
In general, it must be recognized that the greater
eort, and the large amount of material required, for
the currently available cyclone attrition test methods
will generally prevent them from being used as a pure
friability test, i.e., a test for a sole ranking of various
materials without any reference to a particular process.
However, the tests are useful for both the study of
general attrition mechanisms in cyclones and the
evaluation of a given materials susceptibility to the
particular attrition stress inside cyclones, which may
be needed as an input parameter in a cyclone attrition
model.
5 SOURCES OF ATTRITION IN A FLUIDIZED
BED SYSTEM
A rst approach to nding attrition in uid beds was
made by Zenz (1971). He pointed out that there are
various regions in a uidized bed reactor system in
which the stress acting on the bed particles and the
corresponding attrition mechanisms are quite dierent.
In the subsequent works (Zenz, 1974; Vaux and
Figure 12 Loss rate obtained with a screened spent FCC catalyst as a function of the number of passes through the cyclone
(u
c;in
: gas velocity at the cyclone inlet;
c
: solids to gas loading ratio, 90 mm ID cyclone). (After Reppenhagen and Werther,
1999a.)
Keairns, 1980), attempts have been made to identify
these individual sources. Later on, Zenz and Kelleher
(1980) revealed the necessity of studying each of these
sources in isolation in order to get detailed information
about the particular attrition mechanisms. They sug-
gested that we investigate the vicinity of a multihole
gas distributor, the bubbling uidized bed, the
cyclones, the conveying lines, and the solids feeding
devices as separate regions where the attrition
mechanisms are dierent. However, usually the gas
distributor, the bubbling bed, and the cyclones are
regarded as the most relevant attrition sources (Pell,
1990). The present section is therefore also conned
to these sources.
The investigation of the gas distributor and the bub-
bling bed attrition were both almost exclusively carried
out in various derivations from the Gwyn-type test
facility. Hence the material loss is the only attrition
result considered, and the derived model approaches
thus exclusively deal with the production of elutriable
nes. There is consequently a lack of direct informa-
tion on the role of attrition in the adjustment of the
bed particle size distribution.
5.1 Grid Jet as a Source of Attrition
Gas distributors of uidized beds are often designed as
perforated or nozzle plates. Since a minimum pressure
drop is required to obtain a uniform gas distribution
over the beds cross-sectional area, the open surface
area is rather small, and the gas jets issuing from the
distributor holes are at high velocity. Particles are
entrained by these jets, accelerated to high velocities,
and impacted onto the uidized bed suspension at the
end of the jets, resulting in particle degradation similar
to that in jet grinding processes (Kutyavina and
Baskakov, 1972).
One peculiarity of the jet-induced attrition is that
the jets aect only a limited bed volume above the
distributor, which is dened by the jet length. Hence
as soon as the jets are fully submerged, their contri-
bution to attrition remains constant, with further
increasing bed height. Figure 13 shows some experi-
mental results obtained by Werther and Xi (1993).
The jet penetration length can be estimated by var-
ious correlations, such as those given by Zenz (1968),
Merry (1975), Yates et al. (1986), or Blake et al. (1990).
However, in most commercial uidized bed processes
the bed is much higher than the jet penetration length.
Hence jet-induced attrition cannot be investigated in
isolation, because there is always some additional attri-
tion of the bubbling bed. For this reason many authors
(Blinichev et al., 1968; Kutyavina and Baskakov, 1972;
Arastoopour and Chen, 1983; Contractor et al., 1989)
considered the overall attrition rate resulting from
both attrition sources. However, in order to get direct
insights into the mechanisms of jet attrition it is neces-
sary to separate the jet contribution from the measured
overall attrition rate. This can be done in two dierent
ways.
Seville et al. (1992) and Ghadiri et al. (1992b) mea-
sured the attrition rates at various static bed heights.
Assuming a linear increase of the attrition rate with the
bed height above the jetting region, they extrapolated
the measured attrition rates back to the jet length cal-
culated from one of the available correlations. The
extrapolated value was then taken as the attrition
rate of the jetting region. Unfortunately, no compre-
hensive study of the attrition mechanisms were carried
out on the basis of this strategy. Only a few parameter
eects were tested, which did not lead to an unambig-
uous description of the jet attrition mechanisms.
An alternative strategy to study the jet-induced
attrition is suggested by Werther and Xi (1993). They
used a Gwyn-type test apparatus with a particular dis-
tributor design in which a separately fed nozzle was
integrated into a porous plate (Fig. 14). At rst the
bed was only aerated via the porous plate. In this
way the contribution of the bubbling bed attrition
could be measured without any additional attrition
sources. In a second step, the apparatus was operated
with a chosen jet gas velocity. To maintain the cut size
of the gravity separator above the bed at some
prescribed level, they kept the supercial gas velocity
constant by supplying auxiliary air through the porous
plate. The resulting loss rate of the system originates
Figure 13 Inuence of the static bed height on jet attrition
of spent FCC catalyst in a submerged jet test facility
(D
t
0:05 m, u
or
100 m=s, d
or
2 mm). (Werther and Xi,
1993.)
from both the jets and the bubbling bed. And the jet
attrition rate can be now calculated by subtracting the
bubble attrition rate measured before from the overall
attrition rate. This is a certain oversimplication, since
it assumes that the bubble attrition will always be the
same regardless of the ratio of jet air mass ow to
auxiliary mass ow. However, this method holds fairly
well for low jet gas velocities, where the contribution of
the jet gas ow to the total gas ow is rather small. For
higher jet gas velocities, the jet attrition is so high that
the contribution of bubble attrition may be neglected.
Using the above-described experimental setup,
Werther and Xi (1993) carried out a comprehensive
experimental program to study grid-jet attrition.
They found that at steady-state attrition conditions
(cf. Fig. 6), the jet-induced attrition exclusively occurs
in the mode of pure abrasion. On the basis of this
result they suggested and validated a model that con-
siders the energy utilization of this abrasion process by
relating the surface energy created to the kinetic energy
that has been spent to produce this surface area.
Assuming that there is no interaction between the indi-
vidual jets of a given distributor, they derived the fol-
lowing relationship for the grid-jet-induced generation
of nes:
_ mm
abr:;fines;j
n
or
K
j
r
f
d
2
or
u
3
or
7
where n
or
is the number of orices in the distributor, r
f
is the density of the jet gas, d
or
is the diameter of the
orices, and u
or
is the jet gas velocity inside an orice.
K
j
is a constant that characterizes the solids suscept-
ibility to abrasion. It turns out that this parameter can
be expressed as the product of the surface mean
diameter d
pb
of the bed solids and a particle-size-
independent material specic constant C
j
, i.e., the
solids susceptibility to abrasion increases linearly
with its surface mean diameter
K
j
C
j
d
pb
8
A substitution of Eq. (8) into Eq. (7) yields
_ mm
abr:;fines;j
n
or
C
j
d
pb
r
f
d
2
or
u
3
or
9
With regard to the distributor design, Werther and Xi
(1993) have shown that the attrition eect of a hori-
zontal jet equals that of an upward facing jet, i.e., the
same equation and an identical attrition rate constant
can be used for the modeling. In contrast to this,
downward issuing jets were found to generate signi-
cantly higher attrition rates because of the impact of
entrained particles onto the xed bed of particles at the
nose of the jet, i.e., there is probably a dierent attri-
tion mechanism involved requiring a dierent attrition
model or at least a dierent attrition rate constant for a
given material.
Another issue that should be considered with
respect to distributor design is the inuence of its
open surface area A
o
. For such a consideration,
Eq. (9) can be complemented by the following depen-
dencies:
A
o
/ n
or
d
2
or
10
u
or

_
VV
A
o
11
leading to the expression
_ mm
abr:;fines;j
/
_
VV
3
A
2
o
12
According to Eq. (12), for a given gas volumetric ow
rate, the decisive quantity for the distributor attrition
rate is the open surface area A
o
, which suggests that it
is unimportant with respect to attrition whether A
o
originates from a few large or from many small ori-
ces. But this conclusion is somehow in contrast to the
ndings of Boerejn et al. (2000). They observed a
change in the attrition stress with increasing orice-
to-particle size ratio from pure shear to impact stress
and thus a change in the predominant attrition mode
from pure abrasion to the fracture of microspheroids.
Figure 14 Gwyn-type test apparatus suggested by Werther
and Xi (1993) to investigate the jet-induced attrition sepa-
rately from the bubble-induced attrition.
5.2 Bubble-Induced Attrition
Bubble-induced attrition originates from low-velocity
interparticle collisions. Since the bubble rise velocity is
of the order of 1 m/s, the energy is generally not high
enough to shatter the bed particles into fragments. For
this reason, most laboratory experiments have shown
the bubbles to be a minor source of attrition. However,
in a deep uidized bed, with several meters of height,
the contribution of bubble-induced attrition may be a
signicant factor.
Again, as in the case of jet attrition, attention must
be paid to the isolation of that part of the attrition that
is due to bubbles. There are basically two ways to do
this. One is to use a porous plate distributor in order to
avoid any grid jets. The other is the above described
procedure suggested by Seville et al. (1992) and
Ghadiri et al. (1992b): the measurement of the produc-
tion rate of nes at dierent values of the static bed
height permits us to eliminate the grid jet eects (cf.
Sec. 5.1).
Even though most of the attrition tests presented in
the literature deal with bubble-induced attrition, the
respective attrition mechanisms are not quite clear
yet. There are various theoretical and empirical
approaches that can in accordance with Eq. (3) be
summarized in the following denition of a bubble-
induced steady-state attrition rate
:
r
b

_ mm
loss;b;steady-state
m
b
_ mm
attr:;fines;b
m
b
K
b
m
n
b
u u
min

z
13
where _ mm
loss;b;steady-state
is the bubble-attrition-induced
loss rate from a system under steady-state condi-
tions, which can be assumed to be identical to the
bubble-attrition-induced production rate of elutriable
material, _ mm
attr:;fines;b
at this state. m
b
is the bed mass,
the exponent of which varied between 0 and 1. In
the case of n 0, it is implied that the attrition
stress does not change with the bed height; there is
simply a linear increase with the amount of treated
material. In contrast to this, an exponent n > 0
implies an increase of the attrition stress with the
bed height. The velocity u
min
is regarded as a thresh-
old velocity above which the bubble-induced attrition
occurs. Its value varied from u
min
u
mf
to
u
min
u
mf
. The exponent z is in the vast majority
of studies set to a value of one, but in some cases it
is even set to a value of 3, i.e., a linear dependence
of the attrition extent on the input of kinetic energy
with the uidizing gas is assumed. In the following,
the various theoretical and experimental ndings are
briey summarized.
Merrick and Highley (1974) have modeled bubble-
induced attrition as a comminution process. According
to Rittingers law of size reduction (cf. Perry, 1973),
the rate of creation by abrasion of new surface area
S=t is proportional to the rate of energy input
E=t:
S
t
/
E
t
14
Since the size distribution of the nes produced by
abrasion is approximately constant, the rate of produc-
tion of new surface can be taken to be proportional to
the mass rate of production of nes,
S
t
/ _ mm
attr:;fines;b
r
b
m
b
15
The total rate of input of energy to the uidized bed is
given by the product of the volumetric ow rate of gas
(u A
t
) and the pressure drop, which may be expressed
as weight of the bed divided by the beds cross-sec-
tional area A
t
. However, only part of the input energy
is available for bubble formation and thus for commi-
nution. The input energy (u
mf
m
b
g) is required per
unit time for keeping the particles in suspension. That
part of the rate of input of energy that remains for
bubble formation, and thus for attrition, is then
given by u u
mf
m
b
g. Insertion into Eq. (15)
yields
r
b

_ mm
attr:;fines;b
m
b
K
b
u u
mf
16
where K
b
is an abrasion rate constant. With a similar
approach, Ray et al. (1987a) arrived at the same result.
This approach supports the idea of a linear dependence
of the bubble-induced attrition rate on the bed mass,
i.e., the attrition stress does not change with the bed
height, and n 0 in Eq. (13). In contrast to this Vaux
(1978), Ulerich et al. (1980), and Vaux and Schruben
(1983) proposed a mechanical model based on the
kinetic energy of particles agitated by the bubble
motion, where they took the increase of the bubble
velocity with bed height into account. The authors
conclude that the bubble-induced attrition rate is pro-
portional to the product of excess gas velocity and bed
height or bed mass, respectively:
r
b

_ mm
attr:;fines;b
m
b
K
0
b
m
b
u u
mf
17
This inconsistency with respect to the inuence on the
bed height is also reected by the available experimen-
tal data on bubble-induced attrition: Merrick and
Highley (1974) and Pis et al. (1991) have found the
attrition rate r
b
to be independent of bed height.
Kono (1981) found this value to be proportional to
the static bed height with the exponent 0.78, and
Ulerich et al. (1980) and Xi (1993) found r
b
to be
proportional to the bed height.
Obviously the role of bed height is not yet fully
understood. Ray et al. (1987a) have explained these
discrepancies by a consideration of bubble growth
with height above the distributor. They argue that as
long as the bubble size increases with height, the e-
ciency of the transformation of kinetic energy to free
surface energy might increase, thus leading to an attri-
tion rate that increases with bed height. However, as
the bed height reaches the limits of slugging in small-
diameter columns or maximum attainable bubble size
in a large diameter bed, an extra bed height will not
vary the conditions of bubbling and thus will result in
r
b
becoming independent of bed height. However,
further experiments are certainly needed in this area.
Although the dependence on the gas velocity
appears to be reasonably explained by the above
described model approaches, the experimental data
available in the literature are giving even in this respect
an inconsistent picture: Merrick and Highley (1974),
Arena et al. (1983), and Pis et al. (1991) also found
the linear dependence on the excess gas velocity
(u u
mf
) to be valid. As an example, Fig. 15 shows
the results of Pis et al. (1991), which were obtained
in a uidized bed column of 0.14 m in diameter. The
distributor had orices of 1 mm in diameter on a 5 mm
square pitch. Unfortunately, no distinction was made
between the measured attrition rate and the inuence
of the grid jets. However, their inuence might be neg-
ligible in the present case due to the relatively small jet
velocity.
Ray et al. (1987a) obtained a fairly dierent result
as they considered the attrition of narrow fractions of
limestone with particle sizes between 1.09 and 0.77 mm
in a 0.1 m diameter bed, which was equipped with a
porous plate (Fig. 16). Similar results were obtained by
Xi (1993), who investigated the attrition of ne catalyst
particles with a minimum uidization velocity u
mf
of
0.002 m/s (Fig. 17). As is obvious from Figs 16 and 17,
the attrition rate extrapolates to zero at a uidizing
velocity u
min
which is signicantly larger than u
mf
.
This means that a minimum kinetic energy or a mini-
mum extent of bubbling is necessary to cause attrition
in these cases.
More recently, Werther and Reppenhagen (1999)
correlated their bubbling bed attrition test results
obtained under pure abrasion conditions in a 200 mm
ID Gwyn-type test plant even to the excess gas velocity
raised to a power of 3 (Fig. 18), which indicates that
the bubbling bed attrition is linearly increasing with
the excess kinetic energy supplied to the system:
r
b

_ mm
loss;b;steady-state
m
b
_ mm
abr:;fines;b
m
b
K
b
u u
mf
3
18
5.3 Cyclones as Attrition Sources
For the reasons explained above there is only a limited
amount of work published in the open literature on
cyclone attrition. In fact there are results from only
Figure 15 Variation of the steady-state attrition-induced
loss rate of coal ash (0.2 to 0.315 mm) with the excess gas
velocity (u-u
mf
). (Pis et al., 1991.)
loss ow of limestone fractions with (u-u
mf
). (Ray et al.,
1987a.)
two groups (Zenz, 1974; Zenz and Kelleher, 1980;
Reppenhagen and Werther, 1998, 1999a; Werther
and Reppenhagen, 1999). The experimental procedures
of both groups have already been explained in Sec.
4.3.4.
As a result of their experiments with FCC catalyst
material in isolated cyclones, Zenz (1974) and Zenz
and Kelleher (1980) observed even after a few passes
a signicant change in the particle size distribution
(cf. Fig. 7). The content of coarser particles was
decreased in comparison to their initial one, whereas
the content of smaller particles was increased. From
many experiments of this kind, Zenz and Kelleher
(1980) derived design correlations for the estimation
of cyclone attrition for uidized bed systems.
Unfortunately the details of these estimations
remained proprietary.
Hence the only detailed reports yet available in the
open literature on attrition in cyclones are those of
Reppenhagen and Werther (1998, 1999a), who exclu-
sively concentrated on the attrition of uidized bed
catalyst particles. Assuming that the catalysts attrit in
a cyclone exclusively in the mode of pure abrasion,
they suggested a model that regards cyclone attrition
under steady-state conditions as a comminution pro-
cess: it considers the eciency of such a process by
relating the surface energy created by comminution
to the kinetic energy, which has been spent to create
this new surface area. According to the model, the
above dened cyclone attrition rate r
c
[cf. Eq. (6)]
can be calculated from
r
c

_ mm
loss;c;steady-state
_ mm
c;in
_ mm
abr:;fines;c
_ mm
c;in
K
c
m
n
c
u
2
c;in
19
where K
c
is the cyclone attrition rate constant, which
summarizes all particle properties that are relevant to
the abrasion process, and n is an exponent assuming a
power law dependence of the attrition rate on the
solids-to-gas loading ratio m
c
at the entrance of the
cyclone. According to the procedure described in Sec.
4.3.4, the authors carried out comprehensive attrition
tests with a spent FCC catalyst and a 90 mm ID
cyclone in order to scrutinize this model approach. In
these experiments the gas velocity u
c;in
at the entrance
of the cyclone was varied between 8 and 24 m/s with
the solids loading m
c
ranging from 0.05 to 1. In Fig. 19
the measured cyclone attrition rates r
c
obtained under
steady-state conditions are plotted against the cyclone
inlet velocity on a double logarithmic grid.
The relationship between r
c
and u
c;in
predicted by
Eq. (19) is seen to be conrmed over a wide range,
which is indicated by the solid lines. However, as
indicated by the dashed lines, there are some distinct
deviations when the gas velocity exceeds a certain
threshold. The threshold on its part seems to be depen-
dent on the solids loading, i.e., the smaller the solids
loading the smaller the gas velocity from which the
deviation occurs. Reppenhagen and Werther (1999a)
explain this deviation by the occurrence of another
attrition mechanism in addition to abrasion, namely
the so-called surface fragmentation of the catalyst par-
ticles, which results from the combination of increasing
kinetic energy at increasing velocities and increasing
loss ow per unit bed mass of a catalyst with supercial
gas velocity (AVN 802, u
mf
0:002 m=s). (Xi, 1993.)
Figure 18 Bubbling bed attrition test results obtained by
Werther and Reppenhagen (1999) in a 200 mm diameter
Gwyn-type test plant under pure abrasion conditions with
fresh FCC catalyst.
single-particle/wall interactions with decreasing solids
loading. However, the authors could clearly identify
the threshold for the onset of this additional attrition
mode and thus the range of applicability of Eq. (19) by
means of scanning electron microscope images (cf. Sec.
2.1.5, especially Figs. 4 and 5). Furthermore, it should
be noted here that the operating conditions leading
to fragmentation are rarely encountered in industrial
uidized bed applications since high cyclone inlet
velocities are normally avoided in order to keep the
cyclone pressure drop at a reasonable level.
After identication of the thresholds, Reppenhagen
and Werther (1999a) derived the value of the exponent
n in Eq. (19) from all measurements taken under con-
ditions of pure abrasion (straight lines drawn in
Fig. 19) to n 0:5. They explained the negative
value of n by some kind of cushioning eect, i.e.,
the chance for a given particle to impact on the wall
decreases with increasing solids concentration in the
ow. With n 0:5, the model equation can now
nally be written as
r
c
K
c

u
2
c;in
m
c
p 20
In further experiments, where the additional attrition
mode of surface fragmentation was avoided by keeping
the inlet gas velocity below 20 m/s and the gas solids
loading above 0.1, this model equation could be vali-
dated for various types of catalyst and dierently
designed and sized cyclones with tangential inlet. As
an example, the test results for ve dierent catalysts
are shown in Fig. 20, where the cyclone attrition rate is
plotted as a function of u
2
c;in
=m
0:5
.
In complete agreement with Eq. (20), the attrition
rate for each material is proportional to u
2
c;in
=m
0:5
.
However, the absolute value of the attrition rate
depends on the properties of the individual material,
which are summarized in the rate constant K
c
.
Moreover, the authors found that this constant
depends not only on the type of material but also on
the surface mean diameter d
pc
of the solids that enter
the cyclone. Hence they suggested that we subdivide
the attrition rate constant K
c
into
K
c
C
c
d
p;c
21
where a particle-size-independent attrition rate
constant C
c
characterizes now solely the material prop-
erties as, for example, strength, shape, and surface
roughness. Substituting Eq. (21) into Eq. (20) nally
leads to
r
c
C
c
d
pc

u
2
c;in
m
c
p 22
6 ATTRITION IN THE OVERALL FLUIDIZED
BED SYSTEM
In conclusion from the sections above, two main dis-
tinctions must be made when attrition in an overall
uidized bed system is considered: Primarily it must
be recognized that there are several attrition sources
in a uidized bed system with distinctly dierent attri-
tion mechanisms, which must be described separately.
Figure 19 Inuence of the cyclone inlet velocity u
c;in
on the
cyclone attrition rate r
c
at dierent solids loadings measured
by Reppenhagen and Werther (1999a) in a 90 mm ID
cyclone. Material: spent FCC catalyst; u
c;in
: cyclone inlet
velocity;
c
: solids-to-gas loading ratio.
Figure 20 Cyclone attrition test results obtained with var-
ious types of catalysts in a 90 mm ID cyclone. (Reppenhagen
and Werther, 1999a.)
Furthermore, a distinction must be made between the
various attrition modes that may even occur at a single
attrition source depending on the energy level applied.
Although this concept is widely known, up to now it
has rather seldom been applied. Even most of those
publications that are just focusing on a Gwyn-type
test facility do not distinguish between grid-jet attrition
and bubble-induced attrition. This may be the main
reason for the strong discrepancies in the previously
published literature where experimental ndings are
compared, which are obtained from dierent systems
and dierent solids under quite dierent operating
conditions. Ghadiri et al. (1992a) gave a simple exam-
ple of such a discrepancy by comparing the published
ndings for the inuence of the supercial gas velocity
on the attrition-induced loss ow from uidized bed
systems: Some authors (Seville et al., 1992) relate
their experimental results to u u
mf
, while others
assume the attrition rate to be proportional to u
n
,
where the exponent varies from n 1 (Patel et
al., 1986) to n 5:8 (Blinichev et al., 1968).
However, more recently the present authors
(Werther and Reppenhagen, 1999; Reppenhagen and
Werther, 1999b) have demonstrated that a strict
observance of the above concept dramatically helps
to overcome these discrepancies. They considered the
attrition-induced loss ow of catalyst material from the
cold model uidized bed unit with external solids recir-
culation that is schematically shown in Fig. 21. The
cyclone overow is connected to a lter that collects
the elutriated material. Under conditions of steady-
state attrition, the loss rate was then obtained from
the increase in the lter weight per unit time. As bed
material a fresh FCC catalyst material was used. In
order to ensure that no original particle but only
attrition-produced debris is collected on the lter,
the catalyst was previously sieved to remove its frac-
tion of elutriable nes.
The results of two dierent test series are shown in
Fig. 22, where they are plotted as a function of the
supercial gas velocity in a double logarithmic grid.
The rst series of measurements was taken by using a
porous plate as a gas distributor, which was substi-
tuted by a perforated plate (640 orices of 0.7 mm
ID) in the second series. For both series a strong sen-
sitivity against the gas velocity can be observed. In the
case of the porous plate distributor, a 10% increase in
the supercial gas velocity leads almost to a doubling
of the loss ow. In case of the perforated plate, the
sensitivity against the gas velocity is lower, but the
absolute values of the loss ows are higher.
In order to describe such parameter dependencies,
the conventional approach would be simply to corre-
late the measured overall loss ows in the form of a
power law. For the data shown, this would lead to a
dependence on the supercial gas velocity raised to a
power of 4.2 or even 7, respectively. A physically
sound explanation for these exponents is certainly dif-
cult, and it is obvious that such a correlation obtained
with the one system design is not directly transferable
to the other even though only the distributor was
exchanged, which reveals that such a model approach
Figure 21 Experimental setup for the investigation of the
attrition-induced loss ow of fresh FCC catalyst from a ui-
dized bed system. (After Werther and Reppenhagen, 1999;
Reppenhagen and Werther, 1999b.)
Figure 22 Measured loss ows from the uidized bed sys-
tem shown in Fig. 21 operated with previously screened fresh
FCC catalyst. (After Werther and Reppenhagen, 1999;
results in exactly those discrepancies described by
Ghadiri et al. (1992a). In contrast to such a simple
correlation of the overall loss ow, Werther and
Reppenhagen (1999) have chosen to consider the rele-
vant mechanisms with respect to the attrition-induced
loss ow of the system individually. In the following it
will be briey described:
In a rst step, they identied three regions as main
attrition sources, namely the grid jets, the bubbling bed
itself, and the cyclone section. Accordingly, they
described the quasi-stationary loss rate of the overall
system as the sum of their individual contributions:
Since the stress conditions throughout all the experi-
ments could be regarded to be under pure abrasion
conditions, the model equations Eqs. (9), (18), and
(22) could be applied to describe the generation of
elutriable nes; the following reasoning was made:
Since the abrasion-produced nes are rather
small, i.e., typically smaller than about 1 to 3
microns, the respective grade eciency of the
cyclones is distinctly smaller than unity.
Therefore it can be assumed that the nes are
immediately lost after their production, even if
they are produced in the jetting region or in the
bubbling bed. In contrast to this, it can be
assumed that after an initial phase where the ori-
ginal elutriable material is sifted o, the mother
particles are entirely kept inside the system. Due
to the mode of pure abrasion the change in their
particle size is negligible, and so is their contribu-
tion to the loss rate. Hence some kind of steady
state can be assumed where the production of
nes, which originate from the almost unchanged
mother particles, is balanced by the loss rate of
nes.
Hence we can write:
_ mm
loss;tot;steady-state
_ mm
loss;j;steady-state
_ mm
loss;b;steady-state
_ mm
loss;c;steady-state
_ mm
abr:;fines;j
_ mm
abr:;fines;b
_ mm
abr:;fines;c
n
or
C
j
d
pb
r
f
d
2
or
u
3
or
K
b
m
b
u u
mf
3
_ mm
c;in
C
c
d
pc

u
2
c;in
m
c
p
23
However, for a consistent description of the overall
process it is reasonable to substitute the orice velocity
u
or
, the solids loading m
c
and the cyclone inlet velocity
u
c;in
by
u
or
u
D
2
t
n
or
d
2
or
24
with D
t
being the diameter of the uidized bed column,
m
c

_ mm
c;in
r
f
u
c;in
A
c;in
25
and
u
c;in
u
A
t
A
c;in
26
which leads to the following completed model equation
for the overall process:
_ mm
loss;tot
C
j
d
pb
r
f

D
6
t
d
4
or
n
2
or
u
3
K
b
m
b
u u
mf
3
C
c
d
pc

_ mm
c;in
_

r
f
p

A
2:5
t
A
2
c;in
u
2:5
27
From this model equation (27), it is obvious that the
attrition eects in the overall system depend not only
on the prevailing attrition mechanisms but also on the
solids transport in the freeboard which determines the
solids ow into the cyclone and thus the amount and
the particle size distribution of the material that is sub-
jected to cyclone attrition. Hence in a second step of
the overall modeling, these transport eects must be
taken into account, which has in a rst approach
been simply done by measuring the necessary data.
Figure 23 shows a comparison of the experimental
data depicted in Fig. 22 with the calculation from the
model equation (27). The required attrition rate
constants C
j
, K
b
, and C
c
that describe the materials
susceptibility to attrit in the respective regions have
been determined by the corresponding attrition tests
as described in Sec. 4.3. C
j
has been determined from
exactly that Gwyn-type test facility that is shown in
Fig. 14 and was set to zero in the case of the porous
plate distributor; K
b
has been measured in a 200 mm
ID Gwyn-type test apparatus, and C
c
has been
determined from exactly that cyclone attrition-test
procedure that is described in Sec. 4.3.4 using the
equipment sketched in Fig. 11. The parameters _ mm
c;in
and d
pc
were measured in the apparatus sketched in
Fig. 21 under the assumption that _ mm
c;in
may be
approximated by the solids circulation rate measured
in the return line. Figure 23 shows a very good agree-
ment between the measured and the calculated overall
loss ows even though two dierent process designs are
considered.
However, as mentioned above, Eq. (27) requires
information on the solids transport eects. Hence for
an a priori modeling, the implementation of an
entrainment/elutriation model is required. Werther
and Reppenhagen (1999) have given an example of
such a modeling approach: the bed particle size
distribution and thus d
pb
were simply assumed to be
constant, and m
c;in
and d
pc
were substituted by
_ mm
c;in
A
t
G
s
A
t

i
G
si
A
t

i
w
i
K
i
28
d
pc
u
G
s
i
G
si
x
i
G
s
i
w
i
K
i
x
i
29
where G
s
is the entire solids elutriation rate from the
bed, x
i
is the mean diameter of the size interval i, G
si
is
the fractional elutriation rate for the size interval i, w
i
is the weight fraction of the size interval i in the bed
material, and K
i
is the respective elutriation rate
constant, which was calculated from a correlation
suggested by Tasirin and Geldart (1998):
K
i
14:5 r
f
u
2:5
exp 5:4
u
ti
u
_ _
30
where u
ti
is the terminal settling velocity of particles in
the size interval i. As a result, Eq. (27) can be written
as
_ mm
loss;tot
C
j
d
pb
r
f

D
6
t
d
4
or
n
2
or
u
3
K
b
m
b
u u
mf
3
3:81 C
c
d
pc
u r
f

A
3
t
A
2
c;in
i
w
i
exp 5:4
u
ti
u
_ _
_
u
3:75
31
On the basis of Eq. (31), the total loss rate and the
contributions of the individual sources of a given sys-
tem can be a priori calculated in dependence on the
supercial gas velocity. Using the same values for the
materials attrition rate constants and the system
design as in the above considerations, this is demon-
strated in Fig. 24.
Obviously, there is a strong sensitivity of cyclone
attrition inside a uidized bed system against the
supercial gas velocity. It is signicantly higher than
the sensitivities of jet- and bubble-induced attrition,
respectively. Furthermore, it is obvious that owing to
these dierent dependencies, the role of the main attri-
tion source changes with the gas velocity. In the lower
velocity rangein this particular example below
0.55 m/sthe gas distributor is the main attrition
source, whereas at higher velocities cyclone attrition
is dominant.
Figure 23 Comparison of measured and calculated loss
ows for the two dierent system designs. The input data
for the calculation that are aected by entrainment had
been measured. (After Werther and Reppenhagen, 1999;
Figure 24 Inuence of the supercial gas velocity on the
extent of attrition in the individual regions and in the overall
uidized bed system (uidized bed facility from Fig. 21,
screened FCC catalyst, bed mass 5.5 kg). (After Werther
and Reppenhagen, 1999.)
From the above-summarized work of Werther and
Reppenhagen (1999), it must be concluded that attri-
tion phenomena in an overall uidized bed system can
be only modeled when the relevant attrition sources are
identied and separately described. Each individual
attrition source model should take into account the
specic design of the system, the operating conditions,
and the relevant material properties of the solids. The
latter can be obtained from attrition tests using for
each source a particular designed experimental setup
that simulates the relevant process stress. Finally, the
solids transport and segregation eects must be taken
into account when combining the individual model
results again to a description of the overall process.
6.1 Continuous Operation
Up to now, attrition has been considered in batch
processes only. But many industrial uidized bed
reactors are operated in a continuous mode. This
means that the cyclone loss as well as the reacted
material (e.g., in all noncatalytic gassolid reactions)
is compensated for by the addition of freshly fed
material. Furthermore, in some processes there is an
additional withdrawal of bed material, which is also
compensated for by fresh solids in order to keep the
bed inventory at a required level. As a consequence,
there is a residence time distribution of the solids, and
the time dependence of attrition has to be taken into
account. According to Sec. 2.2, there will be both
high initial attrition of freshly fed material and
steady-state attrition of old particles. Zenz and
coworkers (1971, 1972, 1980) were among the rst
to suggest calculation procedures for the content of
nes, which is attained at equilibrium as a result of
attrition and the addition of new catalyst in a ui-
dized bed system. Levenspiel et al. (1969) and Kunii
and Levenspiel (1969) suggested the consideration of
particle balances in the system. Newby et al. (1983)
proposed a simplied mass balance for the entire sys-
tem and distinguished between continuous and instan-
taneous attrition, which was connected to the feed
rate. Fuertes et al. (1991) coupled a description of
time dependence derived from batch processes with
a residence time distribution function of a continuous
stirred tank reactor. Ray et al. (1987b) and Werner et
al. (1995) presented population balance models to
describe process attrition. In all these models, how-
ever, the attrition mechanism is again treated in a
rather supercial way without making a distinction
between the dierent mechanisms prevailing in the
respective parts of the system.
6.2 Changes in the Bed Particle Size Distribution
As mentioned in the introduction, the eect of attrition
on the particle size distribution is quite often as rele-
vant as the attrition-induced loss is. The reason is quite
obvious: it is the strong dependence of the process
performance on the bed particle size distribution. In
the chemical industry, for example, the content of
nes, i.e., the mass of particles below 44 microns, has
often been observed to have a strong eect on the
uidized bed reactor performance. de Vries et
al. (1972) reported an increase in the conversion of
gaseous hydrogen chloride in the Shell chlorine process
from 91 to 95.7% with an increase of the nes content
in the bed material from 7 to 20%. The same eect was
observed by Pell and Jordan (1988) with respect to the
propylene conversion during the synthesis of acryloni-
trile. They reported on an increase of the conversion
from 94.6 to 99.2% as the nes content was changed
from 23 to 44%.
Dealing with FCC processes, Zenz (1971) has given
the following statement:
Ideally it should be possible to predict simply
from the fresh feed catalyst size analysis and a
specic reactor and cyclone geometry how the
bed analysis and reactor losses will change with
time and how these will converge to an equili-
brium. If attrition were not a signicant factor
it is obvious that the addition of fresh make-up
catalyst coarser than the losses would cause the
bed size distribution to become continually coar-
ser until theoretically losses would be reduced to
zero.
Despite its practical relevance the number of pub-
lications dealing with this topic is rather small. In fact,
the few publications that focus on an overall system
with solids recirculation (such as Zenz, 1971, 1974;
Ewell et al., 1981; Gierse, 1991) describe attrition
only by an undened term of particle degradation, to
which neither a particular mechanism nor a particular
region is assigned. However, more recently the present
authors (Reppenhagen and Werther, 2001) suggested a
particle population balance for a uidized bed system
that allows a description of the fate of the individual
particles in terms of attrition and transport eects. It
thus provides a description of the dynamic adjustment
in the steady-state particle size distribution and the
solids loss rate of a given system. In the following
this work is briey summarized.
As in the authors previous work (cf., e.g., Sec. 6)
they considered a process where the attrition mode is
pure abrasion. Focusing on a discretized particle size
distribution they sketched the particular eect of this
attrition mode in Fig. 25.
From each size interval K
i
there is a mass ow _ mm
loss;i
of nes into the smallest size interval K
1
, which is the
fraction of the abrasion-produced nes. It can be
assumed that all particles of the size interval K
i
are
involved in the generation of these nes. As a conse-
quence, all particles shrink, and some of them become
smaller than the lower boundary of their size interval.
They must thus be assigned to the smaller size interval
K
i1
. This mass transfer between neighboring intervals
is denoted as a mass ow _ mm
i;i1
. On the other hand, the
rest of the material remains in its original size interval,
even though the particles are also reduced in size.
These phenomena of mass transfer can be summar-
ized in a set of mass balances for the individual size
fractions,
dm
i
dt
_ mm
loss;i
_ mm
i;i1
_ mm
i1i
i 2 n 1
32
when n is the number of size intervals. An exception is
made for both the interval of the nest particles K
1
and
the interval of the coarsest particles K
n
. The interval K
1
receives all abrasion-produced nes originating from
the other size intervals and the shrunk particles from
the neighboring size interval K
2
. But there is no mate-
rial loss due to a further attrition of the particles within
the interval. Even if there were a further particle degra-
dation, the attrition products would remain within the
size interval K
1
. On the other hand, the particles within
the size interval K
n
undergo abrasion. This results in
both the loss of ne material and the loss of shrunk
particles. But there is no coarser size fraction from
which shrunk particles could be received.
dm
1
dt

n
i2
_ mm
loss;i
_ mm
2;1
33
dm
n
dt
_ mm
loss;n
_ mm
n;n1
34
In order to solve the set of mass balances, Eqs. (32),
(33), and (34), both the fractional loss of nes _ mm
loss;i
and the resulting mass transfer between the neighbor-
ing size intervals _ mm
i;i1
must be known for each size
interval. Hence the inuence of the individual particle
sizes must be taken into account. Based on the descrip-
tion of the abrasion-induced loss ow from an overall
system that is summarized in Eq. (27) this can be done
as follows.
Assuming that the particle size dependent rate con-
stant K
b
for bubble-induced attrition canin analogy
to the other sourcesbe written as the product of a
particle size independent constant C
b
and the surface
mean diameter of the bed material d
pb
:
_ mm
loss
C
d
p

b
;
jd
p
d
pb
;
j
;
c
d
pc
; d
pj
C
C
b
; C
j
; C
c
with d
pj
d
pb
35
with
b
m
b
u u
mf
3
;
j
r
f
d
2
or
u
3
or
;
c

_ mm
c;in
r
f
A
c;in
_
u
2:5
c;in
representing the respective inu-
ences of geometry and operating parameters. With the
denition of the surface mean diameter, Eq. (35) can
be written as
_ mm
loss

_ mm
loss
j
C
xx
i
Q
2;i
36
Figure 25 Abrasion-induced mass transfer between particle size intervals. (From Reppenhagen and Werther, 2001.)
Q
2
is the cumulative surface area distribution of the
particle sizes. From this the material loss of a size
fraction can be deduced as
_ mm
loss
j
C
xx
i
Q
2;i
37
According to Eq. (37) for a given size interval the
material loss due to nes generation increases with its
geometric mean particle size xx
i
and its fraction of the
entire particle surface, i.e., Q
2;i
.
The mass transfer between the neighboring size
intervals, _ mm
i;i1
, can be derived from the natural cou-
pling with the fractional nes production: the mass of
produced nes originating per unit time from a size
interval K
i
corresponds to the material loss of the
initial parent particles of this fraction. Provided that
the loss is evenly contributed by all particles of this size
interval, and moreover assuming that the particles are
spherical, the mass loss per unit time of a single particle
can be derived by
_ mm
loss;i;p

_ mm
loss;i
N
p;i
_ mm
loss;i

r
s;a
p xx
3
i
Q
3;i
m
tot
6
38
with N
p;i
being the number of particles in the size inter-
val K
i
. r
s;a
is the apparent density of the solids, and
m
tot
is the total mass of solids. Provided that _ mm
loss;i
is
constant in the time interval t, the mass balance for a
single particle can be written as
m
i;p;t
m
i;p;tt
_ mm
loss;i;p
t 39
This can be transferred to
x x
i;t
3
6
p

p
6
xx
3
i;tt

_ mm
loss;i;p
t
r
s;a
_ _
_
40
A rearrangement of Eq. (40) and insertion of Eq. (38)
yields
xx
i;t
xx
i;tt
3
1
_ mm
loss;i
t
Q
3;i
m
tot
_
41
which can now be transferred to an abrasion-induced
reduction in the particle diameter d
i
:
d
i
xx
i;t
xx
i;tt
xx
i;tt
3
1
_ mm
loss;i
t
Q
3;i
m
tot
_
1
_ _
42
From this a critical particle size x
i
might be derived,
which characterizes the upper bound of the particular
particle fraction within the size interval K
i
that shrinks
during the time interval t into the smaller size inter-
val K
i1
:
x
i
x
i
d
i
43
with x
i
being the smallest particle size in the size inter-
val K
i
. According to Fig. 26, the transferring mass
fraction can then be calculated as
m
i;i1
m
i
m
loss;i
_ _
i
x
i
x
i1
x
i
44
However, as shown in Sec. 6, in addition to these local
attrition phenomena, the movement of the solids must
be described. The ow sheet in Fig. 27 summarizes the
mechanisms that are taken into account in the authors
work.
In the uidized bed itself ideal mixing of the solids is
assumed. Both freshly fed particles and reentering par-
ticles from the return line are thus evenly distributed in
the bed. The entrainment from the uidized bed into
the cyclone section is described by the correlation sug-
gested by Tasirin and Geldart (1998). The cyclone is
modeled as a series connection of an attrition unit and
a subsequent gassolids separator. Since in typical ui-
dized bed processes the solids loading is distinctly
higher than its critical value, the gassolids separation
is described according to the critical load hypothesis
suggested by e.g. Trefz and Muschelknautz (1993),
which again divides the cyclone separation into a series
of a spontaneous separations of the surplus mass at the
cyclone inlet and a subsequent so-called inner separa-
tion of the remaining critical mass inside the vortex.
The German standard design procedure (VDI Heat
Atlas, 1993) assumes an empirically found distribution
of the particles that are subjected to separation in the
vortex. This procedure cannot, however, be used in
connection with population balancing. The authors
have therefore considered two dierent extreme
approaches: one assumes a particle-size-sensitive
Figure 26 Approach to derive the mass transfer between
neighboring size intervals. (From Reppenhagen and
Werther, 2001.)
separation at the inlet that leaves the ne particles in
the critical mass ow entering the vortex
_ mm
crit
_ mm
fluid
m
G
_ mm
c;in

_
x
limit
0
q
3
x dx 45
and causes the coarser particles as the surplus mass
ow to form the strand. This separation is assumed
to be ideal. The alternative approach assumes that
the particles entering the vortex have the same size
distribution as the material entering the cyclone.
The part of the entering solids mass ow that
exceeds the critical load is completely attributed to
the strands owing directly into the catch of the
cyclone section. The catch feeds the standpipe, where
a certain amount of material is stored. It passes in the
downward direction, until it is nally refed to the bed.
From a previous work (Reppenhagen and Werther,
1999a), the results of a long-term cyclone attrition
experiment are available for comparison with the
above-derived approach for the abrasion-induced
shift in the PSD. In this experiment, a batch of catalyst
has been fed 600 times to an isolated 90 mm ID cyclone
operated at an inlet gas velocity of u
c;in
18 m=s and a
solids loading of m
c
c 0:3. Figure 28 shows the mea-
sured and the calculated PSD after the experiment in
comparison to the initial one.
It is obvious that the developed description of the
particle shrinking is basically suitable to describe the
abrasion eect on the particle size distribustion quali-
tatively, i.e., the shift and the narrowing of the PSD
with the resulting increase of its modal value are pre-
dicted. But it can be seen that the shift is underesti-
mated by the model. In particular, the experimentally
observed strong shift in the coarse particle size range
is not suciently described. One reason for this
underestimation might be the common problem of
discretized particle balances, that either the mass or
the number of particles cannot be conserved (e.g.,
Hill and Ng, 1995). In the above approach, this leads
to a wrong increase in the number of particles: prior to
its transfer a mass, m
i;i1
, is assigned to the mean par-
ticle diameter xx
i
; and afterwards the same mass is
assigned to the smaller mean particle diameter xx
i1
.
This can only be compensated by a wrong increase in
the number of particles. According to Eq. (38), the
increased number of particles results in a wrong
decrease in the material loss per single particle and
with it to an underestimation of the shrinking.
In the experiment shown above, the loss rate was
measured for each pass of the material through the
cyclone. In Fig. 29, this data set is now compared
Figure 27 The particle population balance model of a ui-
dized bed system. (Reppenhagen and Werther, 2001.)
Figure 28 Measured and calculated shift in the size distribu-
tion when a batch of catalyst was fed 600 times to a cyclone.
(Reppenhagen and Werther, 2001.)
Figure 29 Measured cyclone loss rate in comparison with
the simulation. (Reppenhagen and Werther, 2001.)
with numerical predictions of the separation process in
the cyclone. As is obvious from this gure, the assump-
tion of a particle-size-sensitive separation at the
cyclone inlet gives an excellent representation of the
measurements, whereas the nonclassifying mechanism
is far from being suitable.
As there were no experimental data available to
evaluate the entire above-derived particle population
balance, a plausibility check with a ctitious industrial
scale system has been made instead. The chosen data
of the system are given in the list here:
Bed design and inventory
Column diameter D
t
4:0 m
Column height higher than TDH
Bed mass m
b
18,000 kg
Mass in return line m
return
500 kg
Distributor design
Type: perforated plate
Number of orices n
or
5,000
Orice diameter d
or
0:005 m
Design of the cyclone section
Single stage
Number of cyclones in parallel N
c
1
Outer diameter D 1:35 m
p 2000 Pa at operating conditions
Operating conditions
Fluid: air at T 293K and p 1:0 10
5
Pa
Supercial gas velocity u 0:61 m=s
No catalyst discharge
Initial catalyst material
PSD of the initial material shown in Fig. 31
Surface mean diameter of initial material
d
p
59:10
6
m
Jet attrition constant C
j
9:5 10
6
s
2
=m
3
Bubble attrition constant C
b
0:4 10
3
m
2
Cyclone attrition constant C
c
1:2 10
3
s
2
=m
3
For this system, the particle population balance is
solved in discretized time steps t. At the beginning
of each time step, except the very rst one, the fresh
material and the material from the return line are fed
and mixed with the bed material. It follows the calcu-
lation of jet-induced and bubble-induced attrition.
Afterwards the particular material mass fractions are
determined that are entrained from the bed in the
course of this time step. These material fractions are
then subjected to the combination of attrition and gas
solid separation inside the cyclone section. Finally, the
catch of the cyclone is added to the return line, which
in the next time step feeds its excess mass to the bed. At
the starting time t 0, the particle size distributions in
both the bed and the return line are identical to that of
the initial material. As computational time step, a
value of t 10 s was determined by numerical
experiments. As a rst result, Fig. 30 shows the calcu-
lated loss ow of the system during the initial 10 days.
After a very high initial value of approximately
3 t/d, the loss ow rapidly decreases and approaches
asymptotically a signicantly smaller value. The high
initial loss rate can be explained by a strong sifting of
the bed material due to the elutriation of the initial
nes. However, the stepped shape of the graph is
much more noticeable than this initial value. It mainly
results from the combination of the discretized descrip-
tion of the particle size distribution and the cyclone
model with its idea of an absolutely particle-size-sensi-
tive inlet separation: Owing to the continuous sifting in
the cyclone there is a depletion of the ne particles in
the inventory, and the threshold diameter x
limit
from
Eq. (45) thus increases with time. This results in an
increasing content of larger particles with higher
grade eciency in the vortex, and the systems loss
rate decreases accordingly. Since there is only one
grade eciency assigned to one size interval, a step
in the loss rate occurs whenever x
limit
is moving, into
a larger size interval. However, with decreasing loss
rate and the thus slower change in the size distribution,
the time between these steps becomes longer until x
limit
stays within a size interval. From this time on, the
changes in the size distribution are even more slight
and the system tends to a steady state wherein the
elutriated particles are balanced by both the freshly
fed and the attrition-produced particles.
Figure 31 shows a comparison of the calculated size
distribution at steady state with that of the initial mate-
rial. As can be seen, the major changes occurred in the
ner particle size range, i.e. the sifting eect of the
cyclone dominates the attrition-induced shrinking.
Figure 30 Calculated loss ow of the system during the
initial 10 days. (Reppenhagen and Werther, 2001.)
However, despite the underestimation of the particle
shrinking by the present approach (see above), the
dominating eect of the gassolid separation is fairly
well conrmed by industrial ndings. Figure 32 shows
a comparison of the initial and the equilibrium catalyst
material of an industrial FCC process. As in the simu-
lation, there are no changes in the coarse particle size
range but only in the small particle size range.
However, the sifting of the ner particles is not entirely
that indicated by the simulation, which can again be
explained by the assumption of a single and compara-
tively large cyclone in the simulation.
NOMENCLATURE
A
c;in
= cross section of the cyclone inlet area, m
2
A
o
= open surface area of a perforated plate
distributor, m
2
A
t
= cross sectional area of the uidized bed
column, m
2
b = exponent in the Gwyn Equation, Eq. (1)
Bx; y = breakage function
C
b
= particle size independent rate constant of
bubble-induced attrition, s
2
=m
3
C
c
= particle size independent rate constant of
cyclone attrition, dened by Eq. (21), s
2
=m
3
C
j
= particle size independent rate constant of jet-
induced attrition, dened by Eq. (8), s
2
=m
3
C
= particle size independent rate constant,

dened by Eq. (35), s
2
=m
3
d
i
= abrasion-induced reduction in the particle
diamter in interval i, dened by Eq. (42), m
d
or
= diameter of an orice in a multihole gas
distributor, m
d
p
= surface mean diameter, m
d
p
= surface mean diameter, dened by Eq. (35),
m
d
pc
= surface mean diameter of the material that
enters the cyclone, m
d
pc
u = surface mean diameter of the elutriated
material, dened by Eq. (29), m
d
pb
= surface mean diameter of the bed material, m
d
pj
= surface mean diameter for jet-induced
attrition, dened by Eq. (35), m
D
t
= diameter of the uidized bed column, m
E
tot
= total attrition extent of a given system,
dened by Eq. (4)
G
s
= total rate of solids elutriation from the bed,
kg=m
2
s
G
si
= fractional elutriation rate for the size interval
i, kg=m
2
s
K
a
= attrition rate constant in the Gwyn equation,
Eq. (1), s
b
K
b
= rate constant of bubble-induced attrition,
dened by Eq. (13), m
1
kg
n
K
0
b
dened by Eq. (17), m
1
kg
n
K
b
dened by Eq. (18), s
3
=m
3
K
c
= particle size dependent rate constant of
cyclone attrition, dened by Eq. (19), s
2
=m
2
K
i
= elutriation rate constant of the size interval i,
kg=m
2
s
K
j
= particle size dependent rate constant of jet-
induced attrition, dened by Eq. (7), s
2
=m
2
_ mm
abr:;fines
= mass of abrasion-produced nes per unit
time, kg/s
m
b
= bed mass, kg
m
b;0
= initial bed mass, kg
_ mm
c;in
= solids mass ow rate into the cyclone, kg/s
_ mm
crit
= critical solids mass ow rate, dened by Eq.
(45), kg/s
_ mm
fluid
= mass ow of the uid, kg/s
m
i
= material mass in the size interval i, kg
_ mm
i;i1
= mass transfer ow rate due to particle
shrinking from the size interval i to i 1,
kg/s
Figure 31 Simulation of the bed particle size distribution in
a uidized bed system. (Reppenhagen and Werther, 2001.)
Figure 32 FCC process in a renery. (Reppenhagen and
Werther, 2001.)
m
loss
= mass lost from a given system, kg
_ mm
loss;c
= produced mass of nes per unit time by
cyclone attrition, kg/s
_ mm
loss;i
= lost mass of nes per unit time from the size
interval i, kg/s
_ mm
loss;i;p
= lost mass of a single particle per unit time,
kg/s
_ mm
loss;j
= produced mass of nes per unit time by jet-
induced attrition, kg/s
_ mm
loss;tot
= attrition-induced loss ow rate of the entire
uidized bed system, kg/s
m
return
= mass in return line, kg
m
tot
= total mass of solids in a given system, kg
N
c
= number of primary cyclones in parallel within
a uidized bed system
n
or
= number of orices in a multihole gas
distributor
N
p;i
= number of particles in the size interval K
i
p = pressure, Pa
p = pressure drop, Pa
Q
2
x = cumulative particle size distribution in
particle surface
Q
2;i
= fraction of the size interval i on the entire
particle surface
q
3
x = mass density particle size distribution, m
1
Q
3
x = cumulative particle size distribution in mass
Q
3;i
= fraction of the size interval i on the entire
material mass
r
b
= bubble-induced attrition rate, dened by Eq.
(13)
r
c
= cyclone attrition rate, dened by Eq. (6)
r
tot
= overall attrition rate, dened by Eqs. (2) and
(3), 1/s
Sx = selection function
t = time, s
u = supercial gas velocity, m/s
u
c;in
= gas velocity at the cyclone inlet, m/s
u
min
= threshold velocity for bubble-induced
attrition, dened by Eq. (13), m/s
u
mf
= minimum uidization velocity, m/s
u
or
= gas velocity in the orice of a multihole gas
distributor, m/s
u
ti
= terminal velocity of the size interval i, m/s
_
VV = volumetric ow rate, m
3
=s
w
i
= weight fraction of the size interval i in the
bed material
x = particle size, m
xx
i
= geometric mean particle size of the size
interval i, m
x
limit
= critical particle size in the absolutely particle-
size-sensitive inlet separation, dened by Eq.
(45), m
Greek Symbols
= parameter dened by Eq. (35)
m
c
= solids loading at the cyclone inlet, dened by Eq.
(19)
m
G
= critical solids loading for the carrying capacity of
the gas
r
s;a
= apparent density of the catalyst material, kg=m
3
r
f
= density of the uid, kg/m
3
t = time, s
Subscripts and Indices
0 = value at initial state
abr = caused by abrasion
attr. = caused by attrition
b = bed
c = cyclone
feed = related to the feed
nes = related to nes
h = index to number a certain size interval
i = index to number a certain size interval
j = jet
k = index to number a certain size interval
loss = lost material
or = orice
steady-state = under steady state conditions
tot = related to the entire system
REFERENCES
Anthony EJ, Granatstein DL. Sulfadation phenomena in
uidized bed combustion systems. Progress in Energy
and Combustion Science 27:215236, 2001.
Arastoopour H, Chen C-Y. Attrition of char agglomerates.
Arena U, DAmore M, Massimilla L. Carbon attrition dur-
ing the uidized combustion of coal. AIChE J 29:4048,
1983.
Austin LG. A reviewintroduction to the mathematical of
grinding as a rate process. Powder Technol 5:117, 1972.
Bemrose CR, Bridgwater J. A review of attrition and attri-
tion test methods. Powder Technol 49:97126, 1987.
Blake TR, Webb H, Sunderland PB. The nondimensionaliza-
tion of equations describing uidization with application
to the correlation of jet penetration height. Chem Eng Sci
45:365371, 1990.
Blinichev VN, Streltsov VV, Lebedeva ES. An investigation
of the size reduction of granular materials during their
processing in uidized beds. Int Chem Eng 8(4):615618,
1968
Boerejn R, Zhang SH, Ghadiri M. Analysis of ISO uidized
bed test for attrition of uid cracking catalyst particles.
In: Fan LS, Knowlton TM, eds. Fluidization IX. New
Boerejn R, Gudde NJ, Ghadiri M. A review of attrition of
uid cracking catalyst particles. Adv Powder Technol
11(2):145174, 2000.
Braca RM, Fried AA. Operation of uidization processes.
In: Othmer DF, ed. Fluidization. New York: Reinhold,
1956, pp 117138.
British Materials Handling Board. Particle Attrition. Trans.
Tech. Publications Series on Bulk Materials Handling 5,
1987.
Chirone R, Massimilla L, Salatino P. Comminution of car-
bons in uidized bed combustion. Progress in Energy and
Combustion Science 17:297326, 1991.
Clift R. Powder technology and particle science. Powder
Technol 88:335339, 1996.
Contractor RM, Bergna HE, Chowdhry U, Sleight AW.
Attrition resistant catalysts for uidized bed-systems. In:
Grace JR, Shemilt LW, Bergougnou MA, eds.
Fluidization VI. New York: Engineering Foundation,
1989, pp 589596.
Dart JC. Mechanical tests to determine strength and abra-
sion resistance of catalysts. AIChE Symposium Series
143(70):59, 1974.
Davuluri RP, Knowlton TM. Development of a standardized
attrition test procedure. In: Fan LS, Knowlton TM, eds.
1998, pp 333340.
Dessalces G, Kolenda F, Reymond JP. Attrition evaluation
for catalysts used in uidized or circulating uidized bed
reactors. Preprints of the First International Particle
Technology Forum, Part II, Denver, CO, 1994, pp 190
196.
de Vries RJ, van Swaaij WPM, Mantovani C, Heijkoop A.
Design criteria and performance of the commercial reac-
tor for the shell chorine process. 5th Europ Symp Chem
Reac. Eng, Amsterdam, 1972, (B9) pp 5969.
Ewell RB, Gadmer G, Turk WJ. FCC catalyst management.
Hydrocarbon Processing 60:103112, 1981.
Forsythe WL, Hertwig WR. Attrition characteristics of uid
cracking catalysts. Ind Eng. Chem 41:12001206, 1949.
Fuertes AB, Pis JJ, Garcia JC, Rubiera F, Artos V.
Prediction of attrition in a continous uid-bed system.
Geldart D. Gas Fluidization Technology. Chichester: John
Wiley, 1986.
Ghadiri M, Cleaver JAS, Tuponogov VG. Modelling attri-
tion rates in the jetting region of a uidized bed. Preprints
of the Symposium Attrition and Wear, Utrecht, 1992a, pp
7988.
Ghadiri M, Cleaver JAS, Yuregir KR. Attrition of sodium
chloride crystals in a uidized bed. In: Potter OE, Nicklin
DJ, eds. Fluidization VII. New York: Engineering
Foundation, 1992b, pp 604610.
Ghadiri M, Cleaver JAS, Tuponogov VG, Werther J.
Attrition of FCC powder in the jetting region of a ui-
dized bed. Powder Technol 80:175178, 1994.
Ghadiri M, Cleaver JAS, Tuponogov VG. Inuence of dis-
tributor orice size on attrition in the jetting region of
uidized beds. In: Preprints Fluidization VIII(2). Tours,
1995, pp 799806.
Gierse M. Kornhaushalt in Zirkulierenden Wirbelschichten.
Brennsto-Wa rme-Kraft 43:459462, 1991.
Gwyn JE. On the particle size distribution function and the
attrition of cracking catalysts. AIChE J 15:3538, 1969.
Hansen S, Ottino JM. Fragmentation with abrasion and
cleavage: analytical results. Powder Technol 93:177184,
1997.
Hilligardt K, Werther J. The inuence of temperature and
solids properties on the size and growth of bubbles in gas
uidized beds. Chem Eng Technol 10:272280, 1987.
Hill J, Ng KM. New discretization procedure for the break-
age equation. AIChE J 41:12041216, 1995.
Hirschberg B. Solids mixing and segregation in a circulating
uidized bed. PhD dissertation, Technical University
Hamburg-Harburg, 1997.
Knight PC, Bridgwater J. Comparison of methods for asses-
sing powder attrition. Powder Technol 44:99102, 1985.
Kokkoris A, Turton R. The reduction of attrition in uidized
beds by the addition of solid lubricants. AIChE Symp Ser.
87:2031, 1991.
Kokkoris A, Turton R. A phenomenological model predict-
ing the attrition and the reduction of attrition due to the
addition of solid lubricants in slugging beds. Powder
Technol 84:3947, 1995.
Kono H. Attrition rates of relatively coarse solid particles in
various types of uidized beds. AIChE Symp Ser 77:96
106, 1981.
Kraft WW, Ullrich W, OConnor W. The signicance of
details in uid catalytic cracking units: engineering
design, instrumentation and operation. In: Othmer DF,
ed. Fluidization. New York: Reinhold, 1956, pp 184
211.
Krambrock W. Die Berechnung des Zyklonabscheiders und
praktische Gesichtspunkte zur Auslegung.
Aufbereitungstechnik 12(7):391401, 1971.
Kunii D, Levenspiel O. Fluidization engineering. New York:
Robert E. Krieger, 1969.
Kutyavina TA, Baskakov AP. Grinding of ne granular
material with uidization. Chem Techn Fuels Oils
8:210213, 1972
Levenspiel O, Kunii D, Fitzgerald T. The processing of solids
of changing size in bubbling uidized beds. Powder
Technol 2:8796, 1969.
Lin L, Sears JT, Wen CY. Elutriation and attrition of char
from a large uidized bed. Powder Technol 27:105115,
1980.
Merrick D, Highley J. Particle size reduction and elutriation
in a uidized bed process. AIChE Symp Ser 137(70):367
378, 1974.
AIChE J 21:507510, 1975.
Molerus O. Heat transfer in gas uidized beds. Powder
Technol 70:120, 1992.
Mothes H, Lo er F. Zur Berechnung der
Partikelabscheidung in Zyklonen. Chem Eng Process
18:323331, 1984.
Muschelknautz E. Auslegung von Zyklonabscheidern in der
technischen Praxis. Staub-Reinhaltung der Luft 30:187
195, 1970.
Neil AU, Bridgwater J. Attrition of particulate solids under
shear. Powder Technol 80:207219, 1994.
Newby RA, Vaux WG, Keairns DL. Particle attrition in
uidized-bed systems. In: Kunii D, Toei R, eds.
Fluidization IV. New York: Engineering Foundation,
1983, pp 233239.
Overturf BW, Reklaitis GV. Fluidized-bed reactor model
with generalized particle balances. AIChE J 5:813829,
1983.
Patel K, Nienow AW, Milne IP. Attrition of urea in a gas-
uidised bed. Powder Technol 47:257262, 1986.
Pell M, Jordan SP. Eects of nes and velocity on uidized
bed reactor performance. AIChE Symp Ser. 84:6873,
1988.
Pell M. Gas uidization. In: Handbook of Powder
Technology. Amsterdam: Elsevier, 1990, p 8, pp 97-106.
Pis JJ, Fuertes AB, Artos V, Suarez A, Rubiera F. Attrition
of coal and ash particles in a uidized bed. Powder
Technol 66:4146, 1991.
Paramanathan BK, Bridgwater J. Attrition of solidsII.
Chem Eng Sci 38:207224, 1983.
Patience GS, Mills PL. Modelling of propylene oxidation in a
circulating uidized-bed reactor. New Developments in
Selective Oxidation II, 1994, pp 118.
Ray YC, Jiang TS, Wen CY. Particle attrition phenomena in
a uidized bed. Powder Technol 49:193206, 1987a.
Ray YC, Jiang TS, Jiang TL. Particle population model for a
uidized bed with attrition. Powder Technol 52:3548,
1987b.
Reppenhagen, J. Catalyst attrition in uidized bed systems.
PhD dissertation, Technical University Hamburg-
Harburg, Aachen: Shaker, 1999.
Reppenhagen J, Werther J. Particle attrition inuences
cyclone performance. Preprints of the 4th European
Symposium Separation of Particles from Gases,
Nu rnberg, 1998, pp 6372.
Reppenhagen J, Werther J. Catalyst attrition in cyclones.
Powder Technol 113:5569, 1999a.
Reppenhagen J, Werther J. Catalyst attrition in uidized bed
systems with external solids circulation. In: Werther J, ed.
Circulating Fluidized Bed Technology VI. Frankfurt a.
M.: DECHEMA e.V.,1999b, pp 971976.
Reppenhagen J, Werther J. The role of catalyst attrition in
the adjustment of the steady-state particle sizes distribu-
tion in uidized bed systems. In: Kwauk M, Li J, Yang
Schubert H. Aufbereitung fester mineralischer Rohstoe.
Band I, 3. Au. Leipzig: VEB Deutscher Verlag fu r
Grundstondustrie, 1975.
Seville JPK, Mullier MA, Adams MJ. Attrition of agglom-
erates in uidized beds. In: Potter OE, Nicklin DJ, eds.
Fluidization VII. New York: Engineering Foundation,
1992, pp 587594.
Shipway PH, Hutchings IM. Attrition of brittle spheres by
fracture under compression and impact loading. Powder
Technol 76:2330, 1993.
Sishtla C, Findlay J, Chan I, Knowlton TM. The eect of
temperature and gas velocity on nes generation in non-
reactive uidized beds of coal char. In: Grace JR., Shemilt
LW, Bergougnou MA, eds. Fluidization VI. Alberta,
1989, pp 581588.
bedsa new correlation for the elutriation rate constants.
Trefz M, Muschelknautz E. Extended cyclone theory for gas
ows with high solids concentration. Chem Eng Technol
16:153160, 1993.
Ulerich NH, Vaux WG, Newby RA, Keairns L.
Experimental/engineering support for EPAs FBC pro-
gram: Final report. Vol. 1. Sulfur Oxide Control. EPA-
600/7-80-015a, 1980.
Vaux WG. Attrition of particles in the bubbling zone of a
uidized bed. Proc Am Power Conf 40:793802, 1978.
Vaux WG, Fellers AW. Measurement of attrition tendency in
uidization. AIChE Symp Ser 77:107115, 1981.
Vaux WG, Keairns DL. Particle attrition in uid-bed
processes. In: Grace JR, Matsen JM, eds.
Fluidization III. New York: Engineering Foundation,
1980, pp 437444.
Vaux WG, Schruben JS. Kinetics of attrition in the bubbling
zone of a uidized bed. AIChE Symp Ser 222(79):97102,
1983.
VDI-Heat Atlas. Fullarton JW. 1st English ed. Du sseldorf:
VDI-Verlag, 1993.
Weeks SA, Dumbill P. Method speeds FCC catalyst attrition
resistance determinations. Oil Gas J 88:3840, 1990.
Wei J, Lee W, Krambeck FJ. Catalyst attrition and deactiva-
tion in uid catalytic cracking systems. Chem Eng Sci
32:12111218, 1977.
Werner A, Haider M, Linzer W. Modelling of particle popu-
lation in uidized beds of particles diering in size and
physico-chemical behaviour. In: Preprint Fluidization
VIII(1), Tours, 1995, pp 557564.
Werther J. Scale-up modeling for uidized bed reactors.
Chem Eng Sci 47:24572462, 1992.
Werther J, Reppenhagen J. Catalyst attrition in uidized-bed
systems. AIChE J 45:20012010, 1999.
Werther J, Xi W. Jet attrition of catalyst particles in gas
uidized beds. Powder Technol 76(1):3946, 1993.
Xi W. Katalysatorabrieb in Wirbelschichtreaktoren. PhD
dissertation. Technical University Hamburg-Harburg,
1993.
Yates JG, Bejcek V, Cheesman DJ. Jet penetration into ui-
dized beds at elevated pressures. In: Ostergaard K,
Sorensen A, eds. Fluidization V. New York:
Yuregir KR, Ghadiri M, Clift R. Observation of impact
attrition of granular solids. Powder Technol 49:5357,
1986.
Yuregir KR, Ghadiri M, Clift R. Impact attrition of sodium
chloride crystals. Chem Eng Sci 42:843853, 1987.
Zenz FA. Bubble formation and grid design. Int Chem Eng
Symp Ser 30:136139, 1968.
Zenz FA. Find attrition in uid beds. Hydrocarbon
Processing 50:103105, 1971.
Zenz FA. Help from project E-A-R-L. Hydrocarbon
Processing 53:119, 1974.
Zenz FA, Kelleher GH. Studies of attrition rates in uid
particle systems via free fall, grid jets, and cyclone impact.
J Powder Bulk Tech 4:1320, 1980.
Zenz FA, Smith R. When are nes at equilibrium.
Hydrocarbon Processing 51:104106, 1972.
9
Modeling
Thomas C. Ho
Lamar University, Beaumont, Texas, U.S.A.
1 INTRODUCTION
The conversion in gassolids uidized bed reactors has
been observed to vary from plug ow, to well below
mixed ow, mainly depending on reaction and uidi-
zation properties (Levenspiel, 1972). Historically, two
classes of models have been proposed to describe the
performance of uidized bed reactors; one is based
on a pseudohomogeneous approach and the other on
a two-phase approach. The pseudohomogeneous
approach, where the existence of more than one
phase is not taken into account, proposes that we use
the conventional multiphase ow models for the ui-
dized bed reactors. These conventional models may
include ideal ow models, dispersion models, residence
time distribution models, and contact time distribution
models. The two-phase approach, however, considers
the uidized bed reactors to consist of at least two
phases, a bubble and an emulsion, and proposes a
separate governing equation for each phase with a
term in each equation describing mass interchange
between the two phases. Among the two-phase models,
the bubbling bed model proposed by Kunii and
Levenspiel (1969) and the bubble assemblage model
proposed by Kato and Wen (1969) have received the
most attention. Figure 1 illustrates the development
and evolution of these various ow models for ui-
dized bed reactors.
2 PSEUDOHOMOGENEOUS MODELS
The rst attempt at modeling gassolids uidized bed
reactors employed ideal or simple one-parameter
models, i.e., plug ow, complete-mixed, dispersion,
and tank-in-series models. However, the observed
conversion of sometimes well below mixed ow in
the reactors could not be accounted for by these mod-
els, and the approach was dropped by most research-
ers. The next attempt considered residence time
distribution (RTD) model in which all the gas in the
bed is considered equal in terms of gas residence time.
However, since an operating uidized bed consists of a
bubble phase and an emulsion phase of completely
dierent gas contacting hydrodynamics, the approach
is inadequate and was also dropped.
Gilliland and Kundsen (1970) then modied the
RTD models and assumed that the faster gas stayed
mainly in the bubble phase and the slower in the emul-
sion. Their approach was to distinguish the eect of
the two classes of gas with dierent eective rate
constant. The overall conversion equation in their
proposed model is
C
A
C
A
0
=
_
o
0
e
kt
E dt (1)
and the following equation was proposed to describe
the eective rate constant, i.e.,
k = k
0
t
m
(2)
where m is a tted parameter. Its value is small for
short staying gas and high for long staying gas.
Inserting Eq. (2) into (1) yields
C
A
C
A
0
=
_
o
0
exp k
0
t
m1
_ _
E dt (3)
The above equation describes the conversion and is
referred to as the contact time distribution (CTD)
model. Although improvement was made over the
RTD models, the problem with this approach involves
obtaining a meaningful E function to use in Eq. (3)
from a measured C curve obtained at the exit of a
bed. Questions remain as to how the measured C
curve represents the necessary E function in the calcu-
lation due to the considerable backmixing between
the faster and slower gas occurring in an operating
uidized bed.
3 TWO-PHASE MODELS
The discouraging result with the previous approach
has then led to the development of a sequence of
models based on the two-phase theory of uidization
Figure 1 Development and evolution of uidized bed models.
originally proposed by Toomey and Johnstone (1952).
The two-phase theory states that all gas in excess of
that necessary just to uidize the bed passes through
the bed in the form of bubbles. The term two-phase
model, however, represents a broad range of models
with various basic assumptions that may or may not
directly follow the original two-phase theory. For
example, some models consider the wakes and clouds,
while others do not; some models propose the use of
single-size bubbles, while others allow for bubble
growth, and some models use the two-phase ow dis-
tribution following the two-phase theory, while others
neglect the percolation of gas through the emulsion. In
addition, dierent models may propose dierent inter-
phase mass transfer mechanisms.
In the following subsections, the general two-phase
models are briey reviewed, and then we give a more
detailed description of two of the more popular two-
phase models, namely the bubbling bed model
proposed by Kunii and Levenspiel (1969) and the
bubble assemblage model proposed by Kato and
Wen (1969).
3.1 General Two-Phase Models
In most two-phase models, a uidized bed is consid-
ered to consist of two distinct phases, i.e., a bubble
phase and an emulsion phase. Each phase is repre-
sented by a separate governing equation with a term
in each equation describing mass interchange between
the two phases. A general expression of the two-phase
models, therefore, consists of the following two equa-
tions (from Wen and Fan, 1975). For the bubble phase
we have
F
oC
A.b
ot
_ _
FD
b
o
2
C
A.b
oh
2
_ _
FU
oC
A.b
oh
_ _
F
0
C
A.b
C
A.e
_ _
F
s
kC
A.b
= 0
(4)
and for the emulsion phase, we have
f
oC
A.e
ot
_ _
fD
e
o
2
C
A.e
oh
2
_ _
fU
oC
A.e
oh
_ _
F
0
C
A.e
C
A.b
_ _
f
s
kC
A.e
= 0
(5)
where the term FU represents the rise velocity of gas in
the bubble phase and the fU represents the rise velocity
of gas in the emulsion phase (= U
e
).
In the development of the two-phase models, most
investigators used a simplied form of the two-phase
model by either assuming or estimating some of the
terms in the above two equations. Table 1 summarizes
experimental investigations of model parameters asso-
ciated with the two-phase model. As indicated in Table
1, most of the studies assumed steady state (oC
A.b
,ot =
0 and oC
A.e
,ot = 0) with D
e
= 0 (plug ow in emulsion
phase) or D
e
= o (completely mixed in emulsion
phase); also D
b
= 0 (plug ow in bubble phase) and
U
e
= U
mf
. The parameters investigated include gas
interchange coecient (F
o
), particle fraction in bubble
phase (g
b
), and reaction rate constant (k). Among the
investigations, several authors reported that the ow
patterns in the emulsion phase, whether it is assumed
to be plug ow or completely mixed, do not signi-
cantly aect the model prediction (Lewis et al., 1959;
Muchi, 1965). However, Chavarie and Grace (1975)
reported that the two assumptions do aect their pre-
dictions based on the model of Davidson and Harrison
(1963).
A number of theoretical studies on the two-phase
model were also carried out and reported in the litera-
ture. They are listed in Table 2. The ones proposed by
Davidson and Harrison (1963) and Partridge and
Rowe (1966) are briey reviewed below.
3.1.1 Model of Davidson and Harrison (1963)
One of the representative two-phase models is the one
proposed by Davidson and Harrison (1963). This
model follows the two-phase theory of Toomey and
Johnstone (1952) and has the following assumptions:
1. All gas ow in excess of that required for inci-
pient uidization passes through the bed as
bubbles.
2. Bubbles are of uniform size throughout the
bed.
3. Reaction takes place only in the emulsion
phase with rst-order kinetics.
4. Interphase mass transfer occurs by a combined
process of molecular diusion and gas
throughow.
5. Emulsion phase (dense phase) is either per-
fectly mixed (DPPM) or in plug ow (DPPF).
Note that the key parameter in their model is the
equivalent bubble diameter, which was assumed to be
a constant.
Chavarie and Grace (1975) reviewed the model and
stated that the DPPM model is too conservative in its
estimate of overall conversion and that the concentra-
tion proles are in no way similar to their experimental
counterparts. Regarding the DPPF model, they stated
that while the DPPF model oers a fair dense phase
Table 1 Experimental Investigation of Two-Phase Mode Parameters
Authors Model Experimental conditions Experimental results
Shen & Johnstone
(1955)
,
b
= 0
U
e
= U
mf
D
e
= 0 or o
decomposition of nitrous oxide
d
t
= 11.4 cm
L
mf
= 26 32 cm
d
p
= 60 200 mesh
parameter F
o
k = 0.06 0.05 (1/s)
Mathis & Watson
(1956)
D
e
= 0
U
e
= U
mf
decomposition of cumene
d
t
= 5 10.2 cm
L
mf
= 10 31 cm
d
p
= 100 200 mesh
parameter: F
o
, ,
b
k = 0.64 (1/s)
Lewis et al.
(1959)
U
e
= 0
D
e
= 0 or o
hydrogenation of ethylene
d
t
= 5.2 cm
L
mf
= 11 53 cm
d
p
= 0.001 0.003 cm
parameter: F
o
, ,
b
k = 1.1 15.6 (1/s)
,
b
= 0.05 0.18, F = 0.4 0.8
Gomezplata & Shuster
(1960)
,
b
= 0
U
e
= U
mf
D
e
= 0
decomposition of cumene
d
t
= 5 10.2 cm
L
mf
= 3.8 20 cm
d
p
= 100 200 mesh
parameter: F
o
, ,
b
k = 0.75 (1/s)
Massimila & Johnstone
(1961)
,
b
= 0
U
e
= U
mf
D
e
= 0
oxidation of NH
3
d
t
= 11.4 cm
L
mf
= 26 54 cm
d
p
= 100 325 mesh
parameter: F
o
k = 0.071 (1/s)
Orcutt et al.
(1962)
,
b
= 0
U
e
= 0
decomposition of ozone
d
t
= 10 15 cm
L
mf
= 30 60 cm
d
p
= 0.001 0.003 cm
k = 0.1 3.0 (1/s)
Kobayashi et al.
(1966a)
U
e
= U
mf
D
e
= 0
d
t
= 8.3 cm
L
mf
= 10 100 cm
d
p
= 60 80 mesh
parameter: ,
b
k = 0.1 0.8 1/s
,
b
= 15 (L,L
mf
1)
Kobayashi et al.
(1966b)
U
e
= U
mf
D
e
= 0
d
t
= 20 cm
L
mf
= 10 100 cm
d
p
= 60 80 mesh
parameter: ,
b
k = 0.2 3.5 1/s
,
b
= 0.1 0.3
Kato
(1967)
U
e
= U
mf
D
e
= 0
packed uidized bed
hydrogenation of ethylene
d
t
= 8.7 cm
L
mf
= 10 30 cm
d
p
= 100 200 mesh,
d
p
= 1 3 cm
parameters: ,
b
k = 1.1 3.3 1/s
,
b
= 0.35 0.45
Kobayashi et al.
(1967)
D
e
= 0
U
e
= U
mf
residence-time curve
gas: air
tracer: He
particle: silica, gel
d
t
= 8.4 cm
F
o
= 11,d
b
De Grout
(1967)
D
e
= 0
U
e
= U
mf
residence time curve
gas: air
tracer: He
particle: silica
d
t
= 10 150 cm
H
K
= 0.67 d
0.25
t
L
0.5
where L = bed height (m)
H
K
= U,F
o
; U[m/s]
Kato et al.
(1967)
U
e
= 0
D
e
= 0.68
U U
mf
U
mf
_ _
d
p
c
p
residence-time curve
gas: air, H
2
, N
2
tracer: H
2
, C
2
H
4
, C
3
H
5
particle: silica-alumina, glass
d
t
= 10 cm, d
p
= 1 3 cm
F
o
= 5 3 1/s for
U,U
mf
= 2 30
M = 0.4 0.2 1/s for
U,U
mf
= 2 30
Table 2 Theoretical Study of the Two-Phase Model
Authors Parameter assumed Method ,
b
or F
0
Remarks
Van & Deemter
(1961)
,
b
= 0
U
e
= 0
D
e
= D
s
A steady state analysis of gas
backmixing and residence time
curve and rst-order reaction
by two-phase model
H
k
=
F
o
L
U
H
k
= 0.5 2.5. ,
b
= 0
Parameter F
o
is not
related to the bubble
movement in the bed
Davidson & Harrison
(1963)
D
e
= 0 or o
,
b
= 0
Estimation of conversions for
a rst-order reaction
F
o
=
5.85T
1,2
g
1,4
d
5,4
b
4.5U
mf
d
b
Parameter d
b
model
does not account for
bubble growth in the
bed
Muchi
(1965)
U
e
= U
mf
0 - D
e
- o
A study of eect of F
o
, ,
b
, D
e
U
e
on conversion of a rst-
order reaction
No relation between
bubble movement and
parameter
Mamuro & Muchi
(1965)
U
e
= U
mf
,
b
= 0
Analysis of a rst-order reaction
based on the two-phase cell
model
F
o
, = 0.05
= shape factor of bubble
Kobayashi & Arai
(1965)
U
e
= 0
D
e
= 0
A study of the eect of k, ,
b
,
D
e
, and F
o
on conversion of a
rst-order reaction
Parameters ,
b
, F
o
, D
e
are not related to the
bubble movement
Partridge & Rowe
(1966)
D
e
= 0
,
b
,= 0
Consider bubble-cloud phase
reaction in bubble-cloud
phase with non-rst-order
kinetics
Sh
c
= 2 0.69Sc
0.33
Re
0.5
c
Sh
c
=
K
g.c
d
c
T
Allow the variation
of bubble population
with height
Van & Deemter
(1967)
D
e
= 0 Analysis of backmixing
residence time curve of tracer
gas and the rst-order
reactions
F
o
= 0.4 1.2 (1/s) Parameters ,
b
, F
o
, U
e
,
are not related to the
bubble growth in the
bed
concentration prole, predicted dense phase concentra-
tions are far too high and too close to the bubble phase
concentrations. They concluded that the mass transfer
rate in the model is too high.
3.1.2 Model of Partridge and Rowe (1966)
Another representative two-phase model is the one
proposed by Partridge and Rowe (1966). In this
model, the two-phase theory of Toomey and
Johnstone (1952) is still used to estimate the visible
gas ow, as in the model of Davidson and Harrison
(1963). However, this model considers the gas inter-
change to occur at the cloud-emulsion interphase,
i.e., the bubble and the cloud phase are considered to
be well-mixed, the result being called bubble-cloud
phase. The model thus interprets the ow distribution
in terms of the bubble-cloud phase and the emulsion
phase. With the inclusion of the clouds, the model also
allows reactions to take place in the bubble-cloud
phase. The rate of interphase mass transfer proposed
in the model, however, considers the diusive mechan-
ism only (i.e., without throughow) and is much lower
than that used in the model of Davidson and Harrison
(1963).
Chavarie and Grace (1975) also reviewed the model
and found that the model has a tendency to overesti-
mate the cloud size which deprives the concentration
proles and related features obtained from the model
of any physical meaning, owing to the inherent physi-
cal incompatibility. A similar problem regarding the
model was also reported by Ellis et al. (1968).
3.1.3 Modications and Applications
In recent years, several modied versions of the two-
phase model were proposed for modeling uidized bed
reactors. They include a model proposed by Werther
(1980) for catalytic oxidation of ammonia, in which the
mass transfer process is expressed in terms of lm
theory, as described in Danckwerts (1970); a model
proposed by Werther and Schoessler (1986) for cataly-
tic reactions; a model proposed by Borodulya et al.
(1995) for the combustion of low-grade fuels; a
model proposed by Arnaldos et al. (1998) for vacuum
drying; and a model proposed by Srinivasan et al.
(1998) for combustion of gases. The modications
include the consideration of axial mass transfer prole,
the inclusion of a wake phase in addition to the bubble
and emulsion phases, and the consideration of the
growth of bubbles in the bubble phase.
3.2 Bubbling Bed Model
The bubbling bed model proposed by Kunii and
Levenspiel (1969) can be considered a modied version
of the two-phase model where, in addition to the bub-
ble and the emulsion phases, a cloud-wake phase is
also considered. The model represents a group of mod-
els often referred to as backmixing or dense phase ow
reversal models (see also Van Deemter, 1961; Latham
et al., 1968; Fryer and Potter, 1972). A key dierence
between this model and the rest of the two-phase mod-
els is that the interphase mass transfer considers two
distinct resistances, one from the bubble phase to the
cloud-wake phase, and the other from the cloud-wake
phase to the emulsion phase.
The derivation of the model involves the following
background theory and observations reported by
Davidson and Harrison (1963) and Rowe and
Partridge (1962), i.e.,
1. Bubble gas stays with the bubble, recirculating
very much like smoke rising and only penetrat-
ing a small distance into the emulsion. This
zone of penetration is called the cloud since it
envelops the rising bubble.
2. All related quantities such as the velocity of the
rise, the cloud thickness, and the recirculation
rate, are simple functions of the size of rising
bubble.
3. Each bubble of gas drags a substantial wake of
solids up the bed.
3.2.1 Derivation of the Model
Based on the above observations, the bubbling bed
model assumes that
1. Bubbles are of one size and are evenly distrib-
uted in the bed.
2. The ow of gas in the vicinity of rising bubbles
follows the Davidson model (Davidson and
Harrison, 1963).
3. Each bubble drags along with it a wake of
solids, creating a circulation of solids in the
bed, with upow behind bubbles and down-
load in the rest of the emulsion.
4. The emulsion stays at minimum uidizing con-
ditions; thus the relative velocity of gas and
solid remains unchanged.
With the above assumptions, material balances for
solids and for gas give in turn
(Upflow of solids with bubble)
= (Downflow of solids in emulsion) (6)
(Total throughflow of gas)
= (Upflow in bubble) (Upflow in emulsion)
(7)
Letting
u
br
= 0.711(gd
b
)
0.5
(8)
the above material balances give (1) The rise velocity of
bubbles, clouds, and wakes, u
b
:
u
b
= U U
mf
u
br
= U U
mf
0.711(gd
b
)
0.5
(9)
(2) the bed fraction in bubbles, d:
d =
U [1 d ad]U
mf
( )
u
b
U U
mf
u
b
(10)
(3) the bed fraction in clouds, b:
b =
3d(U
mf
,c
mf
)
u
br
(U
mf
,c
mf
)
(11)
(4) the bed fraction in wakes, o:
o = ad (12)
(5) the bed fraction in downowing emulsion including
clouds, oo:
oo = 1 d ad (13)
(6) the downow velocity of emulsion solids, u
s
:
u
s
=
adu
b
1 d ad
(14)
(7) the rise velocity of emulsion gas, u
e
:
u
e
=
U
mf
c
mf
u
s
(15)
Using Davidsons theoretical expression for bubble-
cloud circulation and the Higbie (1935) theory for
cloud-emulsion diusion, the interchange of gas
between bubble and cloud is then found to be
K
bc
= 4.5
U
mf
d
b
5.85
T
0.5
g
0.25
d
1.25
b
(16)
and between cloud and emulsion
K
ce
= 6.78
(c
mf
Tu
b
)
d
3
b
_ _
0.5
(17)
The above expressions indicate that if c
mf
, o, U
mf
, and
U are known or measured, then all ow properties and
regional volumes can be determined in terms of one
parameter, the bubble size. The application of this
model to chemical conversion is described below.
3.2.2 Model Expression for First-Order Kinetics
For a rst-order catalytic reaction occurring in a gas
solid uidized bed with c
A
= 0, the rate equation may
be expressed as
r
A.s
=
1
V
s
dN
A
dt
_ _
= kC
A
(18)
If the bed is assumed to be operated at a fairly high gas
ow rate with vigorous bubbling of large rising bub-
bles, then both the gas ow in the emulsion and the
cloud volume become so small that we can ignore the
throughow of gas in these regions. Consequently, as
an approximation, ow through the bed occurs only in
the bubble phase. The disappearance of A in rising
bubble phase, therefore, can be formulated as (see
Fig. 2):
(Disapparance from bubble phase) = (Reaction in
bubble)+(Transfer to cloud and wake)
(Transfer to cloud and wake)=(Reaction in cloud
cloud and wake)+(Transfer to emulsion)
(Transfer to emulsion)=(Reaction in emulsion)
(19)
In symbols, the above expressions become
r
A.b
=
1
V
b
dN
A
dt
_ _
= g
b
kC
A.b
K
bc
(C
A.b
C
A.c
)
K
bc
(C
A.b
C
A.c
) = g
c
kC
A.c
K
ce
(C
A.c
C
A.e
)
K
ce
(C
A.c
C
A.e
) = g
e
kC
A.e
(20)
Experiment has shown that
g
b
= 0.001 0.01 (21)
and
a = 0.25 1.0 (22)
Also, by the material balance expressions, Eqs. 6 to 15,
it can be shown that
g
c
= (1 c
mf
) 3
U
mf
c
mf
_ _
u
br
u
mf
c
mf
_ _ a
_
_
_
_
_
_
(23)
and
g
e
=
(1 c
mf
)(1 d)
d
g
b
g
c
(24)
On eliminating all intermediate concentration in Eqs.
20a, b, and c, we nd that
r
A.b
= g
b
k
1
1
_ _
C
A
(25)
where
1
=
1
K
bc
2
(26)
2
= g
c
k
1
3
(27)
and
3
=
1
K
ce
1
(g
e
k)
(28)
Inserting for plug ow of gas through the bed yields
the desired performance expression, or
ln
C
A
0
C
A
_ _
= g
b
k
1
1
_ _
L
fluidized
u
b
(29)
where approximately
L
fluidized
u
b
=
1 c
packed
1 c
mf
L
packed
u
br
(30)
Note that since the bubble size is the only quantity that
governs all the rate quantities with the exception of k,
the performance of a uidized bed is a function of the
bubble size. For small bubbles, the results from the
bubbling bed model may range between a plug ow
model and a mixed ow model; for large bubbles, how-
ever, they may be well below those predicted by a
mixed ow model.
3.2.3 Examples and Model Applications
Examples illustrating the model and additional discus-
sions on the model are found in Levenspiel (1972) and
Kunii and Levenspiel (1991). Applications of the
model were reported in several recent studies, including
scale-up studies of catalytic reactors by Botton (1983)
and Dutta and Suciu (1989), a gasication study of
coal carried out by Matusi et al. (1983), and a study
of fast uidized bed reactors by Kunii and Levenspiel
(1998).
3.3 Bubble Assemblage Model
Since the development of the bubbling bed model, var-
ious other hydrodynamic models have been proposed
using other combinations of assumptions such as
changing bubble size with height in the bed, negligible
bubble-cloud resistance, negligible cloud-emulsion
resistance, and nonspherical bubbles. Among them,
the bubble assemblage model, proposed by Kato and
Wen (1969), considers changing bubble size with height
in the bed. The model has the following assumptions:
Figure 2 Bubbling bed model. (From Kunii and Levenspiel, 1969.)
1. A uidized bed may be represented by n com-
partments in a series. The height of each com-
partment is equal to the size of each bubble at
the corresponding bed height.
2. Each compartment is considered to consist of a
bubble phase and an emulsion phase. The gas
ows through the bubble phase, and the emul-
sion phase is considered to be completely
mixed within the phase.
3. The void space within the emulsion phase is
considered to be equal to that of the bed at
the incipient uidizing conditions. The upward
velocity of the gas in the emulsion phase is U
e
.
4. The bubble phase is assumed to consist of
spherical bubbles surrounded by spherical
clouds. The voidage within the cloud is
assumed to be the same as that in the emulsion
phase, and the diameter of the bubbles and
that of clouds is given by Davidson (1961) as
d
c
d
b
_ _
3
=
u
br
2(U
mf
,c
mf
)
u
br
(U
mf
,c
mf
)
for u
br
_
U
mf
c
mf
(31)
where
u
br
= 0.711(gd
b
)
0.5
(32)
Note that the calculation proposed above would not be
applicable for large particles where u
br
may be less than
U
mf
,c
mf
.
5. The total volume of the gas bubbles within the
bed may be expressed as (L L
mf
)S.
6. Gas interchange takes place between the two
phases. The overall mass interchange coe-
cient per unit volume of gas bubbles is given by
F
d
= F
o
K
/
M (33)
7. The bubbles are considered to grow continu-
ously while passing through the bed until they
reach the maximum stable size or reach the
diameter of the bed column. The maximum
stable bubble size, d
b.t
, can be calculated by
(Harrison et al., 1961)
d
b.t
=
u
t
0.711
_ _
2
1
g
(34)
8. The bed is assumed to be operating under iso-
thermal conditions since the eective thermal
diusivity and the heat transfer coecient are
large.
3.3.1 Key Equations in the Bubble Assemblage
Model
In addition to the above assumptions, the model has
the following key equations to estimate various
bubbling properties:
1. Bubble size, based on Cooke et al. (1968):
d
b
= 0.14r
p
d
p
U
U
mf
_ _
h d
o
(35)
where
d
o
= 0.025
6(U U
mf
),(n
o
p) [ ]
0.4
g
0.2
(36)
2. Bubble velocity, following Davidson and
Harrison (1963):
u
b
= (U U
mf
) u
br
= (U U
mf
)
0.711(g d
b
)
0.5
(37)
3. Bed expansion, based on Assumption 5 and
Eqs. 35 through 37,
L L
mf
L
mf
=
U U
mf
0.711(g d
b.a
)
0.5
(38)
where d
b.a
is the average bubble diameter of
the bed given by
d
b.a
= 0.14r
p
d
p
U
U
mf
L
mf
2
d
o
(39)
4. Voidage of the bed, c
(a) For h _ L
mf
,
1 c =
L
mf
L
1 c
mf
( ) (40)
(b) for L
mf
_ h _ L
mf
2(L L
mf
),
1 c =
L
mf
L
(1 c
mf
)
0.5
L
mf
(1 c
mf
)(h L
mf
)
2L(L L
mf
)
(41)
5. Supercial gas velocity in emulsion phase, U
e
U
e
U
mf
= 1
c
mf
a
/
yu
b
U
mf
(1 y a
/
y)
(42)
where
y =
L L
mf
L
(43)
and a
/
is the ratio of the volume of emulsion
transported upward behind a bubble (volume
of wake) to the volume of a bubble. The value
of a
/
is approximately 0.2 0.3 according to
the experimental study of Rowe and Partridge
(1965). Therefore under normal experimental
conditions, Eq. (42) yields U
e
,U
mf
= 0.5
for U,U
mf
= 3, and U
e
,U
mf
= 0 for
U,U
mf
= 5 6. However, in the model, U
e
was assumed to be zero based on the experi-
mental ndings of Latham (1968) and the
argument presented by Kunii and Levenspiel
(1969).
6. Interchange coecient, F
d
, based on Eq. (33)
Since no experimental data are available for
the particle interchange rate, M, or the adsorp-
tion equilibrium constant for the reacting gas
on particle surfaces, K
/
, the model neglects gas
interchange due to adsorbed gas on interchan-
ging particles. Equation (33) therefore can be
reduced to
F
d
= F
o
(44)
where the following equation based on the
experimental work of Kobayashi et al. (1967)
was proposed to describe F
o
:
F
o
=
0.11
d
b
(45)
3.3.2 Calculation Procedure Based on Bubble
Assemblage Model
Let the height of the nth compartment be h
n
, where
n = 1. 2. 3, to n (see Fig. 3). Based on an arithmetic
average of the bubble size, the height of the initial
compartment immediately above the distributor
becomes
h
1
=
d
o
(ch
1
d
o
)
2
(46)
or
h
1
=
2d
o
2 c
(47)
where
c = 0.14r
p
d
p
U
U
mf
(48)
which is a proportionality constant relating the bubble
diameter for a given operating condition. The height of
the second compartment then becomes
h
2
= 2d
o
2 c
(2 c)
2
(49)
and that of nth compartment becomes
h
n
= 2d
o
(2 c)
n1
(2 c)
n
(50)
The number of bubbles in the nth compartment
becomes
N =
6S(c c
mf
)
p(h
n
)
2
(1 c
mf
)
(51)
The volume of cloud in the nth compartment can be
computed from Eq. (31) as
V
cn
=
Np(h
n
)
3
6
3(U
mf
,c
mf
)
u
br
U
mf
,c
mf
(52)
where
u
br
= 0.711 g(h
n
) [ ]
0.5
(53)
The total volume of the bubble phase (bubble and
cloud) and that of the emulsion phase in the nth com-
partment are, respectively,
V
bn
=
Np(h
n
)
3
6
u
br
2(U
mf
,c
mf
)
u
br
(U
mf
,c
mf
)
(54)
and
V
en
= S h
n
V
bn
(55)
The distance from the distributor to the nth compart-
ment is then
h
n
= h
1
h
2
h
3
h
n
(56)
The gas interchange coecient based on unit volume
of bubble phase (bubble and cloud) can be shown as
F
/
on
= F
on
u
br
(U
mf
,c
mf
)
u
br
2(U
mf
,c
mf
)
(57)
Hence, the material balance for the gaseous reactant
around the nth compartment becomes, for the bubble
phase,
(SUC
A.b
)
n1
= F
/
on
V
b
C
A.b
C
A.e
_ _ _ _
n
(r
A.c
V
c
)
n
(SUC
A.b
)
n
(58)
and for the emulsion phase,
F
/
on
V
b
C
A.b
C
A.e
_ _ _ _
n
= r
A.e
V
e
_ _
n
(59)
where r
A.c
and r
A.e
are the reaction rates per unit
volume of the cloud and the emulsion phase, respec-
tively. Note that for the rst-order reaction, the expres-
sions for r
A.c
and r
A.e
are r
A.c
= k
r
C
A.b
and
r
A.e
= k
r
C
A.e
, respectively.
3.3.3 Computational Procedure for Conversion
The computational procedure for conversion and con-
centration prole in a uidized bed reactor is given
below. The following operating conditions are needed,
i.e., particle size (d
p
), particle density (r
p
), minimum
uidization velocity (U
mf
), gas supercial velocity (U),
distributor arrangement (n
o
), column diameter (d
t
),
incipient bed height (L
mf
), reaction rate constant (k
r
),
and order of reaction. It should be noted that the
model requires no adjustable parameters.
First, Eq. (38) is used to calculate the expanded bed
height, L. Next, Eq. (50) is used to compute the size of
the nth compartment. Using Eqs. (51) through (55), the
volumes of the cloud, the bubble phase, and the emul-
sion phase for the nth compartment are then calculated.
The nth compartment concentrations, C
A.bn
and C
A.en
,
are computed from (C
A.b
)
n1
and (C
A.e
)
n1
using Eqs.
(58) and (59). The calculations are repeated from the
distributor until the bed height equivalent to L
mf
is
reached. For bed height above L
mf
, the voidage is
adjusted by Eq. (41), and V
cn
, V
bn
, and V
en
are obtained
using the same procedure as that shown for the height
smaller than L
mf
. The calculation is repeated until the
bed height reaches L
mf
2(L L
mf
).
3.3.4 Discussion of the Bubble Assemblage Model
The computation using the bubble assemblage model
indicates that for most of the experimental conditions
Figure 3 Main features of bubble assemblage model. (From Kato and Wen, 1969.)
tested, the number of compartments is usually greater
than 10. This means, in terms of the ow pattern, that
the gas passing through the bubble phase is close to
plugow. In the emulsion phase, because U
e
= 0 is
used, the ow pattern is essentially considered as a
dead space interchanging gas with the bubble phase
similar to that in the bubbling bed model. The ow
pattern in the emulsion phase, however, does not
expect to aect signicantly the reactor behavior
according to Lewis (1959) and Muchi (1965). It is
worth pointing out that when the reaction is slow,
any model, either a plugow, a complete mixing, a
bubbling bed, or a bubble assemblage model will rep-
resent the experimental data well. However, when
the reaction is fast, a correct ow model is needed to
represent the data.
3.3.5 Examples and Applications
More detailed discussion regarding the model perfor-
mance is given by Wen and Fan (1975) and Mori and
Wen (1975). The literature-reported applications of
this model include a combustion study of coal with
limestone injection by Horio and Wen (1975), a coal
gasication study by Mori et al. (1983), a study on
catalytic oxidation of benzene by Jares et al. (1983),
a catalytic ammoxidation of propylent by Stergiou and
Laguerie (1983), a silane decomposition study by Li et
al. (1989), a catalytic oxidation of methane by Mleczko
et al. (1992), and a study on chlorination of rutile by
Zhou and Sohn (1996).
3.4 Comparison of Models
In an attempt to compare the eectiveness of various
two-phase models, Chavarie and Grace (1975) carried
out a study of the catalytic decomposition of ozone in
a two-dimensional uidized bed where the reactor per-
formance and the ozone concentration proles in both
phases were measured. They compared the experimen-
tal data with those predicted from various two-phase
models reviewed above and reported that
1. The model of Davidson and Harrison (1963),
which assumes perfect mixing in the dense
phase (DPPM), underestimates seriously the
overall conversion for the reaction studied.
While the counterpart model that assumes pis-
ton ow in the dense phase (DPPF) gives much
better predictions of overall conversion, still
the predicted concentration proles in the indi-
vidual phases are in poor agreement with the
observed proles.
2. The model of Partridge and Rowe (1966)
makes allowance for variable bubble sizes
and velocities and for the presence of clouds.
Unfortunately, for the conditions of their
work, the overestimation of visible bubble
ow by the two-phase theory of Toomey and
Johnstone (1952) led to incompatibility
between predicted cloud areas and the total
bed cross section. This mechanical incompat-
ibility prevented direct application of these
models to the reaction data obtained in their
work.
3. The bubbling bed model of Kunii and
Levenspiel (1969) provides the best overall
representation of the experimental data in
this study. Predicted bubble phase proles
tend gently to traverse the measured proles,
while dense phase proles are in reasonable
agreement over most of the bed depth.
Overall conversions are well predicted. The
success of the model can be mainly attributable
to (1) the moderate global interphase mass
transfer, (2) the negligible percolation rate in
the dense phase, (3) the occurrence of reaction
within the clouds and wakes assumed by the
model, and (4) the use of average bubble prop-
erties to simulate the entire bed.
4. The Kato and Wen (1969) bubble assemblage
model, though better suited to represent com-
plex hydrodynamics due to allowance for vari-
able bubble properties, fails to account for
observed end eects in the reactor. While this
model was found to give the best t for the
bubble phase prole, dense phase proles and
outlet reactant concentrations were seriously
overpredicted.
It is worth pointing out that a general observation
made by Chavarie and Grace (1975) was that none of
the models (tested correctly) accounted for the consid-
erable end eects observed at both the inlet and the
outlet, i.e., near the distributor (grid region) and in the
space above the bed surface (freeboard region).
4 MULTIPLE-REGION MODELS
The grid region near the distributor and the freeboard
space above the bed surface have bed hydrodynamics
signicantly dierent from the main body of the bub-
bling bed reactor. A number of authors have observed
the abnormally high rate of reaction near the distribu-
tor, apparently due to the very high rate of interphase
mass transfer. Similarly, many observations have been
made of temperature increase in the freeboard region,
indicating additional reaction in the region. A realistic
approach for modeling a uidized bed reactor would
therefore require the consideration of three consecutive
regions in the bed: the grid region near the bottom, the
bed region in the middle, and the freeboard region
above the bed surface.
4.1 Models for the Grid Region
The grid region plays an important role in determining
the reaction conversion of uidized bed reactors, espe-
cially for fast reactions where the mass transfer opera-
tion is the controlling mechanism. Experimental
studies have indicated that changing from one distri-
butor to another, all other conditions remaining xed,
can cause major changes in conversion (Cooke et al.,
1968; Behie and Kehoe, 1973; Bauer and Werther,
1981). It is generally observed that, in the grid region,
additional mass transfer can take place owing to the
convective ow of gas through the interphase of the
forming bubbles. The ow of gas through the forming
bubbles into the dense phase, and then returning to the
bubble phase higher in the bed, represents a net
exchange between the two phases. The experimental
work of Behie and Kehoe (1973) indicated that the
mass transfer coecient in the grid region, k
je
, can be
40 to 60 times that in the bubble region, i.e., k
be
.
Behie and Kehoe (1973) and Grace and De Lasa
(1978) proposed similar sets of equations to describe
uidized bed reactors considering both the grid and
bed regions. The model of Grace and De Lasa (1978)
contains the following three equations: (1) for the jet
phase in the grid region (0 _ h _ J),
U
dC
A.j
dh
_ _
k
je
a
j
(C
Aj
C
A.e
) = 0 (60)
(2) for the bubble phase in the bed region (J _ h _ L),
b
/
U
dC
A.b
dh
_ _
k
be
a
b
C
A.b
C
A.e
_ _
= 0 (61)
and (3) for the emulsion phase in both the grid and bed
regions (0 _ h _ L),
U(1 b
/
)(C
A.e
C
A.jJ
)
_
J
0
k
je
a
j
(C
A.e
C
A.j
)dh
_
L
J
k
be
a
b
(C
A.e
C
A.b
) dh k
r
C
A.e
L
mf
= 0
(62)
It should be noted that the dierence between this
model and the corresponding model for the bed region
only is the replacement of Eq. (61) by Eq. (60) in the
grid region (0 _ h _ L
j
). The model therefore predicts
a much higher conversion in the grid region because
the mass transfer coecient in the region can be 40 to
60 times greater than in the bed region, as reported by
Behie and Kehoe (1973). Among the grid region stu-
dies, Ho et al. (1987) reported dynamic simulation
results of a shallow jetting uidized bed coal combus-
tor using the grid region model.
In another attempt, Sit and Grace (1986) measured
time-averaged concentrations of methane in the entry
region of beds of 120 to 310 mm particles contained in a
152 mm column with a central orice of diameter 6.4
mm and auxiliary tracer-free gas. The following equa-
tions were proposed for the particlegas mass transfer
in this region:
V
b
dC
A.j
dt
_ _
= Q
or
C
A.or
k
be1
S
ex.b
C
A.e
Q
or
k
be1
S
ex.b
_ _
C
A.j
(63)
where Q
or
represents the convective mechanism dened
as
Q
or
= u
or
p
4
_ _
d
2
or
(64)
and k
be1
represents the diusive mechanism that can be
predicted by the penetration theory expression, i.e.,
k
be1
=
4Tc
mf
pt
f
_ _
0.5
(65)
Their results indicated that the convective mechanism
has a greater eect on the grid region mass transfer.
Since bubbles are also smaller near the grid, they con-
cluded that favorable gassolid contacting occurs in
this region primarily due to convective outow fol-
lowed by recapture. Note that Eq. (63) can be used
to replace Eq. (60) for the grid region modeling.
4.2 Models for the Freeboard Region
When bubbles burst and release their gases at the sur-
face of a uidized bed, it is generally observed that
particles are ejected into the freeboard space above
the surface. These particles can originate either from
the dense phase just ahead of the bubble at the moment
of eruption (Do et al., 1972) or from those solids that
travel in the wake of the rising bubbles (Basov et al.,
1969; Leva and Wen, 1971; George and Grace, 1978).
As for studies of freeboard region hydrodynamics
and reactions, Yates and Rowe (1977) developed a
simple model of a catalytic reaction based on the
assumption that the freeboard contained perfectly
mixed, equally dispersed particles derived from bubble
wakes. The fraction of wake particles ejected, f
/
, was a
model parameter. The governing equation of the
model was proposed as
dC
A.cell
dt
=
k
g
A
p
V
cell
C
Ah
C
Ap
_ _
(66)
where V
cell
can be determined by
V
cell
=
3V
p
f
/
(1 c
mf
)
U U
mf
U u
t
(67)
and k
g
can be evaluated from the expression of Rowe
et al. (1965), i.e.,
Sh =
k
g
d
p
T
= 2 0.69 Sc
0.33
Re
0.5
t
(68)
with the Schmidt number, Sc, dened as
Sc =
m
r
g
T
(69)
and the terminal Reynolds number, Re
t
, dened as
Re
t
=
u
t
d
p
r
g
m
(70)
The model equations were solved for various combina-
tions of the parameters, and it was reported that in
many cases freeboard reactions lead to greater conver-
sion than that achieved by the uid bed itself. The
details of the results are also reviewed by Yates (1983).
5 CONCLUDING REMARKS
Numerous theoretical and experimental studies have
been carried out in the past ve decades in an attempt
to model gassolids uidized bed reactors. However,
the modeling of such reactors remains an art rather
than a science. Models that work well for certain reac-
tion processes may not work for others. In other
words, there has not been any single model universally
applicable to all processes carried out in such reactors.
However, all these modeling attempts do generate
valuable insights regarding reactor behavior, which
leads to improved design and operation.
Nevertheless, additional studies are needed to further
our knowledge in the modeling of such reactors.
NOTATION
A
p
= surface of freeboard particles appearing in Eq.
(66), m
2
a
b
= transfer area of bubbles to emulsion per unit
volume of reactor, m
2
,m
3
a
j
= transfer area of jets to emulsion per unit
volume of reactor, m
2
,m
3
C
A
= concentration of A, kg mol/m
3
C
Ah
= concentration of A at periphery of gas cell
appearing in Eq. (66), kg mol/m
3
C
Ao
= initial concentration of A appearing in Eqs. (1)
and (3), kg mol/m
3
C
Ap
= concentration of A at particle surface
appearing in Eq. (66), kg mol/m
3
C
A.b
= concentration of A in bubble phase, kg mol/
m
3
C
A.bn
= concentration of A in bubble phase at nth
compartment, kg mol/m
3
C
A.c
= concentration of A in cloud phase, kg mol/m
3
C
A.cell
= concentration of A in gas cell appearing in Eq.
(66), kg mol/m
3
C
A.e
= concentration of A in emulsion phase, kg mol/
m
3
C
A.en
= concentration of A in emulsion phase at nth
compartment, kg mol/m
3
C
A.j
= concentration of A in jet phase, kg mol/m
3
C
AjJ
= concentration of A at the tip of jets, kg mol/
m
3
C
A.or
= concentration of A at the distributor orice,
kg mol/m
3
D
b
= axial dispersion coecient of reactant in
bubble phase, m
2
/s
D
e
= axial dispersion coecient of reactant in
emulsion phase, m
2
/s
d
b
= bubble diameter, m
d
b.a
= average bubble diameter, m
d
b.t
= maximum stable bubble diameter, m
d
c
= diameter of cloud, m
d
0
= initial bubble diameter at the distributor, m
d
or
= diameter of distributor orice, m
d
p
= particle diameter, m
d
t
= diameter of bed column, m
E = probability density function
F = volumetric fraction of gas in the bubble phase
F
d
= overall gas interchange coecient per unit
volume of gas bubble, 1/s
F
o
= gas interchange coecient per unit volume of
gas bubble, 1/s
F
on
= gas interchange coecient at the nth
compartment per unit bubble volume, 1/s
F
/
on
= F
on
per unit volume of bubble phase (bubble
and cloud), 1/s
F
s
= volume fraction of solids in bubble phase
f = volumetric fraction of gas in the emulsion
phase
f
/
= fraction of wake particles ejected appearing in
Eq. (66)
f
s
= volumetric fraction of solids in emulsion phase
g = gravitational acceleration, m/s
2
h = distance from the distributor, m
h
1
= distance between the distributor and the top of
the rst compartment, m
h
1
= length of the rst compartment, m
h
n
= distance between the distributor and the top of
the nth compartment, m
h
n
= length of the nth compartment, m
J = height of jet phase, m
K
/
= adsorption equilibrium constant
K
bc
= gas interchange coecient between the bubble
and cloud phases, 1/s
K
ce
= gas interchange coecient between the cloud
and emulsion phases, 1/s
k = reaction rate constant, 1/s
k
be
= bubble to emulsion mass transfer coecient,
m/s
k
be1
= bubble to emulsion mass transfer coecient
during bubble formation, m/s
k
g
= mass transfer coecient, m/s
k
je
= jet to emulsion mass transfer coecient, m/s
k
0
= reaction rate constant dened in Eq. (2), 1/s
k
r
= rst-order reaction rate constant based on
volume of emulsion or cloud, 1/s
L = bed height, m
L
fluidized
= bed height under uidization condition, m
L
mf
= bed height at minimum uidization, m
L
packed
= bed height under packed condition, m
M = Solid interchange coecient between bubble
and emulsion phases per unit volume of
bubble, 1/s
m = tted parameter appearing in Eqs. (2) and (3)
N = number of bubbles in nth compartment
N
A
= moles of A, kg mol
n
0
= number of distributor holes
Q
or
= gas ow through orice, m
3
/s
r
A.b
= rate of reaction per unit bubble volume, kg
mol/(m
3
s)
r
A.c
= rate of reaction per unit cloud volume, kg
mol/(m
3
s)
r
A.e
= rate of reaction per unit emulsion volume, kg
mol/(m
3
s)
r
A.s
= rate of reaction per unit solid volume, kg mol/
(m
3
s)
Re
t
= Reynolds number of particle at terminal
velocity
S = cross-sectional area of the bed, m
2
S
ex.b
= surface area of bubbles, m
2
Sc = Schmidt number, dened as Sc = m,(r
g
T)
Sh = Sherwood number, dened as Sh = k
g
d
p
,T
t = time, s
t
f
= bubble formation time, s
U = supercial gas velocity, m/s
U
e
= supercial gas velocity in the emulsion phase,
m/s
U
mf
u
b
= rise velocity of bubbles in a uidized bed, m/s
u
br
= rise velocity of a single bubble in a uidized
bed, m/s
u
e
= rise velocity of emulsion gas, m/s
u
or
= velocity of gas through orice, m/s
u
s
= downow velocity of emulsion solids, m/s
u
t
= terminal velocity of particle, m/s
V
b
= volume of bubble phase, m
3
V
bn
= volume of bubble phase at the nth
compartment, m
3
V
c
= volume of cloud phase, m
3
V
cell
= volume of a freeboard gas cell expressed in Eq.
(67), m
3
V
cn
= volume of cloud phase at the nth
compartment, m
3
V
e
= volume of emulsion phase, m
3
V
en
= volume of emulsion phase at the nth
compartment, m
3
V
p
= volume of freeboard particles appearing in Eq.
(67), m
3
V
s
= volume of solids, m
3
a = ratio of cloud volume to bubble volume
a
/
= ratio of wake volume to bubble volume
b = bed fraction in clouds
b
/
= fraction of gas ow in the bubble or jet phase
g
b
= volume fraction of solids in bubbles in a
uidized bed
g
c
= volume fraction of solids in clouds in a
uidized bed, -
g
e
= volume fraction of solids in emulsion phase in
a uidized bed,
T = diusivity of reactant gas, m
2
/s
d = bed fraction in bubbles
c = void fraction
c
A
= fractional volume change on complete
conversion of A
c
mf
= void fraction at minimum uidization
c
packed
= void fraction under packed bed condition
y = bed expansion fraction dened in Eq. (43)
p = constant, - 3.1416
r
g
= gas density, kg/m
3
r
p
= density of particle, kg/m
3
m = gas viscosity, kg/(m s)
1
= expression dened in Eq. (26)
2
3
o = bed fraction in wake, appearing in Eq. (12)
oo = bed fraction in downowing emulsion,
appearing in Eq. (13)
c = expression dened in Eq. (48)
REFERENCES
Arnaldos J, Kozanoglu B, Casal J. Vacuum uidization:
application to drying. In: Fan LS, Knowlton TM, eds.
Fluidization IX, Durango, Colorado. New York:
Basov VA, Markhevkd VI, Melik-Akhnazarov TK, Orochko
DI. Investigation of the structure of a nonuniform ui-
dized bed. Int Chem Eng 9:263273, 1969.
Bauer W, Werther J. Proc 2nd World Congress Chem Eng
3:69, 1981.
Behie LA, Kehoe P. The grid region in a uidized bed reac-
tor. AIChE J 19:10701072, 1973.
Borodulya VA, Dikalenko VI, Palchonok GI, Stanchits LK.
Fluidization and combustion of solid biological waste and
low-grade fuels: experiment and modelling. In: Large JF,
Laguerie C, eds. Fluidization VIII, Tours, France. New
Botton R, Vergnes F, Bergougnou MA.Validation by means
of industrial data of Kunii-Levenspiel type bubble models
which can be used in the scale-up to commercial size of
uidized bed reactors. In: Kunii D, Toei R, eds.
Fluidization IV, Kashikojima, Japan. New York:
Chavarie C, Grace JR. Performance analysis of a uidized
bed reactor. II. Observed reactor behavior compared with
simple two-phase models. Ind Eng Chem Fund 14:7986,
1975.
Cooke MJ, Harris W, Highley J, Williams DF. Intn Chem
Engrs Symp Ser 30:21, 1968.
Danckwerts PV. Gas Liquid Reaction. London: McGraw
Hill, 1970.
De Groot JH. Proceedings of the International Symposium
on Fluidization, 1967, pp 348361.
Do H, Grace JR, Clift R. Particle ejection and entrainment
from uidized beds. Powder Technology 6:195200, 1972.
Dutta S, Suciu GD. Unied model applied to the scale-up of
catalytic uid bed reactors of commercial importance. In:
Fluidization VI, Ban, Canada. New York: Engineering
Ellis JE, Partridge BA, Lloyd DI. Proc Tripartite Chem Eng
Conf, Montreal, 1968, pp 43.
Fryer C, Potter OE. Bubble size variation in two-phase mod-
els of uidized bed reactors. Powder Technology 6:317
322, 1972.
George SE, Grace JR. Entrainment of particles from aggre-
gative uidized beds. AIChE Symp Ser 74(176):6774,
1978.
Gilliland ER, Knudsen CW. Paper 16d, AIChE Annual
Meeting, Chicago, 1970.
Gomezplata A, Shuster WW. Eect of uniformity of uidi-
zation on catalytic cracking of cumene. AIChE J 6:454
459, 1960.
Grace JR, De Lasa HI. Reaction near the grid in uidized
beds. AIChE J 24:364366, 1978.
Harrison D, Davidson JF, De Kock JW. On the nature of
aggregative and particulate uidisation. Trans Inst Chem
Engrs 39:202211, 1961.
Higbie R. Trans Am Inst Chem Eng 31:365, 1935.
Ho TC, Ko KN, Chang CC, Hopper JR. Dynamic simula-
tion of a shallow jetting uidized bed coal combustor.
Horio M, Wen CY. Analysis of uidized-bed combustion of
coal with limestone injection. In: Keairns DL, ed.
Fluidization Technology, Vol 2. Washington, DC:
Hemisphere, 1976, pp 289320.
Ishii T, Osberg GL. Eect of packing on the catalytic iso-
merization of cyclopropane in xed and uidized beds.
AIChE J 11:279287, 1965.
Jares JL, Chavarie C, Patterson I, Perrier M, Casalegno L,
Laguerie C. Conversion and selectivity modeling of the
oxidation of benzene to maleic anhydride in a uidized
bed reactor. In: Kunii D, Toei R, eds. Fluidization IV,
Kashikojima, Japan. New York: Engineering
Kato K. PhD dissertation. Tokyo Institute of Technology,
Tokyo, Japan, 1967.
Kato K, Imafuku K, Kubota H. Chemical Engineering
(Japan) 31:967, 1967.
Kato K, Wen CY. Bubble assemblage model for uidized
bed catalytic reactors. Chem Eng Sci 24:13511369, 1969.
Kobayashi H, Arai F. Chemical Engineering (Japan) 29:885,
1965.
Kobayashi H, Arai F, Isawa S, Sunagawa T, Miya K.
Chemical Engineering (Japan) 30:656, 1966a.
Kobayashi H, Arai F, Tanaka T, Sakaguchi Y, Sagawa N,
Sunagawa T, Shiba T, Takahashi K. The 6th Reaction
Engineering Symposium (Japan), 1966b, pp 13.
Kobayashi H, Arai F, Sunagawa T. Chemical Engineering
(Japan) 31:239, 1967.
Kunii D, Levenspiel O. Bubbling bed modelmodel for ow
of gas through a uidized bed. Ind Eng Chem Fund
7:446452, 1968.
Kunii D, Levenspiel O. Fluidization Engineering. New York:
John Wiley, 1969.
Kunii D, Levenspiel O. Fluidization Engineering, 2nd ed.
Kunii D, Levenspiel O. Conversion expressions for FF reac-
tors. In: Fan LS, Knowlton TM, eds. Fluidization IX,
Durango, Colorado. New York: Engineering
Latham R, Hamilton C, Potter OE. Back-mixing and chemi-
cal reaction in uidised beds. Brit Chem Eng 13:666671,
1968.
Leva M, Wen CY. Elutriation. In: Davidson JF, Harrison D,
eds. Fluidization, Chap 14. London: Academic Press,
1971, pp 627649.
Levenspiel O. Chemical Reaction Engineering, 2nd ed. New
Lewis WK, Gilliland ER, Glass W. Solid-catalyzed reaction
in a uidized bed. AIChE J 5:419426, 1959.
Li KY, Peng SH, Ho TC. Prediction of silicon powder elu-
triation in a uidized bed reactor for the silane decompo-
sition reaction. AIChE Symp Ser 85(270):7782, 1989.
Mamuro T, Muchi I. J of Ind Chem (Japan) 68:126, 1965.
Massimilla L, Johnstone HF. Reaction kinetics in uidized
Mathis JF, Watson CC. Eect of uidization on catalytic
cuemene dealkylation. AIChE J 2:518528, 1956.
Matsui I, Kojima T, Furusawa T, Kunii D. Gasication of
coal char by steam in a continuous uidized bed reactor.
In: Kunii D, Toei R, eds. Fluidization IV, Kashikojima,
Japan. New York: Engineering Foundation, 1983, pp
655662.
Mleczko L, Rothaemel M, Andorf R, Baerns M. Fluidized
bed reactor performance for the catalytic oxidative cou-
pling of methane to C
2
hydrocarbons. In: Potter OE,
Nicklin DJ, eds. Fluidization VII, Brisbane, Australia,
1992, pp 487494.
Mori S, Wen CY. Simulation of uidized bed reactor perfor-
mance by modied bubble assemblage model. In: Keairns
DL, ed. Fluidization Technology, Vol 1. Washington DC:
Mori S, Nomura S, Kurita M, Hiraoka S, Yamada I.
Modeling for the uidized bed coal gasier. In: Kunii
D, Toei R, eds. Fluidization IV, Kashikojima, Japan.
Muchi I. Memoirs of the Faculty of Engineering, Nagoya
University (Japan) 17(1), 1965.
Orcutt JC, Davidson JF, Pigford RL. Reaction time distribu-
tions in uidized catalytic reactors. Chem Eng Prog Sym
Series 58(38):115, 1962.
Partridge BA, Rowe PN. Chemical reaction in a bubbling
gas-uidised bed. Trans Inst Chem Engrs 44:T335T348,
1966.
Rowe PN, Partridge BA. Proc Symp on Interaction Between
Fluids and Particles, Inst Chem Eng London, 1962, pp
135.
Rowe PN, Partridge BA. An x-ray study of bubbles in ui-
dized beds. Trans Inst Chem Eng 43:157175, 1965.
Rowe PN, Claxton KT, Lewis JB. Heat and mass transfer
from a single sphere in an extensive owing uid. Trans
Inst Chem Eng 43:1431, 1965.
Shen CY, Johnstone HF. Gassolid contact in uidized beds.
AIChE J 1:349354, 1955.
Sit SP, Grace JR. Interphase mass transfer during bubble
formation in uidized beds. In: Ostergaard K,
Soorensen A, eds. Fluidization V, Elsionore, Denmark.
Srinivasan RA, Sriramulu S, Agarwal PK. Mathematical
modeling of the combustion of gases in bubbling uidized
beds. In: Fan LS, Knowlton TM, eds. Fluidization IX,
Durango, Colorado. New York: Engineering
Stergiou L, Laguerie C. An experimental evaluation of ui-
dized bed reactor models. In: Kunii D, Toei R, eds.
Fluidization IV, Kashikojima, Japan. New York:
Toomey RD, Johnstone HP. Gaseous uidization of solid
particles. Chem Eng Progress 48:220226, 1952.
Van Deemter JJ. Mixing and contacting in gassolid uidized
beds. Chem Eng Science 13:143154, 1961.
Van Deemter JJ. Proceedings of the International
Symposium on Fluidization, Eindhoven, 1967, pp 322.
Werther J. Modeling and scale-up of industrial uidized bed
reactors. Chem Eng Science 35:372379, 1980.
Werther J, Schoessler M. Modeling catalytic reactions in
bubbling uidized beds of ne particles. In: Van Swaaij
WPM, ed. Heat and Mass Transfer in Fixed and
Fluidized Beds. New York: Hemisphere, 1986, pp 355
370.
Wen CY, Fan LT. Models for Flow Systems and Chemical
Yates JG. Fundamentals of Fluidized-Bed Chemical
Processes. London: Butterworths, 1983.
Yates JG, Rowe PN. A model for chemical reaction in the
freeboard region above a uidized bed. Trans Inst Chem
Eng 55:137142, 1977.
Zhou L, Sohn HY. Mathematical modeling of uidized-bed
chlorination of rutile. AIChE J 42:31023112, 1996.
10
Heat Transfer
John C. Chen
Lehigh University, Bethlehem, Pennsylvania, U.S.A.
1 INTRODUCTION
In many gas uidized bed applications, heat transfer
takes place either naturally or by design. The heat
transfer could be between the solid and the gas
phase, between the two-phase mixture and a solid sur-
face, or both. An example is the uidized combustion
of coal, wherein coal particles are uidized by air. The
exothermic oxidation of carbon at particle surface
causes an increase in particle temperature, and this
naturally leads to heat transfer from the hot particles
to the uidizing air. To maintain an overall energy
balance of the bed, it is then necessary to transfer
heat from the particlegas medium to some cooling
surface, i.e., heat exchanger tubes. In this example,
both types of heat transfer (between gas and particles,
and between uidized medium and submerged sur-
faces) occur. This chapter deals with these two types
of heat transfer in gas uidized beds. Whether a heat
transfer process occurs naturally or by design is not of
consequence; the engineer often needs to determine the
rate of transfer in either case. In the above example,
rapid heat transfer between particle and gas reduces
the temperature of the burning particle and aects
such aspects as ash agglomeration in the bed. The
rate of heat transfer from the hot bed material to
heat exchanger determines the amount of heat exchan-
ger surface that must be provided. Because heat trans-
fer between particle and gas phases tends to be rapid, it
is often of less concern and will be treated in less detail.
In contrast, the rate of heat transfer between the ui-
dized bed (particles and gas mixture) and submerged
surfaces is often of engineering interest and will be
treated in greater detail.
Gas uidized beds operate in a number of dierent
regimes, as described in Chapter 3 of this book. The
mechanisms of heat transfer are signicantly dierent
for dierent uidization regimes. Consequently, the
correlations and models for prediction of heat transfer
coecients are regime specic and should only be used
within their range of applicability. This chapter deals
with the two regimes of gas uidization most com-
monly encountered in industrial applications:
Dense bubbling uidized beds (BFBs)
Fast circulating uidized beds (FFBs)
For each regime, rst a description is given of the gen-
eral hydrodynamic characteristics. Then heat transfer
characteristics are described and correlations/models
useful for thermal design are presented. In many
instances, the subject is still under research and no
single design method has won general acceptance.
For this reason, several alternative approaches are
presented, for both BFBs and FFBs.
To use the information given in this chapter, one
must rst determine the applicable regime of uidiza-
tion. Kunii and Levenspiel (1991) and Grace (1986a)
have presented excellent reviews of the available infor-
mation, including graphs showing the operational
boundaries for various uidization regimes. The
information presented by Grace can be replotted in
a generalized map using a dimensionless particle
diameter d
+
p
and dimensionless velocity U
+
, as shown
in Fig. 1. These variables are discussed below and
dened in Eqs. (4) and (7), respectively. Knowing
the physical properties of gas and particles, one can
then set the limits of supercial gas velocity for either
bubbling or fast uidization using the regime map of
Fig. 1.
2 DENSE BUBBLING FLUIDIZED BEDS (BFBs)
2.1 Hydrodynamics
Consider the upward ow of a gas through a bed of
packed particles. At some supercial velocity, the
upward drag force exerted by the gas on the particles
balances the downward body force of gravity. This is
the condition of minimum uidization, marking the
transition between packed beds and uidized beds.
For particles with diameters in the range of 50 to 500
microns and densities in range of 0.2 to 5,000 kg/m
3
,
uidization usually can be achieved smoothly with
increasing gas velocity. As denoted by Geldart
(1973), such particles are classied as type A or B
and include the majority of particles encountered in
uidized beds applications. For such particles, gas
velocities above the minimum uidization velocity
result in the occurrence of gas bubbles in the bed,
wherein some fraction of the gas ows through the
suspension of particles as a continuum phase, while
the remaining fraction ows as discrete bubbles rising
through the suspension. This is the regime commonly
called dense bubbling uidization. The upper limit of
gas velocity for this regime is related to terminal velo-
city of the particles, beyond which interfacial drag
becomes sucient to entrain the particles out of the
bed. Lines (a) and (b) in Fig. 1 indicate the usual limits
of gas velocity for the operation of BFBs.
To establish the appropriate uidization regime for
any given application, one needs to calculate the mini-
mum uidization velocity and the terminal velocity of
the bed particles. The supercial velocity of the gas
for minimum uidization (U
mf
) can be calculated by
solving the following equation for Re
mf
:
Ar = 1:75
(Re
mf
)
2
f
s
(e
mf
)
3
150
(1 e
mf
)(Re
mf
)
(f
s
)
2
(e
mf
)
3
(1)
Figure 1 Map of bubbling and at uidization regimes.
where Re
mf
is the Reynolds number at U
mf
,
Re
mf
=
U
mf
d
p
r
g
m
g
(2)
where f
s
is the sphericity of the particles and e
mf
is the
void fraction of the bed at minimum uidization.
Ar is the Archimedes number, dened as
Ar =
gr
g
(r
s
r
g
)
m
2
g
d
3
p
(3)
Note that a dimensionless particle diameter can be
dened as
d
+
p
Ar
1=3
(4)
If f
s
and e
mf
are not known for Eq. (1), an estimate of
the minimum uidization velocity can be obtained by
the equation of Wen and Yu (1966):
Re
mf
= 1140 0:0408Ar [ ]
1=2
33:7 (5)
The terminal velocity (U
t
) is given by the expression
U
t
=
4gd
p
(r
s
r
g
)
3r
g
C
D
_
(6)
where C
D
is the drag coecient for a single particle. In
the case of near-spherical particles, over the range
1 < Re
t
< 1,000, C
D
is given by the relationship
C
D
= 18:5Re
3=5
t
(6a)
where the Reynolds number at terminal velocity (Re
t
)
is dened by Eq. (52). Substituting (6a) into Eq. (6), an
explicit equation for the terminal velocity is obtained,
U
t
= 0:072g
d
8=5
p
(r
s
r
g
)
r
2=5
g
m
3=5
g
_ _
5=7
(6b)
A dimensionless gas velocity can be dened as
U
+

U
g
U
mf
U
t
U
mf
(7)
with values of 0 and 1 at U
mf
and U
t
, respectively. The
two dimensionless parameter d
+
p
and U
+
are used to
delineate regimes of uidization in Fig. 1. As seen
from Fig. 1, some bubbling beds are operated at gas
velocities slightly above the terminal velocity (i.e.,
U
+
> 1), relying on various detrainment devices to
reduce elutriation of particles out of the bed.
2.2 Heat Transfer
2.2.1 Particle Gas Transfer
Heat transfer between particles and gas in a uidized
bed may be compared to gas convection from a single
xed particle, and to gas convection from a packed bed
of xed particles. A common denition of the heat
transfer coecient can be used for all three cases,
based on the surface area of a single particle (a
p
),
h
p

q
a
p
(T
p
T
g
)
(8)
where for spherical particles, a
p
= pd
2
p
. The mechanism
for heat transfer becomes increasingly more complex
as the process changes from single particles, to xed
beds, to uidized beds. In the case of a single particle,
the heat transfer mechanism is single-phase convec-
tion, governed by the boundary layer at the particle
surface. In xed beds, one has the added complication
of particle packing and the complex gas ow pattern
that results. In bubbling uidized beds, the process is
further complicated by the motion of the suspended
particles and by macromixing caused by gas bubbles.
In all three situations, the heat transfer coecient is
found to increase with increasing velocity of the gas
relative to the particles. As may be expected, the heat
transfer coecient also increases with increasing ther-
mal conductivity, increasing density, and decreasing
viscosity of the gas.
The magnitude of the heat transfer coecient
between particles and gas in a bubbling uidized bed
is generally not large. Sample values of h
p
for
common applications are of approximate order 1 to
100 W/m
2
-K. Nevertheless, the rate of heat transfer
between particles and gas per unit bed volume is extre-
mely high, due to the large interfacial surface area.
Consequently, it is common to nd that thermal equili-
brium between particles and gas is reached quickly,
within a short distance from the point of gas injection.
In many designs, the isothermal condition is assumed
throughout the particlegas mixture. Concern with the
actual rate of heat transfer between particles and gas
arise primarily in situations where one phase or the
other is an intense heat source, such as burning particles
in a uidized combustor.
In those situations wherein one is concerned with
the rate of heat transfer between particles and gas,
correlations for the eective heat transfer coecients
are required. While Eq. (8) gives an exact denition of
the heat transfer coecient h
p
, its application is subject
to some ambiguity. The problem arises from the model
utilized to specify the particle and gas temperatures
(T
p
, T
g
) within the uidized bed. Recognizing that par-
ticle diusivity is high in bubbling beds, it is common
to treat the particles as being isothermal (well mixed)
within the bed. Choices for the gas range from the
perfectly mixed model to the plug-ow model. Values
of h
p
derived from any specic experiment, and the
resulting correlation, would be dierent depending on
the gas model selected. Application of the correlations
must be consistent to the derivations and utilize the
same model for gas ow. Choice of a gas-ow model
is somewhat arbitrary, since the well-mixed and the
plug-ow models both represent idealized limiting
cases, and neither is an accurate representation for
the bubbling uidized bed. Kunii and Levenspiel
(1969) pointed out that results based on the plug-
ow model show less scatter and have the advantage
of allowing analogous comparison to mass transfer
between particles and uid. Based on this reasoning,
the data and correlations given below are all based on
this model, with particles taken as isothermal within
the bed and gas passing through the bed in perfect plug
ow.
Experimental measurements of h
p
have been made
by various investigators, using both steady-state and
transient techniques. Kunii and Levenspiel (1969,
1991) collected ndings of 22 studies and presented
the results in terms of a particle Nusselt number and
the particle Reynolds number, respectively dened as
Nu
p

h
p
d
p
k
g
(9)
Re
p

U
g
d
p
r
g
m
g
(10)
All of these measurements were obtained with gas of
Prandtl number Pr 0:7. The data are shown in Fig.
2, bounded between the two dashed curves. For com-
parison, Ranzs correlation (1952) for convection from
a single sphere is also plotted on Fig. 2, as the dotdash
curve. The equation for Ranzs correlation is
Nu
p
= 2 0:6Re
0:5
p
Pr
0:33
g
(11)
It is seen that the Nusselt numbers for BFBs fall below
those for convection from a single sphere, for Reynolds
numbers less than 20. In fact, the magnitude of Nu
p
for
uidized beds drops below the value of 2.0, which
represents the lower limit of conduction heat transfer.
The cause of this is the bubbling phenomenon. Low
Reynolds numbers correspond to beds of ne particles
(small d
p
and U
g
), wherein bubbles tend to be clouded
with entrained particles. This diminishes the eciency
of particlegas contact below that represented by idea-
lized plug ow, resulting in reduced values of Nu
p
. As
particle diameter increases (coarse particle beds), bub-
bles are relatively cloudless and gasparticle contact
improves. This is shown in Fig. 2 where the Nusselt
numbers of uidized beds are seen to increase with
Figure 2 Nusselt numbers for particlegas heat transfer in dense bubbling beds for Pr
g
= 0:7.
increasing Reynolds number, approaching and slightly
exceeding that of single particles at Re
p
> 60.
The experimental data shown in Fig. 2 can be
approximated by the following equations, representing
the two straight lines in the gure:
Nu
p
= 0:0282Re
1:4
p
Pr
0:33
g
for 0:1 _ Re
p
_ 50
(12)
Nu
p
= 1:01Re
0:48
p
Pr
0:33
g
for 50 < Re
p
_ 10
4
(13)
It should be noted that the Prandtl number term in
these two equations is somewhat speculative, since
the experimental data did not cover a sucient range
for good correlation. Equations (12) and (13) include
Pr to the 0.33 power, based on the expectation that
dependence is similar to that for single spheres, as
shown in Eq. (11). In applying heat transfer coecients
calculated by Eqs. (12) and (13), it is important to use a
model that considers the particles to be well mixed and
the gas to be in plug ow, in order to be consistent with
the denition of h
p
as discussed above.
2.2.2 BedSurface Transfer
It is sometimes necessary to cool or heat dense bub-
bling uidized beds, and this is usually accomplished
by the insertion of heat transfer tubes carrying cooling
or heating uids into the bed. Heat transfer occurs
between the uidized particle/gas medium (often
referred to as the bed) and the submerged tube sur-
faces (often referred to as walls). For this situation,
one requires a heat transfer coecient based on the
surface area of the submerged wall:
h
w

q
a
w
(T
b
T
w
)
(14)
where a
w
is the submerged surface area, T
w
is the tem-
perature of the submerged surface, and T
b
is the tem-
perature of the particle/gas medium in the bed. For
this analysis, any small temperature dierence between
the particles and the gas is neglected.
Due to its engineering importance, this bed-to-sur-
face heat transfer coecient has been measured by
many investigators for various geometries and operat-
ing conditions (see Mickley and Trilling, 1949; Wender
and Cooper, 1958; Vreedenberg, 1960; Botterill and
Desai, 1972; Chen, 1976; Ozkaynak and Chen, 1980).
Typical characteristics of h
w
are:
h
w
is several times greater than the heat transfer
coecient for single-phase gas convection.
h
w
increases steeply as gas velocity exceeds the
minimum uidization velocity.
h
w
attains a maximum value at some specic
velocity that depends on the particle size.
Beyond the maximum point, h
w
decreases slightly
with further increase of gas velocity.
h
w
decreases with increasing particle size.
Figure 3 shows data of Chandran et al (1980) for the
average heat transfer coecients around a horizontal
tube submerged in uidized beds of spherical glass
particles with various mean diameters. The coecients
are plotted against the dimensionless velocity dened
by Eq. (7). It can be seen that all the characteristics
noted above are displayed by these data.
With bubbling dense beds, it is recognized that
mechanisms contributing to heat transfer at submerged
surfaces include gaseous convection during times of
bubble contact, particle conduction/convection during
times of particle contact, and radiation in the case of
high-temperature operation. The eective heat transfer
coecient is often represented as the sum of the coe-
cients for various contributing mechanisms:
h = f
l
h
l
(1 f
l
)h
d
h
r
(15)
where
h
l
= heat transfer coecient for lean gas phase
contact
h
d
= heat transfer coecient for dense particle
phase contact
h
r
= heat transfer coecient for radiation
f
l
= time fraction of contact by lean phase
This additive representation implies that the various
mechanisms can be superimposed, weighted by the
time fraction of lean or dense phase contacts at the
heat transfer surface. While not exact, this is a useful
approach.
Convective Heat Transfer. Most researchers
concur that for bubbling uidized beds, the gaseous
convection is signicantly enhanced by the presence
of solid particles (Ozkaynak and Chen, 1980; Kunii
and Levenspiel, 1991). Models often have focused on
the dense/particle phase contribution, or on the total
convective contribution,
h
c
= f
l
h
l
(1 f
l
)h
d
(16)
Several dierent approaches have been proposed for
the estimation of h
c
. Each approach has its advantages
and disadvantages, and all involve some degree of
empiricism. The following paragraphs summarize the
various approaches. Attention is centered on vertical
and horizontal tubes, since these are the geometries of
most practical interest.
The most common approach assigns thermal resis-
tance to a gaseous boundary layer at the heat transfer
surface. The enhancement of heat transfer is then
attributed to the scouring action of solid particles on
the gas lm, decreasing the eective lm thickness. The
works of Leva et al. (1949, 1952), Dow and Jakob
(1951), Levenspiel and Walton (1954), Vreedenberg
(1958), and Andeen and Glicksman (1976) utilized
this approach. Correlations following this approach
generally present a heat transfer Nusselt number in
terms of the uid Prandtl number and a modied
Reynolds number, with either the particle diameter
or the tube diameter as the characteristic length
scale. Some useful examples of such correlations are
given here.
Levas correlation (1952) for vertical surfaces, for
larger particles, is
Nu
c

h
c
d
p
k
g
= 0:525(Re
p
)
0:75
(17)
Wender and Coopers correlation (1958) for vertical
tubes (SI units) is
h
c
d
p
k
g
= 3:51 10
4
C
R
(1 e)Re
0:23
p
c
pg
r
g
k
g
_ _
0:43
c
ps
c
pg
_ _
0:8
r
s
r
g
_ _
0:66
for 10
2
_ Re
p
_ 10
2
(18)
where C
R
is a correction factor for tubes located away
from the bed axis.
An empirical t to the data of Vreedenberg (1952)
gives the following function for C
R
:
C
R
= 1:07 3:04
r
R
b
_ _
3:29
r
R
b
_ _
2
(19)
where r is the radial position of the heat transfer tube
and R
b
is the radius of the bed.
Figure 3 Heat transfer coecients for horizontal tube in bubbling beds of glass spheres, uidized by air at atmospheric pressure.
(Data of Chandran et al., 1980.)
Vreedenbergs correlations (1958) for horizontal
tubes are
h
c
D
t
k
g
= 420
r
s
r
g
Pr
g
m
2
g
gr
2
s
d
3
p
_ _
0:3
Re
0:3
D
for
r
s
r
g
Re
p
_ 2550
(20a)
and
h
c
D
t
k
g
= 0:66Pr
0:3
g
r
s
(1 e)
r
g
e
_ _
0:44
Re
0:44
D
for
r
s
r
g
Re
p
_ 2050
(20b)
where Re
D
D
t
r
g
U
g
=m
g
and Re
p
d
p
r
g
U
g
=m
g
. In the
intermediate range, Vreedenberg recommends using
the average of the values calculated by Eqs. (20a)
and (20b).
Andeen and Glicksman (1976) slightly modied
Vreedenbergs correlation for large particles to account
for dierent void fractions, resulting in the equation
h
c
D
t
k
g
= 900(1 e)
r
s
r
g
Pr
g
m
2
g
gr
2
s
d
3
p
_ _ _ _
0:3
Re
0:3
D
for
r
s
r
g
Re
p
_ 2550
(21)
which gives the same result as Eq. (20a) for a void
fraction of e = 0.53. In this and other correlations
that require a value for void fraction e, it would have
to be either known a priori or estimated from hydro-
dynamic models (see Chapter 3 of this book). For
example, if the volume fraction of bubbles (d) is deter-
mined from hydrodynamic models, the average bed
void fraction (e) is approximated by
e d (1 d)e
mf
(22)
Equations (20a,b) and (21) calculate average values of
the heat transfer coecient around the circumferences
of horizontal tubes. The experiments of Chandran et
al. (1980) showed that the local heat transfer coe-
cients are actually signicantly dierent at dierent
circumferential positions. For small particles, the max-
imum heat transfer coecient was found at the bottom
and sides of the tube. The heat transfer coecient at
the top of the tube was noticeably smaller, implying
the existence of a stagnant cap of solid particles that
insulate that portion of the heat transfer surface. For
the larger particles, a fairly uniform heat transfer coef-
cient is obtained, albeit with smaller magnitude than
the coecient for small particles. Similar observations
have been reported by Berg and Baskakov (1974) and
Saxena et al. (1978). The detailed prediction of local
heat transfer coecients around horizontal tubes is
beyond the scope of this book. The reader is referred
to the article of Chandran and Chen (1985b), which
presents a method for such calculations.
A second approach to modeling of the convective
heat transfer coecient considers combined gaseous
and particle convection. Molerus et al. (1995) devel-
oped a semiempirical correlation of Nu
c
, which is
given by the following equation, for spherical (or
nearly spherical) particles:
h
c
l
k
g
=
0:125e
s;mf
B
1
1 B
2
(k
g
=2c
ps
m
g
)
_ _ 0:165Pr
1=3
g
r
g
r
s
r
g
_ _
1=3
1
B
3
_ _
(23)
where
l =
m
g
r
s
r
g
_ _
2=3
1
g
_ _
1=3
(24)
B
1
1 33:3
U
e
r
s
c
ps
U
mf
gk
g
_ _
1=3
U
e
_ _
1
(25)
B
2
1 0:28e
2
s;mf
U
e
U
mf
r
g
r
s
r
g
_ _
1=2
r
s
c
ps
gk
g
_ _
2=3
(26)
B
3
1 0:05
U
mf
U
e
_ _
(27)
U
e
U
g
U
mf
(28)
the excess gas velocity above the minimum uidization
velocity.
Borodulya et al. (1991) developed a dierent corre-
lation, which emphasizes the parametric eects of bed
pressure and temperature as reected by changes of gas
and particle properties. Their equation is
h
c
d
p
k
g
= 0:74Ar
0:1
r
s
r
g
_ _
0:14
c
ps
c
pg
_ _
0:24
e
2=3
s
0:46Re
p
Pr
g
e
2=3
s
e
(29)
where the rst term represents particle convection and
the second term represents gas convection. Equation
(29) does not dierentiate between vertical and hori-
zontal surfaces and is claimed to be good for the
following ranges of parameters: 0:1 < d
p
< 4 mm;
0:1 < P < 10 MPa; 140 < Ar < 1:1 10
7
.
A third approach is the so-called packet theory
approach, which considers the heat transfer surface
to be contacted alternately by gas bubbles and packets
of closely packed particles. This leads to a surface
renewal process wherein heat transfer occurs primarily
by transient conduction between the heat transfer
surface and the particle packets during their time of
residence on the surface. Mickley and Fairbanks
(1955) treated the particle packet as a pseudo-homo-
geneous medium with solid volume fraction (1 e)
pa
and thermal conductivity k
pa
, obtaining the following
expression for the average heat transfer coecient due
to particle packets:
h
pa
= 2
k
pa
r
s
c
ps
1 e ( )
pa
pt
pa
_ _
1=2
(30)
where t
pa
is the root-square-average residence time for
packets on the heat transfer surface,
t
pa
t
i
(t
i
)
1=2
_ _
2
(31)
summed over a statistical number of packets. For small
or medium size particles, this packet renewal mechan-
ism accounts for the major portion of convective trans-
fer so that one can take h
c
h
pa
in Eq. (16).
To utilize Eq. (30), one needs eective values of the
packet properties in the vicinity of the heat transfer
surface. In recognition of the higher voidage in the
rst layer of particles next to the surface, Chandran
and Chen (1985) rewrote Eq. (30) as
h
pa
= 2C
k
pab
r
s
c
ps
(1 e)
pab
pt
pa
_ _
1=2
(32)
where k
pab
and (1 e)
pab
are thermal conductivity and
volume fraction in the bulk packet (far from the wall),
respectively. C is a correction factor to account for the
variation in packet voidage within the transient con-
duction zone near the wall. It takes the form
C = exp
a
1
Fo
a
2
a
3
ln Fo
_ _
(33)
where
Fo = Fourier modulus =
k
pab
t
pa
c
ps
r
s
(1 e)
pab
d
2
p
a
1
= 0:213 0:117w 0:041w
2
a
2
= 0:398 0:049w
a
3
= 0:022 0:003w
w = ln(k
pab
=k
g
)
(33a)
If specic information on void fraction in the bulk
packet is lacking, (1 e)
pab
may be approximated as
(1 e)
mf
. The thermal conductivity of the bulk packet
then can be estimated by the general model of Kunii
and Smith (1960). The following equation is an empiri-
cal representation of Kunii and Smiths results,
appropriate in the case of gas uidized beds, with
packets of loosely packed particles:
k
pab
= k
g
e
pab
(1 e)
pab
f
k
2k
g
=3k
s
_ _
(34)
where f
k
is given by
f
k
= 0:305
k
s
k
g
_ _
0:25
for 1 _
k
s
k
g
_ 1; 000
(35)
Ozkaynak and Chen (1980) showed that this modied
packet model could predict the measured convective
heat transfer coecient h
c
with good success, when
packet residence time t
pa
is known. When t
pa
is not
known a priori, Kunii and Levespiel (1991) suggest
that it can be estimated from bubble frequency (f
b
)
and the volume fraction of bubbles at the surface (d
w
),
t
pa
=
1 d
w
f
b
(36)
See Chapter 3 of this book for methods of estimating
bubble frequency and volume fraction. It especially
needs to be kept in mind in the case of horizontal
heat transfer tubes that bubbles that encounter the
tube tend to be deected to one side of the tube or
the other. In such a situation, a position on the tube
surface would be exposed to bubbles at approximately
half of the frequency calculated for an unobstructed
bed.
A fourth and dierent approach is that of Martin
(1984), who utilized an analogy between particle
motion in uidized beds and kinetic motion of mole-
cules in gases. From the kinetic theory of gases, Martin
developed a model to account for thermal energy
transport by particle motion across the boundary
layer at surfaces. The resulting Nusselt number for
convection was obtained as
h
d
d
p
k
g
= Z(1 e
b
) 1 e
N=C
c
Z
_ _
(37)
where
Z
r
s
c
ps
d
p
w
p
6k
g
(38)
N = Nusselt number for heat transfer upon
collision of particle with wall
C
c
= dimensionless parameter inversely propor-
tional to contact time of a particle on wall
w
p
= average random particle velocity.
For the common situation of k
s
k
g
,
N 4(1 Kn) ln 1
1
Kn
_ _
4 (39)
where Kn is the Knudsen number for the gas in the gap
between the particle and the wall. From kinetic theory,
for gases with accommodation constant approximately
unity,
Kn =
4k
g
(2pRT=M
g
)
1=2
Pd
p
2c
pg
R=M
g
_ _ (40)
Assuming that the random particle velocity can be
described by Maxwells distribution, the average par-
ticle random velocity is derived as
w
p
=
gd
p
(e
b
e
mf
)
5(1 e
b
)(1 e
mf
)
_
(41)
The dimensionless parameter C
c
was left as an empiri-
cal parameter. From comparison with experimental
data, Martin (1984b) suggested a value of 2:6 for
C
c
. Using Martins model, the convective coecient
h
c
would thus be calculated by Eqs. (37)(41).
All the design relationships given above, whatever
the approach, treat the heat transfer wall as an isolated
surface in the uidized bed. In practice, the geometry
of most interest is a bundle of tubes, composed of rows
of cylindrical tubes placed either horizontally or verti-
cally inside the uidized bed. Wood et al. (1978)
showed that the presence of a horizontal tube bundle
can suppress slugging by breaking up the bubbles,
resulting in a decrease of heat transfer coecients of
up to 20%, at gas velocities that correspond to the
slugging regime. At higher gas ow rates, this eect
is reduced, owing to the uniform nature of the
bubbling with or without tube banks. Tests made at
atmospheric pressure further showed that heat transfer
coecients were independent of location within the
tube bundle. For vertically placed tube bundles, Noe
and Knudsen (1968) reported no dierence in mea-
sured heat transfer coecients for single or multiple
tubes. In a tube bundle, the tubes are usually arranged
in either a triangular or a square array, with sucient
spacing between neighboring tubes (pitch) to permit
the mixing of the uidized medium and the passage
of gas bubbles. Following the information summarized
by Pell (1990), it is suggested that the pitch-to-diameter
ratio for tubes in a bundle be in the range of 2 to 4 or
greater, in order to achieve good mixing within the
uidized bed.
Radiative Heat Transfer. For many years, there
were contradictory opinions regarding the signi-
cance of radiative heat transfer in uidized beds. Esti-
mates of the radiative contribution to total heat
transfer ranged from 2% to 30% for high tempera-
ture beds. Improved experimental measurements since
1980 have resolved much of this uncertainty. With di-
rect measurements of both radiative heat ux and to-
tal heat ux in bubbling beds, Ozkaynak et al. (1983)
showed that the radiant contribution to total heat
transfer was less than 15% at bed temperatures be-
low 500
C. At higher temperatures, radiant

contribution increased linearly with bed temperature,
becoming greater than 35% when bed temperature
exceeded 800
C. This increase was not more rapid,

i.e., proportional to the fourth power of absolute
temperature, because convective/conductive heat
transfer also increased with temperature as a result of
the changes in the gass physical properties.
The radiative heat transfer coecient in Eq. (15) is
dened as
h
r

q
r
a
w
(T
b
T
w
)
(42)
where q
r
is the radiative heat transfer rate on the wall
surface.
Models to calculate q
r
or h
r
can be grouped into two
approaches. The rst and simpler approach is to use
the StefanBoltzman equation for radiant exchange
between opaque gray bodies. Treating the wall and
bed medium as two opposing parallel surfaces with
respective emissivities of e
w
and e
b
, the radiative heat
transfer rate is
q
r
= a
w
e
bw
s T
4
b
T
4
w
_ _
(43)
where s is the StefanBoltzman constant. The eective
emissivity for the bedwall combination (e
bw
) is given
by the expression
e
bw
=
1
1=e
b
1=e
w
1
(44)
The resulting expression for the radiative heat transfer
coecient is then
h
r
=
e
b
e
w
e
b
e
w
e
b
e
w
_ _
s T
4
b
T
4
w
_ _
T
b
T
w
( )
(45)
Equation (45) is applicable for tubes in a bundle with
large pitch-to-diameter ratio, where particle/gas med-
ium separates adjacent tubes. In using this equation,
values of wall and bed emissivity (e
b
, e
w
) are required.
Assuming that e
w
is known for specic heat transfer
surfaces, the need is then to estimate e
b
. For BFBs of
normal dimensions, e
b
would be greater than the emis-
sivity of individual particles and approach unity for
almost all types of particles. The investigation of
Ozkaynak et al. (1983), with experimental measure-
ments of radiant heat ux, provides an indication of
the eective bed/wall emissivity. For beds with parti-
cles similar to sand particles, e
bw
was found to be in the
range of 0.8 to 1.0 at bed temperatures greater than
700
C, where radiation is signicant. It was also found

that this eective emissivity is fairly insensitive to the
supercial gas velocity. Thus a simple, approximate
model for radiant heat transfer in bubbling uidized
beds is Eqs. (43)(45), with e
b
taken as approximately
equal to 0.9.
The second type of approach takes a more mechan-
istic representation of the radiant transport process.
These models recognize that radiant photons are
emitted, absorbed, and scattered by the solid particles
in the uidized bed. Bhattacharya and Harrison (1976)
utilized this approach to account for radiation
exchange from one layer of particles with 25 neighbor-
ing layers. The particles were treated as an absorbing
and emitting medium so that radiation was attenuated
exponentially with distance. A more rigorous model
was presented by Chen and Chen (1981), whereby
the MickleyFairbanks packet model was modied to
include simultaneous radiative and conductive heat
transfer during alternating contact of the heat transfer
surface by gas bubbles and particle packets. The gas
phase was taken to be transparent to thermal
radiation, while the particle packet was treated as a
radiatively participative medium with absorption,
emission, and scattering of photons. During bubble
contact, radiation was directly exchanged between par-
allel surfaces representing the heat transfer wall and
the boundary of the bubble. During packet contact,
Hamakers two-ux formulation of radiant transport
was used to describe the absorption, scattering, and
emission process within the packet:
dI
dy
= (A S)I SJ AsT
4
dJ
dy
= (A S)J SI AsT
4
(46)
where
I; J =forward and backward radiant uxes,
respectively
A = volumetric absorption coecient
S = volumetric scattering coecient
The one-dimensional transient energy equation com-
pleted the system of equations
k
pa
@
2
T
@y
2
A(I J)2AsT
4
= r
s
c
ps
(1 e)
pa
@T
@t
(47)
Radiant heat ux from the bed into the wall surface is
then
q
r
a
w
= J I (48)
The radiation cross sections of the particle packets
(A; S) are required. These would have to be measured,
as done by Cimini and Chen (1987), or estimated from
particle emissivity using the approach of Brewster and
Tien (1982). Numerical solutions of this model have
been shown to be in good agreement with experimental
data (see Chen and Chen, 1981). While this approach is
more rigorous, it requires signicant eort and would
be appropriate only for situations wherein radiative
heat transfer is of special importance. In most design
applications, the rst approach, using Eqs. (43)(45), is
much easier and would be sucient for the purpose.
Freeboard Heat Transfer. In some applications,
e.g., uidized combustors, heat exchanger tubes are
located in the freeboard space above the bed, as well
as within the bed. The coefcients h
w
, h
c
, and h
r
, de-
ned by Eqs. (15, 16, and 42), are also appropriate
to represent the heat transfer process at the surface
of such freeboard tubes. If the tube is placed so that
it sees mostly bed material (not the vessel walls), Eqs.
(44) and (45) may be used to estimate the radiative
contribution. Evaluation of the convective coefcient
is more difcult, since this must account for the phe-
nomenon of particles splashing from the uidized
bed into the freeboard space, some to fall back into
the bed and the rest being entrained out by exiting
gas. The concentration of particles in the freeboard
space decreases with increasing elevation above the
bed, increasing particle size, and decreasing gas velo-
city. The presence of these particles enhances gaseous
convection so that the magnitude of h
c
varies in a
similar manner.
Experimental data and models for the heat transfer
coecients in freeboards space are limited. Biyikli et al.
(1983) measured heat transfer coecients on horizon-
tal tubes in freeboards of uidized beds with particle
mean diameters ranging from 275 to 850 mm. These
investigators discovered that data for particles of dif-
ferent sizes and materials could be unied into a family
of curves by plotting the normalized ratios of heat
transfer coecients against dimensionless velocity:
h
+
c

h
c
h
g
h
cb
h
g
versus U
+

U
g
U
mf
U
t
U
mf
(49)
where
h
c
= the convective coecient in freeboard
h
g
= the single-phase gas convective coecient at
velocity U
h
cb
= the convective coecient for a surface sub-
merged in the bed
Standard correlations may be used to obtain the
single-phase gas convective coecient, h
g
. For the
common case of horizontal tubes, the correlation of
Douglas and Churchill (1956) is recommended:
h
g
D
t
k
g
= 0:46
D
t
r
g
U
g
m
g
_ _
1=2
0:00128
D
t
r
g
U
g
m
g
_ _
for
D
t
r
g
U
g
m
g
_ _
_ 50 (50)
In Eq. (50), gas properties should be evaluated at lm
temperature (average of bed and wall temperatures) in
cases where the temperature dierence is large.
Figure 4 reproduces the results of Biyikli et al
(1983), showing that h
+
c
is independent of particle
material and size. h
+
c
is dependent only on U
+
and
the elevation H in freeboard, as measured from the
surface level of the collapsed (unuidized) bed.
Because the particle splashing process is aected by
many parameters (e.g., distribution of particle sizes,
bubble coalescence, etc.), it is dicult to nd a general
correlation of h
c
for the freeboard region. As a rst
estimate, the following empirical correlation, based
Figure 4 Normalized heat transfer coecients for horizontal tube in freeboard of bubbling beds. H measured from level of
collapsed (packed) bed. (Data of Biyikli et al., 1983.)
on data of Biyikli et al. (1983), may be utilized for
horizontal tubes in freeboard:
h
+
c
= 1:3e
1:53H=U
+
(51)
with the constraint that 0 _ h
+
c
_ 1:0. Note that the
constant 1.53 has the dimension of (1/m), with H
measured in meters from the level of the collapsed
(unuidized) bed. Equation (51) can be used also for
tube bundles if the bundle has open spacing, i.e., on a
pitch-to-diameter ratio greater than 3.
3 FAST CIRCULATING FLUIDIZED BEDS
(FFBs)
3.1 Hydrodynamics
When the gas velocity in a uidized bed exceeds the
terminal velocity of the bed particles, upward entrain-
ment of particles out of the bed occurs. To maintain
solid concentration in the uidized bed, an equal ux
of solid particles must be injected at the bottom of the
bed as makeup. Operation in this regime, with
balanced injection and entrainment of particles, is
termed fast or circulating uidization. Figure 1 pre-
sents a generalized map of this fast uidization regime,
in terms of the dimensionless particle diameter and gas
velocity, d
+
p
and U
+
. Lines (c) and (d) indicate the lower
and upper bounds of gas velocity, respectively, for the
usual FFB operation. It is seen that for Geldart types
A and B particles, fast uidization permits supercial
gas velocities an order of magnitude greater than that
for bubbling uidization. In the case of catalyst parti-
cles (density of 1800 kg/m
3
), supercial gas velocities
approaching 10 m/s may be encountered. Heat transfer
in fast uidized beds is strongly aected by the volu-
metric concentration of solid particles as they are
transported by the gas through the bed. For this
reason, we start with a review of the hydrodynamic
characteristics of FFBs.
In fast uidized beds, solid particles are fed into the
bed at the bottom, to be entrained by the upward ow-
ing gas. The solid and gas phases pass through the bed
in cocurrent two-phase ow. Since the particles enter
with little vertical velocity, they are accelerated by
shear drag of the gas, gaining velocity by momentum
transfer from the high-speed gas. This hydrodynamic
development results in the volumetric concentration of
solids decreasing with axial elevation along the length
of the bed. Measurements obtained by Herb et al.
(1989) and Hartige et al. (1986) indicate that many
meters of axial length may be required to approach
fully developed ow conditions (i.e., constant solids
concentration). The axial height required for this
hydrodynamic development has been found to increase
with increasing particle size, solid mass ux, and bed
diameter. Cross-sectional-averaged solid concentra-
tions typically range from 15 volume % at bed bot-
tom to less than 3% at bed top. This is signicantly less
than the solids concentrations of 3040% commonly
encountered in bubbling uidized beds.
In addition to the axial variation of solid concentra-
tion, fast uidized beds have the complication of sig-
nicant radial (horizontal) variation in solid
concentration. Radial proles of solid volume fraction
measured by Beaude and Louge (1995) indicate there is
increasing nonuniformity of solid concentration with
increasing solid mass ux. At the higher mass uxes,
local solid volume fraction adjacent to the bed wall
approach a magnitude of 30%, in contrast to solid
fractions of 13% near the bed centerline. Herb et al.
(1989b) and Werther (1993) noted that such radial dis-
tributions could be normalized as a general similarity
prole for various operating conditions by utilizing the
dimensionless ratio of local solid concentration to
cross-sectional-averaged solid concentration.
Solid mass ow ux and velocity also vary across
the radius of fast uidized beds. Experimental mea-
surements obtained by Herb et al. (1992) show that
while local solid uxes are positive upward in the
core of the bed, they can become negative downward
in the region near the bed wall. The dierence between
core and wall regions becomes increasingly greater as
total solid mass ux increases. The downward net ow
of solid in the region near the bed wall has signicance
for heat transfer at the wall.
Another hydrodynamic complication found in fast
uidized beds is the tendency for particles to aggregate
into strands or clusters, as reported by Horio et al.
(1988) and Chen (1996). The concentration of solid
particles in such clusters is signicantly greater than
in the bed itself, and it increases with increasing radial
position and with increasing total solid ux (see Soong
et al., 1993). This characteristic also directly aects
heat transfer at walls.
The above is a very condensed discussion of the
hydrodynamic characteristics of fast uidized beds.
As presented below, heat transfer is strongly dependent
on the time-averaged local concentration of solid
particles and is therefore inuenced by these hydrody-
namic characteristics. Almost all the heat transfer
models require information on solid concentration.
The reader is referred to Chapter 19 of this book for
more detailed discussion of the hydrodynamics, and
for models for predicting solid concentrations in fast
circulating uidized beds.
3.2 Heat Transfer
3.2.1 Particle Gas Transfer
There is little information on the process of heat
transfer between particles and gas in fast uidized
beds. One reason for this situation is the expectation
that the heat transfer rates are high and therefore of
little concern. The smaller concentration of particles
(in comparison to bubbling dense beds) implies less
intense heat source concentration in the cases of
exothermic reactions, and this also alleviates potential
concern. At the time of this writing, there is no estab-
lished model or correlation for estimation of the heat
transfer coecient between particles and gas during
fast uidization.
Given this situation, one possible approach is to
treat the particles as if they were individually
suspended by the upward owing gas. With this sim-
plication, the relative velocity between particle and
gas would be equal to the terminal velocity U
t
, as
given by Eq. (6). The heat transfer coecient based
on particle surface area, as dened by Eq. (8), would
be given by Ranzs correlation (Eq. 11). In applying
this correlation for suspended particles, the particle
Reynolds number should be based on the terminal
velocity rather than the supercial gas velocity,
Re
t

d
p
r
g
m
g
U
t
(52)
By using U
t
rather than U
g
, the eective Reynolds
number is reduced so that the calculated Nusselt num-
bers tend toward the conservative conduction limit of
2. Obviously this approach greatly simplies the actual
phenomena, and disregards the complications of par-
ticle acceleration, clustering, and downow at walls.
Estimates of the particle/gas heat transfer coecient
(h
p
) from Eqs. (11) and (52) should be considered
only as approximate.
3.2.2 BedSurface Transfer
Similar to bubbling beds, fast uidized beds often
require cooling or heating in order to maintain a
steady thermal state. Unlike bubbling beds, the use
of submerged heat exchanger tubes is not common
practice in FFBs, because of concern about tube ero-
sion by the faster gas/particle ows. The usual practice
is to use walls of the FFB as the heat exchange sur-
faces, by mounting cooling/heating tubes in the walls.
Additionally, membrane walls of vertical tubes may
be placed within the fast uidized bed, to increase total
heat transfer area. In either case, engineering design is
concerned with the heat transfer coecient for vertical
surfaces in contact with the gas/particle medium in the
fast uidized bed, h
w
as dened by Eq. (14).
Experimental measurements of this heat transfer
coecient have been obtained by many investigators,
including Kiang et al., 1976; Furchi et al., 1988; Dou
et al., 1991, 1993; Ebert et al., 1993; and Wirth, 1995.
Parameters that have been found to aect h
w
include
gas velocity, solids ow rate (mass ux), mean parti-
cle diameter, axial location of the heat transfer sur-
face along the height of the uidized bed, radial
location of the heat transfer surface across the bed,
and physical dimension of the heat transfer surface.
General characteristics noted from the collected data
include
h
w
is higher than that for gas convection at the same
velocity but lower than that found in dense
bubbling beds.
h
w
decreases with increasing particle diameter.
h
w
increases with increasing solid mass ux.
h
w
increases with increasing radial position toward
the bed wall.
h
w
decreases with increasing elevation along the
bed.
At low elevations, h
w
decreases with increasing gas
velocity.
At higher elevations, h
w
increases with increasing
gas velocity.
h
w
is greater for small heat transfer surfaces, than
that for large surfaces.
Most of these parametric eects parallel those for solid
concentration and therefore imply a strong dependence
of h
w
on local concentration of solid particles.
Many dierent models and correlations have been
proposed for the prediction of the heat transfer
coecient at vertical surfaces in FFBs. At time of
this writing, no single correlation or model has won
general acceptance. The following discussion presents
a summary of some potentially useful approaches. It is
helpful to consider the total heat transfer coecient as
composed of convective contributions from the lean-
gas phase and the dense-particle phase plus thermal
radiation, as dened by Eqs. (15) and (16). All correla-
tions based on ambient temperature data, where
thermal radiation is negligible, should be considered
to represent only the convective heat transfer coe-
cient h
c
.
Convective Heat Transfer. The simplest correla-
tions for the convective heat transfer coefcient em-
pirically relate h
c
to the solids concentration,
expressed in terms of the mixture density for the gas/
solid medium in the bed (r
b
),
r
b
r
s
(1 e)
b
r
g
e
b
= r
s
e
s;b
r
g
e
b
(53)
Since gas density is usually negligible compared to
solid density, r
b
is essentially equal to the mass con-
centration of solid per unit mixture volume. Some
examples of this type of correlation, for convective
heat transfer at vertical walls of fast uidized beds,
are given below. Due to accumulating evidence that
the vertical size of the heat transfer surface aects
the average h
c
, the correlations are divided with regard
to short or long surfaces.
For short surfaces of L
h
_ 0:5 m, the correlation of
Wen and Miller (1961) is
h
c
d
p
k
g
=
c
ps
c
pg
_ _
r
b
r
s
_ _
0:3
U
t
gd
p
_ _
0:21
Pr
g
(54)
and the correlation of Divilio and Boyd (1993) is
h
c
= 23:2(r
b
)
0:55
(55)
For long surfaces of L
h
> 0:5 m, the correlation of
Werdmann and Werther (1993) is
h
c
d
p
k
g
= 7:46 10
4
D
b
r
g
U
g
m
g
_ _
0:757
r
b
r
s
_ _
0:562
(56)
and the correlation of Fraley et al. (1983) is
h
c
= 2:87(r
b
)
0:9
(57)
To use any of these correlations, one needs the value of
the cross-sectional-averaged bed density r
b
. Methods
for estimating this parameter are discussed in Chapter
19 of this book. As a precautionary note, Dou et al.
(1991) have shown the heat transfer coecient to be
dependent on local solid concentration in the vicinity
of the surface (r
m
), rather than the cross-sectional-
averaged concentration (r
b
). Therefore the above cor-
relations for convective coecient at the wall of FFBs
can be justied only if r
b
and r
m
are related by a
generic similar prole. The existence of such a similar
prole has been suggested by Herb et al. (1989b) and
Werther (1993). Examination of Eqs. (54) and (57) also
indicates signicant dierences regarding the para-
meters appropriate for the correlation of h
c
. Particle
size (d
p
), bed diameter (D
b
), particle terminal velocity
(U
t
), heat capacity of solid (c
ps
), and gas thermal prop-
erties (m
g
, k
g
, r
g
, c
pg
) are incorporated into some cor-
relations but not in others. Given this situation, the
user is cautioned to refer to the original references
and apply a given correlation only within the range
of parameters covered by that correlations experimen-
tal data base.
In contrast to the purely empirical correlations
given above, an alternate modeling approach is sug-
gested by Molerus (1993),Wirth (1993, 1995), and
Molerus and Wirth (1997). These researchers reasoned
that the wall-to-bed heat transfer in FFBs is governed
by the uid ow immediately near the wall. They
assumed existence of a thin layer near the wall with
low solids concentration, providing signicant thermal
resistance between the wall and the falling strands of
relatively high solids concentrations. Wirth and
Molerus report that a dimensionless pressure-drop
number that measures bed-averaged solids concen-
tration and the Archimedes number are sucient to
characterize this gasparticle ow and to correlate
the convective heat transfer coecient. At low
Archimedes numbers, heat transfer occurs primarily
by gas conduction. At higher Archimedes numbers,
gaseous conduction and convection both contribute
to heat transfer. The correlation suggested by Wirth
(1995) simply assumed that conduction and convection
were additive, which results in the equation
h
c
d
p
k
g
= 2:85
P=L
b
g(r
s
r
g
)(e
s;mf
)
_ _
0:5
3:28 10
3
Re
pa
Pr
g
(58)
where Re
pa
= Reynolds number for falling cluster
packets = r
g
d
p
V
pa
=m
g
and V
pa
= velocity of falling
cluster packets. Based on physical reasoning, Wirth
(1990) obtained the following relationship for the
Reynolds number of falling clusters:
Re
pa
=
2(1 f)
2
Re
mf
fRe
t
2[f (1 f)e
mf
]
1 4
F
1
F
2
_ _
_
(59)
where
F
1
(1 f)
4
Re
2
mf
f(1 f)
2
Re
mf
Re
t
_
1891(1 e
mf
)f
3
(1 f)Ar (60)
F
2
2(1 f)
2
Re
mf
fRe
t
_ _
2
(61)
f 1
2:3P=L
b
g(1 e
mf
)(r
s
r
g
)
(62)
In contrast to the above correlations for total con-
vective coecient h
c
, some models estimate the sepa-
rate heat transfer contributions from lean and dense
phases as represented by h
l
and h
d
in Eq. (15). It is
generally believed that dense phase convection is the
dominant mechanism, and lean phase convection is a
relatively unimportant mechanism. This simplication
is valid where local solids concentration is greater than
5% by volume, as at the bottom regions of FFBs, close
to vertical walls, or in operations with high mass ux
of solid ow. Measurements of Ebert et al. (1993)
indicate that the lean phase convection can become
signicant, contributing up to 20% of total convective
heat transfer, when the average solids concentration
becomes less than 3%, as is often found in upper
regions of fast uidized beds. Rather than neglecting
lean phase convection, it is better to include an esti-
mate of h
l
in total convection. A conservative estimate
for h
l
is the DittusBoelter correlation for single phase
gas convection,
h
l
D
b
k
g
= 0:023
D
b
r
g
U
g
m
g
_ _
0:8
Pr
g
_ _
0:33
(63)
There is evidence that this approach underestimates
the lean phase convection, since gaseous convection
is enhanced by suspended particles. Lints (1992) sug-
gested that this enhancement could be partially taken
into account by increasing the gas thermal conductivity
in Eq. (63) by a factor of 1.1.
To estimate the contribution of dense phase convec-
tion, one approach considers the heat transfer surface
to be contacted periodically by clusters or packets of
closely packed particles. Experimental evidence sup-
porting this concept is reported by Dou et al. (1993).
Their dynamic measurements at the wall of a fast
uidized bed found signicant correlation between
instantaneous solids concentration and heat transfer
coecient, indicating a direct dependence of the
dense phase heat transfer on cluster contact. This
approach leads to a surface renewal model whereby
the heat transfer to/from the dense phase occurs by
transient conduction between the heat transfer surface
and the particle clusters during their time of residence
at the surface. This is analogous to the packet theory
for heat transfer in bubbling beds, and so Eqs. (30)
(35) can be utilized to calculate the dense phase con-
vective coecient by taking h
d
= h
pa
. Application of
this cluster model requires information on solids
concentration in the cluster (e
s;pab
) and the eective
thermal conductivity of the cluster (k
pa
). Unlike the
emulsion phase in BFBs, which have solids fractions
close to that at minimum uidization (e
s;mf
), clusters in
FFBs have solids fractions signicantly less than e
s;mf
.
Experimental data for e
s;pab
are very sparse. Some lim-
ited data may be found in the articles of Wu, 1989;
Louge et al., 1990; Lints, 1992; and Soong et al.,
1993
.
From the available data, Gu and Chen (1998)
developed the following correlation to estimate the
solids concentration in clusters (e
s;pab
) as a function
of the local solids concentration in the bed medium
(e
s;m
),
e
s;pab
= 0:57 1 1
e
s;m
0:57
_ _
3:4
_ _
(64)
Once the solids concentration in clusters is determined,
the thermal conductivity of clusters (k
pa
) can be esti-
mated by the expression of Chandran and Chen (1985)
as
k
pa
= k
pab
C
2
(65)
where k
pab
is given by Eq. (34) with
e
pab
= 1 e
s;pab
_ _
(66)
and C as given by Eq. (33). The result is same as Eq.
(32), written for the dense-phase convective coecient
in FFB,
h
d
= 2C
k
pab
r
s
c
ps
e
s;pab
pt
pa
_ _
1=2
(67)
This surface renewal model also requires the average
residence time of clusters on the heat transfer surface
(t
pa
). Unfortunately, there is insucient information
to correlate this residence time directly with operating
parameters (e.g. U
g
, D
p
, r
s
, G
s
, L
b
). If the average
cluster falls down along the heat transfer surface with
a velocity V
pa
, over a contact length L
c
, the residence
time for that cluster would be the ratio L
c
=V
pa
. This
would be the eective mean value of t
pa
on a large heat
transfer surface. Cluster velocities measured by Hartge
et al. (1988), Horio et al. (1988), Lints (1992), and
Soong et al. (1995) range from 0.3 to 3 m/s. In absence
of specic information, it is suggested that Wirths
correlation, Eq. (59), be used to calculate the
Reynolds number for falling clusters (Re
pa
), from
which V
pa
can be obtained from the denition of Re
pa
.
Measurements of contact length L
c
are lacking, but
an inference can be drawn from the experiments of
Burki et al. (1993) in fast uidized beds of sand and
catalyst particles. They found that local heat transfer
coecients on a vertical wall decreased with distance
down the heat transfer surface, over an entry length
of approximately 0.30.5 m. This result implies that L
c
is approximately 0.4 m, for their operating conditions.
At this stage of development, we can only take an
approximate estimate for the cluster contact time:
t
pa
=
L
c
V
pa
=
L
c
Re
pa
r
s
d
p
m
g
_ _
(68)
with L
c
0:4 m and Re
pa
from Eq. (59). Clearly this
will require additional research. It should be noted that
in cases where the vertical length of the heat transfer
surface is less than the contact length (L
b
< L
c
), L
h
should be substituted for L
c
in Eq. (68).
Finally, to utilize this approach in the prediction of
the total convective heat transfer coecient h
c
, one
needs to determine the average fraction of wall surface
covered by the dense phase (clusters). By denition this
is equal to the fraction of time that a spot on the heat
transfer surface is covered by the dense phase, and
denoted as (1 f
l
) in Eq. (16). Measurements of
(1 f
l
) have been obtained by Wu et al. (1989),
Louge et al. (1990), Rhodes et al. (1992), and Dou
et al. (1993). Figure 5 plots (1 f
l
) against the cross-
sectional-averaged bed concentration of solids (e
s;b
),
showing the range of experimental data. The following
equation gives an empirical t to these data and may
be used to estimate (1 f
l
) for vertical walls in fast
uidized beds.
1 f
l
= 1 0:018
1
e
s;b
_ _
0:6
for 0:002 _ e
s;b
_ 0:3
(69)
In summary, the various variables required for the
estimation of the convective heat transfer coecient
(h
c
) by the surface renewal model, for vertical walls
in FFBs, may be determined by the equations listed
in following tabulation:
Figure 5 Fraction of wall surface in FFBs contacted by dense phase.
Variable Equation No.
h
c
16
h
l
63
h
d
67
k
pab
34
e
s;pab
64
t
pa
68
f
l
69
Radiative Heat Transfer. Models for estimating
the radiative heat transfer in fast uidized beds
parallel those proposed for dense bubbling beds, as
reviewed above. From a phenomenological point of
view, the suspension of particles in gas acts as a ra-
diatively participative medium, with absorption, scat-
tering, and emission of radiant energy. Only a few
researchers have attempted to model this physics in
FFBs. Chen et al. (1988) used the two-ux radiative
model of Eqs. (46)(48) to analyze simultaneous con-
vection and radiation in high temperature FFBs. Tur-
bulent convection was combined with discrete
radiative uxes in the three transport equations, in-
cluding a term for volumetric heat generation.
Sample results for uidized combustor conditions
predicted a signicant interdependence of radiation
and convection. This conclusion was veried by the
experimental measurements of Han (1992), who
showed that the contribution of radiative heat trans-
fer was strongly effected by bed-averaged solid con-
centration, decreasing with increasing solid mass ux
and decreasing gas velocity. While this participative
media model is mechanistically sound, it requires in-
formation on the absorption and scattering coef-
cients for the particle/gas medium and is relatively
difcult to apply. Its use is recommended only for
cases where detailed estimates of local radiative trans-
fer is desired; in such cases the reader is referred to
the papers of Chen et al. (1988) and Radauer et al.
(1996) for the necessary equations and methods of
solution.
The more common approach for engineering design
is to treat the particlegas suspension as an equivalent
gray surface parallel to the heat transfer surface; see,
for example, Palchonok et al. (1995), Anderson and
Lechner (1993), and Mahalingam and Kolar (1991).
In this approach, Eqs. (42)(45) are used to calculate
the radiative heat ux (q
r
) and the coecient (h
r
). It is
necessary rst to estimate the eective bed emissivity
(e
b
) in FFBs. Grace (1986b) suggests that the emissiv-
ity of the particlegas suspension can be approximated
as
e
b
= 0:5(1 e
s
) (70)
where e
s
is the surface emissivity of the solid particles.
This is a reasonable approximation for large FFBs,
where the mean free path of radiation photons is
short compared to bed dimensions, i.e., the gas/particle
medium is optically thick. Experimental measure-
ments reported by Han et al. (1996) and Han (1992)
indicate that when the particle/gas medium is not opti-
cally thick, the bed emissivity is dependent on the solid
concentration and bed dimension, as well as on particle
emissivity. From the theory of radiant transmission
through absorbing media, the eective emissivity of
the bed would be expected to be of functional form
e
b
= 1 exp
C
e
e
s;b
e
s
D
b
d
p
_ _
(71)
where e
s;b
is the solid volume fraction in the bed, D
b
is
the bed diameter between opposing heat transfer sur-
faces, and C
e
is a dimensionless parameter. Figure 6
plots the range of data measured by Han (1992) for
sand particles of e
s
= 0:7, in an FFB of D
b
= 5 cm. A
value of 0.95 for C
e
in Eq. (71) is seen to t the data
and is recommended for design calculations.
One cautionary note should be kept in mind when
using Eqs. (42)(45) and (71) to calculate radiative heat
transfer in FFB. The beds absolute temperature T
b
is
normally assumed to be uniform across the bed and is
used as the source or sink temperature in Eqs. (42) and
(43). This assumption may be inappropriate in those
cases in which a dense annular region of particles
shields the FFB wall from the bulk bed. In such situa-
tions, it is the average temperature of the particles in
the annular layer that should be taken as the source/
sink temperature for calculation of radiant heat ux
to/from the wall. This requires a mass and heat balance
analysis for the material owing in the annulus, and
the reader is referred to Chapter 19 for necessary
hydrodynamic models.
4 EXAMPLES
4.1 Bubbling Fluidized Bed
Consider a bed of diameter 0.203 m and loaded with
spherical glass beads of mean diameter 125 mm. We
wish to uidize the particles with ambient air, at a
supercial gas velocity of U
g
= 0:16 m=s. Properties
of the two phases are
Gas Particles
Density, r (kg/m
3
) 1.18 2,480
Heat capacity, c
p
(kJ/kg K) 1.00 0.753
Thermal conductivity, k (W/m K) .0262 0.890
Viscosity, m (kg/m s) 1:85 10
5
Pr 0.71
According to Geldarts classication, these particles
are of type B. The void fraction at minimum uidiza-
tion is found to be e
mf
= 0:42.
4.1.1 Fluidization Regime
At the intended operating condition, would the bed
uidize in the bubbling regime?
The answer is found by spotting the operating point
on the map of uidization regimes plotted in Fig. 1.
Calculating the necessary parameters, we nd the
Archimedes number from Eq. (3):
Ar =
gr
g
(r
s
r
g
)
m
2
g
d
3
p
=
(9:81)(1:18)(2480 1:18)(125 10
6
)
3
(1:85 10
5
)
2
= 164
The Reynolds number at minimum uidization is
found by solving Eq. (1):
Ar = 1:75
(Re
mf
)
2
f
s
(e
mf
)
3
150
(1 e
mf
)(Re
mf
)
(f
s
)
2
(e
mf
)
3
164 = 1:75
(Re
mf
)
2
(1:0)(0:42)
3
150
(1 0:42)(Re
mf
)
(1:0)
2
(0:42)
3
Re
mf
= 0:139
from which we obtain the minimum uidization
velocity by Eq. (2),
U
mf
= Re
mf
m
g
r
g
d
p
_ _
= 0:139
1:85 10
5
1:18(125 10
6
)
_ _
= 0:0175 m=s
The terminal velocity is obtained from Eq. (6b),
U
t
= 0:072g
d
8=5
p
(r
s
r
g
)
r
2=5
g
m
3=5
g
_ _
5=7
= 0:072 9:81
(125 10
6
)
8=5
(2480 1:18)
(1:18)
2=5
(1:85 10
5
)
3=5
_ _
5=7
= 0:730 m=s
and the dimensionless velocity is, by denition, from
Eq. (7),
U
+

U
g
U
mf
U
t
U
mf
=
0:160 0:0175
0:730 0:0175
= 0:20
The dimensionless particle diameter is, from Eq. (4),
d
+
p
Ar
1=3
= 164
1=3
= 5:46
This locus is within the bubbling regime on the map of
Fig. 1, so the proposed operation will be a dense
bubbling uidized bed.
Figure 6 Eective emissivities for fast uidized beds for particles of e
s
= 0:7. (Data of Han et al., 1992, 1996.)
4.1.2 ParticleGas Heat Transfer
If the gas enters the bed 30K cooler than the tempera-
ture of the particles in the bed, how long a distance
must the gas travel within the bed before that dier-
ence is reduced to 0.5K? At the supercial gas velocity
of U
g
= 0:16 m/s, take the average void fraction in the
bubbling bed to be e
b
= 0:585, and the average solid
fraction to be e
s;b
= 0:415:
The heat transfer coecient between particle and
gas is given by Eq. (12) or Eq. (13), depending on
the particle Reynolds number, which is dened by
Eq. (10),
Re
p

U
g
d
p
r
g
m
g
=
0:16 125 10
6
1:18
1:85 10
5
= 1:28
For this small Re
p
Eq. (12) is appropriate
h
p
=
k
g
d
p
Nu
p
=
0:0262
125 10
6
0:0282Re
1:4
p
Pr
0:33
g
_ _
=
0:0262
125 10
6
0:0282 1:28
1:4
0:71
0:33
_ _
= 7:46 W=m
2
K
Using the plug-ow model for gas, and assuming the
particles to be well mixed, the energy balance for the
gas gives
r
g
c
pg
U
g
dT
g
= h
p
a
p
n
p
(T
p
T
g
) dL
b
where
a
p
= surface area of single particle = pd
2
p
n
p
= number of particles per unit bed volume =
6(1 e)=(pd
3
p
)
a
p
n
p
=6(1 e)=d
p
= 6(1 0:585)=(125 10
6
)
= 1:99 10
4
particles/m
3
.
Integration of the energy balance equation results in
the following expression for length (L
b
) required to
change gas temperature from T
1
to T
2
:
L
b
=
r
g
c
pg
U
g
a
p
n
p
h
p
ln
T
p
T
1
T
p
T
2
_ _
=
1:18 1; 000 0:16
1:99 10
4
7:42
ln
30
0:5
= 5:23 10
3
m
Thus in 5.2 mm, the gas reaches within
1
2
degree K of
the particle temperature. For all practical purposes,
the bed of gas and particles can be considered to be
isothermal.
4.1.3 Bed-to-Tube Heat Transfer
If a heat transfer tube of a 2.9 cm diameter is sub-
merged horizontally within the uidized medium at
the above operating conditions, what would be the
surface heat transfer coecient estimated by various
empirical correlations?
Since the bed temperature is low, radiative heat
transfer may be neglected; thus only the convective
coecient needs to be calculated. For a horizontal
tube, the correlations of Vreedenberg (1958),
Borodulya (1991), and Molerus and Schweinzer
(1989) can be utilized. For Vreedenbergs correlation,
we rst need to calculate the quantity.
r
s
r
g
Re
p
_ _
=
2480
1:18
1:28 = 2,690 _ 2550
At this magnitude, Eq. (20a) is appropriate.
h
c
D
t
k
g
= 420
r
s
r
g
Pr
g
m
2
g
gr
2
s
d
3
p
_ _ _ _
0:3
Re
0:3
D
= 420
2480
1:18
0:71
_
(1:85 10
5
)
2
9:81 2480
2
(125 10
6
)
3
_
0:3
Re
0:3
D
= 82:1 Re
0:3
D
where
Re
D

D
t
r
g
U
g
m
g
=
0:029 1:18 0:16
1:85 10
5
= 296
and the heat transfer coecient is obtained as
h
c
=
k
g
D
t
82:1Re
0:3
D
_ _
=
0:0262
0:029
82:1 296
0:3
= 409 W=m
2
K
The correlation of Borodulya et al. (1991), Eq. (29), is
also appropriate for this case:
h
c
d
p
k
g
= 0:74Ar
0:1
r
s
r
g
_ _
0:14
c
ps
c
pg
_ _
0:24
e
2=3
s
0:46Re
p
Pr
g
e
2=3
s
e
= 0:74 164
0:1
2480
1:18
_ _
0:14
0:753
1
_ _
:24
0:415
2=3
0:46 1:28 0:71
0:415
2=3
0:585
= 1:87 0:398 2:27
which shows that particle convection (rst term)
dominates over gas convection (second term).
The resulting heat transfer coecient is then
calculated as
h
c
= 2:27
k
g
d
p
= 2:27
0:0262
125 10
6
= 476 W=m
2
K
The correlation of Molerus et al. (1995) calls for
Eq. (23). The necessary parameters are given by Eqs.
(24)(28):
U
e
U
g
U
mf
= 0:160 0:0175 = 0:143 m=s
l =
m
g
r
s
r
g
_ _
2=3
1
g
_ _
1=3
=
1:85 10
5
2480 1:18
_ _
2=3
1
9:81
_ _
1=3
= 1:77 10
6
m
B
1
1 33:3
U
e
r
s
c
ps
U
mf
gk
g
_ _
1=3
U
e
_ _
1
= 1 33:3
0:143 2480 753
0:0175 9:81 0:0262
_ _
1=3
0:143
_ _
1
= 1:60
B
2
1 0:28e
2
s;mf
U
e
U
mf
r
g
r
s
r
g
_ _
1=2
r
s
c
ps
gk
g
_ _
2=3
= 1 0:28 0:58
2
0:143 0:0175
1:18
2480 1:18
_ _
1=2
2480 753
9:81 0:0262
_ _
2=3
= 1:194
B
3
1 0:05
U
mf
U
e
_ _
= 1 0:05
0:0157
0:143
_ _
= 1:005
and the Nusselt number is obtained from Eq. (23),
h
c
l
k
g
=
0:125e
s;mf
B
1
1 B
2
k
g
=2c
ps
m
g
_ _ _ _ 0:165Pr
1=3
g
r
g
r
s
r
g
_ _
1=3
1
B
3
_ _
h
c
l
k
g
=
0:125 0:58
1:60 1 1:194 0:0262=2 753 1:85 10
5
_ _ _ _
0:165 0:71
1=3
1:18
2480 1:18
_ _
1=3
1
1:005
_ _
= 0:0328
which gives the value for the heat transfer coecient
h
c
= 0:0328
k
g
l
= 0:0328
0:0262
1:77 10
6
= 486 W=m
2
K
4.1.4 Heat Transfer Coefcient from Packet Model
Use the packet model to estimate the convective heat
transfer coecient for the above case of a horizontal
tube in a bubbling bed. Additionally, the bed has a
porous plate distributor, and the center line of the
heat transfer tube is located of L
t
= 0:19 m above the
distributor.
To apply the packet model, we rst need to calculate
some bubble characteristics, utilizing hydrodynamic
theory. The necessary equations can be obtained
from Chapter 3, or from the book of Kunii and
Levenspiel (1991). First, to estimate the bubble dia-
meter (d
b
) at the elevation of the tube,
d
b
= d
bm
d
bm
d
bo
e
0:3L
t
=D
b
where
d
bo
= 2:78(U
g
U
mf
)
2
g = 2:78(0:16 0:0175)
2
9:81 = 0:00575 m = 0:575 cm
d
bm
= 0:65
pD
2
b
(U
g
U
mf
)
4
_ _
0:4
= 0:65
p 20:3
2
(16 1:75)
4
_ _
0:4
= 19:0 cm
so that
d
b
= d
bm
d
bm
d
bo
e
0:3L
t
=D
b
= 19:0
19:0 0:575
e
0:3(0:19=0:203)
= 5:09 cm = 0:0509 m
The rise velocity of the bubble, relative to the gas/
particle emulsion, is
u
br
= 0:853
gd
b
_
e
1:49d
b
=D
b
_ _
= 0:853
9:81 0:0509
_
e
1:490:0509=0:203
_ _
= 0:415 m=s
and the actual velocity of the bubble approaching the
tube is
u
b
= U
g
U
mf
u
br
= 0:160 0:0175 0:415
= 0:558 m=s
Since U
g
U
mf
, the bed is vigorously bubbling and
the volume fraction of the bubbles in the bed is
d
w
=
U
g
u
b
=
0:160
0:558
= 0:287
The frequency of bubbles at moderate-to-high eleva-
tions in a bubbling bed is 2 per second. Since bubbles
are equally likely to divert to one or the other side of a
horizontal tube, the frequency of bubble contact by
any spot on the tube would be approximately 2/2, or
1 per second. Equation (36) then estimates the average
residence time of particle packets at the tube surface as
t
pa
=
1 d
w
f
b
=
1 0:285
1:0
= 0:715 s
Next we calculate the eective thermal conductivity of
the bulk packet by Eqs. (35) and (34):
f
k
= 0:305
k
s
k
g
_ _
0:25
= 0:305
0:89
0:0262
_ _
0:25
= 0:126
k
pab
= k
g
e
pab
(1 e)
pab
f
k
2k
g
=3k
s
_ _
= 0:0262
0:42
1 0:42
0:126 2 0:0262=3 0:89
_ _
= 0:115 W=ms
Knowing the residence time, the Fourier number for
transient conduction into the packet is calculated by
Eq. (33a),
Fo =
k
pab
t
pa
c
ps
r
s
(1 e)
pab
d
2
p
=
0:115 0:715
0:753 2480(1 0:42)(125 10
6
)
2
= 4860
For this large Fourier number, Eq. (33) shows C 1:0.
The packet model, Eq. (32), then gives a value for the
convective heat transfer coecient:
h
c
h
pa
= 2C
k
pab
r
s
c
ps
(1 e)
pab
pt
pa
_ _
1=2
= 2 1
0:115 2480 753 0:58
p 0:715
_ _
1=2
= 466 W=m
2
K
4.1.5 Heat Transfer Coefcient from Kinetic Model
Martins (1984) model uses an analogy to the kinetic
theory of gases to calculate the convective heat transfer
coecient for the particle phase in uidized beds.
Using this model, predict h
c
for above case.
First, we calculate the Knudsen number for a gas in
the gap between particle and wall, using Eq. (40),
Kn =
4k
g
(2pRT=M
g
)
1=2
Pd
p
(2c
pg
R=M
g
)
where, for ambient air,
M
g
= 0.029 kg/mol
T = 293K
P = 1:01 10
5
Pa
and the gas content is R = 8:31 J mol
1
K
1
.
Hence
Kn =
4k
g
(2pRT=M
g
)
1=2
Pd
p
(2c
pg
R=M
g
)
=
4 0:0262
2p(8:31 293=0:029)
_
1:01 10
5
125 10
6
(2 753 8:31=0:029)
= 0:00495
The Nusselt number for particle/wall collision is
obtained from Eq. (39):
N 4(1 Kn) ln 1
1
Kn
_ _
4 = 4(1 0:00495)
ln 1
1
0:00495
_ _
4 = 17:2
and the random particle velocity is obtained from Eq.
(41):
w
p
=
gd
p
(e
b
e
mf
)
5(1 e
b
)(1 e
mf
)
_
=
9:81 125 10
6
(0:585 0:42)
5(0:415)(0:58)
_
= 0:013 m=s
The value for Z is calculated by Eq. (38):
Z =
r
s
c
ps
d
p
w
p
6k
g
=
2200 753 125 10
6
0:013
6 0:0262
= 17:1
and taking a value of 2.6 for the empirical constant C
c
,
Eq. (37) gives the convective coecient as
h
d
=
k
g
d
p
_ _
Z(1 e
b
) 1 e
N=C
c
Z
_ _
=
0:0262
125 10
6
_ _
17:1 (0:415) 1 e
17:2=2:617:1
_ _
= 477 Wm
2
K
4.1.6 Comparison of Calculated Coefcients with
Data
The specications given above represent the bubbling
uidized bed of Chandran et al. (1980). Their experi-
mental measurement of heat transfer coecients on the
specied horizontal tube is plotted in Fig. 3. Compare
the values of h
c
calculated by the various models to the
experimental data of Chandran et al.
At the specied gas velocity of U
g
= 0:16 m=s, the
experimental value of the surface heat transfer coe-
cient is, (h
c
)
exp
= 395 W=m
2
K. The following tabula-
tion shows the corresponding values calculated from
the various models, and the percent deviation from
experimental value:
Model
(h
c
)
calc
/
d
(W=m
2
K)
Deviation
(%)
Vreedenberg (1958) 409 3:5
Borodulya et al. (1991) 476 20:5
Molerus and Schweinzer (1989) 486 23:0
Packet model 466 18:0
Martin (1984) 477 20:8
Average deviation 17.2
It is seen that for this single case, the older
Vreedenberg correlation came closest to the measured
value. Needless to say, a single case such as this should
not be taken to be a general assessment of models.
While there are dierences in the calculated coe-
cients, the average deviation of 17% is reasonable
and is representative of current predictive capability.
4.1.7 Freeboard Heat Transfer
For the BFB operating at the above conditions, what
would be the heat transfer coecient (average) around
a horizontal tube of diameter 0.029 m, located in the
freeboard with its centerline 0.2 m above the level of
the collapsed (packed) bed?
Equation (51) describes the exponential decrease of
the heat transfer coecient with increasing elevation in
the freeboard of bubbling beds, for horizontal tubes.
For the present case, H = 0:2 m, and U
+
= 0:2. Hence
h
+
c
= 1:3e
1:53H=U
+
= 1:3e
1:530:2=0:2
+
= 0:281
From the denition of h
+
c
, Eq. (49),
h
+
c

h
c
h
g
h
cb
h
g
where
h
cb
= the convective coeent for tube submerged
in the bubbling bed.
= 395 W/m
2
K from Chandrans data (see
example in Sec. 4.1.6)
h
g
= the gas convective coecient, obtained from
correlation of Douglas and Churchill, Eq.
(50):
h
g
D
t
k
g
= 0:46
D
t
r
g
U
g
m
g
_ _
1=2
0:00128
D
t
r
g
U
g
m
g
_ _
= 0:46
0:029 0:160 1:18
1:85 10
5
_ _
1=2
0:00128
0:029 1:18 0:160
1:85 10
5
_ _
= 8:29
h
g
= 8:29
k
g
D
t
= 8:29
0:0262
0:029
= 7:49 W=m
2
K
Hence the heat transfer coecient for the freeboard
tube is
h
c
= h
+
c
(h
cb
h
g
) h
g
= 0:281(395 7:49)
7:49 = 116 W=m
2
K
Note that this is 15 times greater than the single-phase
gas convection coecient, the increase being attributa-
ble to particles splashing into the freeboard region.
4.2 Fast Fluidized Bed
Consider a small-scale FFB, with sand particles of
diameter 125 mm mean, uidized by ambient air. The
diameter of the bed is D
b
= 0:15 m, with a vertical
length of 11 m. Properties of the air and particles are
as in the table:
According to Geldarts classication, these particles
are of type B. Minimum uidization with ambient air
occurs with void fraction of e
mf
= 0:45. In examples
below, we consider the operation of this bed with a
supercial gas velocity of U
g
= 5:0 m=s and a particle
mass ux of G
s
= 40 or 60 kg/m
2
s.
4.2.1 Fluidization Regime
At the intended operating condition, would the bed
uidize in the fast regime?
The generalized regime map of Fig. 1 can be used to
determine the operational characteristic of this case.
The Archimedes number is found from Eq. (3),
Ar =
gr
g
(r
s
r
g
)
m
2
g
d
3
p
=
(9:81)(1:18)(2200 1:18)(125 10
6
)
3
(1:85 10
5
)
2
= 145
The Reynolds number at minimum uidization is
found by solving Eq. (1),
Ar = 1:75
(Re
mf
)
2
f
s
(e
mf
)
3
150
(1 e
mf
)(Re
mf
)
(f
s
)
2
(e
mf
)
3
145 = 1:75
(Re
mf
)
2
(0:95)(0:45)
3
150
(1 0:45)(Re
mf
)
(0:95)
2
(0:45)
3
Re
mf
= 0:144
from which we obtain the minimum uidization
velocity by Eq. (2),
U
mf
= Re
mf
m
g
r
g
d
p
_ _
= 0:144
1:85 10
5
1:18(125 10
6
)
_ _
= 0:0182 m=s
The terminal velocity is obtained from Eq. (6b),
U
t
= 0:072g
d
8=5
p
(r
s
r
g
)
r
2:5
g
m
3=5
g
_ _
5=7
= 0:072 9:81
(125 10
6
)
8=5
(2200 1:18)
(1:18)
2=5
(1:85 10
5
)
3=5
_ _
5=7
= 0:67 m=s
and the dimensionless velocity is by denition,
U
+

U
g
U
mf
U
t
U
mf
=
5:0 0:0182
0:671 0:0182
= 7:63
The dimensionless particle diameter is, from Eq. (4),
d
+
p
Ar
1=3
= 145
1=3
= 5:25
This locus is within the fast regime, near its upper
boundary, on the map of Fig. 1. Therefore the
proposed operation will be a fast uidized bed.
4.2.2 Convective Heat Transfer at the Wall
Dou et al. (1993) measured the convective heat transfer
coecient at the wall of the FFB described above.
They used a small heat transfer surface of
L
h
= 3:2 cm, located 5.5 m above the air inlet port.
These investigators also obtained measurements of
the cross-sectional bed density (r
b
) at this elevation.
Their data, for two mass uxes of particles, are:
Use models that give empirical correlations of h
c
as
functions of r
b
for short surfaces to calculate the
values of the convective heat transfer coecients, and
compare them with the above experimental data.
The correlation of Wen and Miller (1961), Eq. (54),
gives
h
c
=
k
g
d
p
c
ps
c
pg
_ _
r
b
r
s
_ _
0:3
U
t
gd
p
_ _
0:21
Pr
g
h
c
=
0:0262
125 10
6
0:8
1:0
_ _
12
2200
_ _
0:3
0:671
9:81 125 10
6
_ _
0:21
0:71 = 93:7 W=m
2
K
for G
s
= 40 kg=m
2
s
Gas Particles
Density, r (kg/m
3
) 1.18 2,200
Heat capacity, c
p
(kJ/kg K) 1.00 0.800
Thermal conductivity, k (W/m K) 0.0262 0.900
5
Pr 0.71
Sphericity, f
s
0.95
G
s
(kg=m
2
s) r
b
(kg=m
3
) h
c
(W=m
2
K)
40 12 80
60 24 120
=
24
12
_ _
0:3
93:7 = 115 W=m
2
K
for G
s
= 60 kg=m
2
s
The correlation of Divilio and Boyd (1993), Eq. (55),
gives
h
c
= 23:2(r
b
)
0:55
= 23:2(12)
0:55
= 91:0 W=m
2
K
for G
s
= 40 kg=m
2
s
=
24
12
_ _
0:55
91:0 = 133 W=m
2
K
for G
s
= 60 kg=m
2
K
The table at foot of this page compares the calculated
values to the experimental data.
While these results are quite satisfactory, the reader
is cautioned that comparison for a single case should
not be taken as a denitive assessment. As for all
empirical correlations, application should be limited
to conditions within the range of the correlations
original database.
4.2.3 Convective Coefcient Calculated from
Pressure Drop
The model of Molerus and Wirth (1993, 1995) corre-
lates the convective heat transfer coecient with bed
pressure drop. Calculate h
c
for the above case by this
model and compare with the data of Dou et al. (1993)
at G
s
= 60 kg=m
2
s.
Since the heat transfer surface of Dou is 5.5 m
above the gas and particle inlets, acceleration and
wall friction eects would be negligible and bed
pressure drop related directly to bed density:
P
L
b
gr
b
= 9:81 24 = 235 Pa=m
for G
s
= 60 kg=m
2
s
Several other required quantities have been calculated
in example 2.1:
Ar = 144
Re
mf
= 0:145
U
t
= 0:671 m=s
Re
t

r
g
U
t
d
p
m
g
=
1:18 0:671 125 10
6
1:85 10
5
= 5:35
Additional required quantities are obtained from Eqs.
(59)(62):
f 1
2:3P=L
b
g(1 e
mf
)(r
s
r
g
)
= 1
2:3 235
9:81 (1 0:45)(2200 1:18)
= 0:0456
F
1
(1 f)
4
Re
2
mf
f(1 f)
2
Re
mf
Re
t
189
_
1 e
mf
)f
3
(1 f)Ar
_ _
= (1 0:0456)
4
0:144
2
_
0:0456(1 0:0456)
2
0:145
5:35 189(1 0:45)0:0456
3
(1 0:0456)145
= 1:31
F
2
2(1 f)
2
Re
mf
fRe
t
_ _
2
= 2(1 0:0456)
2
0:144 0:0456 5:35
_ _
2
= 0:258
Re
pa
=
2(1 f)
2
Re
mf
fRe
t
2[f 1(1 f)e
mf
]
1 4
F
1
F
2
_ _
_
=
2(1 0:0456)
2
0:145 0:0456 5:35
2[0:0456 (1 0:0456)0:45]
1 4
1:31
0:258
_ _
_
= 2:48
Source
h
c
W/m
2
K % di. from exp.
G
s
= 40 kg=m
2
s G
s
= 60 kg=m
2
s
Experimental, Dou et al. (1993) 80 120
Calcd by Eq. (54), Wen and Miller (1961) 94 (17%) 115 (4%)
Calcd by Eq. (55), Davilio and Boyd (1993) 91 (14%) 133 (11%)
The convective Nusselt number is given by Eq. (58) as
h
c
d
p
k
g
= 2:85
P=L
b
g(r
s
r
g
)(e
s;mf
)
_ _
0:5
3:28
10
3
Re
pa
Pr
g
= 2:85
235
9:81(2200 1:18)(0:55)
_ _
0:5
3:28
10
3
2:48 0:71
= 0:401 0:0058 = 0:407
from which the heat transfer coecient is obtained as
h
c
=
k
g
d
p
_ _
0:407 =
0:0262
125 10
6
_ _
0:407
= 85:3 W=m
2
K
which is 29% below the experimental value of
120 W=m
2
K.
4.2.4 Convective Coefcient from Surface Renewal
Model
The surface renewal model accounts for convective
heat transfer by the dense-packet phase (clusters) and
by the lean-gas phase. Calculate h
c
for the above case
by this model and compare with the data of Dou et al.
(1993) at G
s
= 60 kg=m
2
s
To calculate the dense phase coecient (h
d
), we
need to determine several parameters of the packet/
cluster model. From example 2.3, we have
Re
pa
= 2:48, whence the velocity of falling clusters is
obtained as
V
pa
= Re
pa
m
g
r
g
d
p
= 2:48
1:85 10
5
1:18 125 10
6
= 0:311 m=s
For the case of interest, the length of heat transfer
surface (L
h
) is smaller than the cluster contact length
(L
c
), so we use the heater length in Eq. (68) to obtain
the cluster contact time,
t
pa
=
L
h
V
pa
0:032
0:311
= 0:103 s
The average solid fraction is obtained from the average
bed density,
e
s;m

r
b
r
s
=
24
2200
= :0110
Gu and Chens correlation, Eq. (64), gives the solids
concentration in clusters,
e
s;pab
= 0:57 1 1
e
s;m
0:57
_ _
3:4
_ _
= 0:57
1 1
0:0110
0:57
_ _
3:4
_ _
= 0:0365
Next, Eqs. (35) and (34) calculate the thermal conduc-
tivity for clusters of this concentration,
f
k
= 0:305
k
s
k
g
_ _
0:25
= 0:305
0:900
0:0262
_ _
0:25
= 0:126
k
pab
= k
g
e
pab
(1 e)
pab
f
k
2k
g
=3k
s
_ _
= 0:0262
(1 0:0365)
0:0365
0:126 2 0:0262=3 0:90
_ _
= 0:0318 W=mK
Assuming that the Fourier number is suciently large
to take C 1:0 from Eq. (33), the coecient for dense
phase convection is then obtained from Eq. (67) as
h
d
= 2
k
pa
r
s
C
ps
e
s;pa
pt
pa
_ _
1=2
= 2
0:0318 2200 800 0:0365
p 0:103
_ _
1=2
= 159 W=m
2
K
Equation (63) is used for an estimate of the lean phase
convective coecient,
h
t
= 0:023
k
g
D
b
_ _
D
b
r
g
U
g
m
g
_ _
0:8
(Pr
g
)
0:33
= 0:023
0:0262
0:15
_ _
0:15 1:18 5
1:85 10
5
_ _
0:8
(0:71)
0:33
= 19:9 W=m
2
K
The weighting factor (f
l
) is obtained from Eq. (69),
1 f
l
= 1 0:018
1
0:0110
_ _
0:6
= 0:731
and the combined convective heat transfer coecient is
calculated by Eq. (16),
h
c
= f
l
h
l
(1 f
l
)h
d
= 0:269 19:9 0:731 159
= 122 K=m
2
K
which is within 2% of experimental value.
4.2.5 Heat Transfer in a High-Temperature Bed
Consider the sand particles described above uidized
in a combustor of diameter D
b
= 0:5 m at atmospheric
pressure and 1150K temperature, with supercial gas
velocity of 5 m/s and solid mass ux of 60 kg/m
2
s.
Assume that e
mf
remains at 0.45. If large areas of the
bed wall were to be used as a heat transfer surface,
maintained at 550K, what would be the eective heat
ux into that surface?
We rst require the physical properties of the bed
material at the operating conditions. Properties of the
particles are little changed by temperature, and so will
be taken at the values given above. Gas properties will
be taken to be those of air at 1 atm and 1150K:
We rst check the uidization regime of the
intended operation.
Using Eqs. (3), (1), (2), (6b), and (4), the operating
parameters are determined as
Ar =
gr
g
(r
s
r
g
)
m
2
g
d
3
p
=
(9:81)(0:307)(2200 0:307)(125 10
6
)
3
(4:51 10
5
)
2
= 636
Re
mf

Ar(f
s
)
2
(e
mf
)
3
150(1 e
mf
)
=
6:36(0:95)
2
(0:45)
3
150(1 0:45)
= 6:34 10
3
U
mf
= Re
mf
m
g
r
g
d
p
_ _
= 0:00634
4:51 10
5
0:307(125 10
6
)
_ _
= 0:00745 m=s
and the dimensionless velocity is by denition, Eq. (7),
U
t
= 0:072g
d
8=5
p
(r
s
r
g
)
r
2=5
g
m
3=5
g
_ _
5=7
= 0:072 9:81
(125 10
6
)
8=5
(2200 0:307)
(0:307)
2=5
(4:51 10
5
)
3=5
_ _
5=7
= 0:673 m=s
The dimensionless velocity and particle diameter are
then obtained from Eqs. (7) and (4),
U
+
=
U
g
U
mf
U
t
U
mf
=
5:0 0:00745
0:673 0:00745
= 7:50
d
+
p
Ar
1=3
= 6:36
1=3
= 1:85
This locus is within the fast regime, near its center, on
the map of Fig. 1. Therefore the proposed operation
will be a fast uidized bed. In actual operation, the bed
average density, at elevation of the heat transfer sur-
face, is found to be r
b
= 45 kg=m
3
. The values for
volume fractions of solid and gas phases are then
e
s;b
=
r
b
r
s
=
45
2200
= 0:0205
e
b
= 1 e
s;b
= 1 0:0205 = 0:980
We now have a choice of several models for the esti-
mation of the heat transfer coecient. The operating
conditions are close to those used by Werdmann and
Werther (1993) in developing their correlation, which
is particularly appropriate for large heat transfer sur-
faces. Thus from Eq. (56),
h
c
= 7:46 10
4
k
g
d
p
_ _
D
b
r
g
U
g
m
g
_ _
0:757
r
b
r
s
_ _
0:562
= 7:46 10
4
0:0743
125 10
6
_ _
0:5 0:307 5:0
4:51 10
5
_ _
0:757
45
2200
_ _
0:562
= 79:4 W=m
2
K
Due to high temperatures, we also need to calculate
radiative heat transfer. The eective emissivity of the
bed is obtained from Eq. (71) using the recommended
value of 0.95 for C
e
and taking the particle emissivity
to be e
s
= 0:7:
Gas Particles
Density, r (kg/m
3
) 0.307 2,200
Heat capacity, c
p
(kJ/kg K) 1.17 0.800
Thermal conductivity, k (W/m K) 0.0743 0.900
5
Pr 0.71
Sphericity, f
s
0.95
e
b
= 1 exp
C
e
e
s;b
e
s
D
b
d
p
_ _
= 1 exp
0:95 0:0205 0:7 0:5
125 10
6
_ _
= 1:00
Assuming the bed wall to be oxidized steel, a reason-
able value for wall emissivity is e
w
0:8. The bedwall
combined emissivity is calculated by Eq. (44) as
e
bw
=
1
1=e
b
1=e
w
1
=
1
1=1:00 1=0:8
= 0:80
The radiative heat transfer coecient is obtained from
Eq. (45),
h
r
=
e
bw
s(T
4
b
T
4
w
)
T
b
T
w
=
0:80 5:67 10
8
(1150
4
550
4
)
1150 550
= 125 W=m
2
K
The total heat transfer coecient and heat ux are,
respectively,
h
w
= h
c
h
r
= 79:4 125 = 204 W=m
2
K
q
w
a
w
= h
w
(T
b
T
w
) = 204(1150 550)
= 1:22 10
5
W=m
2
Note that this high heat ux is representative of opera-
tions for FFB combustors. Note also that 61% of the
heat transfer is due to thermal radiation.
NOMENCLATURE
a = interfacial surface area; m
2
A = radiation absorption coecient, Eq. (41); 1/m
Ar = Archimedes number, Eq. (3); dimensionless
c
p
= specic heat capacity; J/kg K
C = correction factor for conductivity, Eqs. (32 and
33); dimensionless
C
c
= contact parameter in Martins model, Eq. (64);
dimensionless
C
e
= emissivity parameter, Eq. (70); dimensionless
C
D
= coecient of drag, Eq. (6); dimensionless
C
R
= correction factor for radial position, Eq. (18);
dimensionless
d
b
= bubble diameter, cm or m
d
bm
= maximum bubble diameter, cm or m
d
bo
= initial bubble diameter at distributor, cm or m
d
p
d
+
p
= dimensionless particle diameter, Eq. (4)
D = diameter of bed or of tube; m
e = eradiative emissivity; dimensionless
f = time fraction of contact by lean or dense phase,
Eq. (16); dimensionless
f
b
= frequency of bubbles; 1/s
F
v
= view factor for radiation, Eq. (38); dimensionless
F
o
= Fourier modulus, Eq. (33); dimensionless
g = gravitational acceleration; 9.81 m/s
2
h = heat transfer coecient; W/m
2
K
h
+
c
= dimensionless heat transfer coecient, Eq. (44)
h
p
= heat transfer coecient based on particle surface
area, Eq. (8); W/m
2
K
h
w
= heat transfer coecient based on submerged wall
area, Eq. (14); W/m
2
K
H = height in freeboard, measured from surface level
of collapsed bed; m
I = forward radiative ux, Eq. (41); W/m
2
J = backward radiative ux, Eq. (41); W/m
2
k = thermal conductivity; W/m K
Kn = Knudsen number for gas in particlewall gap, Eq.
(67); dimensionless
l = length scale in Eqs (23) and (24); m
L
b
= axial elevation in FFB; m
L
c
= contact length of clusters on wall, Eq. (62); m
L
h
= vertical length of heat transfer surface in FFB; m
M
g
= molecular weight of gas; kg/mol
Nu = Nusselt number; dimensionless
Nu
p
= Nusselt number based on particle diameter, Eq.
(9); dimensionless
P = pressure; Pa
Pr = Prandtl number (c
p
m=k), dimensionless
q = heat transfer rate; W
r = radial position within bed; m
R = universal gas constant
R
b
= radius of bed; m
Re
D
= Reynolds number based on tube diameter, Eq.
(20); dimensionless
Re
mf
= Reynolds number at minimum uidization
velocity, Eq. (2), dimensionless
Re
p
= Reynolds number based on particle diameter, Eq.
(10); dimensionless
Re
t
= Reynolds number at terminal velocity, Eq. (46);
dimensionless
S = radiation scattering coecient, Eq. (41); 1/m
t = time, s
T = temperature; K
U = supercial gas velocity; m/s
U
e
= excess supercial velocity over minimum
uidization, Eq. (28); m/s
U
+
= dimensionless velocity, Eq. (7)
V
pa
= velocity of particle packet or cluster; m/s
w
p
= average random particle velocity, Eq. (65); m/s
y = coordinate distance away from surface; m
Greek Symbols
d
w
= bubble volume fraction at wall, Eq. (36);
dimensionless
e = volume fraction gas (void fraction); dimensionless
e
s
= volume fraction solids, equal to (1 e);
dimensionless
f
s
= sphericity of particles, Eq. (1); dimensionless
m = viscosity; kg/(m s)
r = density; kg/m
3
r
b
= density of particlegas medium, Eq. (47); kg/m
3
s = StefanBoltzmann constant; 5:67 10
8
W=m
2
K
4
t = contact (residence) time; s
t
pa
= residence time of packets, Eq. (31); s
Subscripts
b = bed, or bubble
c = convective
d = dense phase
g = gas
l = lean phase
m = local gasparticle medium
mf = minimum uidization
p = particle
pa = packet or cluster
pab = bulk packet
w = wall or submerged surface
r = radiative
s = solid
t = terminal, or tube
REFERENCES
Andeen BR, Glicksman LR. Heat transfer to horizontal tube
in shallow uidized beds. National Heat Transfer
Conference, St Louis, MO, paper no. 76-HT-67, 1976.
Anderson BA, Lechner B. Local lateral distribution of heat
transfer on tube surface of membrane walls in CFB boiler.
In: Avidan AA, ed. Circulating Fluidized Bed Technology
IV. New York: AIChE, 1993, pp 311318.
Battacharya, Harrison D. Heat transfer in high temperature
uidized beds. Proc of Europ Conf Particle Tech, 1976, p.
K7.
Beaude F, Louge M. Similarity of radial proles of solid
volume fraction in a circulating uidized bed. In:
Fluidization VIII. New York: Engineering Foundation,
1995, pp 245253.
Berg BV, Baskakov AP. Investigation of local heat transfer
between a xed horizontal cylinder and a uidized bed.
Intern Chem Eng 14, no. 3:440443, 1974.
Biyikli S, Tuzla K, Chen JC. Heat transfer around a hori-
zontal tube in freeboard region of uidized beds. AIChE J
29, no. 5:712716, 1983.
Biyikli S, Tuzla K, Chen JC. Freeboard heat transfer in high-
temperature uidized beds. Powder Tech 53:187194,
1987.
Borodulya VA, Teplitsky YS, Sorokin AP, Markevich II,
Hassan AF, Yeryomenko TP. Heat transfer between a
surface and a uidized bed: consideration of pressure
and temperature eects. Int J Heat Mass Transf 34, no.
1:4753, 1991.
Botterill JSM, Desai M. Limiting factors in gasuidized bed
heat transfer. Powder Tech 6, no. 4:231-238, 1972.
Brewster MO, Tien CL. Radiative heat transfer in packed
uidized beds: dependent versus independent scattering.
J Heat Transf 104, no. 4:574580, 1982.
Burki V, Hirschberg B, Tuzla K, Chen JC. Thermal devel-
opment for heat transfer in circulating uidized beds.
AIChE Annual Meeting, St. Louis, MO, 1993.
Chandran R, Chen JC. Inuence of the wall on transient
conduction into packed media. AIChE J 31, no. 1:168
170, 1985a.
Chandran R, Chen JC. A heat transfer model for tubes
immersed in gas uidized beds. AIChE J 31, no. 2:244
252, 1985b.
Chandran R, Chen JC, Staub FW. Local heat transfer coe-
cients around horizontal tubes in uidized beds. AIChE J
102, no. 2:152157, 1980.
Chen JC. Heat transfer to tubes in uidized beds. Award
lecture. National Heat Transfer Conference, St. Louis,
MO, paper no. 76-HT-75, 1976.
Chen, JC. Thomas Baron Plenary LectureClusters. AIChE
Symp Series, 92, No. 313:16, 1996.
Chen JC, Chen KL. Analysis of simultaneous radiative and
conductive heat transfer in uidized beds. Chem Eng
Commun 9:255271, 1981.
Chen JC, Cimini RJ, Dou SH. A theoretical model for simul-
taneous convective and radiative heat transfer in circulat-
ing uidized beds. In: Basu P, Large JF, eds. Circulating
Fluidized Bed Technology II. Oxford: Pergamon Press,
1988, pp 255262.
Cimini RJ, Chen JC. Experimental measurement of radiant
transmission through packed and uidized media. Exp
Heat Transfer 1:4556, 1987.
Divilio RJ, Boyd TJ. Practical implications of the eect of
solids suspension density on heat transfer in large-scale
CFB boilers. In: Circulating Fluidized Bed Technology
IV. New York: Engineering Foundation, 1993, pp 334
339.
Dou S, Herb B, Tuzla K, Chen JC. Heat transfer coecients
for tubes submerged in circulating uidized bed.
Experimental Heat Transfer 4:343353, 1991.
Dou S, Herb B, Tuzla K, Chen JC. Dynamic variation of
solid concentration and heat transfer coecient at wall of
circulating uidized bed. In: Potter OE, Nicklin DJ, ed.
1993, pp 793801.
Douglas WYM, Churchill SW. Recorrelation of data for
convective heat transfer between gases and single cylin-
ders with large temperature dierences. Chem Eng Prog
Symp Ser 51, no. 17:57, 1956.
Dow WM, Jacob M. Heat transfer between a vertical tube
and a uidized airsolid mixture. Chem Eng Progr 47, no.
12:637648, 1951.
Ebert T, Glicksman L, Lints M. Determination of particle
and gas convective heat transfer components in circulat-
ing uidized bed. Chem Eng Sci 48:21792188, 1993.
Fraley LD, Lin YY, Hsiao KM, Solbakken A. ASME Paper
83-HT-92, 1983.
Furchi JCL, Goldstein L Jr, Lombardi G, Mohseni M.
Experimental local heat transfer in a circulating uidized
bed. In: Basu P, Large JF, eds. Cirulating Fluidized Bed
Technology II. Oxford, UK: Pergamon Press, 1988, pp
263270.
Geldart D. Types of gas uidization. Powder Tech 7, no.
5:285292, 1973.
Glicksman LR. Heat transfer in circulating uidized beds. In:
Grace JR, Avidan AA, Knowlton TM, eds. Circulating
Fluidized Beds. London: Chapman and Hall, 1997, pp
261311.
Grace JR. Can J Chem Eng 64:353, 1986a.
Grace J. Heat transfer in circulating uidized beds. In:
Circulating Fluidized Bed Technology. New York:
Engineering Foundation, 1986b, pp 6381.
Gu WK, Chen JC. A model for solid concentration in circu-
lating uidized beds. In: Fan LS, Knowlton T, eds.
1998, pp 501508.
Han GY. Experimental study of radiative and particle con-
vective heat transfer in fast uidized beds. PhD disserta-
tion, Dept. of Chem. Eng., Lehigh University, Bethlehem,
PA, 1992.
Han GY, Tuzla K, Chen JC. Performance of an entrained
particle heat exchanger. Heat Transfer Eng 17, no. 4:64
71, 1996.
Hartige EU, Li Y, Werther J. Flow structures in fast uidized
beds. In: Fluidization V. New York, Engineering
Foundation, 1986, pp 345352..
Herb BE, Dou S, Tuzla K, Chen JC. Axial solid concentra-
tion in CFBs: experimental measurements and model pre-
dictions. AIChE Annual Meeting, 1989a.
Herb B, Tuzla K, Chen JC. Distribution of solid concentra-
tions in circulating uidized bed. In: Fluidization VI. New
York, Engineering Foundation, 1989b, pp 6572.
Herb BE, Dou S, Tuzla K, Chen JC. Solid mass uxes in
circulating uidized beds. Powder Tech 70, no. 3: 197
205, 1992.
Horio M, Morshita K, Tachibana O, Murata M. Solid dis-
tribution and movement in circulating uidized beds. In:
Basu P, Large JF, eds. Circulating Fluidized Bed
Technology II. Oxford, UK: Pergamon Press, 1988, pp
147154.
Kiang KD, Lin KT, Nack H, Oxley JH. Heat transfer in fast
uidized beds. In Fluidization Technology. New York:
Kunii D, Smith JM. Heat transfer characteristics of porous
rocks. AIChE J, no. 1:7178, 1960.
Kunii D , Levenspiel O. Fluidization Engineering. 1st ed.
New York: John Wiley, 1969.
Kunii D , Levenspiel O. Fluidization Engineering. 2nd ed.
Leva M, Grummer M. A correlation of solids turnovers in
uidized systems. Chem Eng Progr 48, no. 6:307313,
1952.
Leva M, Weintrub M, Grummer M. Heat transmission
through uidized beds of ne particles. Chem Eng Progr
45, no. 9:563572, 1949.
Levenspiel O, Walton JS. Chem Eng Progr Symp Series 50,
no. 9:1, 1954.
Lints M. Particle to wall heat transfer in circulating uidized
beds. PhD dissertation, Mass Inst Tech Cambridge, MA,
1992.
Lints MC, Glicksman LR. The structure of particle clusters
near wall of a circulating uidized bed. AIChE Symp
Series, 89, no. 296:3547, 1993.
Louge M, Lischer J, Chang H. Measurements of voidage
near the wall of a circulating uidized bed riser. Powder
Tech 62:269276, 1990.
Mahalingam M, Kolar AK. Heat transfer model for mem-
brane wall of a high temperature circulating uidized bed.
In: Basu P, Horio M, Hasatani M, ed. Circulating
Fluidized Bed Technology III. Toronto: Pergamon
Press, 1991, pp 239246.
Martin H. Heat transfer between gas uidized beds of solid
particles and the surfaces of immersed heat exchanger
elements, part I. Chem Eng Processes 18:157169, 1984a.
Martin H. Heat transfer between gas uidized beds of solid
particles and the surfaces of immersed heat exchanger
elements, part II. Chem Eng Processes 18:199223, 1984b.
Mickley HS, Trilling CA. Heat transfer characteristics of
uidized beds. Ind Eng Chem 41, no. 6:11351147, 1949.
Mickley HS, Fairbanks DF. Mechanism of heat transfer to
uidized beds. AIChE J 1, no. 3:374384, 1955.
Molerus O. Fluid dynamics and its relevance for basic fea-
tures of heat transfer in circulating uidized beds. In:
Avidan AA, ed. Circulating Fluidized Bed IV. New
York: AIChE, 1993, pp 285-290.
Molerus O, Burschka A, Dietz S. Particle migration at solid
surfaces and heat transfer in bubbling uidized beds II,
Prediction of heat transfer in bubbling uidized beds.
Chem Eng Sci 50, no. 5:879885, 1995.
Molerus O, Schweinzer J. Prediction of gas convective part of
the heat transfer to uidized beds. In Fluidization IV.
Molerus O, Wirth KE. Heat Transfer in Fluidized Beds.
Noe AR, Knudsen JG. Local and average heat transfer coef-
cients in a uidized bed heat exchanger. Chem Eng Progr
Symp Series 64, no. 82:202211, 1968.
Ozkaynak TF, Chen JC. Emulsion phase residence time and
its use in heat transfer models in uidized bed. AIChE J
26, no. 4:544550, 1980.
Ozkaynak TF, Chen JC, Frankeneld TR. An experimental
investigation of radiant heat transfer in high temperature
uidized bed. Fluidization IV. New York, Engineering
Palchonok GI, Breitholz C, Anderson BA, Lechner B. Heat
transfer in the boundary layer of a circulating uidized
bed boiler. In: Large JF, Laguerie C, eds. Fluidization
VIII. New York: Engineering Foundation, pp 291299,
1995.
Pell M. Gas Fluidization. Amsterdam: Elsevier, 1990.
Radauer HG, Glatzer A, Linzer W. A model combining con-
vective and radiative heat transfer in CFB boilers. In:
Circulating Fluididized Bed Technology V. New York:
Engineering Foundation, 1996, p 9.
Ranz WE. Friction and transfer coecients for single particle
and packed beds. Chem Eng Progress 48, no. 5:247253,
1952.
Rhodes MJ, Mineo H, Hirama T. Particle motion at the wall
of a circulating uidized bed. Powder Tech 70:207214,
1992.
Saxena SC, Grewal NS, Gabor JD, Zabrodsky SS,
Galershtein DM. Heat transfer between a gas uidized
bed and immersed tubes. Adv in Heat Transfer 14:145
247, 1978.
Soong CH, Tuzla K, Chen JC. Identication of particle clus-
ters in circulating uidized bed. In: Avidan AA, ed.
Circulating Fluidized Bed Technology IV. New York:
AIChE, 1993, pp 809814.
Soong CH, Tuzla K, Chen JC. Experimental determination
of cluster size and velocity in circulating uidized beds. In:
Large JF, Laguerie C, ed. Fluidization VIII. New York:
Vreedenberg HA. J Appl Chem 2 (suppl 1):s26, 1952.
Vreedenberg HA. Heat transfer between a uidized bed and
a horizontal tube. Chem Eng Sci 9, no. 1:5260, 1958.
Vreedenberg HA. Heat transfer between a uidized bed and
a vertical tube. Chem Eng Sci 11, no. 4:274285, 1960.
Wen CY, Miller EN. Heat transfer in solidsgas transport
lines. Ind. Eng. Chem. 53, no. 1: 5153, 1961.
minimum uidization velocity. AIChE J 12, no. 3:610
612, 1966.
Wender L, Cooper GT. Heat transfer between uidized-
solids bed and boundary surfacescorrelation of data.
AIChE J 4, no. 1:1523, 1958.
Werdermann CC, Werther J. Solids ow pattern and heat
transfer in an industrial scale uidized bed heat exchan-
ger. Proc 12th Intern Conf on Fluid Bed Combustion
2:985990, 1993.
Werther J. Fluid mechanics of large-scale CFB units. In:
Avidan AA, ed. Circulating Fluidized Bed Technology
Wirth KE. Zirkulierende Wirbelschichten-stro mungsmecha-
nische Grundlagen. Anwendung in der Feuerungstechnik,
1990.
Wirth KE. Prediction of heat transfer in circulating uidized
beds. In: Avidan A, ed. Circulating Fluidized Bed
Technology IV. New York: AIChE 1993, pp 291296.
Wirth KE. Heat transfer in circulating uidized beds. Chem
Eng Sci 50, no. 13:21372151, 1995.
Wood RT, Staub FW, Canada GS, McLaughlin MH. Two-
phase ow and heat transfer. Tech Rep RP 525-1, General
Electric Co, Schenectady, NY, 1978.
Wu R, Grace J, Lim C, Brereton C. Suspension to surface
heat transfer in a circulating uidized bed combustor.
AIChE J 35:16851691,1989.
11
Mass Transfer
Thomas C. Ho
Lamar University, Beaumont, Texas, U.S.A.
1 INTRODUCTION
Gassolid uidized bed reactors are widely used by
industry. Examples include uid catalytic cracking
(FCC), combustion and gasication, uranium proces-
sing, roasting of sulde ore, reduction of iron ores, and
drying (Zenz, 1978). Of all the phenomena aecting the
performance of bubbling uidized beds as chemical
reactors, interphase mass transfer is probably of pri-
mary importance in most cases (Chavarie and Grace,
1996a,b).
The particlegas mass transfer in a gassolid ui-
dized bed has been an essential subject of investigation
since the invention of the uidized bed technology.
Historically, there have been two approaches in mod-
eling the rate of mass transfer in uidized bed reactors.
One approach, called the homogeneous bed approach,
considers the uidized bed reactor to behave like a
xed bed and correlates the uidized bed mass transfer
coecient in a manner similar to that in a xed bed
based on a plug-ow model. The other approach,
called the bubbling bed approach, considers the ui-
dized bed to consist of two phases, a bubble and an
emulsion, and the gas interchange between the two
phases constitutes the rate of mass transfer. The objec-
tive of this chapter is to review the two approaches.
It is essential to point out that the mechanism of
mass transfer, as will be discussed in this chapter,
represents only one of the potential rate controlling
steps governing reactor performance in the system.
Depending on the types of gassolid reactions, the
rate controlling mechanisms may include (Levenspiel,
1972):
1. Particlegas mass transfer (gas lm diusion)
control
2. Pore diusion (ash layer diusion) control
3. Surface phenomenon control
This chapter is devoted to the rst mechanism listed
above, i.e., mass transfer between the gas and the bed
particles at the gasparticle interphase in a bubbling
uidized bed.
2 HOMOGENEOUS BED APPROACH
The homogeneous bed approach correlates the mass
transfer coecient in a uidized bed in a manner simi-
lar to that in a xed bed based on a plug-ow model.
The transfer equation in this approach for a xed or a
uidized bed closely resembles that for a single sphere
suspended in a gas stream. However, the magnitude of
the mass transfer coecient for these various bed par-
ticles may be signicantly dierent depending on par-
ticle size and operating conditions. The following
subsections review separately the particlegas mass
transfer for various bed particles based on this homo-
geneous bed approach.
2.1 Transfer Between Single Spheres and
Surrounding Gas
The rate of mass transfer between well-dispersed single
spheres and the surrounding gas can be described by
the following mass transfer equation (Kunii and
Levenspiel, 1991):
dN
A
dt
k
g.single
S
ex.single
C
i
A
C
A
_ _
1
where dN
A
,dt represents the transfer rate of A from
the particle surface to the gas stream, k
g.single
represents
the mass transfer coecient of the particle, S
ex.single
is
the exterior surface of the particle, C
i
A
is the concen-
tration of A at the gasparticle interphase, and C
A
is
the concentration of A in the gas stream. The mass
transfer coecient k
g.single
under this single-sphere con-
dition has been well established, both theoretically and
experimentally (Froessling, 1938), as
Sh
single

k
g.single
d
sph
y
D
2 0.6Re
sph
0.5
Sc
0.333
2
In the above equation, Re
sph
is the particle Reynolds
number dened as
Re
sph

r
g
ud
sph
m
3
Sc is the Schmidt number dened as
Sc
m
r
g
D
4
y is the logarithmic mean mole fraction of the inert or
nondiusing component, and D is the gas phase diu-
sion coecient. Equation 2 implies that the Shewood
number has a theoretical minimum at Sh 2.0 even
when the particle is exposed to a stationary gas, i.e.,
u (or Reynolds number) equals zero. The equation also
indicates that the mass transfer coecient (k
g.single
) is
proportional to the diusion coecient of the gas (D),
and is inversely proportional to the particle diameter
(d
sph
) and the mole fraction of the inert component (y).
Note that for nonspherical particles, d
sph
in Eqs. (2)
and (3) can be replaced by screen size d
p
as suggested
by Kunii and Levenspiel (1991), and the two equations
become Eqs. (5) and (6), respectively:
Sh
single

k
g.single
d
p
y
D
2 0.6Re
p
0.5
Sc
0.333
5
Re
p

r
g
ud
p
m
6
It is essential to note that Eq. (2) [or (5)] is valid only
for a single sphere, or single spheres well dispersed in a
gas stream. For packed bed or uidized bed particles,
the particlegas mass transfer coecient may be higher
or lower than that estimated from Eqs (2) [or (5)],
depending on the particle Reynolds number.
2.2 Transfer Between Fixed Bed Particles and
Flowing Gas
The rate of particlegas mass transfer from a dieren-
tial bed segment of a xed bed in the axial direction
can be represented by an equation resembling Eq. (1)
based on a plug-ow model, i.e.,
dN
A
dt
k
g.bed
S
ex.particles
C
i
A
C
A
_ _
7
However, in this equation, dN
A
,dt represents the com-
bined mass transfer rate from all the particles in the
segment, k
g.bed
is the average mass transfer coecient
of the particles, and S
ex.particles
is the total exterior
surfaces of all individual particles in the bed segment.
It is worth pointing out that the average mass trans-
fer coecient for xed bed particles may be higher or
lower than that for single spheres at an identical super-
cial gas velocity. On one hand, the low xed bed
voidage may enhance the mass transfer coecient
owing to the creation of increased true gas velocities
at the gasparticle interphase for xed bed particles.
On the other hand, the actual particle surfaces involved
in the particlegas mass transfer are smaller for xed
bed particles than those for well dispersed particles
owing to particle blockage in the xed bed
arrangement. The use of total exterior surfaces of all
individual particles in Eq. 7, therefore, overestimates
the active surfaces involved in the mass transfer pro-
cess. The net eect of this overestimation is to decrease
the average mass transfer coecient evaluated based
on Eq. (7).
There have been several reports on the average mass
transfer coecient for xed bed particles. In general,
for particle Reynolds numbers greater than 80
(relatively large particles), the average mass transfer
coecient for xed bed particles is found to be higher
than that estimated by Eq. (2) [or (5)] for well-dis-
persed single spheres. On the other hand, for particle
Reynolds numbers less than 80 (relatively smaller par-
ticles), the trend is reversed, i.e., the average mass
transfer coecient for xed bed particles is much smal-
ler than that estimated by Eq. (2). It is apparent that
the increase in mass transfer coecient for relatively
large particles in the xed bed setup is due to the eect
of the increase in true gas velocity at the gasparticle
interface. However, the decrease in average mass trans-
fer coecient for smaller particles is due to the eect of
particle blockage resulting in the overestimation of
active surfaces in Eq. (7), as discussed previously.
2.2.1 Average Mass Transfer Coefcient, k
g.bed
For particle Reynolds numbers greater than 80, Ranz
(1952) reported that the following equation describes
the particlegas mass transfer coecient for xed bed
particles:
Sh
bed

k
g.bed
d
p
y
D
2 1.8Re
p
0.5
Sc
0.333
for Re
p
> 80
8
In the above equation, Re
p
and Sc are similarly dened
as in Eqs. (6) and (4), but the supercial gas velocity is
used in the calculation of Re
p
, i.e.,
Re
p

r
g
Ud
p
m
9
For smaller particles, both Hurt (1943) and Resnick
and White (1949) reported that the particlegas mass
transfer coecient for xed bed particles can be lower
than that estimated from Eq. (2) for single spheres.
Note that their original results were reported in the
form of j
D
-factors; those results, however, can be easily
converted to the Sherwood number. The converted
Sherwood number from the above two studies along
with those from Eqs. (2) (for single spheres) and (8)
(for large xed bed particles) are plotted against parti-
cle Reynolds numbers in Fig. 1. As indicated in the
gure, the Sherwood number for xed bed particles
can be higher or lower than that estimated by Eq. (2)
for well-dispersed single spheres. For smaller particles,
the Sherwood number can be much lower than the
theoretical minimum of the Sherwood number for
single spheres, i.e., Sh
single
2.0.
2.2.2 Total Particle Exterior Surfaces, S
ex.particles
.
Besides the average mass transfer coecient, another
essential aspect in the application of Eq. (7) for pre-
dicting the rate of mass transfer is the estimation of the
total exterior surfaces of the particles in the bed seg-
ment. Two methods are discussed below for estimating
S
ex.particles
, one by Kunii and Levenspiel (1991) and the
other by Carmon (1941).
Method of Kunii and Levenspiel (1991). The
method proposed by Kunii and Levenspiel (1991) is
based on the equation
S
ex.particles
V
seg
1 e
bed
a
0
10
where V
seg
is the volume of the bed segment, e
bed
is the
bed voidage, and a
0
is the ratio of particle surface to
particle volume dened as:
a
0
(surface of a particle/volume of a particle
pd
2
sph
f
s
pd
3
sph
,6
11
6
d
sph
f
s
where f
s
is the sphericity of bed particles (f
s
1.0 for
spherical particles and 0 - f
s
- 1.0 for all other par-
ticles), and d
sph
is the diameter of spherical particles
having the same volume as the bed particles. Equations
(10) and (11) can be combined to yield
S
ex.particles
V
seg
61 e
bed
f
s
d
sph
12
The sphericity for dierent solids has been calcu-
lated and is given in Table 1. Note that the term
(f
s
d
sph
) appearing in Eq (12) is called the eective par-
ticle diameter (d
eff
) and can be estimated according to
the following approximations:
(a) For large particles > 1 mm), d
sph
can be
calculated from the measurement of the size
of particles by calipers or a micrometer if the
Figure 1 Particlegas mass transfer coecient for xed bed
particles.
particles are regular in shape, or by weighing a
certain number of particles if their density is
known, or by uid displacement if the particles
are nonporous. The sphericity f
s
can be mea-
sured directly, estimated from Table 1, or eval-
uated by the pressure drop experiments as
outlined in Kunii and Levenpiel (1991).
(b) For intermediate particles (40 mm - d
p
- 1 mm), screen analysis is the most conveni-
ent way to measure particle size. The d
eff
is
then estimated based on the following sugges-
tions:
1. For irregular particles with no seemingly
longer or shorter dimension (hence isotropic
in shape),
d
eff
f
s
d
sph
f
s
d
p
13
2. For irregular particles with one somewhat
longer dimension, but with a length ratio not
greater than 2 : 1 (eggs, for example),
d
eff
f
s
d
sph
d
p
14
3. For irregular particles with one somewhat
shorter dimension, but with a length ratio not
greater than 1 : 2, then roughly,
d
eff
f
s
d
sph
f
2
s
d
p
15
4. For very at or needlelike particles, estimate
the relationship between d
p
and d
eff
from the
f
s
values for the corresponding disks and
cylinders. An alternative method for these
Table 1 Sphericity of Particles
Type of particles
Sphericity
s
Source
Sphere 1.00 (a)
Cube 0.81 (a)
Cylinder
h d 0.87 (a)
h 5d 0.70 (a)
h 10d 0.58 (a)
Disks
h d,3 0.76 (a)
h d,6 0.60 (a)
h d,10 0.47 (a)
Activated carbon and silica gels 0.700.90 (b)
Broken solids 0.63 (c)
Coal
anthracite 0.63 (e)
bituminous 0.63 (e)
natural dust 0.65 (d)
pulverized 0.73 (d)
Cork 0.69 (d)
Glass, crushed, jagged 0.65 (d)
Magnetite, FischerTropsch catalyst 0.58 (e)
Mica akes 0.28 (d)
Sand
round 0.86 (e)
sharp 0.66 (e)
old beach as high as 0.86 (f)
young river as low as 0.53 (f)
Tungsten powder 0.89 (f)
Wheat 0.85 (d)
(a) From geometric considerations. (b) From Leva (1959). (c) From
Uchida and Fujita (1934). (d) From Carma n (1937). (e) From Leva et
al. (1948, 1949). (f) From Brown et al. (1949).
Source: Kunii and Levenspiel, 1991.
particles is to conduct the pressure drop experi-
ments to determine the eective particle size.
(c) For very small particles (- 40 mm), there are
two methods to determine the eective particle
size. One is to use the scanning of magnied
photographs of particles and the other is to
conduct sedimentation experiments of particles
in a known uid; the terminal velocity of these
particles will give the diameter of the equiva-
lent sphere.
Method by Carmon (1941). The method pro-
posed by Carmon (1941) for the total exterior areas
of xed bed particles is based on the equation
S
ex.particles
SLA 16
where
S S
a
1 X 17
and
S
a
14
X
3
K
t
1 X
2
_ _
0.5
18
In this equation, K
t
is estimated by
K
t

mLQ
A1 P
p
19
Table 2 summarizes the various properties of particles
in the investigation carried out by Resnick and White
(1949). They reported that the following equation
roughly holds for a quick estimation of S appearing
in Eq. (16) for the particles involved in their experi-
ments, i.e.,
S
S
d
d
p
20
where S
d
was reported to have an average value of
4.476. The S
d
values for various size particles in their
experiments are also summarized in Table 2.
2.3 Transfer Between Fluidized Bed Particles and
Fluidizing Gas
The homogeneous bed approach proposes the same
governing equation for a uidized as that for a xed
bed. In this approach, the rate of particlegas mass
transfer in a dierential segment of the bed is again
expressed in the following mass transfer equation:
dN
A
dt
k
g.bed
S
ex.particles
C
i
A
C
A
In the above equation, the term S

ex.particles
represents
the total exterior surfaces of the uidized particles in
the segment of the bed, and k
g.bed
represents the
average mass transfer coecient associated with these
uidized particles. As will be shown below, for the
same group of particles, k
g.bed
is always higher under
uidized bed operations than that under xed bed
operations.
It should be pointed out that, in this homogeneous
bed approach, the mass transfer coecient between
uidized particles and uidizing gas is dicult to eval-
uate and that the evaluated coecient is essentially
empirical in nature. This is especially due to the com-
plex bubbling behavior and the highly nonuniform
hydrodynamic phenomena within a uidized bed,
which make the accurate determination of the terms
in Eq. (7) almost impossible. Nevertheless, several
groups of workers have measured the average mass
transfer coecient based on the homogeneous bed
approach. Table 3 briey summarizes several of the
studies reviewed by Kunii and Levenspiel (1991).
Additional review of the studies has been provided
Table 2 Properties of Naphthalene Particles
Mesh size
r
(g,cm
3
)
X
()
S
a
(cm
2
,cm
3
)
S
(cm
2
,cm
3
)
d
p
(cm)
S
d
()
1420 0.646 0.391 88.1 53.65 0.089 4.775
2028 0.626 0.411 106.5 62.73 0.073 4.579
2835 0.615 0.422 167.7 96.93 0.045 4.362
3548 0.651 0.389 187.8 114.75 0.039 4.475
4865 0.607 0.430 262.5 149.63 0.028 4.189
Average 0.629 0.409 4.476
Source: Resnick and White, 1949.
by Krell et al. (1990). Typical results from these past
studies are summarized in the following subsections.
2.3.1 Average Mass Transfer Coefcient, k
g.bed
The reported average mass transfer coecients for
uidized bed particles are summarized in Fig. 2. As
indicated, the average mass transfer coecient for ui-
dized bed particles, again, can be higher or lower than
that estimated from Eq. (2) for well-dispersed single
spheres. The Sherwood number for uidized bed par-
ticles is generally lower than that for single spheres at
lower Reynolds numbers; however, it can be higher at
higher Reynold numbers (Re
p
> 80). Similar to that
for xed bed particles, the Sherwood number for
uidized particles can be well below the theoretical
minimum of the Sherwood number for single spheres,
i.e., Sh
single
2.0. This is due to the potential overesti-
mation of active surfaces involved in the mass transfer
operation. In a bubbling uidized bed, most of the
particles are expected to stay in the emulsion phase
with C
i
A
at equilibrium with C
A
in the emulsion gas.
Many of these particles may thus be considered as
inerts from the mass-transfer point of view because
they do not contribute to signicant amount of mass
transfer to the bubbling gas. The fact that all the
particle surfaces are included in the mass transfer cal-
culation as described by Eq. (7), therefore, overesti-
mates S
ex.particles
, which in turn decreases the average
mass transfer coecient to a value even below the
theoretical minimum for single spheres.
An attempt was made to correlate the average mass
transfer coecient reported by Resnick and White
(1949) for small size uidized particles, and the follow-
ing equations are obtained (for air system, Sc 2.35):
1. For particles with size between mesh 14 and 20
(d
p
1000 mm),
Sh
bed
0.200 Re
0.937
p
for 30 - Re
p
- 90
21
(d
p
711 mm),
Sh
bed
0.274 Re
0.709
p
for 15 - Re
p
- 80
22
(d
p
570 mm),
Sh
bed
0.773 Re
1.107
p
for 8 - Re
p
- 60
23
(d
p
410 mm.
Sh
bed
0.071 Re
0.926
p
for 6 - Re
p
- 40
24
(d
p
275 mm),
Sh
bed
0.041 Re
1.036
p
for 4 - Re
p
- 15
25
Table 3 Mass Transfer Experiments Included in Figure 2
Investigators Gas Process
Particles
d
p.m
(mm)
d
t
(cm)
L
m
(cm)
Resnick and White Air s. naphthalene Naphthalene 2101700 2.2 1.32.5
(1949) H
2
4.4
CO
2
Kettenring et al.
(1950)
Air v. water Silica gel
Alumina
3601000 5.9 1015
Chu et al.
(1953)
Air s. naphthalene Glass
Lead
7101980 10.2 0.39.2
Rape seed
Richardson and Szekely
(1961)
Air
H
2
a. CCl
4
a. water
Active carbon
Silica gel
882580 3.0 5d
p
Thodos et al. Air s. dichlorobenzene Alumina 18003100 3.8 0.67.0
(1961, 69, 72) v. water Celite 9.5
v. nitrobenzene 11.3
v. n-decane
s.=sublimation, v.=vaporization, a.=adsorption.

Source: Kunii and Levenspiel, 1991.
2.3.2 Total Particle Exterior Surfaces, S
ex.particles
The procedure for determining the total particle exter-
ior surfaces in a bed segment of a uidized bed is
essentially identical to that in a xed bed, as described
previously, i.e., using Eqs. (10) through (15). The e
bed
appearing in Eqs. (10) and (12), however, represents
bed voidage in a uidized bed (or e
f
) in this applica-
tion.
2.3.3 Comparison of K
g.bed
between Fixed Bed and
Fluidized Bed Particles
The experimental study of Resnick and White (1949)
included the interesting observation that, although the
bed hydrodynamics is signicantly dierent, the
measured mass transfer coecients for the xed and
uidized beds can be correlated with Re
p
by the same
set of constants as shown in Fig. 3. In the gure, the
dashed lines represent xed bed operations and the
solid lines represent uidized bed operations. As
indicated, the Sherwood number (or mass transfer
coecient) for the same group of particles continues
to increase with the Re
p
, even during the transition of
the operation from a xed bed to a uidized bed. It
appears that Eqs. (21) through (25), although derived
for uidized bed operations, can be extrapolated to
xed bed operations as well.
2.4 Transfer Between Immersed Isolated Spheres and
a Fluidized Bed
Another type of mass transfer encountered in a ui-
dized bed involves the transfer between an immersed
active particle and a uidized bed. Fluidized bed com-
bustors are a typical example where active coal parti-
cles burn in the matrix of chemically inert particles of
sand and ash which constitute 97 to 99% of the whole
bed mass. The mass transfer between the immersed
active particles and the uidizing gas is expected to
be dierent from that in a bed consisting of only active
particles, because the inert particles may serve as an
impermeable barrier for mass ow.
The subject of mass transfer involving isolated
active particles and the uidizing gas has been studied
by several groups of researchers. Hsiung and Thodos
(1977) measured the mass transfer coecients of a
batch of naphthalene particles uidized in a matrix
of inert particles similar in size and density.
Vandershuren and Delvossalle (1980) experimentally
studied the drying process of moist particles in a ui-
dized bed of similar dry particles. Palchonok and
Tamarin (1983, 1985) measured the mass transfer coef-
cients for an isolated active particle of various sizes
(d
p.a
3.530 mm) freely moving within a uidized
bed of large inert particles (d
p.i
1.26.3 mm). Kok
et al. (1986) measured the particlegas mass transfer
Figure 2 Particlegas mass transfer coecient for uidized
bed particles. (From Kunii and Levenspiel, 1991.)
Figure 3 Comparison of mass transfer coecient in xed
and uidized beds.
coecient in a 30 cm ID air uidized bed consisting of
inert glass beads and a varying mass fraction of sub-
liming naphthalene beads. The above studies were
comprehensively reviewed by Agarwal and La Nauze
(1989) and Linjewile and Agarwal (1990).
In an an attempt to provide a better correlation for
the mass transfer coecient, Palchonok et al. (1992)
derived an interpolation formula that couples an equa-
tion derived by Palchonok and Tamarin (1984), i.e.,
Sh
a
0.117Ar
0.39
Sc
0.33
26
with an expression suggested by Avedesian (1972) for
the limiting Sherwood number, i.e.,
Sh
limit

k
g.a
d
p
y
D
2e
mf
27
to become
Sh
a
2e
mf
0.117Ar
0.39
Sc
0.33
28
where Ar is the Archimedes number, dened as
Ar gd
3
p
r
p
r
g
n
2
r
g
29
The authors reported that Eq. (28) describes the
experimental results of their own and those from
other research groups well. The Sherwood number esti-
mated from Eq. (28) is plotted against Re
p
in Fig. 4
along with that estimated from the equation for single
spheres, i.e., Eq. 2. The gure indicates that, at higher
Reynolds numbers, the Sherwood number for active
particles in a uidized bed of inert particles is higher
than that for single spheres suspended in a gas stream.
However, at low Reynolds numbers, the trend is
reversed, i.e., the Sherwood number for active particles
in a uidized bed is seen to be slightly lower than that
for single spheres well-dispersed in a gas stream. This is
possibly due to the resistance by inert particles at low
Reynolds numbers. Again, the shaded area in Fig. 4
represents the range of particlegas mass transfer coef-
cient measured in a uidized bed as reviewed earlier.
Additional correlation equations for the immersed
active particles may be found elsewhere (see, e.g.,
Cobbinah et al., 1987; Hemati et al., 1992).
2.5 Summary on Homogeneous Bed Approach
The homogeneous bed approach correlates the parti-
clegas mass transfer coecients in a uidized bed in a
manner similar to those in a xed bed based on a plug-
ow model. The experimentally measured average
mass transfer coecients for the bed particles under
either the xed- or the uidized-bed operation can be
higher or lower than those estimated from the theore-
tically derived and experimentally conrmed correla-
tion equation for well-dispersed single spheres. For
ne particles, the measured average mass transfer coef-
cients can be well below the theoretical minimum for
single spheres. These low values of average mass trans-
fer coecients, however, should not be interpreted as
the violation of theoretical laws. They are caused by
the overestimation of the transfer potentials and/or the
active surfaces involved in the mass transfer. The
measured mass transfer coecients in this approach,
therefore, should be treated as empirical in nature.
Figure 5 illustrates comments made by Kunii and
Levenspiel (1991) regarding the homogeneous bed
approach.
3 BUBBLING BED APPROACH
The bubbling bed approach takes into consideration
the heterogeneous nature of a uidized bed, in parti-
cular, the coexistence of bubble and emulsion phases.
In this approach, the bubble phase is assumed to con-
sist of spherical bubbles surrounded by spherical
clouds. Historically, there have been three classes of
models proposed for describing the mass transfer pro-
cess in a uidized bed based on the bubbling bed
Figure 4 Mass transfer coecients between immersed iso-
lated active particles and a uidized bed of inert particles.
(Shaded area: range of particlegas mass transfer coecient
measured in a uidized bed of only active particles).
approach. One class is based on bubbleemulsion
transfer as represented by Kunii and Levenspiel
(1969, 1991); the second class is based on cloud
emulsion transfer as represented by Partridge and
Rowe (1966); the third class is based on empirically
correlated bubbleemulsion transfer as represented
by Chavarie and Grace (1976a). The three classes
of models are described below.
3.1 Model of Kunii and Levenspiel (1991)
Kunii and Levenspiel (1969, 1991) considered the
vaporization or sublimation of A from all particles in
the bed. They assumed that fresh gas enters the bed
only as bubbles, and that at steady state the measure of
sublimation of A is given by the increase in C
A
with
height in the bubble phase. They further assumed that
the equilibrium is rapidly established between C
A
at
the gasparticle interphase and its surroundings. The
above assumptions lead to a mass transfer equation in
terms of a bubbleemulsion mass transfer coecient,
K
GB
:
dN
A
dt
u
b
V
bubble
dC
A.b
dz
_ _
K
GB
V
bubble
C
A.c
C
A.b
30
It is worth pointing out that, in this approach, the
cloud phase and the emulsion phase are assumed to
be perfectly mixed, which leads to C
i
A
C
A.e
C
A.c
.
Figure 6a denes the term K
GB
graphically.
3.1.1 Relation Between K
GB
and k
g.bed
The above equation proposed by Kunii and Levenspiel
(1991), i.e., Eq. (30), can be derived from the tradi-
tional mass transfer expression described previously
based on the homogeneous bed approach, i.e., Eq. (7):
dN
A
dt
k
g.bed
S
ex.particles
C
i
A
C
A
7
Figure 5 Summary of particlegas mass transfer coecient.
(From Kunii and Levenspiel, 1991.)
Figure 6 Graphical description of K
GB
and K
GC
. (a)
Bubble-cloud interchange (from Kunii and Levenspiel,
1991), (b) cloud-emulsion interchange (from Partridge and
Rowe, 1966.)
With the assumption that fresh gas enters the bed only
as bubbles, Eq. (7) can be rewritten as
dN
A
dt
k
g.bed
S
ex.particles
C
i
A
C
A.b
31
where C
A.b
is the concentration of A in the bubble
phase. Equation (31) can then be rewritten as
dV
bubble
C
A.b
dt
k
g.bed
S
ex.particles
C
i
A
C
A.b
32
The above equation can be rearranged to become
V
bubble
dC
A
dt
_ _
k
g.bed
S
ex.particles
C
i
A
C
A.b
33
or
dC
A.b
dt
k
g.bed
S
ex.particles
V
bubble
C
i
A
C
A.b
34
Note that Eq. (32) implies that
dN
A
dt
V
bubble
dC
A.b
dt
35
or
dN
A
,dt
V
bubble
dC
A.b
dt
36
Then, with the nomenclature given in Fig. 7, for a
segment of bed of height dz, we can write
dt
dz
u
b
37
Inserting Eq. (37) into Eq. (34) yields
dC
A.b
dt
u
b
dC
A.b
dz
_ _
k
g.bed
S
ex.particles
V
bubble
_ _
segment
C
i
A
C
A.b
38
where the terms S
ex,particles
and V
bubble
in the bed
segment can be formulated respectively by
S
ex.particles
Adz1 e
f
a
0
39
and
V
bubble
Adz d 40
In the above equations, a
0
is the ratio of particle
surface to particle volume dened previously as
a
0

surface of a particle
volume of a particle
_ _
pd
2
p
f
s
pd
3
p
,6
6
d
p
f
s
11
Figure 7 Nomenclature used in developing the model of Kunii and Levenspiel. (From Kunii and Levenspiel, 1991.)
e
f
is the void fraction in a uidized bed as a whole, and
d is the bubble fraction in the bed segment. Equations
(39) and (40) can be combined to yield
S
ex.particles
V
bubble
_ _
segment
1 e
f
a
0
d
41
dC
A.b
dt
u
b
dC
A.b
dz
_ _
k
g.bed
1 e
f
a
0
d
C
i
A
C
A.b
42
and upon combining this with Eq. (36), the equation
becomes
dN,dt
V
bubble
u
b
dC
A.b
dz

k
g.bed
1 e
f
a
0
d
C
i
A
C
A.b
43
With the assumption that equilibrium is rapidly estab-
lished between C
A
at the gasparticle interphase and its
surroundings, i.e., C
i
A
C
A.e
C
A.c
, Eq. (43) can be
expressed as
dN
A
dt
u
b
V
bubble
dC
A.b
dz
_ _
k
g.bed
1 e
f
a
0
d
V
bubble
C
A.c
C
A.b
44
A comparison between Eqs. (44) and (30) yields
K
GB

k
g.bed
1 e
f
a
0
d
45
Note that Eq. (45) implies that
k
g.bed

d
1 e
f
a
0
K
GB
46
Combining with Eq. (11) the equation yields
Sh
bed

k
g.bed
d
p
y
D

d, 1 e
f
a
0
_ _
K
GB
d
p
y
D
yf
s
d
2
p
d
6D1 e
f
K
GB
47
It is worth mentioning that, according to Kunii and
Levenspiel (1991), K
GB
represents the transfer of A
from the particle phase to the bubble phase via two
sources, i.e., transfer from particles dispersed in the
bubble phase and transfer of gas across the bubble
cloud boundary. These are discussed below.
3.1.2 K
GB
for Nonporous and Nonadsorbing
Particles
For a uidized bed with nonporous and nonadsorbing
particles, the particles dispersed in the bubble phase
will not contribute to any additional mass transfer;
the transfer of gas across the bubblecloud boundary
therefore is the only source of mass transfer, i.e.,
K
GB
K
bc
48
where K
bc
is the bubble-cloud interchange coecient
derived by Davidson and Harrison (1963), i.e.,
K
bc
4.5
U
mf
d
b
5.85
D
0.5
g
0.25
d
1.25
b
49
Note that the above expression for K
bc
considers the
combined eects of through-ow of gas (q) and the
mass transfer coecient between bubble and cloud
(k
bc
), i.e.,
K
bc
q k
bc
S
ex.bubble
50
where
q
3p
4
U
mf
d
2
b
51
and
k
bc
0.975D
0.5
g
d
b
_ _
0.25
52
Sh
bed

yf
s
d
2
p
d
6D1 e
f
K
bc
53
The above equation provides a correlation for estimat-
ing Sh
bed
(or k
g.bed
) based on particle and uidization
properties, which then can be used to determine the
rate of mass transfer employing Eq. (7) based on the
homogeneous bed approach. In the application, the
following equations provided by Kunii and
Levenspiel (1991) are needed for calculating uidiza-
tion properties appearing in Eq. (53)
d
U U
mf
u
b
54
u
b
U U
mf
u
br
55
u
br
0.711g d
b
0.5
56
1 e
f
1 e
mf
1 d 57
The above equations indicate that, besides gas and
particle properties, the uidization properties are a
function of bubble size only.
3.1.3 K
GB
for Highly Adsorbing Particles
For highly adsorbing or sublimable particles, both
the particles dispersed in the bubble phase and the
bubblecloud gas interchange can contribute to the
particlegas mass transfer, and the expression for
K
GB
takes the form
K
GB
g
b
a
0
k
g.single
K
bc
g
b
6
d
p
f
s
k
g.single
K
bc
58
Since k
g.single
can be related to Sh
single
by
k
g.single

Sh
single
D
d
p
y
59
Equation (58) can be rearranged to become
K
GB
g
b
6
d
p
f
s
Sh
single
D
d
p
y
_ _
K
bc
g
b
6 Sh
single
D
f
s
d
2
p
y
K
bc
60
Inserting Eq. (60) in Eq. (47) yields:
Sh
bed

yf
s
d
2
p
d
6D1 e
f
g
b
6 Sh
single
D
f
s
d
2
p
y
K
bc
d
1 e
f
g
b
Sh
single
f
s
d
2
p
y
6D
K
bc
_ _ 61
Like Eq. (53) this equation provides a correlation for
estimating Sh
bed
(or k
g.bed
) based on particle and
uidization properties, which then can be used to
determine the rate of mass transfer employing Eq.
(7). Eqs. (54) through (57) are also needed for calculat-
ing uidization properties appearing in Eq. (61).
Again, all these equations indicate that, besides gas
and particle properties, the uidization properties are
a function of bubble size only. Note that, for a given
bed of solids and constant bubble size, Eq. (61) reduces
to the form (Kunii and Levenspiel, 1991):
Sh
bed
aRe
p
b 62
The equation indicates that Sh
bed
is linearly related to
the particle Reynolds number with a slope a and an
intercept b. This conclusion is similar to but not in
complete agreement with that generated empirically
based on the curve tting of experimental results,
where an exponential relationship was derived as
expressed in Eqs. (21) through (25). It may be that
the bubbles do not have a constant size over a wide
range of Re
p
. However, for a short range of Re
p
, this
conclusion seems reasonable.
3.1.4 K
GB
for Porous or Partially Adsorbing
Particles
For porous but nonadsorbing or partially adsorbing
particles, Kunii and Levenspiel (1991) derived the
following equation involving Z
d
:
Sh
bed

d
1 e
f
g
b
Sh
single
Z
d
f
s
d
2
p
y
6D
K
bc
_ _
63
where
Z
d

1
1 a,m
64
and
a
6k
g.single
t
p.mean
f
s
d
p
65
Note that, in Eq. (64), m is the adsorption equilibrium
constant dened as
C
As
mC
i
A
66
where C
As
is the concentration of tracer A within the
particle in equilibrium with the concentration C
i
A
of
tracer gas at the gasparticle interphase. Note that
for nonporous and nonadsorbing particles, m 0
and Z
d
0, and Eq. (63) reduces to Eq. (53). For
highly adsorbing particles, m is on the order of thou-
sands, in which case Z
d
!1.0 and Eq. (63) becomes
Eq. (61). For porous but nonadsorbing particles,
m e
p
, and Eq. (63) can be used to estimate Sh
bed
.
In demonstrating the eect of the particle adsorbing
property on the mass interchange coecient,
Wakabayashi and Kunii (1971) reported that the
K
GB
can be greatly enhanced even with a small value
of (g
b
Z
d
). Their reported values are shown in Table 4.
Rietema and Hoebink (1975) also reported that their
measured gas interchange coecients increase in bub-
ble size and commented that this was just the reverse of
the prediction of Eq. (49) for K
bc
. According to Kunii
and Levenspiel (1991), this is because the rst term of
Eq. (63) dominates for large bubbles, caused by more
vigorous splitting and coalescing of bubbles.
3.2 Model of Partridge and Rowe (1966)
In this class of models, the cloud surrounding the bub-
ble is considered the primary mass transfer boundary,
and the bubble and the cloud phases are considered a
perfectly mixed single phase, as indicated in Fig. 6(b).
The mass transfer equation then has the form
dN
A
dt
V
cloud
dC
A.c
dt
_ _
k
gc
S
ex.cloud
C
A.e
C
A.c
67
where V
cloud
is the volume of gas in the bubblecloud
phase, k
gc
the cloud emulsion mass transfer coecient,
and S
ex.cloud
the cloud exterior surface area. By analogy
with the transfer from a drop of one immiscible liquid
rising through another, Partridge and Rowe (1966)
proposed the mass transfer coecient to be correlated
by the Sherwood number equation
Sh
c

k
gc
d
c
D
2 0.69 Sc
0.33
Re
0.5
c
68
where Sc is the Schmidt number, dened previously as
Sc
m
r
g
D
4
and Re
c
is a Reynolds number dened in terms of the
relative velocity, u
R
, between rising cloud and emulsion
(u
R
u
c
u
e
), dened as
Re
c

u
R
d
c
r
g
m
69
In this equation, d
c
is the diameter of the sphere with
the same volume as the cloud, while D, m. and r
g
represent diusivity, viscosity, and density of the ui-
dizing gas, respectively.
The mass transfer equation given in Eq. (67) can
also be expressed in terms of cloud-to-emulsion gas
interchange coecient, K
GC
, as follows:
dN
A
dt
V
cloud
dC
A.c
dt
_ _
K
GC
V
cloud
C
A.e
C
A.c
70
where K
GC
can be related to k
gc
by Partridge and
Rowe, 1966)
K
GC

k
gc
pd
2
c
e
mf
V
cloud
3.9e
mf
DSh
c
V
cloud
0.67
71
Equations (70) and (71) were used in the models of
Rowe (1963) and Partridge and Rowe (1966). It is
essential to note that the equation assumes a purely
diusive mechanism and does not consider gas
through-ow between the cloud and the emulsion
phases.
3.3 Model of Chavarie and Grace (1976a)
Chavarie and Grace (1976a) measured mass transfer
rates for bubbles containing ozone injected into an
air-uidized two-dimensional bed and proposed the
following empirical equation for their results:
dN
A
dt
V
bubble
dC
A.b
dt
_ _
k
gt
S
ex.bubble
C
A.e
C
A.b
72
They compared the measured k
gt
with the predictions
of various models available in the literature and found
that those which assumed a combination of through-
ow and diusion tend to overestimate the rate of
transfer, while those which assume a purely diusive
mechanism underestimate it. Table 5 illustrates the
results from various models. They concluded that the
through-ow equation proposed by Murray (1965)
best t their data, which lead to
k
gt

U
mf
p
73
In a follow-up attempt, Sit and Grace (1978)
derived the following equations to include the diusive
mechanism in the expression. For a circular two-
dimensional bubble, the equation has the form
k
gt

U
mf
p

4De
mf
u
b
pd
b
_ _
0.5
74
Table 4 Effect of g
b
Z
d
on K
GB
d
b.m
(m) 0.04 0.06 0.08 0.11
K
GB.measured
(s
1
) 1116 918 814 711
calculated, with g
b
Z
d
0 5 3 2 1.6
K
GB
calculated, with g
b
Z
d
4 10
4
13 12 12 11
Source: Kunii and Levenspiel, 1991
and for a spherical three-dimensional bubble, it has the
form
k
gt

U
mf
4

4De
mf
u
b
pd
b
_ _
0.5
75
The above equation can be converted to K
GT
, the mass
interchange coecient per unit volume of bubble,
based on the appropriate surface-to-volume ratio, i.e.,
K
GT

6
d
b
k
gt
1.5U
mf
d
b
12
d
1.5
b
De
mf
u
b
pd
b
_ _
0.5
76
The corresponding mass transfer equation for K
GT
appearing in Eq. (76) is
dN
A
dt
_ _
V
bubble
dC
A.b
dt
_ _
K
GT
V
bubble
C
A.e
C
A.b
77
Note that this equation is essentially in the same form
as that of Eq. (30) proposed by Kunii and Levenspiel
(1991), i.e., K
GT
reported by Sit and Grace (1978) is
directly comparable to K
GB
described previously by
Kunii and Levenspiel (1991).
In a similar attempt, Hatano and Ishida (1986) stu-
died the particlegas mass transfer coecient in a
three-dimensional uidized bed of nonadsorbing glass
beads with d
p
0.18 mm using optical ber probes.
Tracer gas concentrations in and around rising single
bubbles was measured continuously by the penetrative
probes, while the bubble boundary and the zone with
the prominent particle movement were detected by
Table 5 Interphase Mass Transfer Models and Coefcients for Ozone Exchange in an Air-Fluidized Two-Dimensional
Bed
Model Reference Coecient
k
gt
equivalent
Calculated
k
gt
(m/s)
1. Penetration Chavarie
(1973)
k
gc

4De
mf
u
b
d
c
k
gc 0.0051
2. Cloud two lm Walker
(1970)
k
gc

0.93De
mf
U
mf
a 1
p
d
c
2
3
k
gc
0.0017
3. Partridge and Rowe Partridge and Rowe
(1966)
k
gc

0.26DRe
0.6
c
Sc
0.33
d
c
c
mf
k
gc 0.0020
4. Chiba and Kobayashi Chiba and Kobayashi
(1970)
k
gc

4De
2
mf
u
b
a 1
pd
c
a
_
k
gc 0.0030
5. Kunnii and Levenspiel Kunii and Levenspiel
(1969)
k
gb
0.6D
1,2
g,d
b
1,4
q 2U
mf
d
b
w
Q q k
gb
S
b
k
gc

4De
mf
u
b
pd
c
kgcQ
k
gc
S
b
Q
0.0051
6. Toei et al. Toei et al. (1969)
k
gc

1.02e
mf
1 2e
mf
,a 1
u
b
Do 1
d
b
a
a 1
a 1
_
_
k
gc
k
gs 0.0056
7. Murray through-ow,
no diusion
Murray (1965) k
gb
0
q U
mf
d
b
w
q,S
b 0.0160
Empirical value (from Charvarie and Grace, 1976a) 0.0160
Source: Chavarie and Grace, 1976a.
reective probes. The measurements provided neces-
sary data for determining the interphase mass transfer
coecient. The following equation resulted from their
study:
k
gt
0.127D
0.33
g
0.33
u
b
e
mf
U
_ _
0.22
78
where u
b
can be calculated from Eq. (55). Note that, in
addition to single-bubble experiments, they also
measured the interphase mass transfer coecient asso-
ciated with paired bubbles. However, they found that
the mass transfer coecient remains unchanged in
both single-bubble and paired-bubble conditions.
This nding apparently is dierent from that reported
in the literature from other researchers, as reviewed
below.
3.3.1 Correlation for Multiple-Bubble Beds
The correlation equations described previously for
K
GC
and K
GT
are all based on single-bubble experi-
ments. Although they provide valuable insight and
lead to useful comparisons, these correlation equations
may not be directly applicable to freely bubbling three-
dimensional beds. In such beds, complex bubbling
phenomena such as growth, splitting, coalescence,
and wake shedding occur constantly along the bed,
and the interphase mass transfer mechanism is
expected to be much more complex than those in sin-
gle-bubble beds. Sit and Grace (1981) used a noninter-
fering technique to measure the concentration of ozone
in pairs of bubbles injected into a bed of 390 mm glass
beads uidized by ozone-free air. They found that the
transfer of ozone tracer from the bubble phase to the
emulsion phase is enhanced over the transfer from iso-
lated bubbles in the same particles and the same col-
umn. Bubble growth is also greater for the case where
pairs of bubbles are introduced than when bubbles are
present in isolation. They also found that enhancement
of interphase mass transfer for interacting bubbles
increases with particle size and can be explained in
terms of enhancement of the through-ow component
of transfer, while the diusive component remains
unaltered.
Their study has resulted in the following two mod-
ied forms of Eqs. (74) and (75) for a freely bubbling,
two- or three-dimensional uidized bed. For two-
dimensional beds, the modied equation has the form
k
gt.m
0.4U
mf

4De
mf
u
b.m
pd
b.m
_ _
0.5
79
and for three-dimensional beds, the modied equation
is
k
gt.m

U
mf
3

4De
mf
u
b.m
pd
b.m
_ _
0.5
80
Again, as with Eq. (76) for three-dimensional beds,
k
gt.m
can be converted to K
GT.m
by the expression
K
GT.m

6
d
b.m
_ _
k
gt.m
2.0U
mf
d
b.m
12
d
1.5
b.m
De
mf
u
b.m
pd
b.m
_ _
0.5
81
According to Yates (1983), Eqs. (80) and (81) would
seem to give the best available values of interphase
mass coecients in three-dimensional beds of nonad-
sorbing particles. Note again that K
GT.m
appearing in
Eq. (81) is directly comparable to K
GB
appearing in
Eq. (30) proposed by Kunii and Levenspiel (1991).
3.4 Mass Transfer in the Grid Region
The grid region plays an important role in determining
the reaction conversion of uidized bed reactors, espe-
cially for fast reactions where the mass transfer opera-
tion is the controlling mechanism. Experimental
studies have indicated that changing from one distri-
butor to another, all other conditions remaining xed,
can cause major changes in conversion (Cooke et al.,
1968; Behie and Kehoe, 1973; Bauer and Werther,
1981). It is generally observed that, in the grid region,
additional mass transfer can take place because of the
convective ow of gas through the interphase of the
forming bubbles. The ow of gas through the forming
bubbles into the dense phase and then returning to the
bubble phase higher in the bed represents a net
exchange between the two phases. The experimental
work of Behie and Kehoe (1973) indicated that the
mass transfer coecient in the grid region, k
je
, can be
40 to 60 times that in the bubble region, i.e., k
be
.
Sit and Grace (1986) measured time-averaged con-
centrations of methane in the entry region of beds of
120 to 310 mm particles contained in a 152 mm column
with a central orice of diameter 6.4 mm and auxiliary
tracer-free gas. The following equations were proposed
for the particlegas mass transfer in this region:
V
bubble
dC
A.b
dt
_ _
Q
or
C
A.or
k
be1
S
ex.bubble
C
A.e
Q
or
k
be1
S
ex.bubble
_ _
C
A.b
82
where Q
or
represents the convective mechanism dened
as
Q
or
u
or
p
4
_ _
d
2
or
83
and k
be1
represents the diusive mechanism, which can
be predicted by the penetration theory expression, i.e.,
k
be1

4De
mf
pt
f
_ _
0.5
84
Their results indicated that the convective mechanism
has a greater eect on the grid region mass transfer.
Since bubbles are also smaller near the grid, they con-
cluded that favorable gassolid contacting occurs in
this region primarily because of convective outow
followed by recapture.
3.5 Effect of Particle Adsorption
As pointed out by Kunii and Levenspiel (1991), the
mass transfer rate in a uidized bed can be substan-
tially enhanced for highly adsorbing particles even
though the amount of particles dispersed in the bubble
phase may be small. Similar observations have also
been reported in the literature, see, e.g., Yoshida and
Kunii (1968), Drinkenburg and Rietema (1973), Yates
and Constans (1973), Chiba and Kobayashi (1970),
Nguyen and Porter (1976), and Gupalo et al. (1978).
Among these investigators, Chiba and Kobayashi
(1970) derived a theoretical expression for the ratio
of interphase mass transfer coecients in the presence
(K
0
be
) and absence (K
be
) of adsorption eects:
K
0
be
K
be
1 0.67m
1 e
mf
e
mf
1.5
a
0
a
0
1
_ _ _ _
0.5
85
where m is the adsorption equilibrium constant dened
in Eq. (66) and a
0
is dened as
a
0

u
b
e
mf
U
mf
86
Equation (85) was later conrmed experimentally by
Bohle and van Swaaij (1978), who measured mass
transfer coecients for a number of adsorbing (e.g.,
propane) and nonadsorbing (e.g., helium) gases in a
uidized bed of silica-alumina. A typical comparison
is shown in Fig. 8, where the eect of adsorption is
obvious with the enhancement being as high as 100%.
4 RELATION BETWEEN MASS AND HEAT
TRANSFER COEFFICIENTS
The correlation equations for particlegas mass and
heat transfer coecients are closely related when the
Sherwood number (Sh) is equivalent to the Nusselt
number (Nu), and the Schmidt number (Sc) is equiva-
lent to the Prandtl number (Pr). All the particlegas
correlation equations for the Sherwood number and
the Nusselt number are therefore interchangable. For
example, the correlation equation for estimating the
Nusselt number for particles in a xed bed is expressed
as
Nu
bed

hd
p
k
g
2 1.8Re
p
0.5
Pr
0.333
for Re
p
> 80
87
whereas the Prandtl number (Pr) is dened as
Pr
C
p
m
k
g
88
Note that the corresponding equation for the
Sherwood number has the form
Sh
bed

k
g.bed
d
p
y
D
2 1.8Re
p
0.5
Sc
0.333
for Re
p
> 80
89
Figure 8 Dependence of interphase mass transfer rate on
has adsorptivity. (From Bohle and van Swaaij, 1978.)
The above discussion implies that the correlation equa-
tions derived for Nu
bed
based on heat transfer experi-
ments can also be converted to the Sherwood number
for estimating k
g.particles
.
5 CONCLUDING REMARKS FROM KUNII
AND LEVENSPIEL (1991)
It is worth bringing up the concluding remarks of
Kunii and Levenspiel (1991) at this point to conclude
the chapter regarding particlegas mass transfer in a
uidized bed. Three points were made: rst, particles
dispersed in bubbles should be taken into account
when kinetic processes, such as mass transfer, are car-
ried out in uidized beds; second, when dealing with a
gaseous component that is adsorbed or somewhat cap-
tured by the bed solids (or else desorbed), K
bc
should
be used carefully to represent the movement of these
adsorbed gaseous components; third, the mass transfer
coecient measured for the bed as a whole, k
g.bed
, is
model dependent. For large-particle cloudless bubble
beds, the plug ow model closely matches the ow
conditions in the bed, and the k
g.bed
should match
the single particle coecient, i.e., k
g.single
. However,
for ne-particle clouded bubble beds, the k
g.bed
may
be well below that of k
g.single
.
6 EXAMPLE CALCULATIONS
Two examples are given here to demonstrate the eects
of d
b
and g
b
on the prediction of the particlegas mass
transfer coecient based on Eq. (61). The calculated
results are then compared with the experimental data
of Resnick and White (1949).
6.1 Example 1 Effect of d
b
on Sh
bed
based on
Eq. (61)
Data. r
s
1.06 g,cm
3
, e
mf
0.5, f
s
0.4, g
b

0.005, r
g
1.18 10
3
g,cm
3
, m 1.8 10
4
g,cms, D 0.065 cm
2
,s, Sc 2.35, Z
d
1.0,
y 1.0, U
mf
1.21 cm,s, u
t
69 cm,s.
Determine. The relation between Sh
bed
and Re
p
based on Eq. (61) with d
b
0.1, 0.2, 0.37, 0.5, and
1 cm.
Solution. Equation (61) has the form
Sh
bed

d
1 e
f
g
b
Sh
single
f
s
d
2
p
y
6D
K
bc
_ _
From Eq. (6)
Re
p

r
g
ud
p
m
At u
t
69 cm,s.
Re
p.t

r
g
u
t
d
p
m

0.00118690.028
0.00018
12.66
Then Eq. (2) gives
Sh
single
2 0.612.66
0.5
2.35
0.33
4.84 A
From Eq. (49) (using d
b
0.37 cm),
K
bc
4.5
U
mf
d
b
5.85
D
0.5
g
0.25
d
1.25
b
4.5
1.21
0.37
_ _
5.85
0.065
0.5
980
0.25
0.37
1.25
43.63 s
1
B
Combining Eqs. (54) through (57), i.e.,
d
U U
mf
u
b
u
b
U U
mf
u
br
u
br
0.711gd
b
0.5
1 e
f
1 e
mf
1 d
we obtain
d
1 e
f
U U
mf
u
br
1 e
mf
U 1.21
0.711980 0.37
0.5
1 0.5
U 1.21
6.77
C
Also
f
s
d
2
p
y
6D

0.40.028
2
1
6 0.065
8.04 10
4
D
Combining (A) through (D) yields
Sh
bed

d
1 e
f
g
b
Sh
single
f
s
d
2
p
y
6D
K
bc
U 1.21
6.77
0.0054.84 0.00080443.63
0.0088U 0.011 E
Then, since
Re
p

r
g
Ud
p
m

0.028U0.00118
0.00018
0.184U F
Combining expressions (E) and (F) gives
Sh
bed
0.048 Re
p
0.011 G
Note that the above equation is generated based on
d
b
0.37 cm. This calculation procedure can be easily
repeated for dierent bubble sizes at 0.1, 0.2, 0.5, and
1.0 cm. The generated equations for these dierent
bubble sizes are plotted in Fig. 9 along with the experi-
mentally observed data of Resnick and White (1949).
The results indicate that the correlation based on d
b

0.37 cm describes the experimental data best. The
results in Fig. 9, however, indicate that the Sherwood
number is sensitive to the bubble size selected; a smal-
ler bubble size generates a higher Sherwood number,
i.e., a higher mass transfer coecient.
6.2 Example 2. Effect of c
b
on Sh
bed
based on Eq.
(61)
Data. Same as Example 1; but, with d
b

0.37 cm.
Determine. The relation between Sh
bed
and Re
p
b
0.37 cm and g
b
0.001,
0.005, and 0.01.
Solution. The identical procedure as is demon-
strated in Example 1 can be used to evaluate Sh
bed
at
different g
b
. The results are shown in Fig. 10 where,
as expected, a higher g
b
will generate a higher Sher-
wood number. The results in Fig. 10 clearly indicate
that, in addition to bubble size, the selection of g
b
is
also essential in the estimation of the particlegas
mass transfer coeffcient based on Eq. (61) proposed
by Kunii and Levenspiel (1991).
NOMENCLATURE
A = cross-sectional area of bed, m
2
Ar = Archimedes number dened in Eq. (29),
Ar gd
3
p
r
p
r
g
,n
2
r
g
)
a
0
= surface areas of solid per volume of solid,
m
1
C
A
= concentration of A in bulk gas phase, kg-mol/
m
3
C
i
A
= concentration of A at gasparticle interphase,
kg-mol/m
3
C
A.b
= concentration of A in bubble phase, kg-mol/
m
3
Figure 9 Eect of bubble size on Sh
bed
calculated in
Example 1.
Figure 10 Eect of g
b
b
0.37 cm.
C
A.c
= concentration of A in cloud phase, kg-mol/m
3
C
A.e
= concentration of A in emulsion phase, kg-
mol/m
3
C
As
= concentration of A within a particle dened in
Eq. (65), kg-mol/m
3
C
p
= heat capacity at constant pressure,
d = diameter, m
d
b
= eective bubble diameter, m
d
b.m
= mean eective bubble diameter dened in Eqs.
(79) and (80), m
d
c
= cloud diameter appearing in Eq. (68), m
d
eff
= eective particle diameter dened in Eqs.
(13)(15), m
d
or
= orice diameter, m
d
p
d
p
= particle diameter dened in Eq. (20), cm
d
p.a
= particle diameter of active particles, m
d
pi
= particle diameter of inert particles, m
d
p.m
= mean particle diameter appearing in Table 3,
cm
d
sph
= equivalent spherical diameter of a particle, m
d
t
= bed diameter appearing in Table 3, cm
g = gravitational acceleration, 9.8 m/s
2
h = height or thickness, m
h = particlegas heat transfer coecient, Wm
2
K
1
K
bc
= coecient of gas interchange between bubble
and cloud phase, m
3
/s
K
0
be
= K
be
associated with adsorbing particles, m
3
/s
K
GB
= interchange coecient between bubble and
cloudemulsion, s
1
K
GC
= interchange coecient between bubble-cloud
and emulsion, s
1
K
GT
= K
GB
appearing in Eqs. (76) and (77), s
1
K
GT.m
= mean K
GB
appearing in Eq. (81), s
1
K
t
= permeability dened in Eq. (19)
k
bc
= mass transfer coecient between bubble and
cloud, m/s
k
be
= bubble-to-emulsion mass transfer coecient,
m/s
k
be1
= bubble-to-emulsion mass transfer coecient
during bubble formation, m/s
k
g
= thermal conductivity of gas, Wm
1
K
1
k
g.a
= mass transfer coecient of active particles
immersed in a uidized bed, m/s
k
g.bed
= average mass transfer coecient of bed
particles, m/s
k
g.single
= mass transfer coecient of well-dispersed
single spheres, m/s
k
gc
= mass transfer coecient dened in Eq. (66)
based on cloud surface, m/s
k
gt
= mass transfer coecient dened in Eq. (72)
based on bubble surface, m/s
k
gt.m
= mean mass transfer coecient dened in Eqs.
(79) and (80), m/s
k
je
= jet to emulsion mass transfer coecient, m/s
L
m
= bed height appearing in Table 2, cm
L = bed height dened in Eq. (16), cm
m = Adsorption equilibrium constant, dened in
Eq. (65)
N
A
= mole of A, kg-mol
Nu
bed
= Nusselt number of bed particles,
Nu
bed
hd
p
,k
g
P
p
= pressure drop through porous beds appearing
in Eq. (19), g/cm
2
Q = air ow rate appearing in Eq. (19), cm
3
/s
Q
or
= gas ow through orice, m
3
/s
q = through-ow of gas dened in Eq. (51), m
3
/s
Re
c
= Reynolds number based on cloud properties,
Re
c
r
g
u
c
d
c
,m
Re
p
= particle Reynolds number, Re
p
r
g
Ud
p
,m
Re
sph
= particle Reynolds number based on d
sph
,
Re
p
r
g
Ud
sph
,m
S = surface area per unit bulk volume of bed
dened in Eqs. (16) and (17), cm
1
S
a
= specic surface area per unit volume of solid
as dened in Eqs. (17) and (18), cm
1
Sc = Schmidt number, Sc m,r
g
D)
S
d
= S
d
Sd
p
) as dened in Eq. (20)
S
ex.single
= exterior surface of a single sphere or well-
dispersed single spheres, m
2
S
ex.particles
= total exterior surface of bed particles, m
2
S
ex.bubble
= exterior surface of bubbles, m
2
S
ex.cloud
= exterior surface of clouds, m
2
Sh = Sherwood number, Sh k
g
d
p
y,D
Sh
a
= Sherwood number of isolated active particles,
Sh
a
k
g.a
d
p
y,D
Sh
bed
= average Sherwood number of bed particles,
Sh
bed
k
g.bed
d
p
y,D
Sh
limit
= limit Sherwood number dened in Eq. (27)
Sh
single
= Sherwood number of single spheres, Sh
single

k
g.single
d
p
y,D
t = time, s
t
f
= bubble formation time, s
t
p.mean
= mean particle residence time staying in bubble
phase, s
U
mf
u = gas velocity, m/s
u
b
= bubble rise velocity, m/s
u
b.m
= mean bubble rise velocity dened in Eqs. (79)
and (80), m/s
u
br
= rise velocity of a bubble with respect to the
emulsion phase, m/s
u
c
= cloud (= bubble) rise velocity, m/s
u
e
= velocity of the emulsion gas, m/s
u
or
= velocity of gas through orice, m/s
u
R
= relative velocity, dened as u
R
u
c
u
e
u
t
= particle terminal velocity, m/s
V
bubble
= volume occupied by bubble phase, m
3
V
cloud
= volume occupied by bubblecloud phase, m
3
V
seg
= volume of a bed segment, m
3
X = void fraction of bed appearing in Eqs. (17)
and (18)
y = mole fraction of inert component
z = distance above the distributor, m
a = parameter of Eq. (64), dened in Eq. (65)
a
0
= parameter of Eq. (85), dened in Eq. (86)
g
b
= volume of solids dispersed in bubbles per unit
bubble volume
D = gas diusion coecient, m
2
/s
d = bubble fraction in a uidized bed
e
bed
= void fraction in a xed bed
e
f
= void fraction in a uidized bed as a whole
e
mf
= void fraction in a bed at minimum uidization
velocity
e
p
= void fraction in a uidized bed particle
m = gas viscosity, kg m
1
s
1
n = kinematic viscosity, m
2
/s
r
g
= density of gas, kg/m
3
r
p
= density of particle, kg/m
3
Z
d
= dispersion parameter dened in Eq. (64)
f
s
= sphericity of a particle
p = constant, 3.1416
REFERENCES
Agarwal PK, La Nauze RD. Transfer processes local to the
coal particle: a review of drying, devolatilization and mass
transfer in uidized bed combustion. Chem Eng Res Des
67:457480, 1989.
Avedesian MM. Combustion of char in uidized beds. PhD
dissertation: University of Cambridge, 1972.
Bauer W, Werther J. Proceedings of 2nd World Congress
Chemical Engineering, 1981, p 69.
Behie LA, Kehoe P. The grid region in a uidized bed reac-
tor. AIChE J 19:10701072, 1973.
Behie LA, Bergougnou MA, Baker CGJ. Mass transfer from
a grid jet in a large gas uidized bed. In: Keairns DL, ed.
Fluidization Technology, Vol 1. Washington DC:
Bohle H, Van Swaaij WPM. In: Davidson JF, Keairns DL,
eds. Fluidization. London: Cambridge University Press,
1978, p 167.
Brown GG. Unit Operations. New York: John Wiley, 1950.
Carmon PC. Fluid ow through granular beds. Trans Inst
Chem Eng 15:150166, 1937.
Carmon PC. Symposium on New Methods for Particle Size
Determination in the Subsieve Range. Amer Soc Testing
Materials, 1941, p 24.
Chavarie C. Chemical reaction and interphase mass-transfer
in gas uidized beds. PhD dissertation, McGill University,
1973.
Chavarie C, Grace JR. Interphase mass transfer in a uidized
bed. In: Keairns DL, ed. Fluidization Technology, Vol 1.
Washington DC: Hemisphere, 1976a, pp 251259.
Chavarie C, Grace JR. Interphase mass transfer in a gas-
uidized bed. Chem Eng Sci 31:741749, 1976b.
Chiba T, Kobayashi H. Gas exchange between the bubble
and emulsion phases in gassolid uidized beds. Chem
Eng Sci 25:13751385, 1970.
Chu JC, Kalil J, Wetteroth WA. Mass transfer in a uidized
bed. Chem Eng Prog 49:141149, 1953.
Cobbinah S, Laguerie C, Gilbert H. Simultaneous heat and
mass transfer between a uidized bed of ne particles and
immersed coarse porous particles. Int Heat and Mass
Transfer 30:395400, 1987.
Cooke MJ, Harris W, Highley J, Williams DF. Int Chem
Engrs Symp Ser 30:21, 1968.
Davidson JF, Harrison D. Fluidised Particles. London:
Froessling N. The evaporation of falling drops. Gerland
Beitr Geophys 52:170, 1938.
Gupalo YP, Ryazantsez YS, Sergeez YA. In: Davidson JF,
Keairns DL, eds. Fluidization. London: Cambridge
University Press, 1978, p 162.
Hatano H, Ishida M. In: Ostergaard K, Soorensen A, eds.
Fluidization V, Elsionore, Denmark. New York:
Hemati M, Mourad M, Steinmetz D, Laguerie C.
Continuous and intermittent drying of maize in a otation
uidized bed. In: Potter OE, Nicklin DJ, eds. Fluidization
VII, Brisbane, Australia. New York: Engineering
Hsiung TH, Thodos G. Mass transfer in gas-uidized beds:
measurement of actual driving forces. Chem Eng Sci
32:581592, 1977.
Hurt DM. Principles of reactor design: gas-solid interface
reactions. Ind Eng Chem 35:522528, 1943.
Kettenring KN, Mandereld EL, Smith JM. Heat and mass
transfer in uidized systems. Chem Eng Prog 46:139145,
1950.
Kok J, Stark NL, van Swaaij WPM. In: Ostergaard K,
Krell L, Kunne HJ, Morl L. Flow regimes and heat and mass
transfer in gas-uidized beds of solids. Internat Chem Eng
30:4556, 1990.
Kunii D, Levenspiel O. Fluidization Engineering. New York:
John Wiley, 1969.
Kunii D, Levenspiel O. Fluidization Engineering, 2nd ed.
Leva M, Grummer M, Weintraub M, Pollchick M.
Introduction to uidization. Chem Eng Prog 44:511
520, 1948.
Leva M, Grummer M, Weintraub M, Storch HH. A study of
uidization of an iron schertropsch catalyst. Chem Eng
Prog 44:707716, 1948.
Leva M, Weintraub M, Grummer M, Pollchick M.
Fluidization of an anthracite coal. Ind Eng Chem
41:12061212, 1949.
Levenspiel O. Chemical Reaction Engineering, 2nd ed. New
Linjewile TM, Agarwal PK. The 23rd Symposium on
Combustion. Combustion Inst, 1990, p 917.
Murray JD. On the mathematics of uidization: Part 1.
Fundamental equations and wave propagation. J Fluid
Mech 21:465493, 1965.
Nguyen HV, Potter OE. Adsorption eects in uidized beds.
In: Keairns DL, ed. Fluidization Technology, Vol 2.
Washington DC: Hemisphere, 1976, pp 193200.
Palchonok GI, Tamarin AI. Mass transfer at a moving par-
ticle in a uidized bed of coarse material. J Eng Phys
47:916922, 1984.
Palchonok GI, Dolidovich AF, Andersson S, Leckner B.
Calculation of true heat and mass transfer coecients
between particles and a uidized bed. In: Potter OE,
Nicklin DJ, eds. Fluidization VII, Brisbane, Australia.
Partridge BA, Rowe PN. Chemical reaction in a bubbling
gas-uidised bed. Trans Inst Chem Engrs 44:335348,
1966.
Ranz WE. Friction and transfer coecients for single parti-
cles and pack beds. Chem Eng Prog 48:247253, 1952.
Resnick WE, White RR. Mass transfer in systems of gas and
uidized solids. Chem Eng Prog 45:377390, 1949.
Richardson JF, Szekely J. Mass transfer in a uidised bed.
Trans Inst Chem Engrs 39:212222, 1961.
Rietema K, Hoebink J. In: Keairns DL, ed. Fluidization
Technology, Vol 1. Washington DC: Hemisphere, 1976,
pp 279288.
Sit SP, Grace JR. Interphase mass transfer in an aggregative
uidized bed. Chem Eng Sci 33:11151122, 1978.
Sit SP, Grace JR. Eect of bubble interaction on interphase
mass transfer in gas uidized beds. Chem Eng Sci 36:327
335, 1981.
Sit SP, Grace JR. Interphase mass transfer during bubble
formation in uidized beds. In: Ostergaard K,
Thodos G, Riccetti RE. Mass transfer in the ow of gases
through uidized beds. AIChE J 7:442444, 1961.
Thodos G, Wilkin GS. In: Drinkenburg AAH, ed. Proc Int
Symp on Fluidization. Amsterdam: Netherlands
Thodos G, Wilkin GS. Mass transfer driving forces in packed
and uidized beds. AIChE J 15:47-50, 1969.
Thodos G, Yoon P. Mass transfer in the ow of gases
through shallow uidized beds. Chem Eng Sci 27:1549
1554, 1972.
Toei R, Matsuno R, Miyigawa H, Nishitani K, Komagawa
Y. Gas transfer between a bubble and the continuous
phase in a gas-solid uidized bed. Internat Chem Eng
9:358364, 1969.
Uchida S, Fujita S. J Chem Soc, Ind Eng Section (Japan)
37:1578 (1583, 1589, 1707), 1934.
Vandershuren I, Delvossalle C. Particle-to-particle heat
transfer in uidized bed drying. Chem Eng Sci 35:1741
1748, 1980.
Wakabayashi T, Kunii D. Contribution of solids dispersed in
bubbles to mass transfer in uidized bed. J Chem Eng
(Japan) 4:226230, 1971.
Walker BV. Gas-solid contacting in bubbling uidised beds.
PhD dissertation, Cambridge University, 1970.
Yates JG. Fundamentals of Fluidized-Bed Chemical
Processes. London: Butterworths, 1983.
Yates JG, Constans JAP. Residence time distributions in a
uidized bed in which gas adsorption occurs: stimulus
response experiments. Chem Eng Sci 28:13411347, 1973.
Yoshida K, Kunii D. Stimulus and response of gas concen-
tration in bubbling uidized beds. J Chem Eng (Japan)
1:1116, 1968.
Zenz FA. Encyclopedia of Chemical Technology, Chap 10,
3rd ed. Kirk Olhmer, 1978, p 548.
12
General Approaches to Reactor Design
Peijun Jiang, Fei Wei,* and Liang-Shih Fan
The Ohio State University, Columbus, Ohio, U.S.A.
1 INTRODUCTION
Fluidized bed reactors have a number of characteristics
ideal for industrial applications and hold advantages
over xed bed reactors (e.g., Kunii and Levenspiel,
1991; Fan and Zhu, 1998). Among these characteristics
are good particle mixing, good temperature control,
and adaptability to high-pressure and high-tempera-
ture operations. Fluidized solid particles behave like
liquids, allowing online particle addition/removal to
adjust the catalyst activity and reducing downtime
for catalyst replacement. Furthermore, this behavior
permits easy transport of particles between reactors,
adaptation of continuous operation of solids, more
precise and automatic control, and greater exibility
in system conguration. Fluidized bed reactors can
be operated over a wide range of gas ow rates,
thereby allowing selection of the optimum contact
time or residence time of gas and solid particles. Due
to simple geometric congurations, uidized bed
reactors are suitable for large-scale operations. Other
features of uidized bed reactors are
The rapid mixing of solids, due to bubbles and
strong turbulent ow, leads to isothermal and
hot-spot-free operations.
A bed of well-mixed solids represents a large
thermal ywheel and thus responds slowly to
abrupt changes in operating conditions.
A exible process temperature control enables
optimization of other process variables to
increase product yields and minimize wastes
and by products.
The circulations of solids between two uidized
beds allows continuous catalyst regeneration.
The use of small particles gives rise to eectiveness
factors close to unity
A high gas-to-particle mass and heat transfer rates
can be established.
Some cohesive particles can be processed due to
strong particle motion and interactions.
Disadvantages of uidized bed reactors for industrial
applications include
The gas bypass, in the form of bubbles, reduces the
gassolid contact eciency in bubbling beds.
Poor gassolid contact results from particle accu-
mulation near the wall in risers.
A complicated bubble hydrodynamics and mixing
patterns require special development eorts for
process scaleup.
Nonuniform products are produced owing to wide
distributions of gas and solids residence time in
the bubbling regime.
The strong backmixing for both particles and gas
results in low conversion and poor selectivity.
The strong solids motion leads to erosion of internal
parts.
The attrition of catalyst particles leads to catalyst
loss in cyclones.
*Current aliation: Tsinghua University, Beijing, Peoples
Republic of China
Particle entrainment leads to loss of particles and air
pollution.
1.1 Features and Commercial Processes of Fluidized
Bed Reactors
Fluidized beds have been applied to physical, chemical,
metallurgical, mineral, and other operations. Examples
of these applications are given in Table 1.
Fluidized bed reactors are often considered for
implementation because of the solids mixing require-
ment for temperature uniformity, highly endothermic
or exothermic reactions, small size of particles, contin-
uous solids ow, and/or short gassolid contact time.
Catalysts are commonly the uidizing particles in the
uidized reactor for chemical synthesis. Solid reactants
are commonly utilized in metallurgical and mineral
processing. Operating conditions required for each
application primarily depend on the process rate
derived from the transport and kinetics of the reactions
concerned. For fast reactions, high-velocity uidiza-
tion or a fast uidization regime is used. For slow
reactions, a low-velocity uidization or bubbling ui-
dization regime is used. In the bubbling beds, gas ows
through the interstitial space between particles and
predominantly bypasses the bed in the form of bubbles
without eectively contacting solid particles. On the
other hand, bubbles act as agitators to enhance gas
solid mixing. Fluidized bed reactors typically consist of
a vertical column (reactor), a gas distributor, cyclones,
heat exchangers, an expanded section, and baes, as
shown in Fig. 1a. The gas distributor provides desired
distributions of uidizing gas and support for particles
in the bed. Generally, a minimum pressure drop across
the distributor is required to ensure the uniformity of
gas distribution in the bed. Gas distributors can be
designed in many ways, examples being sandwiched,
stagger perforated, dished, grated, porous, tuyere,
and cap types (Karri and Knowlton, 1999). In the
bubbling regime, the distributor geometry strongly
inuences the initial bubble size from the distributor,
which may then aect the bubble size in the bed. The
cyclone separates solid particles from the outlet gas
and returns solid particles back into the dense bed
through the dipleg. Several cyclones may be combined
to form a multistage cyclone system that resides either
inside or outside the bed. Although cyclones are the
most widely adopted gassolid separators in uidized
beds, other types of gassolid separators, such as fast
separator and lter, are also employed. The expanded
Table 1 Examples of Industrial Applications of Fluidized Bed Reactors
Physical operations Chemical syntheses
Metallurgical and
mineral processes Other
Heat exchange,
catalyst cooler
Phthalic anhydride
synthesis
Uranium processing Coal combustion
Solids blending Propylene ammoxidation
to acrylonitrile
Reduction of iron
oxide
Coal gasication
Particle coating Maleic anhydride
synthesis
Metal heat treatment Fluidized catalytic
cracking
Drying (PVC,
MBS, ABS)
Ethylene dichloride
synthesis
Roasting of sulde
ores
Incineration of
solids waste
Adsorption Methanol to gasoline
and olen processes
Crystalline silicon
production
Cement clinker
production
Granulation,
drying of yeast
Syngas to gasoline Titanium dioxide
production
Microorganism
cultivation
Binding Dimethylbenzene (m)
ammoxidation to
dinitrilebenzene (m)
Calcination of
AlOH
3
Coal plasma pyrolysis
to acetylene
Particle separation Vinyl acetate synthesis Pyrolysis of oil shale Biomass pyrolysis
Nitrobenzene hydrogenation
to aminobenzene
Silicon chloride to
SiCl
3
H
Olen polymerization
section on the top of the uidized bed reduces the
linear gas velocity in the freeboard, thereby enhancing
particle settling. The expanded section may not always
be necessary and depends on the operating conditions
and design of the gassolid separator. The heat
exchanger removes generated heat or adds required
heat to the uidized bed by a cooling or heating
uid. Heat exchangers can be either immersed in the
dense bed or placed both in the dense bed and the
freeboard. Heat exchangers are also placed along the
wall as in uidized bed coal combustors. The proper
design of baes or other types of internal structure
reorganize ow, enhance the breakup of bubbles,
promote gassolid contact, and reduce particle entrain-
ment. Commonly, baes refer to any internal
structures other than diplegs and immersed heat
exchangers, although the latter may also perform the
bae function. Baes can be designed with a number
of variations: horizontal and vertical grates, ns of
dierent sizes and directions, mesh, or even pagoda-
like shapes (Jin et al., 1980). The benets of baes are
more distinct for coarse particle (Groups B and D)
beds than for ne particle beds (Group A), because
the bubbles in the former case are larger.
In the high-velocity uidized bed, the bed can be
operated in turbulent, fast uidization, dilute trans-
port, and downer ow regimes. In the turbulent
regime, the bubble/void phase gradually becomes
indistinguishable from the emulsion phase, and the
particle entrainment rate increases signicantly with
increasing gas velocity. Upon further increase in gas
velocity, the bubble/void phase eventually disappears
and the gas evolves into a continuous phase in the fast
uidization regime. As the gas ow rate increases
beyond the point corresponding to the disappearance
of the bubble/void phase, a drastic increase in the
entrainment rate of the particles occurs so that a con-
tinuous feeding of particles into the uidized bed is
required to maintain a steady operation. As a result,
high-velocity operations of uidized beds require a
solids recycle loop. The most widely used uidized
bed is a solids recycle loop in a riser or a circulating
uidized bed (CFB). The riser (or downer), gassolid
separator, standpipe, and solids ow control device are
the four integral parts of a CFB loop (Fig. 1b) with the
riser (or downer) acting as the main component of the
system. In the riser, gas and solids ow concurrently
upward. During this operation, the uidizing gas is
introduced at the bottom of the riser, where solid par-
ticles are fed via a control device from a standpipe and
carried upward in the riser. Particles exit at the top of
the riser into gassolid separators, after which the
separated particles ow to the standpipe and return
to the riser.
In a CFB, particle separation is typically achieved
by cyclones. The eciency of the solids separator can
aect the particle size distribution and solids circula-
tion rate in the system. The standpipe provides holding
volume and a static pressure head for particles recy-
cling to the riser. The standpipe may be connected to a
heat exchanger or spent catalyst regenerator. In this
situation, the standpipe is a transport line providing
direct solids passage from the riser. The entrance and
exit geometries of the riser have signicant eects on
the gas and solids ow behavior in the riser.
Gas velocity and solids circulation rate are two
operating variables for circulating uidized bed reac-
tors. The key to smooth operation of a CFB system is
eective control of the solids circulation rate to the
riser. The solids ow control device serves two major
functions: sealing riser gas ow to the standpipe and
controlling the solids circulation rate. Both mechanical
and nonmechanical valves are used to perform these
functions. Typical mechanical valves are rotary, screw,
buttery, and sliding valves. Nonmechanical valves
include L-valves, J-valves, V-valves, seal pots, and
other variations. Blowers and compressors are com-
monly used as gas suppliers. Operating characteristics
of these gas suppliers are directly associated with the
dynamics and instability of riser operation and must be
taken into consideration.
Both the low and high velocity operations of
uidized beds can be conducted in a variety of cong-
urations as exemplied in Fig. 2. Each conguration
has advantages and disadvantages. Figure 2a shows
Figure 1 Typical uidized bed reactors. (a) Low-velocity
uidized bed, and (b) high-velocity uidized bed.
the uidized bed with vertical baes that reduce the
bubble size, yielding more uniform ow structures in
the uidized bed. For drying processes involving sticky
particles or particles with high moisture contents, shal-
low bed operation is essential to smooth uidization.
Shallow beds are also operated with small bubble sizes,
providing ecient gassolid contact. Figures 2b and c
show the congurations of multistage shallow bed
operations. For multistage drying operation, particles
are introduced from the top stage and dried through
multistage uidized beds while hot gas enters at the
bottom stage. This countercurrent gassolid contact
improves the eciency of the uidized bed operation.
Particle exchange between uidized beds or between
regions in a uidized bed can be accomplished through
external or internal solids circulation, as shown in Figs.
2d and e. The external or internal solids circulation
allows the coupled reaction and regeneration scheme
to be implemented for the solid particles, thereby
prolonging particle on-stream time. Reactors in series
and parallel are common arrangements for uidized
bed operations (Figs. 2e and f). These systems can be
used in conjunction with multistep reaction processes.
The crosscurrent multicell type of uidized bed (Fig.
2f) allows recovery of rapidly released gas products
from solid particles where the particle residence time
is to be long. A uidized bed can be also operated in
the presence of an external force eld other than the
gravitational eld, such as centrifugal and magnetic
elds. The conguration shown in Fig. 2h is a typical
centrifugal uidized bed. For fast reactions and pro-
cesses requiring on-line catalyst regeneration, the cir-
culating uidized bed is usually used as shown in Figs.
2g and i. The downer reactor oers advantages for
ultrafast reactions and processes with good product
selectivity. Through gas spouting, spouted beds (Fig.
2j) oer means to uidize large particles (Group D)
with well-structured internal particle circulation.
Figure 2 Variations of uidized bed reactor congurations. (a) Baed bed, (b) multi-stage bed, (c) multi-stage with counter-
current solid ow, (d) multi-stage with external solid circulation, (e) uidized bed with online solid exchange, (f) multi-cell
bubbling bed, (g) CFB, (h) centrifugal uidized bed, (i) downer, (j) spouted bed.
1.2 General Procedure for the Development of
Fluidized Bed Reaction Processes
Many factors aect optimum uidized bed reactor per-
formance, including hydrodynamics, heat and mass
transfer of interparticles and intraparticles, and com-
plexities of reaction kinetics. The design of uidized
bed reactor processes follows the general approach
for multiphase reactor processes. Krishna (1994) and
Jazayeri (1995) outlined the general procedure for this
process development. The design of the processes can
be described by considering various factors as illu-
strated in Fig. 3.
The uidized bed reactor design requires
understanding the reaction chemistry. The essential
knowledge for a design engineer may include the
reaction kinetics, conversion or yield, and selectivity,
thermodynamics, and process parameters (e.g., operat-
ing temperature and pressure as well as heat of reac-
tion) aecting the reaction.
Information on intrinsic kinetics is an essential ele-
ment in the analysis of reactor performance. Kinetic
data should be obtained over a range of temperatures
spanning the entire operation conditions. It is also
essential to obtain the kinetic data under much higher
conversion conditions than those anticipated from the
uidized bed reactor, due to the two-phase ow nature
(see Chapter 3), as most reaction occurs in the emul-
sion/dense phase where the gas reactant is highly
converted. Often the reaction kinetics employed is
obtained from a slugging bed reactor or a uidized
bed reactor in which transport properties are
embedded in the kinetic information; thus such infor-
mation cannot represent the intrinsic kinetics. For
example, Squires (1994) indicated the eects of transi-
ent solids mixing patterns on the reaction kinetics in
bubbling uidized bed reactors. Gas bubbles act as
agitators in the bubbling bed and lead to rapid mixing
and global circulation of solid particles. The bed also
undergoes strong exchange of gas between the emul-
sion phase and the bubble phase by means of gas ow
through the bubble and bubble coalescence and
breakup. As a consequence, the solid particles are
exposed to gas of highly uctuating chemical composi-
tions. Thus the reaction kinetics of a uidized bed
reactor could be dierent from the intrinsic kinetics
commonly obtained in a TGA (thermogravimetric
analyzer) or a dierential reactor with steady varia-
tions in the concentration of chemical compounds.
Ideally, the reaction kinetics should be examined in a
dierential reactor under a chemical environment simi-
lar to that experienced in a uidized bed reactor and
reexamined in a pilot uidized bed reactor.
Also, in many situations, the pseudo-kinetics
expression is used owing to the lack of detailed kinetics
information or for simplicity. Useful pseudo-kinetics
can only be obtained in an environment as close as
possible to that of the proposed commercial opera-
tions. For instance, the catalyst decay in a FCC
(uid catalytic cracking) system is mainly caused by
Figure 3 Schematic diagram of uidized bed reactor developments. (From Krishna, 1994; Jazayeri, 1995.)
coking. The means and extent of coke formation on
the catalysts depend on the reaction environment and
catalyst residence time. Reners commonly employ a
microactivity test (MAT) unit to establish the activity
of catalysts for particular feedstock. The MAT is based
on the concept of continuously contacting a hydrocar-
bon feedstock with a catalyst sample of approximately
1 gram during 75 to 100 seconds of residence time. In
the MAT apparatus, the catalyst/oil ratio used
depends on the catalyst residence time. The detailed
procedure is dened in ASTM (D3907-80). This pro-
cedure would yield a signicant discrepancy from the
conventional riser reactor operation in which the cat-
alyst ow and the hydrocarbon ow are set for a given
operating condition and the catalyst/oil ratio would
not be a function of catalyst residence time. Another
signicant discrepancy between the MAT and the riser
operation is in contact times. In a conventional riser
reactor, the catalyst and the hydrocarbon stay in inti-
mate contact for about 25 seconds before being sepa-
rated in cyclones. In the MAT unit, however, the
catalyst reacts with hydrocarbons for as long as 75
100 seconds. In this sense, the MAT technique only
allows one to establish a relative performance indicator
of catalytic materials and is of limited use in its appli-
cation to catalytic riser reactors. Thus the kinetic mod-
els derived from the data obtained by MAT are of little
practical usage for eectively simulating riser reactors
and scaling up.
A small hot unit is always used for examining the
feasibility of the uidized bed reactor operation for a
new reaction. The hot unit provides information on
conversion, yield, and selectivity as well as by-product
distributions for a given uidization regime. Squires
(1994) developed a vibrated bed microreactor capable
of varying the Peclet number and of back-to-front
particle mixing independently. With such a reactor,
the sensitivity of reaction to Peclet number and
macroscale solids mixing could be expediently exam-
ined.
As shown in the process development diagram
(Fig. 3), hydrodynamic studies are essential. Such stu-
dies would provide information on the basic ow pat-
terns, mixing, particle attrition behavior, and mass and
heat transfer for a specic design of a uidized bed.
Operational stability is also examined. Like the reac-
tion kinetic studies, hydrodynamic information is
obtained under operating conditions similar to those
intended for commercial reactors. In general, uidized
bed reactor design involves studying specic reactions
at progressively increased reactor scales from bench to
demonstration, in conjunction with analysis based on
models that incorporate the kinetics and the transport
processes.
A pilot unit integrates the reactor with all process-
related components, such as feed, catalyst recovery,
product separation and purication, downstream,
and recycles. The pilot unit should be operated in the
same ow regime as the commercial unit to ensure that
the gassolid contact scheme and mixing patterns are
similar. For a deep bed with a moderate uidizing
velocity, a bench or small pilot unit usually operates
in the slugging regime. However, operation in the slug-
ging regime is not possible in a large commercial unit.
Therefore, care should be exercised when extrapolating
data from pilot units to large-scale commercial units.
Operational constraints such as impurities of feed,
length of operating period, and equipment reliability
should be taken into account during pilot testing.
Impurity in the product stream may lead to reduction
of product yield due to catalyst deactivation or poison-
ing.
A pilot reactor needs to be as small as possible to
maintain its exibility of small-scale operation and to
provide a means for readily testing alternative designs.
However, it needs to be large enough to provide scale-
up information for designing the commercial reactor.
Mathematical modeling is commonly used in the pro-
cess of sizing the pilot scale unit. The mathematical
models are also used as tools or guides for the success-
ful scale-up to commercial scale. Model development
has become an integrated approach in a new process
development and in optimal reactor design. The model
can be empirical, semiempirical, or mechanistic. The
mechanistic model takes into account the interplay
between the reaction kinetics and the transport pro-
cess. Due to the complicated nature of uidized bed
ow, proper assumptions are important for the simpli-
cation of model equations. Figure 4 shows a typical
ow diagram for describing the mathematical models
that provide predictions of product rate, product yield,
conversion, species concentration, and temperature
proles in a new process development. Experimental
verication of the model parameter using pilot or com-
mercial plant data is necessary.
Two basic approaches are often used for uidized
bed reactor modeling. One approach is based on com-
putational uid dynamics developed on the basis of the
mass, momentum, and energy balance or the rst prin-
ciple coupled with reaction kinetics (see Chapter 9).
Another approach is based on phenomenological mod-
els that capture the main features of the ow with
simplications by assumption. The ow patterns of
plug ow, CSTR (continuous-stirred tank reactor),
and dispersion are commonly assumed in the model
formulation. Most of the phenomenological models
developed are derived using only mass balance without
considering momentum and/or energy balance.
Additional relationships are obtained through empiri-
cal correlations. The two-phase theory is a typical
example of a phenomenological model. The success
of model prediction often relies on the accuracy of
empirical correlations established for accounting for
transport properties. New commercial reactors are
commonly developed through examining incremental
scale-up units.
2 KEY STEPS IN FLUIDIZED BED REACTOR
DESIGN
Important factors to consider in selecting gassolid
reactors include gassolid contact schemes, nature of
the reactions, temperature, and pressure among many
other factors. Of particular pertinence in determining
desirable reactor performance can be proper selection
of particles, ow regime, and system congurations.
2.1 Particle Selection and Catalyst Development
Particles are the bed material employed in uidized bed
reactors and can be reactants (e.g., coal and limestone),
products (e.g., polyethylene), catalysts, or inert. The
choice of particle size, in general, aects the hydrody-
namics, transport processes, and hence the extent of
reactor conversion. Particles experience particleparti-
cle collisions, friction between particles and walls or
internals, and cyclones. In some cases, the catalyst
material is inherently susceptible to attrition, and spe-
cial preparation to enhance the attrition resistance is
required. For example, the vanadium phosphate metal
oxide (VPO) catalysts developed for butane oxidation
to maleic anhydride in a circulating uidized bed reac-
tor (Contractor et al., 1987) have an attrition problem
that can be treated by spray-drying active catalyst
microspheres together with a polysilicic acid hydrogel
under conditions that allow the silica (SiC or SiO
2
) to
migrate to the outer surface and form a strong porous
layer on the outer surface (Bergna, 1988). This porous
shell allows the reactant and product molecules to dif-
fuse in and out of the catalyst particles without signi-
cantly aecting the maleic anhydride selectivity.
For catalytic reactions, most active sites of the
catalysts are dispersed on the surface of supported
particles. Once appropriate active compositions
are identied, catalyst development can focus on
the selection of support materials. Most reactions in
gas-uidized bed reactors are carried out under high-
temperature conditions to overcome high activation
energies and ensure the presence of a gas phase.
Similarly, elevated pressures are typically utilized to
enhance the concentration of reactants on solid sur-
faces. The catalyst supports should be chemically and
physically stable under high temperature, because the
structure must remain unchanged for a long period of
time. Supports provide an enhanced thermal and
mechanical stability to the catalyst material as well as
reduce the usage of expensive metals. Selection of inert
support can signicantly aect the uidized bed opera-
tion. Inorganic materials generally satisfy the rigid
constraint. Only inorganic lattices have been found
to possess a static and stable three-dimensional back-
bone capable of sustaining a porous structure under
severe reaction conditions. The commonly used
catalysts in uidized bed reactors include zeolites,
amorphous silica, and alumina or other metal oxides.
Aluminum oxides are commercially available in dier-
ent forms. The major phase is g-alumina, which has
small pores, a surface area up to 200 m
2
,g and surface
hydroxyl groups. Amorphous silica can be prepared
with high surface area (up to 500 m
2
,g). Zeolites are
aluminosilicates of well-dened crystal structure and
regular pore size so they can be engineered to catalyze
reactions with shape selectivity. Each of these materials
can serve as either a catalyst or a support. The activity
and selectivity of catalysts are dictated by the nature of
active sites and may be modied by the presence of
promoter or doping agents. For uidized bed reactors,
the supporting particles are tailored to achieve opti-
mum hydrodynamics for a specic reaction. Catalyst
support particle sizes and pore structures are generally
chosen to ensure minimum intraparticle mass transfer
resistance. Another reason to avoid intraparticle
diusion resistance may result from selectivity
Figure 4 Flow diagram of reactor model developments.
considerations. Minimizing internal transport resis-
tance ensures that desired products rapidly diuse
out of the catalyst particle and avoid further undesir-
able reactions. For a consecutive reaction scheme, high
diusion resistance decreases intermediate composi-
tions, while nal product conversion increases. The
specic surface area and pore structures are also very
important, and the eects are reaction specic. The
reaction of ammoxidation of propylene to acrylonitrile
preferably uses catalysts with specic surface areas
below 100 m
2
,g and a pore radius greater than 50 A
.
The catalysts are designed to provide limited oxygen
access to the reactant, and a higher surface area would
only enhance the extent of oxidation. Thus the forma-
tion of desired products while restricting further oxida-
tion is possible. Conventional processes for catalytic
cracking of heavy hydrocarbon feedstock to gasoline
and distillate fractions typically use a large-pore mole-
cular sieve, such as zeolite Y, as the primary cracking
component. Upon the addition of a medium-pore zeo-
lite, such as ZSM-5, to the cracking catalyst composi-
tion, an increase in the octane number of the gasoline
fraction can be obtained. Conventional ZSM-5 crack-
ing additives are implemented with a crystal size in
excess of 0.2 microns, since smaller crystal materials
reduce hydrothermal stability and hence rapidly lose
activity when exposed to the high-temperature steam
generated during FCC regeneration.
Particles in gassolid reactions in uidized bed reac-
tors require appropriate preparation (e.g., grinding,
surface treatment) to achieve optimum eciency.
Coal must be ground to a particle size range suitable
for operation in a uidized bed combustor. Sorbet
powders are synthesized to obtain desirable pore struc-
ture. Limestone is extensively used in the in situ
removal of acid gas species, such as SO
2
from uidized
bed coal combustors, and H
2
S from advanced dual-
cycle gasication systems. Removal of SO
2
involves
the injection or uidization of dry calcium-based pow-
ders in the high-temperature environment (800
1150
C) of the combustor, calcination of the sorbent

to produce CaO, and further reaction with SO
2
to form
higher molecular volume CaSO
4
.
CaCO
3
s $CaOs CO
2
g
CaOs SO
2
g
1
2O
2
g $CaSO
4
s
The optimum particle size and temperature range of
operation dier slightly for each of the above steps.
The initial reactivity of the sorbent and the ultimate
sulfur capture are strongly inuenced by available sur-
face area, pore size, and volume characteristics.
Commercially mined limestone powder used in SO
2
capture suers from a very low surface area (less
than 3 m
2
,g) and negligible porosity. As a result, a
signicant loss of internal pore volume is observed
due to the pore lling/pore plugging by the large mole-
cule, CaSO
4
. Specically, Ghosh-Dastidar et al. (1996)
have shown that the pores generated by the calcination
of limestone are typically less than 50 A
and are very

susceptible to pore blockage and plugging. This leads
to premature termination of sulfation and incomplete
utilization of sorbent. Ghosh-Dastidar et al. (1996)
reported that carbonate powders with high-surface-
area and porosity can exhibit very high reactivity and
conversion when compared to high-surface-area
hydrates. Fan et al. (1998) have developed a novel
sorbent based on optimization of the surface area
and pore size distribution of the calcium carbonate
powder. Such optimized calcium carbonate powders
lead to the generation of CaO with pores in the 50
200 A
size range. This powder is generated by a pre-

cipitation process in a three-phase reactor system by
bubbling CO
2
through a CaOH
2
suspension in the
presence of anionic surfactants.
On an individual particle level, particle size and den-
sity play a dominant role in dictating the heat and mass
transfer rates and hydrodynamics. Usually, particle
sizes are chosen so that resistance of mass and heat
transfer between particles and surrounding ow is neg-
ligible. Mass transfer between the gas and a single
sphere can be estimated by (Froessling, 1938):
Sh 2.0 0.6Sc
1,3
Re
1,2
p
1
For catalytic reactions, particles used in uidized bed
processes are usually in the range of 40 to 100 mm in
mean diameter. Similarly, particle-to-gas heat transfer
coecients in dense phase uidized beds can be esti-
mated by (Kunii and Levenspiel, 1991):
h
gp
d
p
z
2 0.6 1.8Re
1,2
p
Pr
1,3
Re
p

Ud
p
r
g
m
g
2
For high-activity catalysts, the required heating time of
a particle has to be relatively short. This is especially
true for systems in which the products and selectivity
are sensitive to temperatures. For high-velocity
operations, solid particles are recycled internally or
externally back into the reactors. The temperature of
the particles in the reactor is usually dierent from that
from recycle. It is imperative that particles of two
dierent temperatures rapidly mix so that the tempera-
ture variations in the entrance zones can be minimized.
Small particle size can often eliminate these concerns.
Selection of particle size and density is further com-
plicated by hydrodynamics, which strongly depends on
particle physical properties. From a hydrodynamic
point of view, particle selection must consider the
interplay among ow regime, mixing characteristics,
and particle residence time. Particle design provides
the most eective tools to improve reactor perfor-
mance for existing plants through modications of
hydrodynamic behavior. Approaches for particle
design include
Optimum particle size and density
Mixture of two groups of particles (e.g., add ne
particles to a coarse particle bed and vice versa)
Modication of particle size distributions
Use of ne particles (with interparticle forces to be
smaller than the hydrodynamic forces)
These approaches are discussed in conjunction with the
ow regime behavior, which is given in the following
section.
The particle size distribution (PSD) signicantly
impacts the reactant conversion in a uidized bed reac-
tor. Sun and Grace (1990) examined the three dierent
particle size distributions, wide, narrow, and bimodal,
on the performance of a catalytic uidized bed reactor
using the ozone decomposition reaction. They found
that a uidized bed with particles of wide size distribu-
tion yields the highest reactant conversion. Of further
interest, the property of particle entrainment and elu-
triation is a function of particle size, density, and
shape. Both entrainment and elutriation rates increase
with decreasing particle size and density. Particle
separation eciency may serve as another measure
for particle selection.
2.2 Flow Regime Relevancy
The factors considered for choosing a uidization/ow
regime include
1. Gassolid contact pattern and interphase mass
transfer
2. Backmixing characteristics for both gas and
particle phases
3. Reaction kinetics
4. Heat transfer
5. Residence time and reactor height/diameter
6. Ability of solid handling in large quantity
7. Productivity or throughput
Contact schemes of gassolid systems in uidized beds
are classied by the state of gas and solids motion. For
noncirculating systems, the gas at low velocity merely
percolates through the voids between packed particles,
while the particles remain motionless in the xed bed
state. With an increase in gas velocity, particles move
apart and become suspended; the bed has entered the
uidization state. Further increase of gas velocity sub-
jects the ow to a series of transitions from a bubbling
uidization regime at low velocities to a dilute trans-
port regime at high velocities accompanied by signi-
cant variations in gassolid contact behavior. Some
gassolid ow structures associated with given ow
regimes are presented in Fig. 5.
Figure 5 Flow regime diagram of uidized bed reactors. (a) Fixed bed, (b) bubbling uidized bed, (c) turbulent uidized bed, (d)
circulating uidized bed, and (e) dilute transport bed.
2.2.1 Characteristics of Flow Regimes
The bubbling regime is characterized by the coexis-
tence of a bubble phase and a dense/emulsion phase
as shown in Fig. 5b. The majority of the uidizing gas
is present in the form of bubbles, and as a result, the
gas velocity through the dense phase is very low (on the
order of 0.01 m/s for Group A particles). The dense
phase occupies about 4060% of the bed, depending
on the particles and operating conditions. Figure 6
shows typical values for dense phase fraction, bubble
fraction, and overall voidage for FCC particles under
dierent gas velocities. Once formed, bubbles rise,
grow, coalesce, break up, and nally erupt on the
bed surface. Bubbles induce the drift eect on sur-
rounding particles and entrain particles in the wake
region, thereby inducing vigorous local and global cir-
culation of solid particles in the dense phase. Although
bubbles in uidized beds resemble those in gasliquid
systems, the interface of bubbles is permeable to gas
ow. The exchanges of gas between bubble and emul-
sion phases are accomplished by through-ow, bubble
coalescence and breakup, and diusion. A small frac-
tion of particles may be entrained into the bubble
phase depending on the ne content and operating
conditions. The solid particles in the bubble phase
play an important role in the reactant conversion
when fast reactions and/or slow interphase exchange
are encountered.
The turbulent regime is often regarded as a transi-
tion regime between the bubbling and fast uidization
regime. Bubbles or voids are still present, although
they are less distinguishable in the dense suspension.
With an increase in gas velocity, the dense phase struc-
ture varies gradually, and local nonuniformity of solids
concentration emerges. In this regime, interactions
between gas voids and the dense/emulsion phase are
vigorous and provide an eective gassolid contact.
The coexistence of a bottom dense region and an
upper dilute region characterizes the solids ow in a
riser. The solids holdup within the dense phase zone
ranges from 0.1 to 0.3, while the holdup proles in the
dilute region can be approximated by an exponential
decay. A core-annular ow structure along the radial
direction develops with a dense particle layer in the
wall region and a dilute core region. Particle exchanges
occur between the dilute core and the dense wall
region, as temporal and spatial accumulations of par-
ticles in the wall region form a transient dense particle
layer or streamers. Nonuniform particle distributions
or the presence of localized dense zones usually results
from the existence of particle clusters and greatly inu-
ences the hydrodynamic characteristics.
The hydrodynamics of a circulating uidized bed is
further complicated by the existence of signicant var-
iations in solids concentration and velocity in the
radial direction. A more uniform distribution can be
achieved at conditions of lower solids concentrations
under higher gas ow conditions. In the dilute trans-
port regime, the solids concentration is very low and
both gas and solids have short residence times.
A concurrent downward ow circulating uidized
bed, or a downer, is a new alternative ow arrange-
ment for a high-velocity system. A downer reactor
system has similar system congurations to a riser
reactor system except that both the gas and the solid
particles ow downward. Concurrent downward ow
of particles and gas reduces the residence time of solid
particles because the downward ow is in the same
direction as gravity. More uniform radial gas and
solids ow than those in a riser can be achieved. The
downer leads to more uniform contact time between
the gas and solids. With these advantages, downer
reactors have been proposed for processes such as
uid catalytic cracking, which requires short contact
time and uniform gas and solids residence time distri-
butions.
As discussed, gassolid uidized bed behavior varies
not only with ow regime but also with other factors
such as particle properties, operating pressures, and
temperatures. Grace (1986a) summarized the eects
of particle properties and operating conditions on ui-
dization behavior and prepared a ow regime diagram.
Figure 6 Eects of gas velocity on the voidage of dense and
dilute phases in a uidized bed. (From Yang et al., 1997.)
The owregime diagramwas further modied by Kunii
and Levenspiel (1997) and is shown in Fig. 7. The ow
diagram is presented as a function of non-dimensional
particle diameter and gas velocity dened as
d
p
d
p
,
g
,
p
,
s
g
j
2
g
_ _
1,3
3
and
U
U
,
2
g
j
g
g,
p
,
g
_ _
1,3
4
For given particles and operating velocity, the gas
solid contact pattern can be determined using this dia-
gram. Likewise, for a given ow regime, this diagram
could provide available combinations of particle
properties and gas velocity.
2.2.2 GasSolid Contact and Interphase Mass
Transfer
Gas in the dense and bubble phases plays dierent
roles in a bubbling bed reactor. When gas enters the
uidized bed reactor, the gas in the bed ows to the
dense and bubble phases. The gas reactant reacts in the
dense phase upon contact with the particles. Interphase
mass transfer allows gas reactant and product transfer
between the bubble phase and the dense phase. As a
bubble rises through a dense or emulsion phase region
in a bubbling bed, gas in the surrounding dense phase
could either ow through the bubble phase or circulate
between the bubble phase and surrounding dense
phase, depending on the ratio of bubble rise velocity
to interstitial gas velocity in the dense phase. As
depicted in Fig. 8a, when the bubble rise velocity is
greater than the interstitial gas velocity in the dense
phase, a clouded bubble forms in which the circula-
tory ow of gas takes place between the bubble phase
and the surrounding clouded region. The cloud region
size decreases as the bubble rise velocity increases. On
the other hand, when the interstitial gas velocity in the
dense phase is greater than the bubble rise velocity, gas
ows through the bubble phase yielding a cloudless
bubble, as shown in Fig. 8b. Bubbles in reactors with
coarse particles, e.g., Group D, are typically cloudless
bubbles, while those with ne particles, e.g., Group A,
are typically cloud bubbles. As shown in Fig. 8a, for
gas to transfer from the bubble to the emulsion phase,
resistances exist at the bubblecloud interface and the
cloudemulsion phase interface. As indicated in Fig. 9,
product formation may take place in bubble, cloud, or
dense phase. Davidson and Harrison (1963) assumed
that the gas exchange rate is made up of through-ow
and gas diusion. The through-ow velocity is on the
order of minimum uidization velocity, U
mf
. The volu-
metric mass transfer coecient in the bubbling regime,
dened based on per unit volume of the bubble phase,
can be estimated by (Sit and Grace, 1981):
Figure 7 Generalized map of uidized regimes. (From Kunii and Levenspiel, 1997.)
k
q

U
mf
3

4D
m
e
mf
U
b
pd
b
_ _
1,2
5
The eectiveness of gassolid contact is reected in the
reactant conversions. Figure 10 shows the extents of
reactant conversions in a bubbling bed reactor for
various reactions as a function of dimensionless rate
coecients along with model predictions based on
both the plug ow and the CSTR ow patterns for
bubbling bed reactors (Kunii and Levenspiel, 1991).
The conversions are typically lower than those pre-
dicted by either model, and this is an indication of
inecient gassolid contact.
In the turbulent regime, interactions between gas
voids and the dense phase are strong, yielding extensive
global and local solids mixing, thereby enhancing gas
solid contact. In high-velocity operations of a uidized
bed reactor, contact resistance of the bulk gas with
particles in clusters or wall layers is present. Particle
accumulation in the wall region also imposes a hin-
drance on gassolid contact owing to low gas ow
near the wall. Since signicant amounts of particles
are in the wall layer, this layer plays an important
role in mass and heat transfer. The extent of solid
accumulation in the wall region can be analyzed
using the coreannular model, and the mass balance
of solid particles can be expressed as
p
4
D
2
G
s

p
4
D2d
w
2
V
pc
1 ee
c
r
p
p
4
4Dd
w
4d
2
w
V
pw
1 ee
w
r
p
6
The cross-sectional averaged voidage is
p
4
D
2
1 ee
p
4
D2d
w
2
1 ee
c

p
4
4Dd
w
4d
2
w
1 ee
w
7
Upon rearrangement of Eqs. (6) and (7), the ratio of
solid particles in the wall region to the overall particle
content can be expressed as
1 ee
w
4h1 h
1 ee
1
1 2h
2
1 ee
c
1 ee
V
pc
G
s
,r
p
1 ee
V
pc
V
pw
h
d
w
D
8
Using the correlation of radial voidage proles devel-
oped by Zhang et al. (1991),
er ee
0.191f
2.5
3f
11
f
r
R
9
A ratio of 0.51 is obtained for typical conditions
( ee 0.88 and h 0.05).
Like Fig. 10 for the bubbling bed reactor, the eects
of gassolid contact in a riser reactor can be described
in terms of the extent of reactant conversion for var-
ious reactions along with predictions as given in Fig 11
Figure 8 Bubble congurations and ow patterns around a fast bubble and a slow bubble. (a) Fast bubble (U
b
> V
d
), (b) slow
bubble (U
b
- V
d
).
Figure 9 Gas transfer pathway in the bubbling regime.
by the plug-ow and CSTR models. It is seen that an
ineective gassolid contact induces lower conversions
than predicted from either the plug-ow or the CSTR
ow pattern and results from the passage of a large
fraction of gas through the dilute central core region
of the riser. Gas mixing is governed by the combined
eects of gas diusion and turbulent convection.
Turbulent intensity is indirectly reected by the varia-
tions of solids concentration uctuations, as shown in
Fig. 12. High gas entrainment in the core region
diminishes the agitation or eddy turbulence in the lat-
eral direction. The dense wall layer is of low turbulent
intensity. The strong turbulent actions on the interface
between core and wall regions are not sucient to
sweep accumulated particles away from the wall
under most operating conditions.
2.2.3 Solids Mixing
Bubbles entrain particles during their rise in a bubbling
uidized bed, and the entrained particles are carried up
to the bed surface and erupted to the freeboard. These
particles then move downward in the bed, creating a
solids circulation pattern. On the local scale,
rising bubbles displace surrounding dense phase solid
particles through wakes and drift eects. Furthermore,
irregular motion, coalescence, and breakup of bubbles
cause local particle mixing. Solids mixing has a signi-
cant inuence on (1) gassolid contact, (2) temperature
proles, (3) heat transfer, and (4) design of particle
feeding and discharge. In the bubbling regime, the
solid ow approaches a completely mixed ow pattern,
and the particle residence time distribution can be
expressed as
f t
p
e
t
p
,t
p
t
p

total solid in the bed
particle feeding rate
10
In the turbulent regime, strong interactions between
gas ow and particles lead to a well solids mixing state.
Little, however, has been done to quantify such a state
of mixing. Particle segregation occurs in the radial
direction of a fast uidized bed. Particles move down-
Figure 11 Fraction of unconverted reactant A as a function
of dimensionless reaction rate coecient in a riser reactor.
(From Jiang et al., 1991.)
Figure 12 Radial proles of the standard deviation of local
solids concentration with frequency lower than 12.5 Hz.
Figure 10 Fraction of unconverted reactant A as a function
of dimensionless reaction rate coecient in bubbling bed
reactors. (From Kunii and Levenspiel, 1991.)
ward along the reactor wall under some operating
conditions. Downward particle ow results in local
particle recirculation, leading to the ow deviating
from plug ow with poor gassolid contact. On the
other hand, downward solids ow extends particle resi-
dence time and its distribution. Particle residence time
aects the extent of solids reactions or catalytic reac-
tions. For catalytic reactions, as the reactants are
absorbed on the catalyst surface, the actual reaction
time will depend on the rate processes of reaction
and absorption. For an operation in a circulating ui-
dized bed or pneumatic transport reactor, both the gas
reactant residence time and the particle residence time
in the reactors are important. For reactions sensitive to
catalyst activity, nonuniform products and low selec-
tivity might be a result of solids backow due to the
prolonged residence of low activity catalysts.
2.2.4 Gas Mixing
Fluid ow without any backmixing represents an ideal
plug ow, while that with innite backmixing repre-
sents an ideal CSTR. Gas dispersion is a function of
bubble size and reactor scale. As the gas velocity
increases, bubble motion becomes more vigorous. At
the critical gas velocity, U
c
, the gas dispersion reaches
a maximum as the bed undergoes transition from the
bubbling to the turbulent regime; trends in the gas
dispersion coecient, as shown in Fig. 13, reect the
change of the heterogeneous nature of the bed. The gas
dispersion coecient presented is approximately four
orders of magnitude higher than the molecular diu-
sion due to turbulence-dominated gas mixing. For
high-velocity operations in the riser, the backmixing
in the core region is negligible, while the backmixing
near the wall is extensive.
Tracer techniques are commonly used to determine
the gas dispersion coecients in uidized bed reactors.
The tracer concentration measured at the outlet in
response to a pulse or step input of the tracer at the
inlet can be used to calculate the dispersion coecient
based on the dispersion models in a form similar to Eq.
(11), i.e.,
oC
ot
uu
oC
oz
D
ax
o
2
C
oz
2

D
r
r
o
or
r
oC
or
_ _
11
Empirical correlations for the dispersion coecient in
bubbling uidized bed reactors are available in Wen
and Fan (1975).
2.2.5 Heat Transfer
Three major components contributing to surface heat
transfer are radiation, particle convection, and gas
convection. Particles seldom directly contact heat
transfer surfaces and are usually separated by a thin
gas layer through which particle heat convection
occurs. In gassolid uidized beds, radiation may be
neglected when the bed temperature is below 400
C.
The relative eect of particle convection to gas convec-
tion on heat transfer depends appreciably on the types
of particles used in uidization. Particle convection
is the dominant mechanism for small particles
(d
p
- 400 mm), such as Group A particles. Gas con-
vection becomes dominant for large particles
(d
p
> 1500 mm), such as Group D particles (Maskaev
and Baskakov, 1974) and for high-pressure or high-
velocity uidization. For Group B particles, both par-
ticle and gas convection are signicant. Molerus (1992)
studied the eects of system properties on heat transfer
and the data plotted in Fig. 14a shows the general
trend of heat transfer for three groups of particles.
The increasing and decreasing behavior is a result of
interplay between the particle and gas convective heat
transfer. Figure 14b shows the variations of the wall-
to-bed heat transfer coecient with local solids
concentration in a circulating uidized bed. Heat
transfer coecients are inuenced by other factors
such as particle size and distribution, solids concentra-
tion, and geometry of the heat transfer surface; a
detailed account of heat transfer is given in Chapter
10. In general, heat transfer coecients increase with
the solids concentration for ne particles. Thus heat
Figure 13 Inuence of gas velocity on the axial gas disper-
sion in a turbulent uidized bed. (From Wei et al., 2000a.)
transfer is generally higher in a bubbling bed than that
in a CFB.
2.2.6 Flow Regime Selection
Each ow regime has its unique ow characteristics,
which are illustrated in Table 2 using Group A particles
as an example. Knowing these ow characteristics can
aid in the selection of a ow regime for a given reaction.
The solids holdup in a packed state is typically in the
range of 0.450.65 and decreases with increasing gas
velocity. Typical axial proles of solids holdup for
Groups A and B particles are given in Fig. 15, while
radial proles for the fast uidization and dilute trans-
port regimes are given in Fig. 16. The turbulent ui-
dized bed regime provides improved gassolid contact
eciency with heat transfer and gas backmixing com-
parable to or slightly higher than those in the bubbling
bed regime while maintaining a relatively high solids
concentration. This unique regime has been applied to
many process operations such as the MTG process
developed by Mobil, butane oxidation, and acryloni-
trile production. When the gas velocity to the fast ui-
dization regime is increased, the coreannular ow
structure occurs. The relative dimension of the annular
region to the core region is not proportional to the size
of the riser, which makes riser scale-up a challenging
issue. To minimize the segregated ow between the core
and wall regions, baes and other internal structures
(Jiang et al., 1991; Ran et al., 1999) can be used.
Manipulations of particle size distribution have also
been suggested as a means for improving the gas
solid contact. The ow in a dilute transport reactor is
close to plug ow. At high operating gas velocities the
solids holdup is very low. This ow regime is applicable
for fast reactions with short residence time require-
ments. As noted, uidized bed reactors are often used
due to nearly isothermal operation and intensive mix-
ing. Clearly, this is the case for coal combustion and
catalyst regeneration in FCC systems. Cracking cata-
lysts deactivate rapidly by coking, and the regeneration
process is dominated by highly exothermic coke com-
bustion. The residence time for catalyst regeneration
determines the regenerator size. Operation in the tur-
bulent regime ensures ecient contact between oxygen
and catalysts while avoiding the formation of hot spots
and preventing catalysts deactivation by sintering and
steaming in catalyst regeneration.
For selective reactions requiring isothermal opera-
tion with desired conversion sensitive to backmixing,
such as butane oxidation, partial oxidation of natural
gas, and acrylonitrile production, a compromise
between isothermal operation and the plugow
requirement must be made in regime selection. The
turbulent regime can accommodate this compromise.
However, regime selection also requires consideration
of catalyst characteristics and system congurations.
Oxidation of hydrocarbons is routinely carried out in
xed-bed bundle reactors and uidized bed reactors
with simultaneous presence of oxygen and hydrocar-
bon at the active sites of the catalyst. Sze and Gelbein
(1976) proposed a reactorregenerator process for oxi-
Figure 14 Heat transfer variations with gas velocity and solids concentration. (a) Dependence of heat transfer coecient on gas
velocity for three types of particles. (From Molerus, 1992), (b) dependence of heat transfer coecient on solids concentration in a
CFB. (From Shi et al., 1998.)
dative synthesis of aromatic nitriles. The same type of
process with a circulating uidized bed reactor is used
for selective oxidation of n-butane to maleic anhydride
(Contractor and Sleight, 1987). In this process, n-
butane is oxidized in the riser by the oxygen-loaded
catalyst, which in turn is reoxidized in a separate
regenerator operated in the bubbling bed regime. The
separated reactionregeneration conguration allows
for optimum ow regime to be employed for each of
the reactors associated with dierent types of reaction.
Catalyst regeneration also provides a means of
maintaining overall catalyst activity within systems
with rapid catalyst deactivation. Deactivated catalyst
particles can be continuously removed from a reaction
zone while regenerated catalyst from a regeneration
zone is continuously recycled back to the reaction
zone in a ow scheme similar to that shown in Fig.
17. The averaged activity of a uidized bed can be
maintained even though each catalyst particle under-
goes rapid deactivationregeneration. However, this
process does not apply to situations in which the pro-
duct and selectivity are sensitive to catalyst activity, as
deactivated catalysts may lead to undesirable products.
Advances in catalyst development have led to a shift
in industrial focus from the case of low-throughput
bubbling bed reactors with low catalyst activities to
high-throughput turbulent bed reactors or risers with
high catalyst activities. Catalytic cracking of petroleum
feedstock to light oils, gasoline, and solvents can be
carried out in bubbling beds or riser reactors. In bub-
bling beds, catalyst particles are well mixed and hot
spot formation is avoided. With the development of
new zeolite catalysts with high activities in the 1960s,
riser reactors were subsequently implemented and pro-
Table 2 Comparisons of Fluidization Properties of Bubbling, Turbulent, Fast Fluidization, and Dilute Transport Beds
with Group A Particles
Bubbling bed Turbulent bed Fast uidization Dilute transport
Flow nature Bubbles and
emulsion phase
Dispersed dilute
phase and dense
phase
Core-annular ow,
particle accumulation
near wall and particle
clustering
Dispersed dilute ow
with a thin particle
layer on the wall
Solids holdup 0.5, decrease with
gas velocity
0.30.5, decrease
with gas velocity
Bottom dense region:
0.050.4,
Upper dilute region:
- 0.05, decrease with
gas velocity and increase
with solids ow rate
- 0.05, decrease with
gas velocity and increase
with solids ow rate
Axial proles of
solids holdup
Fig. 16 Fig. 16 Fig. 16 Fig. 16
Radial proles
of solids fraction
Uniform Slightly dense in
wall region
Large variations Uniform except in the
thin wall layer
Gas backmixing
(dispersion)
High Pe
a
0.10.4 Pe
a
58, close to plug
ow in central zone;
strong backmixing near
the wall
Plug ow
Radial gas mixing Pe
r
220 Pe
r
220 Pe
r
1001000 Pe
r
100200
Axial solids mixing Bubble agitation Very well Poor, Pe
a
510 Poor
Radial solids mixing Well Very well Coreannular structure
poor, Pe
a
1001000
Poor
Gassolid contact Poor Well Poor in wall region Well
Temperature prole Uniform May have slight
axial gradient
Slight axial gradient Large axial gradient
Heat transfer Very high Very high Low Poor
Solids ow Entrainment High entrainment Solids circulation Solids transport
Gas velocity
(throughput)
Low Higher Highest Highest
Residence time
distribution
Long/wide Long/wide Short/relatively wide Short/narrow
duced an increase of more than 10% gasoline yield
over conventional bubbling bed reactors. With new
catalyst development, FCCs can also be operated in
a downer reactor due to the short residence time
often associated with a downer. Rapid feed heating is
required and may be attained, at least in part, by hot
catalyst from the regenerator returning to the reactor
through the feed stream. The particle temperature may
be signicantly higher than the average cracking tem-
peratures, and rapid mixing is necessary to avoid
undue thermal cracking. Furthermore, for short resi-
dence time reactors, the initial gas and solids ow
development is important in order to control the reac-
tion selectivity and product distribution. As a result,
the downer distributor design has a signicant eect on
the gassolid ow structure in the downer, especially in
the entrance section. A good feed nozzle design that
provides excellent gassolid mixing and uniform
distributions of gas and solids over the entrance cross
section would enhance the gas and solid ow develop-
ment.
Increasing gas velocity or gas throughput in a reac-
tor eventually leads to a high production rate with
transition of ow regimes to turbulent or riser uidiza-
tion. Some designs and modications have been imple-
mented on bubbling bed reactors to operate as
turbulent bed reactors:
1. Increasing operating gas velocity to greater
than 0.3 m/s
2. Decreasing particle sizes
3. Increasing ne content (d
p
- 45 mm)
4. Increasing bed length-to-diameter ratio
5. Placing internal baes, bubble suppressor for
staging operation
Circulating uidized bed (CFB) risers are particu-
larly advantageous for the following reactor process
characteristics:
1. High rates of reaction
2. Reactions with reversible deactivation requir-
ing continuous catalyst regeneration and high
selectivity
3. Varying feed and product requirements with
staged reactant inlets that require independent
control of gas and solids retention time and
Figure 15 Typical axial proles of solids concentration in
various uidization regimes.
Figure 16 Radial proles of solids concentration in a CFB.
Figure 17 Schematic diagram of separated reaction and
regeneration system.
where heat removal or supply through solids
circulation is essential
In practice, regime selection cannot be accom-
plished without considering other factors. For
instance, velocity selection must be incorporated into
the distributor design. In the bubbling regime, the gas
velocity through a distributor has to be high enough to
overcome orice resistance and prevent weeping. For
reactions taking place near the particle melting and
sticking temperature, strong agitation is necessary
to prevent the particles from agglomeration.
Furthermore, CFB riser operating conditions aect
the eciency of downstream equipment such as
cyclones, lters, and standpipes.
2.3 Process Requirements
Successful process developments rely on the under-
standing of process requirements and constraints.
Fluidized bed reactors have been used in many chemi-
cal processes. The typical reactions can be grouped as
catalytic reactions, gassolid reactions, noncatalytic
gas reactions, and polymerization. The following
describes process requirements for these systems.
2.3.1 Catalytic Reactions
Catalyst selection should be based on catalyst reactiv-
ity, reaction selectivity, and physical properties such as
particle size, density, and resistance to attrition. For
process development, heat and mass transfer phenom-
ena together with reactivity and physical properties of
catalysts must be taken into account. The catalytic
process begins with gas reactant transferring to the
catalyst outer surface and subsequent intraparticle dif-
fusion of the reactant through the pores of the catalyst.
Reactants then absorb onto the catalyst surface and
react to form product. These products desorb from
the surface, and, through intraparticle diusion, the
products exit from the pores and outer catalyst surface.
Consider the example of the ammoxidation of propy-
lene to produce acrylonotrile over multicomponent
molybdenum/bismuth catalysts:
C
3
H
6
NH
3
1
2
O
2
!CH
2
CHCN3H
2
O
H
R.298K
515 kJ,mol
The reaction network of propylene ammoxidation is
not fully understood, but a kinetic model can be
approximated by the network shown in Fig. 18. For
decades, the Sohio acrylonitrile process has utilized
turbulent uidized beds for propylene ammoxidation
to synthesize acrylonitrile. The industrial signicance
of this process was designated by the ACS as a
National Historic Landmark process in 1996. The
Sohio acrylonitrile process is an innovative, single-
step catalytic reaction over metal oxide catalysts and
is implemented, today, in more than 90% of acryloni-
trile production throughout the world. The gas mix-
ture, consisting of propylene and ammonia, is
preheated to 150
C and fed into a uidized bed cata-

lytic reactor that is operated at 420460
C and 1.32
bar. The molar feed ratio of propylene/ammonia/air is
1 : 1.15 : 10 and gives a minimum excess ammonia over
propylene with about 10% stoichiometric excess air in
respect to propylene. Catalyst particles range from 50
to 70 mm in mean diameter, and the supercial gas
velocity ranges from 0.4 to 0.8 m/s with a residence
time between 8 and 10 seconds. As shown in Fig. 18,
this reaction is dominated by the competition between
partial and completed oxidation. The catalyst is regen-
erated in situ for this single region and single reactor
design.
Internal heat exchangers are used to remove heat
and control the reaction temperature of the reactor
within 5
C. Heat transfer tubes are usually arranged

vertically and must be capable of controlling bed tem-
perature for the worst possible scenario including reac-
tion runaway. Heat exchangers can also act as
internals to prevent the formation of large bubbles in
the uidized bed, thereby reducing overoxidation,
which is directly associated with gas backmixing.
Internal structures also allow uidized beds to be oper-
ated in the turbulent regime at lower gas velocities. The
structure of the internals has a signicant inuence on
Figure 18 Reaction network for acrylonitrile by catalytic
ammoxidation of propylene. (From Wei et al., 1996.)
the performance of a turbulent uidized bed reactor
and is discussed further in Chapter 7.
Instead of in situ catalyst activation, a catalyst can
be regenerated in a uidized bed reactor with a sepa-
rated reaction zone and regeneration zone. The unique
features of this type of reactor are reected by the gas
distributor design. These distributors consist of an air
distributor and a propylene/ammonia distributor that
plays three roles: (1) distributing the three reactant
gases uniformly in the reactor, (2) forming a separated
regeneration and reaction zone within the reactor to
facilitate reactions of the redox mechanism, (3) am-
mable controls for the reactant mixtures. The air dis-
tributor in the bottom of the uidized bed reactor is
usually a perforated plate with uniformly distributed
holes. Direct contact of catalyst particle within a high-
velocity air jet from the distributor may cause severe
catalyst attrition. Thus holes on the distributor must
be suciently large (e.g., 2035 mm I.D.) to reduce the
gas velocity lower than the jetting velocity but not
excessively large to yield low pressure drops (caused
by low gas velocity) across the distributor, which
would lead to nonuniform air distribution.
To avoid spontaneous ignition of the reactant mix-
ture, propylene/ammonia is fed into the reactor
through a second distributor placed above the air
distributor, separated from the air feed. In this way,
propylene and air are not mixed prior to contact with
the catalysts. This second distributor is usually a tub-
ular type with a horizontal O-ring or branch tubular
arrangement designed to create a uniform distribution
across the plate while minimizing downward disper-
sion of propylene and ammonia into the regeneration
zone. In order to avoid possible explosion during
mixing of feed components and in case of catalyst
deactivation or absence of catalyst, a proper distance
(e.g., 0.30.6 m) between the two distributors must be
maintained. Creation of an oxygen-rich regeneration
zone between the two distributors provides catalyst
contact with oxygen to reoxidize the reduced catalyst
from the reaction zone. Oxygen-loaded catalysts are
then returned to the reaction zone where the catalysts
come into contact with propylene and ammonia to
release crystal oxygen and produce the desired acry-
lonitrile. The spent catalysts must once again be
reoxidized.
The drawbacks of using a turbulent bed for partial
oxidization reactions are backmixing and low through-
put. A wide distribution in solid residence time also
makes controlling the reaction dicult. With new
developments in catalysts, synthesis of acrylonitrile
can be completed in a few seconds and suggests that
the turbulent uidized beds may be replaceable with
high-density riser type reactors (Wei et al., 1994a).
High gas velocity and high solids circulation rates in
high-density risers can reduce backmixing signicantly
and improved gassolid contact eciency. The small
hot model and pilot plant tests show that the average
reaction rate in a riser can be as high as 0.200.35 kg
propylene/kg catalyst/hour, which is three to ve times
higher than that in a turbulent uidized bed. However,
more eorts are needed to commercialize this concept
as experimental observations indicate that nonuniform
distributions of solids ow in the radial direction wor-
sens under high-density conditions (Wei et al., 1994b,
1997, 1998). The gas and solids mixing in the radial
direction decreases signicantly with increasing solids
concentration and thus implies that results obtained in
small units may not correlate well with large units and
pose scale-up challenges.
2.3.2 GasSolid Reactions
Unlike the catalytic reaction discussed above, gas
solid reactions involve the solid particle as well as the
gas in the reaction. Typical examples of industrial
applications include spent FCC catalyst regeneration,
calcination, coal combustion, gasication, and silicon
chlorination. Owing to the solid particle involvement
in the reaction, signicant changes in the chemical
compositions and physical properties of the particles
occur during the reaction. Particles reduce in size and/
or increase in porosity in some reactions like coal com-
bustion, whereas particles increase in size and/or
decrease in porosity in other reactions such as lime-
stone sulfation. As a result, the particle properties
vary unlike those particle properties in catalytic reac-
tions. However, as with catalytic reactions, gassolid
reactions take place on the particle surface as gas
reactant adsorbs to the surface.
Consider the example of spent FCC catalyst regen-
eration where C
m
H
n
is present as spent catalyst:
C
m
H
n
O
2
!CO
2
COH
2
O
For decades, regeneration has utilized bubbling ui-
dized beds. Air is fed into a uidized bed regenerator
which is operated at 550800
C and 1.32 bar. The

spent FCC catalyst particles range from 50 to 70 mm
in mean diameter, and the supercial gas velocity
ranges from 0.3 to 1.8 m/s with a solids residence
time between 5 and 30 minutes. This reaction is domi-
nated by reaction kinetics or mass transfer depending
on the uidization regimes and temperature. The
unique features of the regenerator are reected by the
various uidization regimes used for the reactor design
to improve eciency. For kinetic-control reactions,
high solids concentration and high temperature are
the primary choice of the reactor conditions; the bub-
bling uidized bed is the usual choice for high solids
concentration and extensive heat transfer. The poor
mass transfer between bubble phase and emulsion
phase yields low regenerator eciency and requires
the unit to be one of the largest among the FCC pro-
cesses. On the other hand, the extensive heat transfer
with high solids concentration allows for easy heat
removal and hence easy reaction control.
With the use of zeolite catalyst in FCC, the require-
ment of complete catalyst regeneration becomes
important in improving the selectivity and yield of
the gasoline product (Avidan et al., 1990). Therefore
the one-stage well-mixed bubbling or turbulent ui-
dized bed regenerator cannot achieve high eciency
with more than 95% of the conversion. Moreover,
the steam generated by burning hydrogen in coke is
detrimental to the stability of catalyst owing to the
high temperatures produced. A two-stage uidized
bed reactor with dierent reactor combinations has
been developed to meet the process requirements.
Among various combinations, the turbulent-circulat-
ing uidized bed combination achieves great success.
This combination uses the turbulent bed for the rst
stage and the circulating uidized bed for the second
stage. The rst stage quickly burns most of the hydro-
gen and 70% of the carbon with high density and low
temperatures. Then, at high temperatures, half-regen-
erated catalyst is separated from ue gas primarily
containing steam and is transported into the circulat-
ing uidized bed. Once it is within the circulating
uidized bed, fresh air is supplied to burn out the
rest of the carbon at high temperature, high oxygen
concentration, and nearly plug-ow condition. Thus
the catalyst can be completely regenerated with high
eciency.
2.3.3 Gas-Phase Olen Polymerization
Olen polymerization through heterogeneous catalysis
is one of the most important processes to produce
polyolen resins. In this process, small catalyst parti-
cles are continuously fed into reactors operated under
controlled temperature, pressure, and chemical compo-
sition. Gas-phase olen polymerization can be carried
out in uidized bed reactors and in vertical and hor-
izontal stirred bed reactors. To compete with other
process types and maintain a high level of eciency,
the monomer mixture must be operated close to its dew
point without condensing to obtain high monomer
concentration and high yields.
Olen polymerization with metallocene catalysts
involves initiation, chain propagation, formation of a
dead chain with a saturated chain end through the
chain transfer agent, b-hydride elimination to form a
dead chain with a vinyl terminal double bond,
insertion of a macromer with a vinyl end group, and
catalyst deactivation. Assuming that all the reactions
associated with each step are rst order, the reaction
processes can be expressed as
Initiation (activation) A M
k
a
!
P
1.0
Propagation P
i.b
M
k
p
!
P
i1.b
Long chain branching P
i.b
R
m.n
k
b
!
P
im.bn1
Chain transfer P
i.b
CTA
k
c
!
R
i.b
A
b-hydride elimination P
i.b
k
[
!
R
i.b
A
Deactivation P
i.b
k
d
!
R
i.b
where M is monomer, P is the living (growing) poly-
mer, R is terminated polymer, CTA is chain transfer
agent, and A is catalyst. Typical consumption for mod-
ern catalysts is on the order of 10,00050,000 gram of
polymer per gram of catalyst. The polymerization reac-
tion is exothermic on the order of 20 cal/mol.
The UNIPOL process, developed by Union Carbide
in 1968, is a gas phase uidized bed polymerization
process and initially developed to make high-density
polyethylene with Ziegler-Natta catalysts. Along with
the development of these catalysts, enhancements to
process eciency and extension of the method scheme
to produce linear low-density polyethylene, polypropy-
lene, elastomers, and other ethylene-a olen
copolymers have been developed (Burdett et al.,
1998). A ow diagram of the UNIPOL process is
shown in Fig. 19. Four primary operations make up
this process: monomer purication, reaction, resin
degassing, and resin pelleting. The operating gas velo-
city is relatively high ( 0.6 m,s) with a conversion of
only about 2% (- 5%) per pass. Upon exiting the
reactor, the unreacted/inert gas mixture is cooled, com-
pressed, and then recycled back into the reactor until
the overall conversion reaches nearly 100%.
Polyethylene polymerization is typically carried out
at a pressure of 2030 atm and a temperature of 75
105
C. In order to avoid agglomeration and sheeting,

the operating temperature is limited by the particle
softening temperature. The uidized bed reactor starts
with an initial charge of polymer particles and estab-
lishes steady operation by supplying inert gas or a
mixture of inert gas and monomers. After the uidized
bed reaches a certain temperature, catalyst particles are
injected into the bed continuously and initiate poly-
merization to form a broad size distribution of poly-
mer particles. In polymerization, reactant diuses from
the gas phase, through the boundary layer surrounding
the particles, and into the catalyst pores to the active
sites where polymerization through a coordination-
insertion mechanism occurs to form a solid polymer.
The catalyst particles are composed of small metal
fragments and explode into a large number of smaller
fragments during the reaction. In the early stage of
polymerization, these smaller particles are quickly
encapsulated by polymer molecules and grow to large
polymer particles. For polyethylene, the reaction
begins with small particles (20200 mm in diameter)
and nishes, 35 hours later, with a mean particle dia-
meter of 2002000 mm. From gas-phase monomer to
solid-state polymer, ethylene experiences a dramatic
physicochemical transition. Similarly, the particle
morphology undergoes a dramatic change, and the
developmental stages are shown in Fig. 20.
Implementation of uidized beds for gas-phase poly-
merization also provides a typical example in which
particles grow in size during a short reaction time.
Key factors to uidized bed polymerization processes
are localized/overall temperature control and handling
of sticky particles due to process conditions operated
near the resin melting temperature. Without proper
cooling, the reactor temperature would increase until
the catalysts become inactive or particles fuse to the
reactor wall. Often, the production rate of the gas-
phase polymerization in a uidized bed reactor is lim-
ited by the maximum heat removal rate. The preferred
methods for removing heat from a uidized bed reactor
involving gas-phase polymerization are
1. Recycling unreacted gas through a heat exchan-
ger before returning the gas to the uidized bed aids in
cooling the reactor. The heat removal rate depends on
the external heat exchanger capacity and gas ow rate.
In order to maintain this cooling sink, monomer con-
sumed in the polymerization reaction is replaced by
adding excess monomer gas to the recycle stream.
2. Liquid, with a boiling point lower than the reac-
tion temperature, is intermittently injected into the bed
with a spray pattern and immediately evaporates,
thereby removing heat. The injected liquid can be
monomer to be polymerized or inert liquid hydro-
carbon.
3. Fluidized bed reactors are connected in series as
a staging operation. The uidizing gas from the prior
reactor is cooled o before entering the next uidized
bed.
4. Part of the gaseous stream leaving the top of the
reactor may be condensed by means of an external heat
exchanger and then reinjected into the uidized bed in
liquid form. The boiling point of condensable liquid
has to be lower than the operating temperature of
polymerization. Thus the reaction heat can be removed
Figure 19 Schematic diagram of uidized bed reactor for gas-phase polymerization. (From Burdett et al., 1998.)
owing to the large latent heat of vaporization and high
heat transfer rates of condensable liquids. The conden-
sable liquid can be monomers and inert liquids (e.g.,
pentane, isopentane, butane, and hexane).
Particle temperature control is another issue
associated with uidized bed polymerization as high
particle temperature will cause localized melting and
fouling. High temperature may also cause catalyst
deactivation and alter the monomer compositions in
the growing polymer particles. For example, the mono-
mer composition becomes more ethylene rich at higher
temperature in polyethylene polymerization. As dis-
cussed earlier, particle temperature varies depending
on the reaction rate (heat generation) and heat transfer
rate between the particle surface and surrounding bulk
ow. Hutchinson and Ray (1987) developed a single-
particle model to predict the temperature prole as a
function of process conditions, reaction kinetics, and
particle growth. The mass balance equation for a
growing particle is:
p
6
d
3
p
r
p
pd
2
p
k
s
M
1
M
p
12
The heat balance for a single particle in the condensa-
tion model of operation can be written as
H
p
p
6
d
3
p
r
p
pd
2
p
h
gp
T
p
T
1
qpd
2
p
H
c
13
Model analysis has revealed the existence of multiple
unstable operational conditions (Hutchinson and Ray,
1987) and that a uniform concentration of active
catalyst helps reduce high initial catalyst activities,
thereby reducing the likelihood of polymer melting.
Uniform temperature distribution across the bed is
achieved through strong particle mixing and thus
favors high-velocity operation. Also, particle size dis-
tributions range widely from 10 to 3,000 mm in uidized
bed polymerization systems, and this wide size distribu-
tion enables smooth uidization. Resin particles are
often characterized by rough surfaces and irregular
shapes (see SEM photo in Fig. 20). Polymers are dielec-
tric materials with low electrical dissipation rates, and
as a result they carry strong electrostatic charges.
Therefore special attention to reactor design and to
the handling of these charged particles is required.
For some processes, carbon particles may be added to
reduce the handling problems of these particles.
The polymerization reactor can be operated under
condensing mode conditions in which the recycled gas
stream from the reaction is partially condensed prior to
its reintroduction to the reactor. The condensing mode
of operation allows the latent heat of vaporization to
be used to absorb the substantial heat of reactions
generated from polymerization reactions, thereby
enhancing production rates of the reactor. Under the
condensing mode of operation, the state of uidization
varies with the extent of gas condensation as illustrated
in Fig. 21. For low liquid injection rates or low liquid
contents, the evaporative liquid vapor is well dispersed
in the uidizing gas yielding a gaseous stream of higher
Figure 20 Morphology developments and particle growth of polyethylene particles during gas-phase polymerization. (From Xie
et al., 1994).
density and viscosity and hence higher drag forces with
the solid particles. Under this state of liquid evapora-
tion, the bed expansion in the uidized bed would be
higher than that of a uidized bed without liquid injec-
tion (Fig. 21a) at a given gas velocity. The bed expan-
sion will not be higher, however, with any further
increase in liquid injection rates; particle agglomera-
tion is seen (Fig. 21b), although smooth uidization
can be maintained. With continued increase in liquid
contents, a segregation of the agglomerates is observed
along the axial direction, even though the bed remains
in a uidized state (Fig. 21c). In the medium range of
liquid contents, channeling in the bottom accompanied
by smooth uidization in the upper region appears in
the bed (Fig. 21d). Higher liquid content leads to
strong binding forces between particles, and channel-
ing occurs throughout the entire bed (Fig. 21e). As a
result, a nonuniform distribution of liquid content is
often observed in the bed, which leads to localized high
liquid content that can cause severe particle
agglomeration and local deuidization. With full
liquid injections, the bed becomes a slurry bed (Fig.
21f). Bottom liquid injections may cause liquid satura-
tion and channeling in the bottom zone, sometimes
even at low overall liquid contents. Liquid injection
plays a key role in enhancing liquid content but may
require injection nozzles to atomize the liquid rapidly
for rapid evaporation and a uniform mix of droplets
with particles. The uidizing gas must be able to pass
through the interstitial voids among particles and exert
sucient drag force on the particle to overcome
capillary forces, to achieve uidization without severe
particle agglomeration. The addition of ne particles
can improve uidization by absorbing the surface
moisture of polymer particles into the interstitial
voids of these ne particles that attach to the polymer
particle surface. This would create a layer of ne par-
ticles on the granular surface and reduce the capillary
forces of the granules.
Fluidization quality in a condensing mode of opera-
tion can be monitored by pressure drop measurements.
As shown in Fig. 22, the uidized bed with dry poly-
ethylene powders exhibits a typical pressure drop curve
of a normal uidized bed. With moderate liquid con-
tents, the pressure drop decreases with an increase in
gas velocity before the bed reaches the uidized state
due to partial channeling. In the uidization state, the
presence of large agglomerates and intermittent col-
lapse of local channeling cause large uctuations in
pressure drops. The pressure drop curve of dry
particles can be used as a baseline for monitoring ui-
dization quality. Pressure drop can also be coupled
with bed expansion to monitor further the uidization
quality. During uidization, the bed expands, and a
low bed expansion is a characteristic of high moisture
contents in the gas stream (Fig. 23). The low bed
expansion, however, may also reveal the existence of
dead zones, channeling, and/or deuidization.
Evaporative liquid injection is a common practice in
the feed nozzle area of the FCC riser. Similar uidiza-
tion phenomena can be observed in the uidized bed
processes like particle drying and particle gradulation
(see Sec. 2.3.4). Proper atomization of injected liquid
and dispersion of the ow to enhance mixing is the key
to smooth uidized bed operation with high liquid
injection rates and low liquid vaporization time.
2.3.4 Physical Operations
Physical operations of uidized bed technology are
very diverse. A brief description of processes and
requirements for selected examples is given as follows.
Figure 21 Particle ow patterns under various liquid contents in a uidized bed reactor. (a) Dry particle uidization, (b) particle
agglomeration, (c) agglomerate segregation, (d) bottom channeling, (e) whole bed channeling, (f) paste or slurry bed.
Fluidized beds have been used for coating
operations for many years in the production of phar-
maceuticals, fertilizers, salts and the solidication of
radioactive waste.
In the coating process, a lm is formed around a
particle by atomizing the coating agent into a uidized
bed. The process consists of spraying a liquid solution
into the bed of uidized particles, coating particles with
liquid solution, and then evaporating the solvent from
the particle surface. The solute forms a layer or lm
around individual particles causing the particle size to
increase. If the liquid spraying rate is high enough, the
coated particles may collide before the solution com-
pletely dries and cause particles to agglomerate. This
can be minimized by using high uidization velocity
and large particles. However, at lower coating ow
rates and higher uidizing velocities, the lm on parti-
cles may undergo an onion ring growth pattern. By
taking into account the hydrodynamics and coating
kinetics, optimized conditions can be achieved.
Fluidized beds are often implemented in industry
for drying granular materials. For each individual par-
ticle, the drying process involves moisture migration
from the inner core regions to the particle surface.
The surface moisture is then vaporized or evaporated
into the uidizing medium. Meanwhile, heat has to be
transferred from the uidizing medium to the inside of
each particle. Diusion, capillary ow, or internal
pressure increase by vaporization or decrease by drying
could be mechanisms of moisture migration. For dry-
ing, in the constant drying rate period, the moisture on
a particle surface is over saturated, and the drying rate
is controlled by uidization characteristics. In the
varying drying rate period, the heat conduction and
moisture diusion to the particle surface are rate-con-
trolling factors. The drying course varies with particle
temperature and heating rate, since drying is often
accompanied by particle shrinkage and deformation.
Optimized uidized bed dryer designs must emphasize
the dierence in drying kinetics. The strategy is to
enhance particle mixing and heat transfer in the con-
stant rate drying period. When materials with a high
internal resistance against moisture migration are
dried, the proper heating rate and system temperature
are essential. Heat can be supplied by the uidization
gas in a uidized bed dryer, but the gas ow need not
be the only source. Heat may be eectively introduced
by heating surfaces such as panels or tubes immersed
in the uidized bed. The system temperature can there-
fore be adjusted by controlling the external heating
source. Uniform temperature distribution is another
requirement for heat-sensitive materials to avoid
overheating. Special attention to the requirements of
uidized bed design must be made when high moisture
or sticky materials are involved. Using stirred uidized
beds and mixing dried material with fresh feeds are
eective ways to prevent overheating and sticking.
3 REACTOR MODELING AND ESTIMATION
OF DESIGN PARAMETERS
As indicated in Fig. 4, the hydrodynamics and reaction
kinetic models, both phenomenological modeling
Figure 22 Eects of moisture content on the overall pres-
sure drop in a uidized bed.
Figure 23 Eects of moisture content on the bed expansion
in a uidized bed.
(Kunii and Levenspiel, 1990) and computational uid
dynamics (CFD) modeling (Gidaspow, 1994), have
been extensively studied in the literature.
Considerable eort remains to be made in regard to
CFD modeling as a predictive tool for uidized bed
performance in terms of conversion and selectivity.
Phenomenological and empirical models are
routinely utilized for predictive purposes in commer-
cial uidized bed reactors in industry today.
3.1 Bubbling Fluidized Bed Reactors
Bubbles are the key features of the bubbling uidized
bed reactor. Two-phase theory, which considers the
bed to be composed of a bubble phase and an emulsion
phase, is often the basis in reactor model formulation.
For the cases of ne particle uidization, bubbles are
surrounded by a cloud region, as implied in Fig. 24;
depending on model consideration, the cloud region
can be treated separately or lumped with either the
bubble or the emulsion. Many models have been devel-
oped with variations in the degree of gas backmixing in
each of the two phases and the interphase gas
exchange. Grace (1986b) provided a thorough descrip-
tion of reactor models. Modeling is also described in
detail in Chapter 9.
The simplest case of two-phase models is shown in
Fig. 25a, in which the cloud region is merged with the
dense phase. Based on the two-phase model, mass bal-
ance equations in each phase for a given species can be
formulated, and the reactor outlet concentration can
be predicted when the hydrodynamic variables and
reaction kinetics are known. Assuming a particle-free
bubble phase and plug ow of gas bubbles, the mass
balance for species A can be expressed as:
U
b
dC
ba
dz
k
q
a
b
e
b
C
dA
C
bA
0 14
with a boundary condition
C
bA
C
Ain
at z 0 15
Assuming complete mixing in the emulsion phase with
respect to gas and solid particles, the mass balance for
species A in the emulsion phase can be written as
U
d
C
Ain
C
dA

_
H
f
0
k
q
a
b
e
b
C
bA
C
dA
dz
1 e
b
e
ds
H
f
r
A
0
16
where r
A
is the reaction rate per unit volume of cata-
lysts. In obtaining Eq. (16), the mass transfer resistance
between the catalyst particles and the surrounding gas
ow is neglected. The concentration of Species A in the
emulsion phase is constant. The concentration proles
of species A in the bubble phase can be obtained by
integration of Eq. (14) as
C
bA
C
dA
C
Ain
C
dA
exp
k
q
a
b
e
b
U
b
z
_ _
17
The concentration of species A at the bed surface can
be obtained from the mass balance at bed height H
f
:
UC
Aout
U
b
C
bA zH
f
U
d
C
dA
18
The analytical solutions of Eqs. (14)(16) for dierent
types of reaction kinetics were given by Grace (1986b).
This simple model captures the key features of bub-
bling uidized bed reactor and is convenient to use.
Figure 24 Congurations of a two-phase model for bubbling uidized bed reactors.
As discussed previously, the addition of ne parti-
cles is commonly practiced in the operation of uidized
bed reactors. In some cases, the addition of ne
particles is necessary in order to provide smooth bed
operations. In systems with ne particles, the assump-
tion of a solids-free bubble phase is no longer valid.
Thus the two-phase model needs to be modied to
account for the ne particle presence and hence reac-
tion in the bubble phase. The mass balance of species
A in the bubble phase can now be expressed by
U
b
dC
bA
dz
k
q
a
b
e
b
C
dA
C
bA
e
b
e
bs
r
A
0 19
The mass balance equation for species A in the emul-
sion phase remains essentially the same as Eq. (16).
Grace (1986b) further simplied the model by assum-
ing that no net gas ows through the emulsion phase,
i.e., U
d
0. Under this assumption, Eqs. (16) and (19)
can be solved analytically for simple reaction kinetics.
Eects of ne particle solids concentration in the
bubble phase on the reactor conversion can also be
examined analytically.
Two-phase models could not be directly applied to
account for the entrance eects of the reactor. For
shallow beds or reactors with strong jetting, the distri-
butor eects dominate the hydrodynamic behavior of
the bed and thus have to be considered dierently from
that of the in-bed region. The two-phase model that
applies to the in-bed region could be modied to
describe the entrance eects, as conceptually depicted
in Fig. 25b. Detailed descriptions of distributor eects
on hydrodynamics are given in Chapter 6.
3.2 Turbulent Fluidized Bed Reactors
In the turbulent uidized bed reactor, the two-phase
ow nature is still distinguishable though signicantly
less distinct than that in the bubbling regime. Thus the
models developed based on the two-phase concept for
the bubbling regime can be extended for the turbulent
regime. Due to an increased presence of particles in the
bubble phase, the bubble phase is referred to as the
dilute phase as shown in Fig. 25a. For catalytic reac-
tions in a turbulent uidized bed reactor with an
assumption of plug ow in the dilute phase, the con-
centration of species A can be expressed as
de
b
V
b
C
bA
dz
k
q
a
b
e
b
C
bA
C
dA
e
b
e
bs
r
A
0
20
Likewise, the concentration of species A in the dense
phase can be written as
d1 e
b
1 e
ds
V
d
C
dA
dz

d
dz
1 e
b
1 e
ds
D
ax
dC
dA
dz
_ _
k
q
a
b
e
b
C
dA
C
bA
1 e
b
e
ds
r
A
0
21
When the ow properties remain unchanged along
the axial direction in the dense phase, Eq. (21) can be
rearranged as
d
2
C
dA
dz
2

V
d
D
ax
dC
dA
dz

k
q
a
b
D
ax
e
b
1 e
b
1 e
ds
C
dA
e
ds
1 e
ds
r
A
D
ax
k
q
a
b
D
ax
e
b
1 e
b
1 e
ds
C
bA
0
22
The term V
b
,D
ax
is associated with the Peclet number,
Pe
a
HV
b
,D
ax
. For the rst order reaction,
r
A
kC
A
, and the reaction rate term is associated
with the Damkholer number, Da kH1 e,D
ax
.
High-velocity operation in the turbulent bed yields
signicant particle entrainment to the upper dilute
region or freeboard. The one-dimensional plug ow
model can be used to describe the reaction in this
region. The mass balance for species A can be
expressed as
dUC
A
dz
e
fs
r
A
0 23
where e
fs
is the solids volume fraction in the freeboard.
The total gas (reactants, products, and inerts) volu-
metric ow rate Q, can be expressed as
Figure 25 Schematic representation of two-phase model
developed for bubbling uidized bed. (a) Two-phase model,
(b) two-phase model with jetting.
Q AU
n
g
R
m
T
P
24
where n
g
is the total molar ow rate for all gas species
at a given axial position, z. The total gas ow rate
varies due to molar ow rate changes of the individual
species through reactions as described by Eqs. (20)
through (23). For convenience, an averaged value can
be assumed, and therefore the molar ow rate for
species A, n
gA
, can be expressed by
UC
A
A n
gA
25
In the upper dilute region,
1
A
dn
g
dz
e
bs
r
i
26
The inlet boundary condition for the dense and dilute
phase equations in a uidized bed can be given respec-
tively as
V
d
C
A0
V
d
C
dA
z0
D
ax
dC
dA
dz
_ _
z0
at z 0
27
C
bA
C
A0
at z 0 28
where C
A0
is the inlet concentration of species A.
The outlet boundary conditions for the dense phase
equation can be written as
dC
A
dz
_ _
zH
0 at z H 29
To exemplify the approach, the butane oxidation to
maleic anhydride process is presented. The required
information on hydrodynamics and mass transfer can
be obtained from the equations given in Table 3. These
equations have been veried for turbulent uidized bed
reactor applications of butane oxidation. Dierent
correlations are required for other applications.
Figure 26 describes a triangular reaction network
for butane oxidation (Buchanan and Sundaresan,
1986). The reactions for each pathway can be
expressed as
r
1
: 2CH
3
CH
2
CH
2
CH
3
7O
2
!2C
4
H
2
O
3
8H
2
O
r
2
: C
4
H
10
mO
2
!132mCO2m9CO
2
5H
2
O
r
3
: C
4
H
2
O
3
bO
2
!6 2bCO2b 2CO
2
H
2
O
where b and m are stoichiometric coecients. The
reaction rate for each pathway is written as
r
1

k
1
C
bta
1 K
B
C
bta
,C
O
2
K
MA
C
ma
,C
O
2
30
r
2

k
2
C
bta
1 K
B
C
bta
,C
O
2
K
MA
C
ma
,C
O
2
31
r
3

k
3
C
ma
1 K
B
C
bta
,C
O
2
K
MA
C
ma
,C
O
2
32
where k
1
, k
2
, k
3
, K
B
, and K
MA
are rate constant para-
meters. The underlying assumptions made in the model
equations are as follows:
The uidized bed reactor is operated under isother-
mal conditions.
Mass transfer resistance on the catalyst particle in
the dense phase is negligible.
Catalyst activity is uniform and remains unchanged
throughout the uidized bed operation.
Reaction takes place in the dense and dilute phases
of the bed as well as in the freeboard.
To summarize the model calculation, the concentra-
tions of each species (reactant and products) in the bed
are calculated using Eqs. (20) and (21). Equation (23)
is used to calculate the concentration proles of each
species in the upper dilute region. The total gas ow
rate can be calculated by addition of the molar ows of
each species at any given axial location within the bed.
The total ow rate can be converted to the volumetric
gas ow rate, and the supercial gas velocity may be
determined using Eqs. (24) and (25). The values for
hydrodynamic parameters are then calculated for the
obtained gas velocity and the set of equations is solved
numerically. Figure 27 shows the axial proles of
butane and maleic anhydride concentrations calculated
based on the model equations. The eects of ne par-
ticle (d
p
- 45 mm) contents on butane conversion are
also predicted as shown in Fig. 28. As surmised from
the gure, the content of ne particles aects uidiza-
tion properties and reactor conversion.
3.3 Riser and Downer Reactors
The hydrodynamic model development for a
circulating uidized bed follows the same approach
as bubbling and turbulent beds. In the macroscale,
the gassolid ow is characterized by a coexistence of
a bottom dense region and an upper dilute region. The
ow in the radial direction can be described by a core-
annular structure with a dense particle region close to
the wall and a central dilute region. In the mesoscale,
the ow is characterized by the presence of particle
clusters moving upward in the core and moving down-
ward near the wall. Phenomenological models attempt
to capture these main ow features and express the
ow behavior in simplied mathematical form to
describable levels. Cluster concepts evolved from the
observations of large slip velocity between gas and
particles. This type of model describes the motion of
clusters and can be combined with reaction models to
predict reactor performance when cluster properties
are known.
Various core-annular models have been developed
to describe the gassolid ow (Horio et al., 1988; Bai et
al., 1995; Bolton and Davidson, 1998). The main
dierence among these models lie in the degree of
complexity and in the assumptions associated with
simplications. Core-annular ow structures become
dominant in the upper dilute region. Thus, when the
dilute ow is predominantly present in the riser, mod-
els based on core-annular structures can be applied to
reactor models. Kunii and Levenspiel (1990) extended
the conventional uidized bed model (a dense lower
region coupled with a freeboard upper region) to cir-
Table 3 Correlations Used for Model Simulation of Butane Oxidation in a Turbulent Fluidized Bed
Reactor
Parameters Correlations Sources
U
d
U
2
d
85.7e
ds
m
g
d
p
r
g
U
d
0.5711 e
ds
3
r
p
d
p
g
r
g
0
This chapter
V
d
V
d
U
d
,1 e
ds
This chapter
V
b
V
b
e
b
U U
d
U V
d
1 e
b
1 e
ds
This chapter
U
c
U
c
gd
p
_
m
20
m
_ _
0.20
r
g20
r
g
r
p
r
g
r
g
D
d
p
_ _
0.27
0.211
D
0.27

2.42 10
3
D
1.27
_ _
Cai et al. (1989)
e
e 3.0
U 1
U 2
_ _
Ar
0.055
r
p
r
g
r
g
_ _
0.132
d
p
D
_ _
0.02
exp0.157F
45
Jiang et al. (1999)

e
b
e
b

e e
ds
1
e
ds
assume e
s
0
This chapter
e
mb
e
mb

0.321 exp0.157F
45
r
0.042
g
m
0.081
g
g
0.205
d
0.176
sv
Xie et al. (1995)
e
ds
1 e
ds
e
mb
0.02Ar
0.13
U
U
c
_ _
3
This chapter
e
bs
e
bs
0.07
U
U
t
_ _
0.3
exp0.25F
45
This chapter
e
fs
e
fs
e
s
e
bf
e
s
expaz H
f
e
bf
1 e
Kunii and Levenspiel (1997)
W
tf
W
tf
Ar
p
_
H
H
f
e
fs
dz e
s
H H
f
e
bf
e
1 expaH H
f
a
This chapter
H
f
W
t
r
p
A1 eH
f
W
tf Jiang et al. (1999)
k
q
a
b
Experiments Jiang et al. (1999)
D
ax
Experiments Jiang et al. (1999)
culating uidized beds. The two-phase theory was used
to describe the lower dense region, while the entrain-
ment model proposed by Wen and Chen (1982) was
adopted to account for the upper dilute region. When
the radial solids concentration variation is not preva-
lent, the entrainment model can provide a sound
account of the macroscale hydrodynamic behavior.
Detailed information on hydrodynamic modeling is
available in Chapter 19.
Although the ow structures have been well recog-
nized, quantitative information on the parameters
describing the structure is far from sucient. As
experimental results show continuous proles for the
solids concentration distributions and gas dispersion
proles in both radial and axial directions, no clear,
denitive demarcation can dene the core and annular
regions, cluster and surrounding suspension, and the
lower dense and upper dilute regions. Since distribu-
tions of the ow properties in the radial and axial
direction are nonuniform in the riser ow, a general
two-dimensional dispersion model can be used to
describe this ow. Considering a reaction under iso-
thermal conditions and steady-state operations, the
general form of the mass balance equation in the 2D
cylindrical coordinate for a species A can be expressed
as
o
oz
V
gz
eC
A
_ _
1
r
o
or
rV
gr
eC
A
_ _
o
oz
eD
ax
oC
A
oz
_ _
1
r
o
or
reD
r
oC
A
or
_ _
1 er
A
33
Note that D
ax
, D
r
, e, and V
g
vary with radial and axial
positions. For an axial symmetric ow, V
gr
0 and
V
gz
f r. With the assumption that D
ax
, D
r
, e, and
V
g
remain unchanged within a dened computational
cell and may vary from element to element, the mate-
rial balance for a species A in a given element, Eq. (33),
can be further simplied to
oC
A
oz

H
Pe
a
o
2
C
A
oz
2

D
Pe
r
1
r
o
or
r
oC
A
or
_ _
1 e
eV
gz
r
A
34
Noting that eV
gz
U, the boundary conditions for
Eq. (34) can be written as
Figure 26 Kinetic network for butane oxidation to maleic
anhydride. (From Buchanan and Sundaresan, 1986.)
Figure 27 Model predictions of concentration in a turbulent
uidized bed. (From Jiang and Fan, 1999.)
Figure 28 Eects of ne particle content on the butane con-
version in a turbulent uidized bed reactor. (From Jiang and
Fan, 1999.)
C
A0
C
A
z0
H
Pe
a
dC
A
dz
_ _
z0
at z 0
35
oC
A
or
0 at r 0 and r
D
2
36
oC
A
oz
0 at z H 37
The mass balance equations for a given species in the
reaction system need to be solved in conjunction with a
hydrodynamic model in order to simulate the reactor
performance.
For illustrative purposes, the following hydrody-
namic model and correlations are used in the model
calculation of catalytic reactions of propylene ammox-
idation to acrylonitrile. These hydrodynamic models
and correlations are appropriate for a high-density
riser under propylene ammoxidation conditions.
Specically, the axial proles of cross-sectional aver-
aged solids concentration was obtained by tting the
cluster-based model proposed by Li and Kwauk (1980)
with the experimental data from Wei et al. (1998). The
axial prole of solids concentration obtained can be
expressed as:
ln
ee e
a
e
ee
_ _

1
Z
0
z Z
i
38
where e
a
, e
, and Z
0
are empirical constants that can be
calculated using the correlations
e
a
A
18Re
s
2.7Re
1.687
s
Ar
_ _
0.0741
Re
s

d
p
r
g
m
g
U
1 e
a
G
s
e
a
r
p
_ _
39
e
B
18Re
s
2.7Re
1.687
s
Ar
_ _
0.02857
40
Z
0
500 exp 69e
e
a
41
where, for acrylonitrile catalyst, A 0.484 and
B 0.95. Z
i
can be obtained by the overall solids
balance in the circulating loop.
The radial proles of solids concentration are
calculated using the correlation (Wei et al., 1998)
1 ez. r
1 ee
2.3
2.1
1 exp 20r,D 7.665
42
Experimental data for gas velocity proles in high-den-
sity risers are very limited. Radial proles of the gas
velocity are expected to follow a trend similar to that
of the particle velocity (Wei et al., 1998). Using the
form of the equation for the radial particle velocity
prole, Wei et al. (1998) suggested that the gas velocity
prole could be obtained by tting the radial solids
concentration proles and considering the mass bal-
ance of the gas ow. Specically, the mass balance of
gas ow at any cross section of the riser is expressed as
_
D,2
0
2rV
gz
rer dr
D
2
U
4
43
The radial prole of gas velocity can be obtained by
substituting the correlation for radial proles of solids
concentration, Eq. (42), in Eq. (43):
V
gz
r
U

2.6
1 exp20r,D 10q
0
44
where q
0
is an empirical constant that can be calculated
from
q
0
2.5
2
1 exp1.71491 ee 1.263
45
Based on the ammoxidation of propylene to the
acrylonitrile reaction kinetic network shown in Fig.
18, a prediction for the reactor performance of a
high-density rise is made. The axial proles of the
main compounds for typical high-density riser condi-
tions are shown in Fig. 29 along with the experimental
data obtained in a pilot-scale reactor. Axial Peclet
numbers for gas and solids in the circulating uidized
bed range from 4 to 8 (Liu et al., 1999) and vary
slightly with operating conditions. Due to the axial
gas backmixing, the changes in concentration along
the axial position are less than those in a reactor
with plug ow. The reaction is rst order with respect
to propylene and zeroth order with respect to O
2
and
NH
3
. Therefore a much higher reaction rate at the inlet
region exists than in a well-mixed reactor. Extensive
gas backmixing in a reactor results in overoxidization
of acrylonitrile and a yield reduction.
Models can also be used for parameter sensitivity
analysis. Due to the complexity of reaction networks
and hydrodynamics, the eects of various factors on
the reactor performance are complex. Model analysis
provides a guiding tool for process development. The
eects of Pe
a
on the yield of acrylonitrile are shown in
Fig. 30. As shown, when Pe
a
is less than 0.05, the yield
of acrylonitrile changes marginally with Pe
a
, and the
reactor can be considered well mixed. When Pe
a
is
greater than 10, the yield of acrylonitrile is almost
the same as that in a plug-ow reactor. Model simula-
tion also reveal the existence of a Pe
a
-sensitive range
from 0.02 to 10, in which the yield of acrylonitrile
increases signicantly with an increase in Pe
a
.
Figure 31 shows the model analysis of the eects of
radial gas dispersion coecient on radial proles of
propylene concentration. The radial mass transfer
has a signicant eect on the conversion and yield.
When the radial Peclet number decreases from 1400
to 200, the conversion of propylene increases by over
10%, and the yield of acrylonitrile increases by about
7%. Since the reaction is rst order with respect to
propylene, risers are operated under dilute conditions
at Pe
r
200, so the radial concentration distribution
of propylene is uniform and radial mass transfer is not
the limiting factor to the reaction. Nonuniform distri-
butions of catalyst in a radial direction develop as the
particle density of the riser increases. The low turbu-
lent intensity and low gas velocity in the wall region
result in poor mass transfer, and the low mass transfer
and high catalyst density yield a reactant-decient zone
in the wall region. Since the reaction network of pro-
pylene ammoxidation is represented by a combination
of parallel and series reactions with the intermediate
product being the desired product as shown in Fig. 18,
poor mass transfer favors complete oxidation instead
of acrylonitrile formation.
4 CONCLUSION
Fluidized bed reactor systems have been proved useful,
reliable, and cost-eective, suitable to many industrial
applications. The general approaches for designing a
uidized bed reactor are described in this chapter with
specic attention given to the knowledge required for
the selection of ow regimes and to the interplay
between transport properties and kinetics. Examples
of several industrial uidized bed reactor processes
are given to illustrate the operational principles. No
single universal approach for designing a uidized
bed reactor system exists. At present, uidized bed
reactor system design relies heavily on correlations,
engineering models, and plant observations. The com-
putational uid dynamic approach may oer viable
and attractive options over the traditional approaches;
however, challenges remain in considering reactive
Figure 29 Axial proles of concentrations for propylene,
oxygen, and carbon dioxide in the process of ammoxidation
of propylene. (From Wei et al., 2000a.)
Figure 30 Eects of axial Peclet number on acrylonitrile
yield in a high-density riser. (From Wei et al., 1997.)
Figure 31 Eects of radial Peclet number on radial proles
of propylene concentration in a high-density riser. (From
Wei et al., 2000a.)
ow that involves the incorporation of kinetics, heat,
and mass transfer properties into uid dynamic calcu-
lation. Continued research eorts are necessary toward
the development of uidized bed processes for new
industrial applications. More measurements on the
microscopic properties of the uidized bed reactor
such as velocities, concentrations, and temperatures
for a given ow regime and reaction kinetics are neces-
sary to allow further renement of the current reactor
models and their predictions.
NOTATION
A = cross-sectional area of a uidized bed
Ar = Archimedes number ( r
g
r
s
r
g
gd
3
p
,m
2
g
a
b
= interfacial surface area of bubble per unit
bubble volume (a
b
6,d
b
C = concentration
C
A
= concentration of species A
C
A0
= inlet concentration of species A
C
b
= reactant concentration in the dilute phase
C
d
= reactant concentration in the dense phase
d
b
= mean bubble diameter based on equivalent
spherical volume
d
p
= particle diameter
d
p
= nondimensional particle size
d
p
r
g
r
p
r
g
g,m
2
g
1,3
d
sv
= surfacevolume equivalent particle diameter
1,x
i
,d
p
D = column diameter
Da = Damkoler number
D
ax
= gas dispersion coecient in axial direction
D
m
= molecular diusion coecient
D
r
= gas dispersion coecient in radial direction
F
45
= percentage of ne contents
g = gravitational acceleration constant
G
s
= solids circulation rate
H = uidized bed height
H
0
= static bed height
H
f
= dense-phase bed height
H
c
= latent heat
H
p
= heat of polymerization reaction
H
R.298K
= heat of reaction at 298K
h
gp
= particle-to-gas heat transfer coecient
k = reaction rate constant
k
0 00
= rst-order catalytic rate constant dened by
Kunii and Levenspiel (1991) in Fig. 10
k
d
= decay coecient of catalyst activity
k
q
= interfacial mass transfer coecient
(volumetric transfer rate per unit bubble
surface area)
k
s
= mass transfer coecient
K
1
. K
2
. K
3
= rate constant parameters
K
b
. K
MA
= rate constant parameters
m = reaction order
n
g
= molar gas ow rate
M
p
= monomer concentration at particle surface
M
1
= monomer concentration in surrounding
medium
P = bed pressure
P = pressure drop
Pe
a
= Peclet number based on axial dispersion
coecient ( UH,D
ax
)
Pe
r
= Peclet number dened based on radial
dispersion coecient ( UD,D
r
Pr = Prandtl number ( n,a)

q = condensated vaporization rate
q
0
= empirical constant dened in Eqs. (44) and
(45)
Q = total gas volumetric ow rate
Re
p
= Reynolds number based on particle
diameter r
g
UD
p
,m
g
Re
s
= Reynolds number dened in Eq. (39)
R = column radius
r = radial coordinate in cylindrical coordinate
system
r
i
= reaction rate per unit volume of catalyst for
species i
r
A
= reaction rate per unit volume of catalyst for
species A
R
m
= gas constant
r
p
= polymerization rate per unit volume of
particle
Sc = Schmidt number ( m
g
,r
g
D
m
)
Sh = Sherwood number ( k
s
d
p
,D
m
)
t = time
T = temperature
T
1
= bulk (or surrounding medium) temperature
T
p
= particle temperature
t
p
= particle residence time
u = velocity
uu = average velocity
U = supercial gas velocity e
b
V
b
1 e
b
V
d
= nondimensional gas velocity

Ur
2
g
,m
g
r
p
r
g
g
1,3
Re,Ar
1,3
U
b
= supercial bubble rising velocity
U
c
= transition velocity from bubbling to
turbulent regimes
U
d
= supercial gas velocity in the dense phase
U
mf
= minimum uidization velocity
U
t
= terminal velocity
V
b
= interstitial (linear) gas velocity in the bubble
phase
V
d
= interstitial (linear) gas velocity in the dense
phase
V
g
= linear gas velocity
V
gz
= linear gas velocity in axial direction
V
gr
= linear gas velocity in the radial direction
V
pc
= particle velocity in the dilute core region
V
pw
= particle velocity in the dense wall region
W
t
= weight of catalyst in the whole bed
W
tf
= weight of catalyst in free board
z = vertical (axial) coordinate
Z
0
= empirical constant in Eq. (41)
Z
i
Greek Letters
a = thermal diusivity
d
w
= thickness of the dense wall region
e = overall or local voidage
ee = cross-sectional averaged voidage
e
a
e
b
= volume fraction of the dilute (or bubble)
phase ( bubble volume/bed volume)
ee
c
= averaged voidage in the dilute core region
ee
w
= average voidage in the dense wall region
e
d
= volume fraction of the dense (or emulsion)
phase ( 1 e
b
)
e
ds
= volume fraction of solid particles in the
dense phase (per unit volume of dense-
fraction)
e
bs
dilute phase (per unit volume of dilute
phase
e
fs
free board region
e
mb
= voidage at incipient bubbling
e
mf
= voidage at incipient uidization
e

l = thermal conductivity
m
g
= gas viscosity
n = kinematic viscosity
r
p
= particle density
r
g
= gas density
t
0 00
= contact time dened by Kunii and
Levenspiel (1991) in Fig. 10
t
p
= mean particle residence time
Subscripts
A = species A in the reaction
b = bubble phase in bubbling bed or dilute
phase in turbulent bed and riser
bta = butane in the reaction of butane oxidation
d = dense phase
in = inlet
ma = maleic anhydride in the reaction of butane
oxidation
O
2
= oxygen in the reaction of butane oxidation
out = exit
s = solid (or particle) phase
REFERENCES
Avidan, AA, Shinnar R. Development of catalytic cracking
technology, a lesson in chemical reactor design. Ind Eng
Chem Res 29:931942, 1990.
Bai DR, Zhu JX, Jin Y, Yu ZQ. Internal recirculation ow
structure in vertical upward owing gassolid suspensions
I, a core/annular model. Powder Technol 85:171178,
1995.
Bergna HE. Method of making maleic anhydride. U.S.
Patent, 4,769,477, 1988.
Bolton LW, Davidson JF. Recirculation of particles in fast
uidized risers. In: Basu P, Large JF, eds. Circulating
Fluidized Bed Technology II. Toronto: Pergamon Press,
1988.
Buchanan JS, Sundaresan S. Kinetics of redox properties of
vanadium phosphate catalysts for butane oxidation.
Applied Catalysis 26:211226, 1986.
Burdett ID, Eisinger RS, Cai P, Lee KH. Recent develop-
ments in uidized bed process for olen polymerization.
AIChE Annual Meeting, Miami, FL, 1998.
Cai P, Chen SP, Jin Y, Yu ZQ, Wang ZW. Eect of operat-
ing temperature and pressure on the transition from bub-
bling to turbulent uidization. AIChE Symp Ser
85(270):3743, 1989.
Contractor RM, Sleight AW. Maleic anhydride from C-4
feedstock using uidized bed reactor. Catalyst Today
1:587607, 1987.
Davidson JF, Harrison D. Fluidized Particles. New York:
Fan LS, Zhu C. Principles of GasSolids Flows. New York:
Fan LS, Ghosh-Dastidar A, Mahuli S. Calcium carbonate
sorbent and methods of making and using same. US
Patent 5, 779,464, 1998.
Froessling N. The evaporation of falling drops. Gerlands
Beitr. Geophys 52:170, 1938.
Ghosh-Dastidar A, Mahuli S, Agnihotri R, Fan LS.
Investigation of high-reactivity calcium carbonate sorbent
for enhanced SO
2
capture. Ind Eng Chem Res 35:598
606, 1996.
Gidaspow D. Multiphase Flow and Fluidization-Continuum
and Kinetic Theory Descriptions. San Diego: Academic
Press, 1994.
gassolid and other two-phase suspension. Can J Chem
Eng 64:353363, 1986a.
Grace JR. Fluidized beds as chemical reactors. In: Geldart
D, ed. Gas Fluidization Technology. New York: John
Wiley, 1986b, pp 285339.
Horio M, Morishita K, Tachibana O, Murata N. Solids dis-
Technology II. Toronto: Pergamon Press, 1988.
Hutchinson RA, Ray WH. Polymerization of olens through
heterogeneous catalysis, VII. Particle ignition and extinc-
tion phenomena. J Appl Polym Sci 34:657676, 1987.
Jazayeri B. Successfully scale up catalytic gas-uidized beds.
Chem Eng Prog 91(4):2631, 1995.
Jiang P, Fan LS. Characteristics of turbulent uidized beds
with ne particles at high temperatures and pressures.
AICHE Annual Meeting, Dallas, 1999.
Jiang P, Bi H, Jean RH, Fan LS. Bae eects on perfor-
mance of catalytic circulating uidized bed reactor.
AIChE J 37:13921400, 1991.
Jin Y, Yu ZQ, Shen JZ, Zhang, L. Pagoda-type of vertical
internal baes in gas uidized beds. Int Chem Eng
20:191196, 1980.
Karri SBR, Knowlton TM. Gas distributor and plenum
design in uidized beds. In: Yang WC, ed. Fluidization
Solids Handling and Processing. Westwood, NJ: Noyes,
1999.
Krishna R. A systems approach to multiphase reactor selec-
tion. Advances Chem Eng 19:201249, 1994.
Kunii D, Levenspiel O. Entrainment of solid from uidized
beds, II. Operation of fast uidized beds. Powder Technol
61:193206, 1990.
Kunii D, Levenspiel O. Fluidization Engineering. 2d ed.
Kunii D, Levenspiel O. Circulating uidized-bed reactors.
Chem Eng Sci 52:24712482, 1997.
Li Y, Kwauk M. The dynamics of fast uidization. In: Grace
JR, Masten, JM, eds. Fluidization. New York: Plenum
Press, pp 537544.
Liu JZ, Grace JR, Bi H, Morikawa H, Zhu J. Gas dispersion
in fast uidization and dense suspension upow. Chem
Eng Sci 54:54415449, 1999.
Maskaev VK, Baskakov AP. Features of external heat trans-
fer in a uidized bed of coarse particles. Int Chem Eng
14:8083, 1974.
Molerus O. Heat transfer in gas uidized beds, Part 2.
Dependence of heat transfer on gas velocity. Powder
Technol 70:1520, 1992.
Ran X, Wang ZW, Wei F, Jin Y. Particle velocity and solids
fraction proles of a swirling ow in a riser. Proceedings
of the Sixth International Conference on Circulating
Fluidized Bed, Wu rzburg, Germany, Aug. 1999, pp 131
139.
Shi D, Nicolai R, Reh L. Wall-to-bed heat transfer in circu-
lating uidized beds. Chem Eng Proc 37:287293, 1998.
Sit SP, Grace JR. Eect of bubble interaction on the inter-
phase mass transfer in gas uidized beds. Chem Eng Sci
36:327335, 1981.
Squires A. Origins of the fast uid bed. Adv in Chem Eng
20:137, 1994.
Sun G, Grace JR. The eect of particle-size distribution on
the performance of a catalytic uidized-bed reactor. Chem
Eng Sci 45(8):21872194, 1990.
Sze MC, Gelbein AP. Make aromatic nitriles this way.
Hydrocarbon Process 3:103106, 1976.
Wei F, Lu FB, Jin Y, Yu ZQ. Petrochemical Engineering (in
Chinese) 23:631636, 1994a.
Wei F, Jin Y, Yu ZQ. Macro visualization of gas solids ow
structure in high density circulating uidized beds. In:
IV. New York: AIChE, 1994b, pp 588596.
Wei F, Lai ZP, Jin Y, Yu ZQ. A CFB reactor model for the
synthesis of acrylonitrile. Proceedings of Asian-Pacic
Conference on Chemical Reaction Engineering, Beijing,
China, 1996, pp 625630.
Wei F, Lu FB, Jin Y, Yu ZQ. Mass ux proles in a high
density circulating uidized bed. Powder Technol 91:189
195, 1997.
Wei F, Lin HF, Cheng Y, Wang ZW, Jin Y. Proles of
particle velocity and solids fraction in a high-density
riser. Powder Technol 100:183189, 1998.
Wei F, Du B, Fan LS. Gas mixing in a turbulent uidized
bed under high temperature/pressure conditions. AIChE
Annual Meeting, Los Angeles, CA, 2000a.
Wei F, Wan XT, Hu YQ, Wang ZG, Yang YH, Jin Y. A
pilot plant study and 2-D dispersion-reactor model for a
high-density riser reactor. Chem Eng Sci, in press, 2000b.
Wen CY, Chen LH. Fluidized bed freeboard phenomena:
entrainment and elutriation. AIChE J 28(1):117128,
1982.
Wen CY, Fan LT. Models for Flow Systems and Chemical
Xie HY, Geldart D. Fluidization of FCC powders in the
bubble-free regime: eect of types of gases and tempera-
ture. Powder Technol 82:269277, 1995.
Xie T, McAuley KB, Hsu JCC, Bacon DW. Gas phase ethy-
lene polymerization: production processes, polymer prop-
erties, and reactor modeling. Ind Eng Chem Res 33:449
479, 1994.
Yang G, Wei F, Jin Y, Yu ZQ, Wang Y. Unique properties
of 30-mm particles as the catalyst of uidized-bed reac-
tors. AIChE J 43:11901193, 1997.
Zhang W, Tung Y, Johnsson F. Radial voidage proles in
fast uidized beds of dierent diameters. Chem Eng Sci
46(12):30453052, 1991.
13
Fluidized Bed Scaleup
Leon R. Glicksman
Massachusetts Institute of Technology, Cambridge, Massachusetts, U.S.A.
1 INTRODUCTION
Typically, the development of a new commercial ui-
dized bed process involves a laboratory bench scale
unit, a larger pilot plant, and a still larger demonstra-
tion unit. Many of the important operating character-
istics change between the dierent size units. The
critical problem of scale-up is how accurately to
account for the performance changes with plant size
to optimize the performance of a full-size commercial
unit. In addition, it would be helpful if the link between
smaller units could be understood, so that the smaller
unit could be used to improve the commercial plant
performance or solve existing problems.
One discouraging problem is the decrease in reactor
or combustor performance that is found when some
pilot plants are scaled up to larger commercial plants.
These problems can be related to poor gas ow pat-
terns, undesirable solid mixing patterns, and physical
operating problems; see Matsen (1985). In most
instances, it is dicult to make any observations or
measurements directly within the commercial uidized
bed. Thus problem solving is hindered by a lack of
understanding of the underlying cause of poor beha-
vior. In the Synthol CFB reactors constructed in South
Africa, rst scale-up from the pilot plant increased the
gas throughput by a factor of 500. Shingles and
McDonald (1988) describe the severe problems initially
encountered and their resolution.
In some scaled-up uidized bed combustors, the
lower combustion zone has been formed into a narrow
rectangular cross section. Sometimes the lower section
is divided into two separate subsections; this is referred
to sometimes as a pant leg design, to provide better
mixing of fuel and sorbent in a smaller eective cross
section, and to reduce the potential maldistribution
problems in the scaled-up plant.
Matsen (1985) pointed out a number of additional
problem areas in scale-up such as consideration of par-
ticle size balances, which change over time due to reac-
tion, attrition, and agglomeration. Erosion of cyclones,
slide valves, and other components owing to abrasive
particles are important design considerations for com-
mercial units that may not show up in pilot plants.
As the diameter of a uidized bed is increased, the
bed hydrodynamics may not remain similar. In some
instances, the ow regime may change between small
and large beds even when using the same particles,
supercial gas velocity, and particle circulation rate
per unit area. The issue of scale-up involves an under-
standing of these hydrodynamic changes and their
inuence on chemical reaction rates, conversion e-
ciency, and thermal conditions through variations in
gas distribution, gassolid contact, residence time, and
solid circulation and mixing.
There are several avenues open to deal with scale-
up. There are a large number of empirical and semi-
empirical correlations that exist in the uidized bed
literature to predict uid dynamic or overall process
behavior. In addition, there are probably a large num-
ber of proprietary correlations used by individual com-
panies. The danger lies in extrapolating these relations
to new geometric congurations of the riser or inlet,
which lead to ow conditions outside the range of pre-
vious data, or to beds that are much larger. Avidan
and coauthors in a 1990 review of FCC summed up the
state of the art: basic understanding of complex ui-
dization phenomena is almost completely lacking.
While many FCC licensors and operators have a
large body of in-house proprietary data and correla-
tions, some of these are not adequate, and fail when
extrapolated beyond their data base; see Avidan et al.
(1990).
As an example, consider the inuence of mean par-
ticle size. In the early work on bubbling uidized bed
combustors, attempts were made to use relations from
the classic uidization literature that had been devel-
oped from experiments with smaller particles, such as
FCC applications. In many cases, the relationships for
small particles gave erroneous results for combustors
with much larger particles. For example, the two-phase
theory equating the excess gas velocity above minimum
uidization to the visible bubble ow was substantially
in error for large particle systems. Jones and
Glicksman (1986) showed that the visible bubble ow
in a bubbling bed combustor was less than one-fth of
u
0
-u
mf
. In other cases, even the trends of the para-
metric behavior with particle size were changed. Heat
transfer to immersed surfaces in ne particle bubbling
beds increases strongly with a decrease in the mean
particle size. For large particle beds, the heat transfer
decreases with decreasing particle diameter in some
instances.
Numerical models have been developed that are
more closely based on fundamental principles. The
models range from simple one-dimensional calcula-
tions to complex multidimensional computational
uid dynamics solutions. There is no doubt that such
numerical models are a great aid in synthesizing test
data and guiding the development of rational correla-
tions. In a recent model evaluation, modelers were
given the geometry and operating parameters for sev-
eral dierent circulating beds and asked to predict the
hydrodynamic characteristics without prior knowledge
of the test results; see Knowlton et al. (1995). None of
the analytical or numerical models could reliably pre-
dict all of the test conditions. Few of the models could
come close to predicting the correct vertical distribu-
tion of solid density in the riser, and none could do it
for all of the test cases. Although it is tempting to think
that these problems can be solved with the next gen-
eration of computers, until there is better understand-
ing and thorough verication of the fundamental
physical models and equations used to describe the
hydrodynamics, the numerical models will not stand
alone as reliable scale-up tools.
Another approach to scale up is the use of simplied
models with key parameters or overall coecients
found by experiments in large beds. For example,
May (1959) used a large-scale cold reactor model dur-
ing the scale-up of the uid hydroforming process. This
technique must be used with care. A large cold model
may not directly simulate the hydrodynamics of a real
process that operates at elevated pressure and tempera-
ture unless care is taken in the selection of ow rates,
particle sizes, and particle densities for the cold tests.
Johnsson et al. (1987) have shown examples of ver-
ication of a model for a 2.13 m diameter industrial
phthalic anhydride reactor. Several bubbling bed mod-
els gave good overall prediction of conversion and
selectivity when proper reaction kinetics were used.
The results were shown to be quite sensitive to the bub-
ble diameter. The comparison was a good check of the
models for the reaction kinetics, but the reactor model
required accurate bubble size estimates obtained from
measurements of overall bed density in the reactor.
As Matsen expresses it, after over a half a century of
scale-up activity in the chemical process industry,
such scale-up is still not an exact science but is rather
a mix of physics, mathematics, witchcraft, history and
common sense which we call engineering; see Matsen
(1997).
Since changes in the bed size primarily inuence
scale-up through changes in the bed hydrodynamics,
one focus of this chapter will be on experimental
results and models that deal explicitly with the inu-
ence of bed diameter on hydrodynamic performance
for both bubbling and circulating uidized beds. The
changes in the bed dynamics will, in turn, impact the
overall chemical conversion or combustion eciency
through changes in the particle-to-gas mass transfer
and the heat transfer from the bed to immersed sur-
faces or the bed wall. Several examples of this inuence
will be presented.
The second focus of this chapter will be on the
design rules for small-scale experimental models that
permit the direct simulation of the hydrodynamics of a
hot, possibly pressurized, pilot plant or commercial
bed. By the use of this modeling technique, beds of
dierent diameters, as well as dierent geometries
and operating conditions, can be simulated in the
laboratory. To date, this technique has been success-
fully applied to uidized bed combustors and gasiers.
Rules for the use of this experimental modeling tech-
nique for uidized bed operations with dierent parti-
cle sizes, as well as for the simulation of bed-to-solid-
surface heat transfer, will also be given. A brief review
of verication experiments and comparisons to large-
scale commercial systems will be shown.
2 REACTOR MODELING: BED DIAMETER
INFLUENCE
Representative results provide a prospective of reactor
modeling techniques that deal with bed size. There are
limited results in the open literature, and in all prob-
ability there are additional unpublished proprietary
materials in this area. Early studies of uidized reac-
tors recognized the inuence of bed diameter on con-
version due to less ecient gassolid contacting. In
bubbling beds that are larger in diameter and deeper,
bubbles can grow to larger size. The larger bubbles
allow more gas to bypass the dense particle-laden
phase. This becomes particularly striking as the bubble
size approaches the bed depth and at large volume
fractions of bubbles within the bed. Smaller bubbles
provide improved gassolid contacting and more uni-
form solids mixing.
Experimental studies have been used to predict reac-
tor performance. Frye et al. (1958) used a substitute
reaction of ozone decomposition to study hydrocarbon
synthesis. The ozone decomposition can be run at low
pressures and temperatures and can be rate controlled
in the same way and by the same catalyst as the reac-
tion under development. Frye and coworkers used
three beds, 2, 8, and 30 inches in diameter to study
the size inuence. We should interject a caution that
the use of pressures and temperatures dierent from
those of the actual reaction may mean that the hydro-
dynamics of the substitute reaction model will dier
from the actual application; this will be illustrated
later in the chapter. Figure 1 shows the apparent reac-
tion rate constant for the dierent bed diameters at
two dierent bed heights, with the other parameters
held constant. Note that the rate constant decreased
by roughly a factor of 3 between the 2 inch and 30 inch
beds.
Figure 1 Apparent reaction-rate constant vs. reactor diameter and bed height. (From Frye et al., 1958.)
May (1959) reports results of tests done in cold
models used to simulate the ow through large reactors
whose performance had been found to be inferior to
that of smaller pilot units. The importance of this pro-
blem can be appreciated from the scale of the equip-
ment used, a 5 foot diameter unit used for the scale-up
tests. This unit was uidized with compressed air at 27
to 38
C (80 to 100
F) and pressures up to 689 kPa (100

psi). Gas residence time in the bed was determined by
the use of tracer gas. Radioactive solid tracers were
introduced into the bed to determine solid mixing.
The data obtained in the larger units are much more
erratic, with evidence of large-scale mixing patterns.
Figure 2 shows the axial mixing coecients obtained
in experiments with dierent size beds. Mixing in the
larger diameter bed is an order of magnitude larger
than that in a small laboratory unit. The measured
hydrodynamic behavior of the gas and solid was com-
bined with a reaction model to predict the reactor
behavior. Here again, there should be concern about
the accuracy with the air experiments done at ambient
temperature. Use of identical bed geometry and bed
solid material does not guarantee identical hydrody-
namics. The shift in gas properties from the cold
model to the hot reactor may cause a marked dier-
ence in behavior. Additional scaling parameters must
be maintained constant between the reactor and the
cold model to insure identical hydrodynamics, and in
some cases just to guarantee identical ow regimes!
Volk et al. (1962) show the eect of bed diameter on
the conversion of CO in the Hydrocol reaction in
which hydrogen and carbon dioxide are converted
over an iron catalyst to hydrocarbons and oxygenated
hydrocarbons in a bubbling or possibly slugging bed.
Figure 3 shows the CO conversion. It is seen that the
conversion rate is reduced as the reactor diameter
increases. This is probably due to larger bubbles in
the larger diameter bed. In general, smaller bubbles
give more uniform solid mixing and a better gassolid
constant. Larger bubbles can allow more gas by passing
through the bed. Volk used vertical tubes within the
reactor to reduce the equivalent diameter of the system,
equal to the hydraulic diameter, dened as four times
the free cross-sectional area divided by the wetted peri-
meters of all surfaces in the cross section. The perfor-
mance correlated with the equivalent diameter. It was
also found that bed expansion was correlated with bed
diameter. The internals serve to suppress the bubble
size, although in some cases bubbles will preferentially
stay near the vertical tubes and aect bed performance.
In their process, larger beds were built with internals
that kept the equivalent diameter the same as that of
smaller units. The large units with internals appeared to
give comparable gas-to-solid contacting. The use of
vertical internals may not be feasible for a number of
reasons, such as tube erosion. The use of the equivalent
diameter approach may not be universally valid.
Van Swaaij and Zuiderwag (1972) used the ozone
decomposition reaction to study the conversion char-
acteristics in a bubbling bed. Studies were made with
beds of 5, 10, 23, 30, and 60 cm diameter, respectively
and up to 300 cm bed heights. The results were com-
Figure 2 Solid diusivity in axial direction for large units. (From May, 1959.)
pared with predictions using a two-phase ow model
with the mass transfer coecient between the bubble
and dense phase derived from residence time distribu-
tion results of gas tracer tests using pulse response. The
results from the ozone conversion and the residence
time distribution interpreted by the two-phase model
gave reasonably similar results. In these cases, the mass
transfer between phases is the limiting resistance for
the reaction. Note that for larger bed diameters the
mass transfer coecient decreases. Van Swaaij and
Zuiderweg (1973) showed that the inclusion of vertical
tubes in a bed gave bubble-to-dense-phase mass trans-
fer results that were roughly equivalent to those of a
smaller open bed with the same hydraulic diameter.
The solids axial mixing was higher than that predicted
using the hydraulic diameter.
Bauer et al. (1981) measured the inuence of bed
diameter on the catalytic decomposition of ozone.
Figure 4 shows the decrease of the conversion with
bed diameter for Bauers data. This gure also
shows the inuence of distributor design on conver-
sion. In many small-scale experiments, a porous
plate is used that will give better performance than
the distributors used in large shallow commercial
bed designs.
Avidan and Edwards (1986) successfully scaled up
from bench scale to demonstration plant from 0.04 m
to 0.6 m diameter while maintaining nearly 100% con-
version for a uid bed methanol-to-gasoline process. In
this case they ran at a high supercial gas velocity; the
bed was in the turbulent ow regime suppressing bub-
bles. By this technique, they eliminated the losses asso-
ciated with gas bypassing in bubbles.
3 INFLUENCE OF BED DIAMETER ON
HYDRODYNAMICS
3.1 Bubbling Beds
In the studies mentioned above, the major objective was
the experimental determination of conversion as a func-
tion of overall design parameters and particle proper-
ties. There have also been studies of the inuence of bed
diameter on the hydrodynamics in an eort to under-
stand the cause of the conversion loss with bed size
increase. These studies have aided in the development
of physical models of reactor performance. De Groot
(1967) measured gas residence time, bed expansion, and
solid axial mixing in a series of beds at dierent dia-
meters uidized with air at ambient conditions. He used
a narrow size range and a broad size range of crushed
silica with sizes below 250 mm. Beds with diameters of
0.1, 0.15, 0.3, 0.6, and 1.5 m were used in the tests.
There was a substantial decrease respectively in bed
expansion and bubble fraction for narrow size range
particles at large bed diameters, indicating the possibi-
lity of gas bypassing in bubble channels, Fig. 5. The
axial diusivity also increased with bed diameter and
was a strong function of particle size distribution.
Werther (1974) measured the bubble characteristics
in cylindrical beds of diameters 100, 200, 450, and 1000
mm for ne particles with a mean diameter of 83 mm.
He showed that for bed diameters commonly used for
laboratory experiments, 200 mm or smaller, the bed
diameter had a strong eect on the bed hydrody-
namics. There was a zone of preferred bubble ow
near the bed walls at lower elevations, Fig. 6. The
Figure 3 CO conversion in Hydrocol reaction for several
reactor diameters. (From Volk et al., 1962.)
Figure 4 Conversion catalytic decomposition of ozone for
dierent bed diameters and distributors. (From Werther,
1992.)
bubbles grew in size and moved toward the center-
line, presumably by coalescence, higher up in the bed.
The transition to slugging occurred higher up in the
larger bed at the same supercial gas velocity. The
bubble velocity increased with height until slug ow
was approached, after which the velocity decreased.
For the only case shown, the 100 mm bed, the max-
imum velocity occurred when the bubble dimension
was about one third of the bed diameter. In larger
beds, the bubble rise velocity was higher for the same
bubble volume, Fig. 7. Hovmand and Davidson
(1971) reviewed data on bubble rise velocity and con-
cluded that when the bubble diameter exceeded one-
third of the bed diameter, the bed diameter rather
than the bubble diameter governed the bubble rise
velocity. Note that Werthers results at dierent
supercial velocities are well correlated by the drift
ux form,
u
b
= u
0
u
mf
( ) c
gd
v
p
(1)
It is curious that the inuence of bed size appeared to
hold even when the bubble was much smaller than the
bed diameter. This may be tied to the local concen-
tration of bubbles within certain sections of the bed.
The increased local bubble ow led to higher coales-
cence rates and higher local bubble velocities. There
was a distinction in bubble velocity between Geldart
group A and B powder while the ratio of visible bub-
ble ow to u-u
mf
seems to be independent of the
Geldart group. Werther also found that the visible
bubble ow, the product of the number of bubbles
per unit time crossing a given surface and their
respective volumes was considerably less than u-u
mf
.
This was especially true in the lower regions of the
smaller bed and throughout the large diameter beds.
The residence time of bubbles was signicantly higher
for small beds, with diameters of 200 mm or less,
than it was for large diameter beds. Werther con-
cludes that the smallest bed diameter that appears
Figure 5 Bed expansion as a function of bed diameter at a
uidization velocity of - 0.20 m,s. (From DeGroot, 1967.)
Figure 6 Bubble gas ow V
b
as a function of the distance r
from the vessel center line in a height of 30 cm above the
distributor in beds of dierent diameters D
B
(u
0
= 9 cm,s,
H = 50 cm. (From Werther, 1974.)
Figure 7 Relationship between local mean bubble size and
local mean bubble rise velocity in beds of dierent diameters
(u
0
= 9 cm/s). (From Werther, 1974.)
suitable for obtaining good scale-up results is 500
mm. It should be pointed out that this criterion
would probably vary with ow conditions, bed
depth, and particle size. Also, Werthers experiments
were carried out with air at ambient conditions.
Although the trends and physical picture may be
similar for beds at elevated temperature and/or pres-
sure, there will probably be some changes in the
numerical limits for this criterion.
Whitehead (1967) found patterns of bubble tracks
in a large 1.2 m square bed similar to the patterns
observed by Werther: preferred bubble tracks near
the walls and corners of a shallow open bed and mer-
ging of bubbles toward the bed center at higher eleva-
tions. Nguyen et al. (1979) also found that a horizontal
tube bank in the large bed caused smaller bubbles that
appeared in more random locations across the upper
surface of the bed. This work was carried out using ne
solids at low supercial velocity, 15 cm/s, and modest
bed depths.
Geldart (1970) showed a substantial distinction
between bubble sizes in two-dimensional and three-
dimensional beds. He used 128 mm river sand in a
30.8 cm round bed and a 68 1.27 cm rectangular
cross section bed. The bubbles in the three-dimensional
round bed were larger. There were dierences in the
visible bubble ow rate at the same supercial velocity.
Geldart ascribes the dierences in bubble diameter to
dierences in visible bubble ow rate as well as to out-
of-line bubble coalescences in the three-dimensional
bed.
Glicksman and McAndrews (1985) determined the
eect of bed width on the hydrodynamics of large par-
ticle bubbling beds. Sand particles with a mean dia-
meter of 1 mm were uidized by air at ambient
conditions. A rectangular-cross-section bubbling bed
was used. The bed width ranged from 7.6 to 122 cm,
while the other cross-sectional dimension was held
constant at 122 cm. Most experiments were carried
out with an open bed. The bubble rise velocity
increased with the bed width; when the bubble velocity
was represented as the drift ux plus the rise velocity of
an isolated bubble,
u
b
= u
0
u
mf

gd
B
p
(2)
varied from 0.4 in a two-dimensional bed to 0.6 in
the three-dimensional bed. The mean vertical chord
length of bubbles decreased with bed width; see Fig.
8. The visible bubble ow decreased dramatically with
an increase of bed width at a xed supercial velocity;
see Fig. 9. Representing the total gas ow as the sum of
the visible bubble displacement, the ow through the
dense bed at minimum uidizing conditions and the
gas throughow in the bubbles can be written as,
u
0
= Q
b
(1 o)u
mf
mou
mf
(3)
The gas throughow coecient m increased from a
mean value of 3.6 for the two-dimensional bed to a
mean value of 11.7 for the largest, 1.22 1.22 m bed
cross section. The bed depths at minimum uidization
were 46 and 76 cm in the tests. For these rather shallow
beds there were no observable preferred bubble tracks,
and the location of erupting bubbles was random
across the bed surface. For the cases observed, with
u
0
/u
mf
varying from 1.3 to 1.8, the inuence of the
wall was absent when the bubble diameters were
roughly one-fth of the bed width or less. For deeper
beds or higher gas velocity, the ratio of bubble dia-
meter to bed width is expected to be the best criterion
for determining when wall eects will be negligible.
For one test series, ve staggered rows of horizontal
tubes with a horizontal center-to-center spacing of 15.2
cm were placed in the 1.22 m square bed. The tube bed
results are shown in Figs. 8 and 9 with a T symbol. The
behavior of bubbles in the bed with the tube bank
resembled that of an open bed with a smaller width.
Dierent tube arrangements and spacing yield dierent
eective bed widths. Glicksman and Yule (1991)
showed that the bed expansion had dierent rates of
change with supercial velocity when dierent hori-
zontal tube arrangements were tested in the large
cold bed. The dierences are due to varying maximum
bubble sizes and the possibility of vertical bubble align-
ments to augment gas bypass through the dense bed
substantially.
3.2 Mixing
Van Deemter (1980) surveyed data on solid mixing
in uidized beds of dierent diameters. Many of the
experiments in large beds were considered inconclu-
sive. The compiled data for longitudinal uid disper-
sion, M
L
, gas back mixing, M
B
, and longitudinal
dispersion of solids, M
S
are shown in Fig. 10.
Note the strong inuence of bed diameter on mix-
ing. The scatter in the data was attributed to dier-
ences in gas velocity and range of particle sizes. It
may also be set up by dierent measurement tech-
niques. Large dierences between small and large
bed diameters may be due to ow regime transi-
tions.
The dierences in behavior between small labora-
tory beds and larger demonstration units can in part
be attributed to a switch from porous plate distributors
in the small bed to discrete hole or bubble caps in the
larger beds. The porous plates give a better quality of
uidization, e.g., smaller bubbles, for shallow beds and
beds of moderate depth; see Rowe and Stapleton
(1961).
Yerushalmi and Avidan (1985) suggest that the
axial dispersion coecient of solids in slugging and
turbulent ow varies approximately linearly with the
bed diameter, as with Thiel and Potter (1978).
3.3 Inuence of Bed Diameter on Circulating or Fast
Fluidized Beds
Arena et al. (1988) measured the hydrodynamic beha-
vior of two circulating uidized beds with riser dia-
meters of 0.041 and 0.12 m ID of roughly the same
height. At the same supercial gas velocity and solid
recirculation rate, the larger diameter column had a
higher solids fraction. The average slip velocities
derived from these data are also higher for the large
diameter riser; see Hartge et al. (1985). Yerushalmi
and Avidan (1985) found a similar trend when com-
paring a 15.2 and 7.6 cm column. Noymer et al.
(1995) also compared two columns of 5.08 and 7.68
cm diameter of the same height, which were used to
simulate larger pressurized uidized bed combustors.
They found higher solids loading for the larger dia-
meter riser at equal gas velocity and solid recircula-
tion. In addition, the fraction of the wall covered by
clusters was higher for the larger diameter column
when the two beds had equal solids ow and when
the two beds had equal cross section averaged solids
concentration. The increased wall coverage should
lead to higher heat transfer rates from the bed to
the wall.
Figure 8 Variation of main chord length with wall spacing. I
m
= 76 cm, probe height = 46 cm, X, 1.3 U
mf
; *; 1.5U
mf
; V;
1.8 U
mf
; T denotes with tubes. (From Glicksman and McAndrews, 1985.)
Rhodes et al. (1992) compared the solids ux pro-
les across the cross sections of a 0.152 and a 0.305 m
diameter circulating bed riser. They found a region
where the solid prole, given by the ratio of local
ux to average ux, had a similar variation over the
cross section and was insensitive to the level of solid
ux. The variation of the local solids ux over the
radius was a function of the gas velocity and the riser
diameter. In the larger riser, the proles were some-
what atter and the thickness of the downowing
region relative to the bed radius was smaller. The com-
parisons were not exact since the cross sections com-
pared for the two beds were at dierent heights.
Zhang et al. (1991) carried out investigations with
three dierent fast bed systems with diameters of 32,
90, and 300 mm. They found that the radial voidage
distribution, given as a ratio to the cross-sectional
average, was independent of bed diameter and solids
recycle rate. The similarity does not hold at transition
to the turbulent regime. The results are for the center
of the riser excluding the entrance and exit regions. It
would be interesting to determine if the similar voidage
proles hold for larger diameter risers.
The thickness of the downowing layers at the wall
of the CFB is typically dened as the distance from the
wall to the position of zero average vertical solid ux.
Zhang et al. (1995) made measurements of the layer
thickness on a 12 MWth and a 165 MW CFB boiler.
They found that the thickness increased for the larger
bed. They related data from many dierent beds, Fig.
11, with the equivalent bed diameter, taken as the
hydraulic diameter, using the form
o = 0.05De
0.74
(4)
The thickness, o, was found to be insensitive to particle
concentration, gas velocity, and height within the fur-
nace. This suggests that the thickness results from a
balance of solids internal circulation that is generally
Figure 9 Visible bubble ow Q
b
, I
mf
= 76 cm, probe height = 46 cm. X, 1.3U
mf
; *; 1.5U
mf
; V; 1.8U
mf
; T denotes with tubes.
Experiment uncertainty = 20%. (From Glicksman and McAndrews, 1985.)
much higher than net throughow. If the local solids
ux prole, as a ratio of the cross-sectional average, is
roughly invariant over the cross section for these larger
beds, the thickness of the wall layer follows from a
mass balance. The upow is proportional to the core
area, D
2
,4, and the downow to the product of peri-
meter and layer thickness, proportional to oD. Thus
the thickness, o, should vary as the ratio of cross-sec-
tional area to perimeter, i.e., proportional to the
hydrodynamic diameter.
Patience et al. (1992) developed a dimensionless cor-
relation for the mean slip factor between gas and solid
by using solid suspension data from various small
laboratory beds. The proposed correlation relates the
slip to the Froude number based on the bed diameter.
It remains to be seen if the correlation will hold at
Froude numbers typical of large beds and if other
dimensionless factors are important for large beds.
Johnsson and Leckner (1995) observed the ow pat-
terns in a large laboratory circulating bed, 1.7 1.4 m
in cross section. The lower zone was characterized by a
dense bottom similar to a bubbling bed with exploding
bubbles. The bed porosity was 0.7 or less. These obser-
vations are in contrast to those made by other investi-
gators in much smaller circulating beds.
3.4 Flow Transition
Hovmand and Davidson (1971) review Stewarts cri-
terion for the transition from bubbling to slug ow,
u u
mf
0.35(gD)
1,2
= 0.2 (5)
and show it gives good agreement with most experi-
ments. When u u
mf
is larger than the value found
from Eq. (5), the bed will be in slug bed. Thus a small
Figure 10 Mixing coecients for dierent vessel diameters;
M
L
, longitudinal dispersion of uid; M
B
, gas backmixing;
M
S
: longitudinal solid dispersion. (From Van Deemter,
1980.)
Figure 11 Empirical correlation and experimental data of thickness of downowing layer at the wall of a CFB as a function of
the equivalent bed diameter. (From Zhang et al., 1995.)
diameter laboratory bed may exhibit behavior that is
far dierent from that of a scaled-up pilot plant. The
small bed may be well into slug ow whereas the larger
bed may be in bubbling bed. There is some data that
suggests the ratio of bed depth to bed diameter may
also inuence the bubble-to-slug transition.
Thiel and Potter (1977) carried out slugging ow
experiments in beds of three dierent diameters, the
largest being 0.22 m in diameter. They found that the
bed aspect ratio inuenced the onset of slug formation.
They also found that the transition to turbulent ow
occurred at much lower supercial gas velocity in the
0.22 m diameter bed than in the 0.1 m bed. With uid
cracking catalyst, the transition velocity decreased
from 20 cm/s to 2.5 cm/s. Yerushalmi and Avidan
(1985) assert that in large shallow beds where slugging
does not occur, transition from bubbling to turbulent
ow should be independent of bed size.
4 EXPERIMENTAL MEANS TO ACCOUNT
FOR SCALE-UP: USE OF SCALE MODELS
Since large pilot plants are costly, it may not be feasible
to undertake an extensive scale-up program.
Furthermore, it is dicult to observe or measure
detailed behavior within large hot beds. Thus many
test results from beds at elevated temperatures and
pressures are conned to overall operating characteris-
tics, leaving the designers to speculate on the cause of
shifts in performance with bed size. Full-scale cold
models are also expensive and require lengthy construc-
tion to modify the bed geometry. As we will see below,
the full-scale cold test may not correctly simulate the
hydrodynamics of the actual process at elevated tem-
perature and pressure. Indeed, a familiar occurrence in
atmospheric bubbling uidized bed combustors is the
marked dierence in ow behavior between a bed ui-
dized with cold air and the same bed, using the same
particles, run at normal operating conditions.
A technique, that can assist in the scale-up of com-
mercial plant designs is the use of scale models. A scale
model is an experimental model that is smaller than the
hot commercial bed but that has identical hydrody-
namic behavior. Usually the scale model is uidized
with air at ambient conditions and requires particles
of a dierent size and density than those used in the
commercial bed. The scale model relies on the theory
of similitude, sometimes through the use of
Buckinghams pi theorem, to design a model that
gives identical hydrodynamic behavior to the commer-
cial bed. Such a method is used in the wind tunnel
testing of small model aircraft or in the towing tank
studies of naval vessels.
Once a technique has been established to design a
small cold model that simulates the hydrodynamics of
a hot (possibly pressurized) uidized bed, then a series
of dierently sized models can be used to determine the
inuence of bed size on the performance of commercial
beds; see Fig. 12. Model A
/
simulates the behavior of
commercial bed A, model B
/
simulates a larger com-
mercial bed B, and so forth. Then by comparing mod-
els A
/
, B
/
, with C
/
we can determine the expected
changes in operating characteristics when commercial
bed A is replaced by larger beds B and C.
Designing a model uidized bed that simulates the
hydrodynamics of a commercial bed requires account-
ing for all of the important mechanical forces in the
system. In some instances convective heat transfer can
also be scaled, but at present, proper scaling relation-
ships for chemical reactions or hydromechanical
eects, such as particle attrition or the rate of tube
erosion, have not been established.
4.1 Development of Scaling Parameters
There are several approaches to developing the correct
scaling relationships. Probably the most straightfor-
ward is the nondimensionalization of the governing
Figure 12 Use of scale models with dierent bed diameters
to simulate the inuence of diameter on the hydrodynamics
of a hot commercial reactor.
equations. If we can write the proper equations gov-
erning uid and particle dynamic behavior, we can
develop the proper scaling relationships even if we can-
not solve the equations (at present we cannot). In
essence, if a model is designed that follows the exact
same equations and boundary conditions as the com-
mercial bed, with the same values of the coecients,
then the model and the commercial bed should have
identical behavior. The trick is to nondimensionalize
all of the terms of the equations so that the model and
the commercial beds have identical nondimensiona-
lized values of the parameters. The complete develop-
ment of this is given in Glicksman et al. (1994) and
Glicksman (1997).
An alternate approach is to design a scale model so
that the ratio of all of the important forces is the same
in the model as it is in the full-scale bed, and a the scale
model uses geometry similar to that of the larger ui-
dized bed. If the ratio of forces acting on a particle in
the scale model bed have the same ratio as they do for
a particle in the full bed, then the trajectory of the
particle motion should be the same for both beds.
The same follows for the motion of the gas and the
motion of clusters of particles or particles in a dense
phase. From this point of view, we must identify all the
important forces in the system. Table 1 lists these
forces.
To characterize the gas ow in the uidized bed,
important forces are viscous and inertia as well as par-
ticle-to-gas forces. For the particles, the important
forces include gravity, particle inertia, gas interaction
with the particles such as drag, collisional forces
between particles and between particles and wall, and
particle surface forces such as electrostatic and adhe-
sion forces. For larger particles, in Geldart groups B or
D, the particle surface forces can be neglected. The
question is complicated for smaller particles because
the surface forces are dicult to quantify. We must
also include the distributions of particle sizes, since
they can inuence the relative magnitude of, say, grav-
ity to drag forces.
For this work we will concentrate on particles of
sucient size and mass so to minimize interparticle
surface forces relative to other forces. Limited results
from Litka and Glicksman (1985), Chang and Louge
(1992), and Glicksman et al. (1991a) suggest that,
within the normal range of particle and wall coecient
of restitution and particle-to-particle sliding friction,
these parameters have a modest inuence on the bed
dynamics. The particle-to-particle and particle-to-wall
collisions may still play an important role in the bed
dynamics. This only assumes that the changes in the
coecients of restitution and friction are small enough
to justify their omission in the list of important dimen-
sionless parameters that can be controlled or modied.
To construct a scale model of a commercial bed,
rst the geometry must be similar. The linear dimen-
sions of all components in the commercial bed, bed
diameter, height, internal tube diameter and spacing,
etc., must be reduced by the same factor in the model.
The location coordinates of a particle within the bed
are also reduced in the same fashion.
When all of the force ratios given in Table 1 are set
equal between a scale model and a commercial bed,
then the scale model should show identical dynamic
behavior. This behavior refers to the ow regime, par-
ticle and gas velocity (nondimensionalized as a ratio to
the supercial velocity), and particle location (the
coordinates nondimensionalized as a ratio to the bed
diameter or bed height). Thus at a given location
X,D
m
, y,D
m
in the model, where D
m
is the model
diameter, the velocity u,u
0m
is the same as u,u
0c
at
XD
c
, y,D
c
in the commercial bed.
The requirement of equal force ratios for all of the
terms given in Table 1 is referred to as the full set of
scaling relations. They can be written as
Table 1 Important ratio of forces acting in a uidized bed.
,
s
U
0
d
p
,m Particle inertia/gas viscous force
,
f
U
0
L,m Gas inertia/Gas viscous force
U
2
0
,gL Inertia/gravity force
,
s
,,
f
Solid inertia/Gas inertia force
Surface ForcesCollision, Adhesion; Form is Debatable
Additional Important Ratios
G
s
,,
s
U
0
Solid recycle volumetric ow/Gas volumetric ow rate
L,D Bed height/Bed diameter
PSD Particle size distribution,
1
(mass fraction of d
i
) for d
i
,d
p
(mean solid diameter) for all d
i
s
Particle sphericity
u
2
0
gL
.
,
s
u
0
d
p
m
.
,
f
u
0
L
m
.
,
s
,
f
.
G
s
,
s
u
0
.
L
D
.
bed geometry, . PSD
(6)
Where is the particle sphericity and PSD is the
dimensionless particle size distribution.
Alternatively, the terms can be rewritten by multi-
plying terms together, although the number of inde-
pendent dimensionless parameters stays the same.
Thus, Eq. (6) can be rewritten, for example, as
,
f
,
s
d
3
p
g
m
2
.
,
s
,
f
.
u
2
0
gL
.
,
f
u
0
L
m
.
G
s
,
s
u
0
.
L
D
.
bed geometry, , PSD
(7)
He et al. (1995) have shown that for spouting beds,
cohesive factors are important, and this group must
be augmented by including the internal friction angle
and the loose packed voidage to achieve similar scale
models. Since interparticle friction can occur in slug-
ging beds, these additional parameters should be
included to scale slugging beds properly.
5 SIMPLIFIED SCALING RELATIONSHIPS
For scaling to hold with the full set of scaling relation-
ships, all of the dimensionless parameters given in Eqs.
(6) or (7) must be identical in the scale model and the
commercial bed under study. If the small scale model is
uidized with air at ambient conditions, then the uid
density and viscosity are xed and there is only one
unique modeling condition that will allow complete
similarity. In some cases, this requires a model that is
too large and unwieldy or will not permit simulation of
a very large bed.
In most situations, one would expect that not all of
the parameters are of rst order importance. By redu-
cing the number of parameters that must be main-
tained in the model it will be possible to model larger
commercial beds with small scale models. This will
involve simplications of the interparticle drag at the
extreme of small and large Reynolds numbers based on
particle diameter. If the same simplication can be
shown to hold in both of these limits, it is reasonable
to consider application of the simplication over the
entire range of conditions.
By representing the particle to air drag force by the
Erqun equation or single particle drag, it can be
shown, Glicksman et al. (1994), that in the limit of
small particle Reynolds numbers the governing para-
meters can be reduced to
u
2
0
g
2
.
,
s
,
f
.
u
0
u
mf
.
L
1
L
2
.
G
s
,
s
u
0
. bed geometry, . PSD (8)
At the other extreme, at large particle Reynolds num-
bers it can be shown that the same set of governing
parameters, Eq. (8), applies.
Since the same simplied set of dimensionless para-
meters holds exactly at both high and low Reynolds
numbers, it is reasonable to expect that they hold, at
least approximately, over the entire range of conditions
for which the drag coecient can be determined by the
Erqun equation or an equation of similar form.
For the more general case, Fig. 13 shows the
value of [ the dimensionless drag coecient relative
to [ at low Re over a range of conditions when u
0
/
u
mf
is 10 and 3, respectively, and Fr and
s
remain
constant. When u
0
/u
mf
and the slip velocity are high
there is a larger variation of dimensionless drag
coecient with Reynolds number. Note that [
does not vary with particle Reynolds number
when the Reynolds number remains above about
10
3
or below about 10. When u
0
/u
mf
is 1000, a
condition is approached with very ne particle bub-
bling beds or circulating beds. The use of the Erqun
relationship is questionable except for the dense
lower part of the bed.
5.1 Range of Validity of Simplied Scaling
To determine the validity of the simplied scaling
laws over a wide range of conditions, the simplied
scaling laws have been used, Eq. (8), to design
hypothetical models whose linear dimensions are
one-fourth and one-sixteenth, of the linear dimensions
of a model designed using the full set of scaling laws,
Eq. (6). To determine the validity of the smaller, sim-
plied models, the dimensionless drag coecient [L,
,
s
u
0
will be compared between the simplied models
and the model using the full set of scaling laws.
Figure 14 shows a comparison of the exact model
and the simplied models for a pressurized uidized
combustor. Using the full set of scaling laws, the
exact model, uidized by ambient air, is approxi-
mately the same size as the combustor. The simplied
models are reduced in size by their respective assumed
length scale. The other parameters of the simplied
model are then calculated to match the simplied
parameters. For example, when the length scale is
reduced to one-fourth that of the exact model, the
velocity is reduced by one-half to keep the Froude
number constant. The particle diameter is then
reduced appropriately to keep the ratio of u
0
/u
mf
con-
stant. These calculations were carried out over a
range of particle Reynolds numbers, Re
pE
, based on
the full scaling law, or exact, model. In the simplied
scaling relationships, the Reynolds number is not
maintained constant. The concern is how much the
drag coecient is impacted by the shift in Reynolds
number. It was found that the particle Reynolds num-
ber for the one-fourth scale simplied model
remained roughly equal to 0.34 Re
pE
over a wide
range of values for Re
pE
, whereas the particle
Reynolds number for the one-sixteenth scale model
was roughly 0.12 Re
pE
.
Using these Reynolds number scale factors, the
error in the dimensionless drag coecient [L,,
s
u
0
using the simplied scaling models is shown on Fig.
15 for u
0
/u
mf
of 10, plotted as a function of Re
pE
based on parameters for the exact scaled bed. For a
particle Reynolds number of 1000 or less, which corre-
sponds to pressurized beds with particles of 1 mm or
less, the error in the drag coecient with the simplied
scaling laws is 20% or less for a one-quarter length
scale model. The error is 40% or less for a one-
sixteenth length scale model. At u
0
/u
mf
of 1000 and
u
slip,
u
mf
of one-ftieth the errors for the one-sixteenth
scale model are 20% or less for Re
pE
less than 10
3
. For
particles of 0.2 mm or less, corresponding to a
Reynolds number of 100 or less, the errors in drag
coecient are minimal. When the Ergun equation
applies for the drag coecient, a one-quarter scale
model based on the simplied scaling laws should be
valid for any conditions. A one-sixteenth scale model
should be valid for particle diameters of about 0.2 mm
or less for a pressurized bubbling bed with u
0
/u
mf
of 10
and u
slip
,U
0
of 0.3. At u
0
/u
mf
of 1000 and u
slip
,U
0
of
one-ftieth, the one-sixteenth scale model should be
valid for pressurized beds with particles up to 1 mm
in diameter. These conclusions apply when the parti-
cle-to-uid drag term is given by the Ergun equation or
similar relationships and the scaled particles are not so
small that interparticle surface forces come into play.
5.2 Clusters
In the freeboard of a bubbling bed or in the upper
portion of a circulating bed where particles generally
are considered to act in clusters or groups, a similar
examination of scaling of the gas-to-solid drag can
be made. Consider all the particles grouped into
clusters with an eective diameter d
c
and the clus-
ters occupying a volume fraction
c
of the bed
volume. The cluster-to-gas drag will be represented
by the drag coecient for a solid sphere of diameter
d
c
, C
D
.
If the reduced scale models faithfully reproduce the
dynamics of the exact case, the cluster dimensions
should scale directly with the linear dimensions of the
bed. Thus, a one-quarter linear scale model that has a
velocity one-half that of the exact case will have a
cluster Reynolds number (Re
dc
) one-eighth that of
the exact bed. From the relationship of C
D
with Re
the change of C
D
with model scale at a given Reynolds
number of the exact bed can be determined. Figure 16
Figure 13 Ratio of drag coecient to low Reynolds number drag coecient, u
0
/u
mf
= 10 and 3. (From Glicksman et al., 1993b.)
shows the shift in C
D
using the C
D
relationship of
White (1974) for length scales of one-fourth, one-
eighth, and one-sixteenth of the exact bed length as a
function of the cluster Reynolds number of the exact
bed. Also shown in the gure is the typical Reynolds
number of an atmospheric combustor with a 0.3 m
cluster diameter, approximately 1.5 10
4
. In a bub-
bling bed, the cluster diameter in the freeboard should
be at least equal in size to the diameter of bubbles
erupting at the bed surface. For beds with horizontal
tubes, the bubble diameter will be equal to or larger
than the horizontal tube spacing. In a bubbling bed
without tubes, the bubbles and clusters can be much
larger. In an open circulating bed, the cluster diameter
is more dicult to determine. It is reasonable to
assume that its diameter is proportional to the bed
diameter, equal in magnitude to the bed diameter or
one order of magnitude smaller. From these considera-
tions, the Reynolds number based on the cluster dia-
meter should be 10
4
or larger in an atmospheric
combustor with a cluster diameter of 0.2 m. The cluster
Reynolds number should be 10
5
or larger in a pressur-
ized combustor. From Fig. 16 it can be seen that a one-
quarter scale or an eighth scale model should have
drag coecients similar to those of the exact bed.
For pressurized beds, the drag coecients should be
very close in magnitude.
Considering the drag acting on a single particle in a
dispersed ow, the validity of the simplied relation-
ships can be examined by comparing the terminal velo-
city predicted by the simplied relationship and the
exact value.
Figure 14 Exact and simplied models of a pressurized uidized bed combustor. (From Glicksman et al., 1993b.)
The errors in u
t
/u
mf
is shown in Fig. 17 for simpli-
ed scale models at two dierent linear dimensions.
Scaling a combustor with comparatively small parti-
cles, 0.2 mm or less, gives good agreement for u
t
/u
mf
even at one-sixteenth linear scale, while for large par-
ticles a linear scale of one-fourth gives fair agreement
for u
t
/u
mf
. Since u
0
/u
mf
is held constant in the simpli-
ed scaling laws, close agreement of u
t
/u
mf
also results
in close agreement of u
t
/u
0
.
6 FURTHER SIMPLIFICATIONS IN THE
SCALING RELATIONSHIP
6.1 Viscous Limit
Glicksman (1984) showed that the list of controlling
dimensionless parameters could be reduced if the uid
particle drag is primarily viscous or primarily inertial.
For very low particle Reynolds numbers the scaling
relationships can be further simplied to
,
s
u
0
d
2
p
mD
.
s
.
gD
u
2
0
.
D
L
.
G
s
,
s
u
0
. bed geometry, PSD (9)
or alternatively, this can be written as
u
mf
u
0
.
s
.
gD
u
2
0
.
D
L
.
G
s
,
s
u
0
. bed geometry, PSD (10)
Note that this is a subset of the simplied scaling laws
presented above with the solid-to-gas density ratio
removed.
Recent results indicate that in most scaling interac-
tions, elimination of the ratio of solid-to-gas density
ratio from the set of scaling parameters leads to dis-
crepancies between the scale models and the larger
beds being simulated; see Glicksman et al. (1998).
7 DESIGN OF SCALE MODELS
7.1 Full Set of Scaling Relationships
We will rst consider the steps to design a model that is
similar to another bed based on the full set of scaling
parameters; see Eqs. (6) or (7).
To construct a model that will give behavior similar
to another bed, for example, a commercial bed, all the
dimensionless parameters listed in Eqs. (6) or (7) must
have the same value for the two beds. The require-
ments of similar bed geometry is met by the use of
geometrically similar beds; the ratio of all linear bed
dimensions to a reference dimension such as the bed
diameter must be the same for the model and the com-
mercial bed. This includes the dimensions of the bed
internals. The dimensions of elements external to the
bed, such as the particle return loop, do not have to be
matched as long as the return loop is designed to pro-
vide the proper external solids ow rate and size dis-
Figure 15 Error in drag coecient for simplied scaling relationships, u
0
/u
mf
= 10. (From Glicksman et al., 1993b.)
tribution, and solid or gas ow uctuations in the
return loop do not inuence the riser behavior; see
Rhodes and Laussman (1992).
Proper conditions must be chosen to design a scale
model to match the dimensionless parameters of the
commercial bed. To model a gas uidized commercial
bed, a scale model using air at standard conditions is
most convenient, although several investigators have
used other gases; see Fitzgerald and Crane (1980),
Fitzgerald et al. (1984), Chang and Louge (1992) or
pressurized scale models, Almstedt and Zakkay
(1990), Di Felice et al. (1992 a,b). The gas chosen for
the model, along with the gas pressure and tempera-
ture, determines the values of ,
f
and m. The particle
density for the model is chosen to match the density
ratio, so that
,
f
,
s

m
=
,
f
,
p

c
(11)
where the subscript m is for the model and c is for the
commercial bed. For the remaining parameters, the
form of Eq. (6) will be chosen for the dimensionless
parameters. Combining the Reynolds number based
Figure 17 Error in terminal velocity using simplied scaling at 1013 kPa and 800
C. (From Glicksman et al., 1994.)

Figure 16 Error in cluster drag coecient for xed u
0
/u
mf
using C
d
for a solid sphere. (From Glicksman et al., 1993b.)
on bed diameter and the square root of the Froude
number, and rearranging, we obtain
,
f
u
0
D
m
f
gD
_
u
0
=
D
3,2

g
_
v
f
!
m
=
D
3,2

g
_
v
f
!
c
(12)
D
m
D
c

=
(v
f
)
m
(v
f
)
c

2,3
(13)
All of the linear dimensions of the model are scaled to
the corresponding dimensions of the commercial bed
by the ratio of the kinematic viscosities of the gas
raised to the two-thirds power. By taking the ratio of
the Reynolds number based on the particle diameter to
the Reynolds number based on the bed diameter,
,
f
u
0
D
m
f
m
f
,
f
u
0
d
p
=
D
d
p

m
=
D
d
p

c
(14)
The particle diameters in the model scale by the same
factor as the bed diameter, by the ratio of the kine-
matic viscosities to the two-thirds power.
Equating the Froude number and rearranging, we
nd
u
0m
u
0c
=
v
fm
v
fc

1,3
=
D
m
D
c

1,2
(15)
Thus the velocity scales are the square root of the
linear dimension scale.
By satisfying both Eq. (13) and Eq. (15), the
Reynolds number and the Froude numbers are kept
identical between the model and the commercial bed.
Combining G
s
,,
s
U
0
and the product of the
Reynolds and Froude numbers along with Eq. (13) it
can be shown that
G
s
,, ( )
m
G
s
,,
s
( )
c
=
v
f
m
v
f
c

1,3
(16)
so that the ratio of solids ow to solids density scales as
the ratio of the cube root of the kinematic viscosity.
Once the model uid and its pressure and tempera-
ture are chosen, which sets the gas density and viscos-
ity, there is only one unique set of parameters for the
model that gives similarity when using the full set of
dimensionless parameters. The dependent variables, as
nondimensionalized as X,D, y,D, u,u
0
are the same in
the respective dimensionless time and spatial coordi-
nates of the model as the commercial bed. The spatial
variables are nondimensionalized by the bed diameter,
so that the dimensional and spatial coordinates of the
model are proportional to the two-thirds power of the
kinematic viscosity, as given by Eq. (13),
x
m
x
c
=
v
m
v
c

2,3
(17)
Since the velocity scales with v
1,3
, the ratio of time
scales can be expressed as
t
m
t
c
=
v
fm
v
fc

1,3
(18)
Similarly, it can be shown that the frequency scales as
f
M
f
c
=
v
fc
v
fm

1,3
(19)
Table 2 gives the values of design and operating para-
meters of a scale model uidized with air at ambient
conditions which simulates the dynamics of an atmo-
spheric uidized bed combustor operating at 850
C.
Fortunately, the linear dimensions of the model are
much smaller, roughly one-quarter those of the com-
bustor. The particle density in the model must be much
higher than the particle density in the combustor to
maintain a constant value of the gas-to-solid density
ratio. Note that the supercial velocity of the model
diers from that of the combustor along with the spa-
tial and temporal variables.
When modeling a pressurized hot bed (Table 3), the
ambient temperature model uidized with air has
dimensions very close to those of the pressurized com-
bustor. If another gas is used in the model, particularly
a gas with a higher density, the model can be made
much smaller than the pressurized combustor (Table
4). Care must be taken to select a safe modeling gas
and one that yields a solid density for the model that is
available.
7.2 Design of Scale Models Using the Simplied Set
of Scaling Relationships
The simplied scaling relationships, Eq. (8) oer some
exibility in the model design, since fewer parameters
must be matched than with the full set of scaling rela-
tionships. When the uidizing gas, and the pressure
and temperature of the scale model, are chosen, the
gas density and viscosity for the scale model is set.
The model must still be geometrically similar to the
commercial bed. There is still one free parameter.
Generally, this will be the linear scale of the model.
For the simplied scaling relationships, the gas-to-
solid density ratio must be maintained constant,
,
f
,
s

m
=
,
f
,
p

c
(20)
With ,
f
of the model set by the uidized gas and its
state, the solid density in the model follows from Eq.
(20). Choosing the length coordinate of the model, D
m
,
the new free parameter, the supercial velocity in the
model is determined so that the Froude number
remains the same,
u
2
0
D
!
m
=
u
2
0
gD
!
c
(21)
so that
u
cm
u
0c
=
D
m
D
c

1,2
(22)
Note that in the simplied case the velocity scaling is
not uniquely tied to just the gas properties as it is in the
full scaling relationship. With U
0
and ,
s
set the solids
recycle rate can be determined by
G
s
,
s
u
o

m
=
G
s
,
s
u
o

c
(23)
G
m
G
sc
=
,
sm
,
sc

u
0m
u
0c

=
,
fm
,
fc

D
m
D
c

(24)
Finally, the mean particle size for the model as well as
the sphericity and particle size distribution must be
determined. The particle size is determined by the
need for equal values of u
0
/u
mf
between the model
and the commercial bed:
u
0
u
mf

=
u
0
u
mf

(25)
Table 2 Atmospheric combustor modeled by a bed uidized
with air at ambient conditions
Given Commercial bed
Scale model,
full scaling laws
Temperature (
C) 850 25
Gas viscosity (10
5
kg,ms) 4.45 1.81
Density (kg,m
3
) 0.314 1.20
Derived from scaling laws
Solid density ,
sc
3.82,
sc
Bed diameter, length, etc. D
c
0.225D
c
Particle diameter d
pc
0.225d
pc
Supercial velocity u
0c
0.47u
0c
Volumetric solid ux (G
s
,,
s
)
c
0.47(G
s
,,
s
)
c
Time t
c
0.47t
c
Frequency f
c
2.13f
c
Table 3 Pressurized combustor modeled by a bed uidized with
air at ambient conditions
Scale model,
full scaling laws
Temperature (
C) 850 25
Gas viscosity (10
5
kg,ms) 4.45 1.81
Density (kg,m
3
) 3.14 1.20
Pressure (bar) 10 1
Solid density ,
sc
0.382,
sc
c
1.05D
c
Particle diameter d
pc
1.05d
pc
0c
1.01u
0c
s
,,
s
)
c
1.01(G
s
,,
s
)
c
Time t
c
1.01t
c
Frequency f
c
0.98f
c
(u
mf
)
m
= (u
mf
)
c
u
cm
u
0c

= (u
mf
)
c
D
m
D
c

1,2
(26)
In general, u
mf
is a function of the particle diameter
and gas properties, as well as and e
mf
. Once the
uidizing gas and the length of scale of the model are
chosen, the proper particle diameter is that which gives
the value of u
mf
needed in Eq. (26).
If both the model and commercial bed are in the
region where the respective Reynolds number based
on particle diameter and gas density is very low, then
a single algebraic relationship can be developed. In
that region,
u
mf

,
s
d
2
p
m
(27)
d
pm
d
pc
=
,
fc
m
m
,
fc
m
m

1,2
=
D
m
D
c

1,4
(28)
when both (Re
dp
)
m
and (Re
dp
)
c
- 20.
When the Reynolds number of the model and commer-
cial bed are both very large,
U
2
mf
d
p
,
s
,
f
(29)
Combining Eq. (29) with Eqs. (22) and (25), we nd
u
2
mf
m
u
2
mf
c
=
d
pm
d
pc
(,
s
,,
f
)
m
(,
s
,,
f
)
c
=
u
2
0m
u
2
0c
=
D
m
D
c
(30)
Since the gas-to-solid density ratio of the model and
the commercial beds must be the same to satisfy the
simplied scaling relationships Eq. (30) becomes
d
pm
d
pc
=
D
m
D
c

1,4
(31)
when both (Re
dp
)
m
and (Re
dp
)
c
> 1000.
Tables 5 and 6 show the values of the mean particle
diameter for models of an atmospheric and pressurized
commercial bed, for dierent selected linear scale
ratios between the model and the commercial bed.
By the use of the simplied scaling parameters, the
linear scale factor can be changed as shown in Tables 5
and 6. Note that as the linear scale factors are changed
the particle diameter changes much more slowly. The
scale model of the 10 atmosphere bed has a mean par-
ticle diameter that is quite close to the mean particle
diameter of the commercial bed. The model particles
have a substantially lower density in this case.
It is not clear where cohesive forces will become
important. The use of very dense particles (for the
models of the one atmosphere bed) will cause a shift
of the boundary of cohesive inuence as given, for
example, by Geldarts classication. However, ade-
quate experimental data are still lacking with such
dense ne particles to set the limits of cohesive inu-
ence.
Note that for completeness the nondimensional par-
ticle size distribution, the sphericity, and the internal
angle of friction (for slugging and spouting beds)
should also be matched between the two beds.
7.3 Hydrodynamic Scaling of Bubbling Beds
Experiments using scaled models of bubbling beds
have been carried out since 1980 using the scaling rela-
tionships presented in previous sections. The earlier
Table 4 Pressurized combustor modeled by a bed uidized with
refrigerant vapor 134a at ambient conditions
Scale model,
full scaling laws
Temperature (
C) 850 20
Gas viscosity (10
5
kg,ms) 4.45 1.19
Density (kg,m
3
) 3.14 4.34
Pressure (bar) 10 1
Solid density ,
sc
1.38,
sc
c
.334D
c
Particle diameter d
pc
.334d
pc
0c
.58u
0c
s
,,
s
)
c
.58(G
s
,,
s
)
c
Time t
c
.58t
c
Frequency f
c
1.7f
c
work was concerned with experimental testing and ver-
ication of the scaling relationships. Hot bed behavior
was compared to cold scaled models in several studies.
In others, cold beds of dierent sizes were compared to
each other. For bubbling beds there is the sense of
what constitutes sucient verication of the scaling
relationships. Since bubbles are the prime motive
agents for both gas and solids displacement, a detailed
verication should involve comparison of bubble prop-
erties through the large bed and the scale model. This
is preferable to comparison of overall performance
parameters for a bed based on input and exhaust mea-
surements.
7.4 Verication of Scaling Relationships for Bubbling
and Slugging Beds
Most early experiments devoted to verifying the scaling
relationships have dealt with the full set of scaling
relationships. Recent experiments have dealt with the
simplied set of dimensionless parameters. In some
experiments, additional scaling parameters were unin-
tentionally matched.
Fitzgerald and Crane (1980) were among of the rst
to evaluate the full set of hydrodynamic scaling para-
meters. They compared the hydrodynamics of two
scaled beds using pressure uctuation measurements
and movies. In one bed cork particles were uidized
with air; the other bed used sand uidized with pres-
surized refrigerant 12 vapor. Movies showed qualita-
tive agreement between bubble growth and the solids
ow in the beds.
Fitzgerald et al. (1984) measured pressure uctua-
tions in an atmospheric uidized bed combustor and a
quarter-scale cold model. The full set of scaling para-
meters was matched between the beds. The autocorre-
lation function of the pressure uctuations was similar
for the two beds but not within the 95% condence
levels they had anticipated. The amplitude of the auto-
correlation function and the experimentally deter-
mined time-scaling factor diered from the expected
value. They suggested that the dierences could be
due to electrostatic eects.
Table 5 Scale models of atmospheric commercial hot bed using the simplied
scaling relationship
Commercial beds Particle diameter of model with bed linear scale factor
d
p
u
mf
Dm,Dc = 1,4 1/9
40 jm 7.45 10
4
m,s 10 jm 8 jm
60 1.68 10
3
15 12
100 4.66 10
3
24 20
200 1.86 10
2
49 40
400 7.42 10
2
98 80
1000 0.441 245 198
Commercial bed t = 800
C, P = 1 atm, ,
s
= 2500 kg,m
3
gas:air. Model bed
,
s
= 8960 kg,m
3
gas:air at STP.
Table 6 Scale model of 10 atm commercial hot bed using the simplied scaling
relationships
Commercial beds Particle diameter of model with bed linear scale factor
d
p
u
mf
Dm,Dc = 1,4 1/9
40 jm 7.44 10
4
m,s 31 jm 25 jm
60 1.66 10
3
46 38
100 4.65 10
3
78 63
200 1.85 10
2
155 126
400 7.18 10
2
310 250
1000 0.329 610 550
Commercial bed t = 800
C, P = 1 atm, ,
s
= 2500 kg,m
3
gas:air. Model bed ,
s
= 896 kg,m
3
gas:air at STP.
Nicastro and Glicksman (1984) experimentally ver-
ied the full set of scaling laws for bubbling uidized
beds. They compared the time-resolved dierential
pressure measurements from a bubbling uidized bed
combustor and a scaled cold model. Good agreement
was obtained between the measurements that were pro-
portional to the bubble size and frequency. Figure 18
presents the comparisons. They concluded that hydro-
dynamic similarity had been achieved between the hot
combustor and the cold model. When actual hot bed
material was used in the cold model, which was a vio-
lation of the scaling laws, the models behavior was
very dierent from that of the hot bed.
Horio et al. (1986) used three geometrically similar
bubbling beds, uidized with ambient air, to verify
their proposed scaling laws. The solid-to-gas density
ratio was maintained constant in the experiments,
although it was not one of the proposed scaling para-
meters. By maintaining a constant density ratio, they
in essence used the simplied set of scaling parameters,
Eq. (8). Video analysis of bubble eruptions at the bed
surface were used to determine that similarity was
achieved for the cross-sectional average bubble dia-
meter, bubble diameter distribution, and radial distri-
bution of supercial bubble velocity. Horio et al.
(1989) veried the bubbling bed scaling relations for
solid mixing and segregation. Sand was used as a bed
material in straight and tapered bed geometries. A bed
sectioning technique was used to measure the transient
radial dispersion coecient and the distribution of
oat tracers. They concluded that bed mixing and the
behavior of oating bodies obey the scaling laws in
both straight and tapered beds. The solid-to-gas den-
sity ratio was again held constant in the tests, satisfying
the simplied set of scaling laws.
Newby and Keairns (1986) made bubbling bed scal-
ing comparisons between two cold models using the
full set of scaling laws. One bed was uidized with
two dierent 200 mm glass powders using ambient air.
The second bed, which was a half-scale model of the
rst, used pressurized air to uidize 100 mm steel
powder. High-speed movies showed good agreement
between the nondimensional bubble frequencies in
the two beds. Figure 19 is a plot of the nondimensional
bubble frequencies as a function of bed Froude
number.
Zhang and Yang (1987) carried out scaling compar-
isons between two two-dimensional beds with u
2
0
,gD
and u
0
,u
mf
matched between them. They also inadver-
tently kept the solid-to-gas density ratio constant; thus
they matched the simplied scaling parameters. They
found through photographs that the beds appeared
qualitatively similar. The beds also had similar dimen-
sionless freeboard entrainment rates and dimensionless
bed heights over a range of u
0
/u
mf
.
Roy and Davidson (1989) considered the validity of
the full and viscous limit scaling laws at elevated pres-
sures and temperatures. The nondimensional domi-
nant frequency and amplitude of the pressure drop
uctuations were used as the basis of the comparison.
They concluded that when the full set of scaling para-
meters is matched, similarity is achieved.
Di Felice et al. (1992a) investigated the validity of
the full set of scaling laws for bubbling and slugging
uidized beds. They used an experimental facility that
permitted the pressurization of dierent diameter test
sections to match the scaling parameters. Minimum
uidization measurements, video measurements of
bed expansion, and pressure uctuation data were
used to compare the similarity of ve dierent bed
congurations. Three of the beds were scaled properly,
Figure 18 Comparison of dimensionless power spectra of
dierential pressure uctuations. Double probe across levels
2 and 3; x,L = 0.0, coal burning bubbling bed combustor.
Full set of scaling laws with iron grit in cold bed; hot bed
material in cold bed violates scaling laws. (From Nicastro
and Glicksman, 1984.)
the fourth had a mismatched particle sphericity, and
the fth was purposefully misscaled relative to the
others. In the bubbling regime, good agreement in
the nondimensional bed expansion measurements was
obtained for all but the bed with the misscaled particle
sphericity. The pressure uctuations for the three prop-
erly scaled beds in the bubbling regime showed good
agreement, while the misscaled beds exhibited poor
agreement with the other three. Figure 20 is a plot of
the dimensionless dynamic pressure variance for the
ve beds in the bubbling regime. The two sets of
data that deviate from the other three correspond to
the misscaled beds.
In the slugging regime, Di Felice et al. (1992a) found
that the bed expansion characteristics were similar to
those in the bubbling regime, but the pressure uctua-
tion characteristics for all ve beds were in poor agree-
ment with each other. They attributed this to the
importance of particle material properties and some
particleparticle interaction eects that are not
accounted for in the full set of scaling laws. This is
discussed further below in the light of recent results
for spouting beds.
Di Felice et al. (1992b) evaluated the full set of
scaling laws for three dierent Geldart powder cate-
gories (A, B, and D) in the bubbling and slugging
uidization regimes. Pressure uctuations were used
as the basis for the scaling comparisons. In the bub-
bling regime, the RMS and dominant frequencies of
the pressure uctuations showed good agreement for
all three powder categories. In the slugging regime,
Geldart groups B and D exhibited fair agreement in
the RMS of their pressure uctuations, but their
dominant frequencies disagreed. They found that the
full set of scaling laws are valid for bubbling beds
uidizing powders in Geldart groups A, B, and D.
They also concluded that the full set of scaling rela-
tionships is not sucient for slugging beds where par-
ticleparticle interactions are also thought to be
important.
7.5 Verication of Scaling Laws for Spouting Beds
He et al. (1997) extended the scaling considerations to
spouting beds. They showed that for spouting beds the
full set of scaling relationships, Eq. (5), must be aug-
mented with two new parameters, the internal friction
angle and the loose packed voidage. By systematic tests
in dierently sized cold beds as well as comparisons
between hot and cold beds, they showed excellent
agreement when the full set of scaling parameters, aug-
mented with the two spouting bed parameters, were
held constant. Close agreement was found for spout
diameter, fountain height, longitudinal pressure pro-
les, and dead zone boundary; see Fig. 21. When the
Figure 19 Comparison of nondimensional bubble frequen-
cies from two cold scaled models. (From Newby and
Keairns, 1986.)
Figure 20 Comparison of dynamic pressure variance for
three properly scaled beds and two misscaled beds in bub-
bling regime. Properly scaled: &, laposorb; ~, sand; *,
bronze. Intentionally misscaled: , iron; ~, sand. (From
DiFelice et al., 1992a)
internal friction angle and sphericity were mismatched,
there was a large disagreement in fountain height.
The internal friction angle is also important for
slugging beds; see Zenz and Othmer (1960). DiFelice
et al. (1992 a,b) did not report their values; it could be
that the disagreement they found in their slugging bed
tests was due to mismatches of the internal friction
angle.
7.6 Verication of Scaling Relationships for
Pressurized Bubbling Beds
Almstedt and Zakkay (1990) made scaling compari-
sons between a hot PFBC with horizontal tubes and
a pressurized cold scale model using the full set of
scaling laws. The cold model had linear dimensions
one-half those of the hot bed. A capacitance probe
was used to measure the mean values of the bubble
frequency, pierced length, bubble rise velocity, and
bubble volume fraction. Scaling comparisons were
made using the dimensionless form of these dependent
hydrodynamic parameters. Three dierent bed materi-
als were used in the cold bed: olivine sand and two
dierent size distributions of the hot bed material,
one properly scaled and one out of scale. Almstead
and Zakkay concluded that behavior that is hydrody-
namically similar to that of a pressurized uidized bed
combustor can be achieved using a properly scaled
cold model.
Glicksman and Farrell (1995) constructed a scale
model of the Tidd 70 MW
c
pressurized bubbling bed
combustor. The scale model was uidized with air at
atmospheric pressure and temperature. They used the
simplied set of scaling relationships to construct a
one-quarter length scale model of a section of the
Tidd combustor. Low-density polyurethane beads
were used to obtain a close t with the solid-to-gas
density ratio for the combustor as well as the particle
sphericity and particle size distribution. Dierential
pressure measurements were made between several ver-
tical elevations within the bed. The solid fraction pro-
les were obtained from the vertical pressure prole
with a hydrostatic assumption. The cold model solid
fraction prole showed very close agreement with data
taken from pressure taps in two dierent locations
within the combustor; see Fig. 22. The probability den-
sity functions of the cold model and combustor give
very close agreement. The power spectral density of the
combustor exhibited several distinct peaks at increas-
ingly higher frequencies. All but the rst peak were not
seen in the cold model. The peaks could be due to tube
vibrations in the hot bed, uctuations upstream or
downstream of the bed, or hydromechanical interac-
tions between the bed and the internals.
Farrell (1996) experimentally evaluated the im-
portance of the solid-to-gas density ratio (,
s
,,
f
) for
scaling the hydrodynamics of bubbling and slugging
uidized beds. Two bed materials, polyethylene plastic
(,
s
= 918 kg,m
3
) and a dolomite/limestone sorbent
mixture (,
s
= 2670 kg,m
3
), were used to create a mis-
match in the density ratio. The size of the particles was
chosen so that the remaining simplied scaling para-
meters were matched. The internal angle of friction
was similar between the two materials.
In one case, the solid fraction of the sorbent mate-
rial was less than the plastic in the lower regions of the
bubbling bed, with good agreement in the upper sec-
tion of the bed. However, for the same conditions, the
dimensionless standard deviation of the time-varying
pressure drop showed the best agreement in the bottom
of the bed with a large discrepancy in the upper por-
tion of the bed. The plastic bed material has a much
broader transition region between its fully bubbling
and fully slugging regimes than the sorbent material,
and the nature of this transition is dierent in the two
materials. Therefore the solid-to-gas density ratio
inuences both the hydrodynamics in the bubbling
regime and the boundary at which the transition to
slugging occurs. This is consistent with the conclusion
Figure 21 CDimensionless spout diameters as a function of
dimensionless height for small columns. Case A, test case;
Case B, all dimensionless parameters matched, bed diameter
halved; Case C, particle Reynolds number mismatched; Case
D, Froude number mismatched; Case E, density ratio,
Reynolds number mismatched; Case F, bed Reynolds num-
ber mismatched; Case G, internal friction angle, loose packed
voidage mismatched. (From He et al. 1995.)
of Glicksman et al. (1993b), who found it essential to
match the density ratio when scaling circulating ui-
dized bed hydrodynamics.
8 APPLICATIONS OF SCALING TO
COMMERCIAL BUBBLING FLUIDIZED BED
UNITS
A substantial number of experimental investigations
have demonstrated the validity of scaling. This has
increased awareness of the concept and condence in
its application. Scaling has many useful applications.
The dynamic characteristics of dierent bed designs
can be quickly compared. The inuence of bed dia-
meter on hydrodynamic behavior can be studied by
the use of several dierent models. The models allow
easy experimental examination of existing operating
characteristics. The beds also can be used to conrm
quickly the inuence of proposed modications in
operating parameters and bed geometry. Since the
models usually operate at ambient conditions, it is pos-
sible to instrument them to observe detailed behavior.
This allows a better understanding of the fundamental
physics as well as the identication of hydrodynamic
factors needed for proper correlation of performance.
The earliest scaling studies were directed at atmo-
spheric bubbling bed combustors. To date, a rich vari-
ety of questions have been addressed.
Jones and Glicksman (1986) constructed a model
of the 20 MW bubbling bed pilot plant jointly spon-
sored by the Tennessee Valley Authority and the
Electric Power Research Institute (EPRI) at
Paducah, Kentucky. Figure 23 shows a photograph
of the model of the in-bed tubes installed in the scale
model. The model, which is roughly 100 by 120 cm in
cross section, simulates two-thirds of the entire 20
MW pilot plant. Care was taken to match the pilot
plant tube bundle geometry and distributor design.
Steel grit particles with the same dimensionless size
distribution and sphericity as the hot bed material
were used. The full set of scaling parameters was
matched in the model and the combustor. The largest
discrepancy was in the solid-to-gas density ratio,
which was 18% smaller in the model than in the
pilot plant.
The measured bubble velocity for an actively bub-
bling bed was found to agree closely with the drift ux
form proposed by Davidson and Harrison (1963). In
contrast, the volumetric ow rate of the bubbles was
found to be far less than that predicted by the two-
phase hypothesis; see Fig. 24.
Figure 22 Solid fraction prole comparisons for Tidd PFBC and cold model based on simplied scaling laws (From Glicksman
and Farrell, 1995.)
The larger particles were thrown high in the splash
zone, higher than predicted by a ballistic trajectory
using the bubble rise velocity as the initial velocity
and neglecting any air drag. Later observations of
this model showed that when bubbles erupt at the sur-
face, the accompanying gas ow has a velocity much
higher than the bubble rise velocity; see Glicksman and
Piper (1987). This sets up a substantial gas bypass from
the distributor to the surface of a relatively shallow
bed. The observations from the model led to a mechan-
istic model for gas throughow aided by the low resis-
tance of the bubble cavity; see Yule and Glicksman
(1988), and an accurate prediction of bubble volume
ow rate and bed expansion; see Glicksman and Yule
(1991).
A major question in the design of a commercial
sized bubbling bed is the need to identify part load
operating techniques. While reducing the total com-
bustion rate, it is desirable to keep the bed operating
temperature constant. This requires a reduction of the
heat transfer to the water lled tubes within the bed.
One technique uses the contraction of the bed that
accompanies a decrease in supercial velocity. As the
bed contracts, some of the tube rows are uncovered,
reducing the net heat transfer. The scale model allowed
many dierent tube arrangements to be tested. The
validity of the scaling technique was conrmed by a
comparison of the bed expansion measured for the
pilot plant and that found in the model equipped
with the same tube bank geometry; see Fig. 25.
A second method of reducing load while maintain-
ing constant bed temperature is to reduce the super-
cial velocity of a portion of the bed distributor to a
value below u
mf
. In this design, the bed does not con-
tain vertical partitions above the distributor. The scale
model was used to determine the rate of growth of the
xed bed in the deuidized zone along with the heat
transfer to tubes in that region. Figure 26 shows a
typical pattern of particle accumulation in a slumped
zone adjacent to an actively uidized zone. Heat trans-
fer coecients are also shown. Note that tubes near the
upper surface of the solids that experience a downow
of solids have a very high heat transfer rate.
The heat transfer from tubes in the freeboard was
also measured for the 20 MW model. The variation of
measured overall heat transfer coecient in the 20
MW pilot plant versus velocity agreed closely with
that predicted from the scale model test. When the
bed height is lowered, uncovering some tubes, the
heat transfer is reduced because there are fewer parti-
cles contacting the tube surface.
Ackeskog et al. (1993) made the rst heat transfer
measurements in a scale model of a pressurized bub-
bling bed combustor. These results shed light on the
inuence of particle sizes, density, and pressure levels
on the fundamental mechanism of heat transfer, e.g.,
the increased importance of the gas convective compo-
nent with increased pressure.
A multisolid bed contains a mixture of large solids
that are contained in a dense region at the bottom of
the bed and ner particles that recirculate through the
bed and external cyclone. Ake and Glicksman (1989)
used a cold scale model of a multisolids combustor to
determine the dense bed expansion. The measured
expansion in a properly scaled quarter-scale model
using steel pellets to simulate the coarse particles and
to satisfy the solid-to-gas density ratio gave good
agreement with eld data; see Fig 27. It was also
demonstrated that an improperly scaled cold model,
using the same coarse material as the hot bed, had
an incorrect gas-to-particle density ratio and substan-
tially overpredicted the bed expansion.
Tube erosion has been observed in both atmo-
spheric and pressurized bed combustors. The scaling
analysis presented earlier can be used to construct an
accurate hydrodynamic simulation of the commercial
bed. This can be used to investigate qualitatively the
factors related to tube wear such as the location of
highest wear around the circumference of an individual
tube and the location within the bed of highest wear.
Quantitative wear rates cannot be obtained from
model tests unless the parameters governing both the
hydrodynamics and the wear phenomena are matched
between the model and the commercial bed. The inclu-
sion of more large particles causes an increase in the
Figure 23 Model of 20 MW bubbling uidized bed combus-
tor showing tube arrangement. (From Jones and Glicksman,
1986.)
wear rate. The second row of tubes is shielded by the
rst row, reducing the incidence of strikes by large
high-velocity bubbles. In the interface between an
actively uidized and a slumped bed, there is a ow
of particles down the slumped surface, which causes
higher wear. The relative changes in wear rate with
tube location as well as the circumferential location
with highest wear agreed closely between the cold
model and the 20 MW combustor.
Other boiler manufacturers have used scale models
to aid in the design of large bubbling bed combustors.
Yang et al. (1995) used atmospheric cold models 30
cm and 3 m in diameter to simulate a large jetting
uidized bed gasier to be operated at about 12 atm.
They used the full set of scaling relations that
required the model to be about the same dimensions
as the hot bed. A half-round bed with a transparent
wall was used to photograph the jet behavior. They
measured and developed correlations for jet penetra-
tion, bubble frequency, vertical bubble size, jet half-
angle, and others. They also found considerable gas
leakage from the bubbles to the emulsion phase. This
data covers much larger bed sizes than had been
examined before.
9 HYDRODYNAMIC SCALING OF
CIRCULATING BEDS
Given the success in scaling bubbling beds, research
has progressed to scaling of circulating fast beds. The
initial research has focused on the verication of both
the full and the simplied scaling relationships for cir-
culating beds. The verication using data from com-
bustors is complicated by the diculty in accurately
measuring the recycle rate of solids, an important para-
meter. Figure 28 presents the range of scaling para-
meters for experimental studies undertaken for
Figure 24 Deviation from two-phase theory for model of bubbling uidized bed combustor. (From Jones and Glicksman,
1986.)
circulating in terms of the Froude number and the
Reynolds number based on particle diameter and the
supercial gas velocity.
Horio et al. (1989) experimentally veried their pro-
posed circulating uidized bed scaling laws. The solid-
to-gas density ratio was not varied in the tests; thus
they eectively veried the simplied set of scaling
laws. Two cold scaled CFBs, uidized using ambient
air, were used in the verication. Good agreement in
the axial solid fraction proles was obtained for most
of the conditions tested. An optical probe was used to
verify similarity in the annular ow structures and the
cluster velocities.
Ishii and Murakami (1991) evaluated Horio et al.
(1989) CFB scaling relationships using two cold CFB
models. Solids ux, pressure drop, and optical probe
measurements were used to measure a large number of
hydrodynamic parameters to serve as the basis for the
comparison. Fair to good similarity was obtained
between the beds.
Tsukada et al. (1991) applied Horio et al. (1989)
CFB scaling laws at several dierent elevated pressures
(viscous limit scaling laws). A single bed and bed mate-
rial were used in the study. A pressure vessel was used
to vary the gas pressure. They found that as the pres-
sure was increased the axial solid fraction prole chan-
ged, indicating a change in the hydrodynamics. In this
study, the gas-to-solid density ratio changes with pres-
sure level. It is likely based on the recent results of
Glicksman et al. (1993b) that the change in gas-to-
solid density ratio led to the inuence of the pressure
level on the bed hydrodynamics. This points out a
deciency of the viscous limit, Horios (1989) scaling
relationship.
Glicksman et al. (1991a) made scaling comparisons
between an experimental circulating uidized bed com-
bustor and a scaled cold model based on the full set of
scaling laws. Due to uncertainties in the hot bed solid
circulation measurements, the cold bed solids ux was
adjusted until the average bed solid fraction matched
that of the hot bed. Good agreement was obtained
between the vertical solid fraction proles except
near the top of the beds. Disagreement may have
Figure 26 Particle accumulation in slumped zone adjacent
to active bed: air velocity through active region = 151 cm,s;
air velocity through slumped region = 7.3 cm,s; numbers
above heater are heat transfer coecient in W,m
2
C after
1530 min. (From Glicksman et al., 1992.)
Figure 25 Bed expansion for TVA 20 MW
e
FBC and for scale model, shown in Fig. 23. (From Glicksman et al., 1989.)
been due to the dierences between the gas-to-solid
density ratio in the cold model and the hot bed.
Good agreement was also obtained in the comparison
of the probability density distribution and the Fourier
transform of the pressure uctuations.
Glicksman et al. (1993a) evaluated the full set of
scaling laws for circulating uidized beds. Solid frac-
tion data were obtained from the 2.5 MW
th
Studsvik
atmospheric CFB prototype. The full set of scaling
laws was evaluated through solid fraction prole com-
parisons between Studsvik and a one-quarter scale cold
model. Fairly good agreement was obtained; the pro-
les most closely matched in the top of the beds.
Dierences between the proles were attributed to
uncertainty in the hot bed solid ux measurements
and to the mismatch in the solid-to-gas density ratio.
A modied set of the simplied scaling relationships
were also evaluated by Glicksman et al. (1993b) in a
series of comparison tests using circulating beds in
which the gas-to-solid density ratio was not held con-
stant. The average solid fraction proles, Fig. 29, solid
fraction probability density functions, and power spec-
tral densities were all in poor agreement. It is believed
the beds were operating near the point of incipient
choking condition as predicted by the Yang (1983)
correlation. Because this correlation indicates that
choking is a strong function of the solid-to-gas density
ratio, it was concluded that this parameter must be
matched to model bed hydrodynamics near the bound-
ary between dierent ow regimes.
The simplied scaling relationships were used by
Glicksman et al. (1993b) to compare two geometrically
Figure 27 Expanded bed height model data compared to eld data for multisolids uidized bed for properly scaled steel pellets
and misscaled gravel particles. (From Ake and Glicksman, 1989.)
similar beds, one having linear dimensions four times
larger than the other. In one series of tests, properly
sized plastic particles were used in both beds; in another
test series, glass particles were used in the two beds. The
average solid fraction proles showed excellent agree-
ment; see Fig. 30. The probability density functions and
power spectral densities also agreed well.
Glicksman et al. (1993b) veried the simplied scal-
ing laws for hot beds by comparing the solid fraction
proles for the Studsvik bed, the one-quarter scale cold
Figure 28 Range of experimental scaling studies for circulating and bubbling uidized beds.
Figure 29 Solid fraction proles, glass/plastic viscous limit scaling; density ratio mismatched: low velocity case. (From
Glicksman et al., 1993a.)
model, and a one-sixteenth scale cold model. The one-
sixteenth scale model had a cross sectional area of
16 cm
2
to simulate a 2.5 MW combustor! The average
solid fraction proles were in good agreement for most
of the conditions tested. The agreement was excellent
between the one-quarter scale cold model, which used
the full set of scaling laws, and the one-sixteenth scale
model that used the simplied set of scaling laws. Thus
any disagreement between the Studsvik bed and the
one-sixteenth scale model is not due to the simplica-
tions of the full set of scaling laws. The density ratio
was not matched exactly between the hot bed and the
two cold beds, which may have aected the agreement.
Figure 31 provides a typical comparison of the solid
fraction proles in the three beds. The authors con-
cluded that the simplied set of scaling laws, which
includes the solid-to-gas density ratio, gives acceptable
results over a wide range of particle densities and bed
sizes, even when the length ratio is as small as one-
sixteenth for an atmospheric combustor.
Figure 30 Solid fraction proles, glass simplied scaling, high velocity case. (From Glicksman et al., 1993a.)
Figure 31 Solid fraction proles, hot bed scaling with simplied scaling laws, low velocity, 49% primary air. (From Glicksman
et al., 1993a.)
Glicksman et al. (1995) used the simplied scaling
parameters to construct a one-half linear scale model
of a Foster Wheeler circulating bed combustor pressur-
ized to 14 bar. The combustor has a 20.3 cm inner
diameter with an overall height of 8.3 m with both a
primary and a secondary air supply. The solids recycle
rate was accurately determined by a calorimetric bal-
ance of a uidized bed heat exchanger in the return
loop of the circulating bed. The cold model, one-half
scale, used polyethylene plastic particles to match the
dimensionless particle size distribution as well as the
gas-to-solid density ratio.
The time averaged vertical pressure dierence was
used to determine the solid fraction distribution in the
combustor of the cold model. Figure 32 shows the
close agreement between the combustor and the cold
model. Three test cases had similar solid fraction ver-
sus height proles. The fourth, operated at low gas
velocity and solids recycle rate, had a more abrupt
decrease of solid fraction with bed height. The cold
model reected the same behavior. The probability
density functions also were in agreement except for
some discrepancy near the secondary air inlets. This
might be due to uctuations set up by the bubbling
bed heat exchanger upstream of the secondary air
inlet to the combustor. The bubbling bed was not
duplicated in the cold scale model.
Subsequent experiments veried that a 1/6.5 linear
scale model of another Foster Wheeler circulating bed
combustor, designed with the simplied scaling rela-
tionships, exhibited very close agreement with the full
scale hot combustor. Figure 33 shows a comparison of
the average solid fraction versus bed height between
the pressurized hot combustor and the 1/6.5 linear
scale model, using plastic particles uidized by ambient
air. Figure 34 shows a comparison of the Fourier
transform versus frequency, based on an equivalent
hot frequency.
Further verication of the scaling laws are needed in
terms of radial solids distribution and solids diusivity.
These have not been compared between hot and cold
beds. In addition, work needs to be carried out to see
the lower limit of solids diameter for which the present
set of scaling parameters holds. At some point, surface
forces will have an important inuence on bed
dynamics. This will require additional scaling para-
meters that include these eects. The range of validity
of the present set of scaling parameters for uidized
beds for FCC operations and for cyclone separators
remains uncertain at the present time.
Figure 32 Solid fraction prole comparison: between pressurized circulating uidized bed combustor and one-half size scale
model based on simplied scaling. Two dierent operation conditions. (From Glicksman et al., 1995.)
10 CONCLUSIONS
As uidized beds are scaled up from bench scale to
commercial plant size, the hydrodynamic behavior of
the bed changes, resulting, in many cases, in a loss of
performance. Although there have been some studies
of the inuence of bed diameter on overall perfor-
mance as well as detailed behavior such as solids mix-
ing and bubble characteristics, generalized rules to
guide scale-up are not available. The inuence of bed
diameter on performance will dier for dierent ow
regimes of uidization.
Small, properly scaled laboratory models operated
at ambient conditions have been shown accurately to
simulate the dynamics of large hot bubbling and circu-
lating beds operating at atmospheric and elevated pres-
sures. These models should shed light on the overall
operating characteristics and the inuence of hydrody-
namics factors such as bubble distribution and trajec-
tories. A series of dierently sized scale models can be
used to simulate changes in commercial bed behavior
with bed size.
The scale models must be carefully designed. Failure
to match one or more important dimensionless para-
meters will lead to erroneous simulation results.
Modeling can be extended to particle convective heat
transfer. Wear or erosion of in-bed surfaces can be
qualitatively studied, although quantitative assessment
requires the identication and simulation of additional
wear-related parameters.
Most of the simulation eort has been applied to
uidized bed combustors that use relatively large par-
ticles. Simulation can also be used for other uidiza-
tion processes in the petrochemical industry. Research
should be undertaken to identify the proper scaling
parameters for beds uidized with smaller particles.
Similar simulations may also apply to components
such as cyclones.
ACKNOWLEDGMENTS
The author would like to acknowledge the contributions
of many present andformer M.I.T. students who carried
out scaling studies of uidized beds. Much of the M.I.T.
research mentioned in this chapter has been sponsored
by the Electric Power Research Institute, the National
Science Foundation, the U.S. Department of Energy,
and the Tennessee Valley Authority.
NOTATION
C
D
= Drag coecient of sphere
D = Bed diameter
d
B
= Bubble diameter
d
c
= Cluster diameter
d
p
= Particle diameter
d
V
= Equivalent bubble diameter
G
S
= Solids recycle rate per unit area
g = Acceleration of gravity
L = Typical bed dimensions
Q
B
= Bubble volume ow rate
t = Time
u = Gas velocity
u
b
= Bubble rise velocity
u
mf
= Minimum uidization velocity
u
0
= Supercial gas velocity
u
f
= Terminal velocity
x. y = Coordinates
Figure 33 Average solid fraction: 11.5 m high CFB com-
bustor at 7.5 bar compared to 1/6.5 scale cold model.
Figure 34 Frequency response: 11.5 m high CFB combustor
at 7.5 bar compared to 1/6.5 scale cold model.
Dimensionless
PSD = Dimensionless particle size distribution
Re = Reynolds number
Greek
[ = Drag coecient
o = Bubble volume fraction
o = Boundary layer thickness in a fast bed
= Distributor plate voidage
e = Void fraction
e
c
= Cluster void fraction
e
mf
= Void fraction at minimum uidization
,
f
= Fluid density
,
S
= Solid density
m = Fluid viscosity
v = Fluid kinematic viscosity
= Particle sphericity
Subscripts
( )
c
= Commercial bed
( )
m
= Model bed
Superscripts
( )
/
= Dimensionless
() = Vector
REFERENCES
Ackeskog HBR, Almstedt AE, Zakkay V. An investigation
of uidized-bed scaling: heat transfer measurements in a
pressurized uidized-bed combustor and a cold model
bed. Chem Eng Sci 48:1459, 1993.
Ake TR, Glicksman LR. Scale model and full scale test results
of a circulating uidized bed combustor. Proceedings of
1988 Seminar on Fluidized Bed Combustion Technology
for Utility Appl. EPRI, 1989, pp 124.
Almstedt AE, Zakkay V. An investigation of uidized-bed
scaling-capacitance probe measurements in a pressurized
uidized-bed combustor and a cold model bed. Chem Eng
Sci 45(4):1071, 1990.
Arena U, Cammarota A, Massimilla L, Pirozzi D. The
hydrodynamic behavior of two circulating uidized bed
units of dierent size. In: Basu P, Large JF, eds.
Circulating Fluidized Bed Technol II. Oxford:
Pergamon Press, 1988.
Avidan A, Edwards M. Modeling and scale up of Mobils
uid bed MTG process. In: Ostergaard K, Sorensen A,
eds. Fluidization V. New York: Engineering Found, 1986.
Avidan A, Edwards M, Owen H. 50 years of catalytic crack-
ing. Oil Gas Journal 33, 1990.
Bauer W, Werther J, Emig G. Inuence of gas distributor
design on the performance of uidized bed reactor. Ger
Chem Eng 4:291, 1981.
Chang H, Louge M. Fluid dynamic similarity of circulating
DeGroot JH. Scaling-up of gas-uidized bed reactors. In:
Proceedings of the International Symposium on
Fluidization. Drinkenburg AAH, ed. Amsterdam:
Netherlands University Press, 1967.
DiFelice R, Rapagna S, Foscolo PU. Dynamic similarity
rules: validity check for bubbling and slugging uidized
beds. Powder Technol 71:281, 1992a.
DiFelice R, Rapagna S, Foscolo PU, Gibilaro LG. Cold
modelling studies of uidised bed reactors. Chem Eng
Sci 47:223, 1992b.
Farrell PA. Hydrodynamic scaling and solids mixing in
pressurized uidized bed combustors. PhD dissertation,
Massachusetts Institute of Technology, Cambridge, MA,
1996.
Fitzgerald TJ, Crane SD. Cold uidized bed modeling.
Proceedings of 6th International Conference on
Fluidized Bed Combustion, III, 1980, p 815.
Fitzgerald TJ, Bushnell D, Crane S, Shieh Y. Testing of cold
scaled bed modeling for uidized-bed combustors.
Frye CG, Lake WC, Eckstrom HC. Gas-solid contacting
with ozone decomposition reaction. AIChE J 4(4):403,
1958.
Geldart D. The size and frequency of bubbles in two- and
three-dimensional gas-uidised beds. Powder Technol
4:41, 1970.
Glicksman LR. Scaling relationships for uidized beds.
Chem Eng Sci 39:1373, 1984.
Glicksman LR. Scaling relationships for uidized beds.
Chem Eng Sci 43:1419, 1988.
Grace JR, Avidan A, Knowlton TM, eds. Circulating
261311.
Glicksman LR, Farrell P. Verication of simplied hydro-
dynamic scaling laws for pressurized uidized beds: part I.
Bubbling uidized beds. Proc. 13th Int. Conference for
Fluidized Bed Combustion, 1995, p 981.
Glicksman LR, McAndrews G. The eect of bed width on
the hydrodynamics of large particle uidized beds.
Glicksman LR, Piper GA. Particle density distribution in a
freeboard of a uidized bed. Powder Technol 53:179,
1987.
Glicksman LR, Yule T. Prediction of the particle ow con-
ditions in the freeboard of a freely bubbling uidized bed.
AIChE J, 1991c.
Glicksman LR, Yule T, Dyrness A, Carson R. Scaling the
hydrodynamics of uidized bed combustors with cold
models: experimental conrmation. Proc. 9th Int. Conf.
on Fluidized Bed Combustors, 1987, p 511.
Glicksman LR, Mullens G, Yule TW. Tube wear tests in the
MIT scaled uidized bed. Proceedings of EPRI
Workshop, Nov, 1987b.
Glicksman LR, Yule T, Carson R, Vincent R. Comparison
of results from TVA 20 MW uidized bed combustor with
MIT cold scale model. Proc. 1988 Seminar on Fluidized
Bed Combustion Technology for Utility Applications,
EPRI GS-6118, 1-20-1, 1989.
Glicksman LR, Westphalen D, Brereton C, Grace J.
Verication of the scaling laws for circulating uidized
beds. In: Basu P, Horio M, Hasatani M, eds.
Circulating Fluidized Bed Technology. Oxford:
Pergamon Press, 1991a.
Glicksman LR, Yule T, Dyrness A. Prediction of the expan-
sion of uidized beds containing tubes. Chem Eng Sci
46:1561, 1991b.
Glicksman LR, Yule T, Arencibia A, Pangan A. Heat trans-
fer to horizontal tubes in a slumped bed adjacent to a
uidized bed. Proceedings of 7th Eng. Foundation Conf.
on Fluidization. New York: Engineering Foundation,
1992, p 813.
Glicksman LR, Hyre MR, Westphalen D. Verication of
scaling relations for circulating uidized beds. Proc.
12th Int. Conf. on Fluidized Bed Combustion, 1993a, p
69.
Glicksman LR, Hyre MR, Woloshun K. Simplied scaling
relationships for uidized beds. Powder Technol 77:177,
1993b.
Glicksman LR, Hyre M, Farrell P. Dynamic similarity in
uidization. Intern J Multiphase Flow, 20(suppl): 331,
1994.
Glicksman LR, Hyre M, Torpey M, Wheeldon J.
Verication of simplied hydrodynamic scaling laws for
pressurized uidized beds: Part II. Circulating uidized
beds. Proc. 13th Int. Conference for Fluidized Bed
Combustion, 1995, p 991.
Glicksman LR, Hyre M, Farrell PA. Importance of the
solid-to-gas density ratio for scaling uidized bed
hydrodynamics. Proc of the 9
th
International
Conference on Fluidization, Fluidization IX
Conference, 1998.
Hartge E-U, Li Y, Werther J. Flow structures in fast
uidized beds. Proc. 5th Engineering Foundation Conf.
on Fluidization, Elsinore Denmark, 1985, p 345.
He Y-L, Lim CJ, Grace JR. Scale-up studies of spouted beds.
Chem Eng Sci 52:329339, 1997.
Horio M, Nonaka A, Sawa Y, Muchi I. A new similarity rule
for uidized bed scale-up. AIChE J 32:1466, 1986.
Horio M, Ishii H, Kobukai Y, Yamanishi N. A scaling law
for circulating uidized beds. J Chem Eng Japan 22:587,
1989.
Hovmand S, Davidson JF. Slug ow reactors. In: Davidson
JF, Harrison D, eds. Fluidization. London: Academic
Press, 1971.
Ishii H, Murakami I. Evaluation of the scaling law of
circulating uidized beds in regard to cluster behaviors.
In: Basu P, Horio M, Hasatani M, eds. Circulating
Fluidized Bed Technology III. Oxford: Pergamon Press,
1991.
Johnsson F, Leckner B. Vertical distribution of solids in a
circulating uidized bed furnace. Proceedings 13
th
International Conference on Fluidized Bed
Combustions. American Society of Mechanical
Engineers, 1995, pp 671702.
Johnsson JE, Grace JR, Graham JJ. Fluidized bed reactor
model verication on a reactor of industrial scale. AIChE
J 33(4):619, 1987.
Jones L, Glicksman LR. An experimental investigation of
gas ow in a scale model of a uidized-bed combustor.
Knowlton TM, Geldart D, Matsen J. Oral presentation.
Fluidization VIII. International Symposium of the
Engineering Foundation, Tours, France, 1995, pp 14
19.
Litka T, Glicksman LR. The inuence of particle mechanical
properties on bubble characteristics and solid mixing in
Matsen JM. Fluidized beds in scaleup of chemical processes:
conversion from laboratory scale tests to successful com-
mercial size design. In: Bisio A, Kabel RL, eds. Scaleup in
Chemical Processes. New York: John Wiley, 1985, p 347.
Matsen JM. Design and scale-up of CFB catalytic reactors.
In: Grace JR, Avidan AA, Knowlton TM, eds.
Circulating Fluidized Beds. London: Chapman and
Hall, 1997, pp 489503.
May WG. Fluidized-bed reactor studies. Chem Eng Progress
55 (12):49, 1959.
Newby RA, Keairns DL. Test of the scaling relationships for
uid-bed dynamics. In: Ostergaard K, Sorensen A, eds.
Fluidization V. New York: Engineering Foundation,
1986.
Nguyen HV, Potter OE, Whitehead AB. Bubble distribution
and eruption diameter in a uidized bed with a horizontal
tube bundle. Chem Eng Sci 34:1163, 1979.
Nicastro MT, Glicksman LR. Experimental verication of
scaling relationships for uidized bed. Chem Eng Sci
39:1381, 1984.
Noymer PD, Hyre MR, Glicksman LR. The inuence of bed
diameter on hydrodynamics and heat transfer in circulat-
ing uidized beds. Fluidization and Fluid-Particle
Systems, AIChE, 1995, pp 8690.
Patience GS, Chaouki J, Berruti F, Wong R. Scaling consid-
erations for circulating uidized bed risers. Powder
Technol 72:31, 1992.
Rhodes MJ, Laussman P. A study of the pressure balance
around the loop of a circulating uidized bed. Can J
Chem Eng 70:625, 1992.
Rhodes MJ, Wang XS, Cheng H, Hirama T, Gibbs BM.
Similar proles of solids ux in circulating uidized bed
risers. Chem Eng Sci 47(7):1635, 1992.
Rowe PN, Stapleton WM. The behavior of 12-inch diameter
fast uidized beds. Trans Instn Chem Engrs 39:181, 1961.
Roy R, Davidson JF. Similarity between gas-uidized beds at
elevated temperature and pressure. Fluidization VI. New
York: Engineering Foundation, 1989.
Shingles T, McDonald AF. Commercial experience with
synthol CFB reactors. In: Basu P, Large BP, eds.
Circulating Fluidized Bed Technology I. Oxford:
Pergamen Press, 1988.
Thiel WJ, Potter OE. Slugging in uidized beds. Ind Eng
Chem Fundam 16(2):242, 1977.
Thiel WJ, Potter OE. The mixing of solids in slugging gas
Tsukada M, Nakanishi D, Takei Y, Ishii H, Horio M.
Hydrodynamic similarity of circulating uidized bed
under dierent pressure conditions. Proc. 11th
International Conference Fluidized Bed Combustion,
1991, p 829.
Van Deemter JJ. Mixing patterns in large-scale uidized
beds. In: Grace JR, Matsen JM, eds. Fluidization. New
York: Plenum Press, 1980.
Van Swaaij WPM, Zuiderweg FJ. Investigation of ozone
decomposition in uidized beds on the basis of a two-
phase model. Chemical Reaction Eng. Proc. of the 5th
European/2nd Int Symp. Chem. Reaction Eng.
Amsterdam/London/New York: Elsevier, 1972.
Van Swaaij WPM, Zuiderweg FJ. The design of gas-solids
uidized bedsprediction of chemical conversion. Proc.
International Symposium on Fluidization and its Appl.
Ste Chimie Industrielle, Toulouse, 1973.
Volk W, Johnson CA, Stotler HH. Eect of reactor internals
on quality of uidization. Chem Eng Prog 58:44, 1962.
Werther J. Inuence of bed diameter on the hydrodynamics
of gas uidized beds. AIChE Symposium Ser 70(141):53,
1974.
White FM. Viscous Fluid Flow. New York: McGraw-Hill,
1974.
Whitehead AB, Young AD. Fluidisation performance in
large scale equipment. Proc. Intern. Symp. on
Fluidisation, Eindhoven, 1967, p 294.
Yang WC. Criteria for choking in vertical pneumatic convey-
ing lines. Powder Technol 35:143, 1983.
Yang WC, Newby RA, Keairns DL. Large-scale uidized
bed physical model: methodology and results. Powder
Technol 82:331, 1995.
Yerushalmi Y, Avidan A. High velocity uidization. In:
Davidson JF, Clift R, Harrison D, eds. Fluidization. 2d
ed. New York: Academic Press, 1985.
Yule T, Glicksman LR. Gas ow through erupting bubbles
in uidized beds. AIChE Symp Ser 262(84):1, 1988.
Systems. New York: Reinhold, 1960.
Zhang MC, Yang RYK. On the scaling laws for bubbling
gas-uidized bed dynamics. Powder Technol 51:159, 1987.
46(12):3045, 1991.
Zhang W, Johnsson F, Leckner B. Fluid dynamic boundary
layers in CFB boilers. Chem Eng Sci 50(2):201, 1995.
14
Applications for Fluid Catalytic Cracking
Ye-Mon Chen
Shell Global Solutions US, Houston, Texas, U.S.A.
1 INTRODUCTION
1.1 What Is FCC?
Fluid catalytic cracking (FCC) is the primary conver-
sion unit in most reneries in the United States. It
converts low-value heavy components of crude oil
into a variety of high-value lighter products. A modern
FCC unit consists of three major sections: the reactor/
regenerator section, the main fractionation/gas plant
section, and the ue gas cleaning/power recovery sec-
tion. This chapter focuses mainly on the reactor/regen-
erator section of the FCC process, paying special
attention to its relationship with uidization and
uidparticle systems. Readers who are interested in
broader aspects of the FCC process can refer to several
recent publications, such as Sadeghbeigi (1995) and
Wilson (1997).
In the reactor/regenerator section of the FCC pro-
cess, liquid hydrocarbon feedstock is preheated, mixed
with steam, and injected into the riser reactor through
feed nozzles. Hot regenerated catalyst is drawn from
the regenerator and contacts the hydrocarbon feed.
The two are thoroughly mixed in the lower riser by
ne atomization through the feed nozzles. The tem-
perature of regenerated catalyst drops rapidly as the
feed vaporizes and reacts, converting the feedstock into
lighter products while a coke layer deposits on the
spent catalyst, temporarily deactivating the FCC cata-
lyst. Product vapor and spent catalyst are separated by
riser termination devices, such as cyclone systems, at
the end of the riser. The product vapor goes to the
main fractionation and gas plants for product separa-
tion. The spent catalyst enters the stripper where steam
is introduced to further recover entrained/adsorbed
hydrocarbon products. The spent catalyst then passes
through a spent catalyst transfer line and enters the
regenerator. As the catalyst activity is restored by
burning o the coke layer on spent catalyst, a large
amount of heat is released, which heats up the regen-
erated catalyst. The ue gas from the regenerator goes
to ue gas cleaning and the power recovery system.
The regenerated catalyst then returns to the reactor
riser. This completes the catalyst circulation loop.
1.2 How the FCC Unit Fits in a Typical US
Renery
The FCC unit usually has the second highest through-
put after the crude distillation unit in most U.S. re-
neries. Crude oil contains hydrocarbons of dierent
boiling points ranging from light gas to residue.
Typically, crude oil is rst distilled in an atmospheric,
or crude, distillation unit to produce a wide range of
hydrocarbon products according to the dierent boil-
ing points in the original crude. The residue from the
atmospheric distillation unit, the atmospheric residue,
is then fed to the vacuum distillation unit, which pro-
duces vacuum gas oil and vacuum residue. The vacuum
gas oil from the vacuum distillation unit is the tradi-
tional FCC feedstock with typical boiling points ran-
ging from 650 to 1050
F. Depending on the type of

crude and the operating conditions of the vacuum dis-
tillation unit, the portion of vacuum gas oil can range
from 25 to 35% of the basic crude. Many reneries use
delayed cokers to take the vacuum residue from the
vacuum distillation unit as feedstock. Liquid products
from cokers, the coker gas oils, are another common
feedstock for the FCC unit.
Over the years, FCC has evolved into a very exible
process that can take a wide variety of other feed-
stocks, including hydrotreated gas oils, deasphalted
oils, slop oils, and lube extracts. In addition, many
modern FCC units are capable of the direct processing
of atmospheric residue or vacuum residue, which is
called residual or resid uid catalytic cracking
(RFCC).
The FCC process also plays a key role as a feed
preparation unit for downstream rening processes.
For example, light petroleum gas (LPG) produced by
the FCC unit is a common feedstock to the alkylation
unit.
1.3 Why is FCC Important?
The FCC unit has several important functions in ren-
ing and petrochemical manufacture. First of all, the
FCC unit is a gasoline machine. About 45% of world-
wide gasoline production comes either directly from
FCC units or indirectly from FCC downstream units,
such as alkylation units.
The second function of the FCC unit is to reduce
the amount of residue in crude oil. In the U.S. market,
the product slate produced directly from the distilla-
tion units contains too little transportation fuel (gaso-
line and diesel) and too much residue. The FCC unit
enables reneries to convert most of the residue into
lighter products, thus fully utilizing the crude oil to
meet the market demands.
The third function of the FCC unit is to provide
exibility to the rening operation. The FCC unit
can adjust operating conditions to maximize the pro-
duction of gasoline, middle distillate (LCO), or light
petroleum gas (LPG). This exibility enables the ren-
ery to adjust its product slate to meet seasonable
demands in the market. Reneries in the U.S. also
take in dierent crude diets according to changes in
the crude market. The exibility of the FCC unit
enables the renery to process dierent crude diets as
the crude market changes.
The fourth function of the FCC unit is to produce
light olens for downstream rening processes, such as
alkylation and other petrochemical processes, as in
MTBE, PE, and PP plants.
1.4 Why Is Particulate Technology Important to
FCC?
FCC is by far the most important petrochemical pro-
cess that involves the broadest scope of particulate
technology. Although the chemistry of the FCC pro-
cess is rather complex, many of the operational pro-
blems are in fact associated with handling uid
particle systems and related mechanical issues, such
as erosion. In particular, the reactor/regenerator sec-
tion of the FCC unit includes many aspects of uid
particle systems that are critical to FCC design and
operation.
For instance, the lower riser is typically operated in
the fast uidization regime, whereas the upper riser is
in the dilute transport regime. In between, the feed
injection section is a complicated transition that
involves rapid mixing, heat transfer, and reaction
between the FCC catalyst and the feedstock. The chal-
lenge is to achieve fast mixing and vaporization of the
liquid feedstock, which plays a key role in achieving
the overall riser performance. In the middle section of
the riser, the phenomena of catalyst segregation and
backmixing are related to cluster formation of the cat-
alyst. This presents another challenge to the riser reac-
tor design. At the end of the riser are riser termination
and reactor cyclone systems, which handle gassolid
separation. The challenge is to achieve a fast and
clean separation that minimizes the degradation of
the hydrocarbon products by minimizing post riser
cracking.
The stripper is typically operated in the bubbling
regime utilizing various internals to improve steam
catalyst contacting eciency. The challenge is to max-
imize the contacting eciency in order to recover the
most hydrocarbons by proper design of stripper inter-
nals and steam injection systems. The regenerator is
typically the largest vessel of the FCC process operated
in the turbulent to fast uidization regime. The chal-
lenge is to maximize catalyst regeneration by proper
distribution of spent catalyst and air.
Both the reactor and the regenerator have cyclone
systems in the dilute phase to separate catalyst and to
return it back to the system via cyclone diplegs.
Prediction of the transport disengaging height (TDH)
is critical to determining the locations of cyclone inlets.
Prediction of bed entrainment is critical to estimating
the catalyst loading to the cyclones. Designs of
cyclones and diplegs are critical to FCC operation in
maintaining proper particle size distribution of catalyst
in the system. Last, but certainly not least, the stand-
pipe is the main driving force for catalyst circulation
between the regenerator and the reactor, which drives
the FCC process. The challenge is to maximize pres-
sure buildup and ow stability in the standpipe by
proper design of the standpipe inlet and aeration.
2 FUNDAMENTALS OF THE FCC PROCESS
2.1 FCC Catalyst
The FCC catalyst is the heart and soul of the process.
Both chemical and physical properties of the catalyst
determine how the FCC unit is designed and operated.
Since fresh catalyst is added to the FCC unit regularly,
and catalyst is also withdrawn and lost through
cyclone systems, the most important catalyst proper-
ties to FCC operation are those of the equilibrium
catalyst.
The physical form of a modern FCC catalyst is a
ne powder with a broad particle size distribution
(PSD), mostly in the range between 10 to 150 microns.
The average particle size is around 70 to 80 microns. It
is a typical group A particle according to the Geldard
classication (1973). The apparent bulk density of
FCC catalyst is about 50 to 60 lb=ft
3
. In addition to
average particle size and density, one parameter critical
to FCC operation is the content of nes in the range of
0 to 40 microns in the equilibrium catalyst. If the con-
tent of the nes drops, the uidization property of the
equilibrium catalyst will deteriorate rapidly even
though the increase in the average particle size is rela-
tively small. This typically leads to catalyst circulation
problems. The loss of nes is also an indication that
the cyclone systems are not functioning properly.
A modern FCC catalyst includes four major com-
ponents. They are zeolite, matrix, binder, and ller.
The rst component, zeolite, is the primary active
ingredient of the FCC catalyst, which can vary in the
range of 15 to 50 wt% of the catalyst. It is a molecular
sieve with a well-dened lattice structure, which pro-
vides the selectivity to allow only a certain size range of
hydrocarbon molecules to enter the catalyst lattice
structure. The acidic sites on the zeolite provide most
of the activity of the FCC catalyst.
The basic building blocks of zeolite are silica and
alumina tetrahedra. Each tetrahedron of the basic
block has either a silicon or an aluminum atom at
the center and four oxygen atoms at the corners. The
pore diameter of the zeolite lattice is approximately 8
angstroms. The unit cell size (UCS) is the distance
between the two repeating cells in the zeolite structure.
Fresh FCC catalyst has a UCS in the range of 24.5 to
24.75 angstroms. In general, zeolite does not allow
hydrocarbon molecules with a molecular diameter
greater than 8 to 10 angstroms to enter the structure.
The acidic sites of zeolite catalyst come from the alu-
mina tetrahedron. The aluminum at the center of the
tetrahedron is at a 3 oxidation state surrounded by
four oxygen atom at the corners, which are shared with
neighboring cells. Thus each alumina tetrahedron car-
ries a net charge of 1, which has to be balanced by a
positive ion. Sodium ion is commonly used in catalyst
production for this purpose, which is later replaced by
an ammonium ion. Upon drying of the catalyst,
ammonia is vaporized, resulting in the formations of
acidic sites of both the Bronsted and Lewis types. The
Bronsted acidic sites can be further exchanged with
rare earth material, such as cerium and lantheium.
The second component, matrix, plays an important
role for cracking large hydrocarbon molecules, thus
contributing signicantly to the overall performance
of FCC catalyst. Matrix also contains alumina but
mostly in amorphous form, which is another source
of the catalytic activity. The function of matrix is to
provide active sites in larger pores in the amorphous
form of alumina, which allow larger hydrocarbon
molecules to diuse in and crack into smaller mole-
cules. This precracking function of the matrix enables
FCC catalyst to process heavier feedstock with large
hydrocarbon molecules, which are otherwise unable to
enter the zeolite structure. The last two components,
the ller and the binder, provide the physical integrity
and mechanical strength of the FCC catalyst.
Metal contamination has detrimental eects on
FCC catalyst performance and should be closely mon-
itored. Sodium, which is present in fresh FCC catalyst,
can also be present in the feedstock. Other metals,
which include nickel, vanadium, iron, and copper, ori-
ginate mostly from the heavy ends of the hydrocarbon
feedstock. As catalyst comes in contact with feed in the
unit, the metal components deposit on the catalyst and
stay there. Sodium is known to neutralize the acidic
sites and causes the zeolite structure to collapse.
Nickel, vanadium, iron, and copper are promoters
for dehydrogenation reactions. Vanadium is also
known to destroy zeolite activity. FCC units, which
process heavy feedstock, typically have higher metal
contamination on the equilibrium catalyst.
2.2 Catalytic Cracking Reactions
Cracking of a hydrocarbon molecule means breaking
the long carbon-to-carbon chain into shorter hydro-
carbon molecules. There are two parallel mechanisms
of cracking reactions. One is thermal cracking, which is
a slower process, and the other is catalytic cracking.
The reaction kinetics of catalytic cracking goes
mainly through the intermediate of a carbenium ion.
The rst step of catalytic cracking begins with the reac-
tion of a hydrocarbon molecule at the acidic sites of
the FCC catalyst, forming a carbenium ion (R-CH
2
.
This can occur from either adding a H
charge to an
olen molecule at the Bronsted acidic site, or removing
a H
charge from a paran molecule at a Lewis acidic

site. The three key catalytic reactions following the
formation of the carbenium ion are isomerization,
beta scission, and hydrogen transfer.
The carbenium ion has the tendency to isomerize,
because the primary ions are the least stable and the
tertiary ions are the most stable. Thus, regardless of
the initial forms of carbenium ions, and hence of the
original hydrocarbon molecules in the feedstock, they
tend to rearrange themselves preferentially toward sec-
ondary and tertiary ions. This has a signicant impact
on the FCC product properties. Because the tertiary
ion is the preferred carbenium ion formation, the pro-
ducts of catalytic cracking reactions contain highly
branched molecules.
The actual cracking step of the catalytic reaction is
the beta scission of the carbenium ions. The cracking
step occurs at the beta position because the carbon-to-
carbon bond beta to the charged carbon on the carbe-
nium ions is the weakest bond of the entire hydrocar-
bon chain and the easiest to break. The beta scission of
a carbenium ion produces an olen and a new carbe-
nium ion. The new carbenium ion produced by the
beta scission can repeat the beta scission or undergo
rst isomerization and then beta scission. The combi-
nation of beta scission and the fact that the primary
carbenium ion is the least stable means that the hydro-
carbon molecule formed by catalytic cracking contains
at least three carbon atoms. This results in high pro-
duction of C3 and C4 in LPG from the catalytic crack-
ing reactions. As the hydrocarbon chain length of the
carbenium ion becomes shorter, the reaction rate of the
beta scission becomes lower. Thus the rate of catalytic
cracking decreases as conversion gets higher in the
upper riser.
Hydrogen transfer is another important mechanism
in FCC catalytic reaction kinetics. Unlike beta scission
and isomerization, which involve only a single mole-
cule, hydrogen transfer is a bimolecular reaction. In
order for hydrogen transfer to occur, two hydrocarbon
molecules have to be adsorbed on two active sits on the
catalyst, and the two active sites have to be close
enough to allow the two adjacent molecules to interact.
One important outcome of hydrogen transfer is that it
enables a shorter, less reactive carbenium ion to trans-
fer its charge to a longer, more reactive molecules, thus
further propagating the catalytic cracking reactions.
Another important outcome is the redistribution of
hydrogen atoms in the FCC products. For example,
hydrogen transfer between an olen and a naphthene
produces a paran and a cyclo-olen. Subsequent
hydrogen transfer can further convert the cyclo-olen
to an aromatic molecule. A rare-earth-exchanged zeo-
lite catalyst increases the tendency to hydrogen trans-
fer. This is because the rare earth forms bridges
between acidic sites on the catalyst. Since hydrogen
transfer reaction requires adjacent acidic sites to
occur, bridging these sites with rare earth promotes
hydrogen transfer. The end results of hydrogen trans-
fer are that it reduces the reactivity of the gasoline
produced by FCC, which improves gasoline stability.
This is because olens are the most reactive species in
the gasoline that are subject to secondary reactions,
and hydrogen transfer reduces olens in the gasoline.
However, hydrogen transfer also has major drawbacks
to the quality of the FCC products. It lowers light
olens in the LPG, lowers the octane number of the
gasoline, and produces more aromatics in gasoline and
light cycle oil (LCO).
Dehydrogenation is an undesirable catalytic reac-
tion, which is promoted by metal contamination on
equilibrium catalyst. The dehydrogenation reaction
extracts hydrogen molecules from hydrocarbon mole-
cules, thus increasing the production of highly aromatic
products and coke. Nickel contamination is known to
promote dehydrogenation. Other metals, such as cop-
per and iron, are also known to increase hydrogen pro-
duction. Metal contamination comes mainly from the
heavy ends of the hydrocarbon feedstock, particularly
the residues, as discussed previously.
2.3 Thermal Cracking Reactions
Thermal cracking reactions are undesirable but inevi-
table parallel reactions to the catalytic cracking reac-
tions under normal FCC reactor conditions. The
hydrocarbon molecules will crack owing to high tem-
perature alone in absence of active FCC catalyst.
Before the development of the FCC process, thermal
cracking used to be the primary conversion unit.
Today, reneries still utilize thermal cracking pro-
cesses, such as delayed coking and visbreaking for ther-
mally cracking heavy residues.
The reaction kinetics of thermal cracking goes
through a dierent intermediate of free radicals. Two
free radicals are formed when a hydrocarbon chain
splits at any carbon-to-carbon or hydrogen-to-carbon
bond. This is the rate-determining step of thermal
cracking reactions. Splitting a hydrocarbon molecule
produces two uncharged molecules with each molecule
having an unpaired electron. Since more energy is
required to split a hydrogen-to-carbon bond, this is
less likely to occur. For splitting primary, secondary,
or tertiary carbon-to-carbon bonds, there is little dif-
ference in energy requirements. Thus the formation of
free radicals is nonselective. Methyl and ethyl free radi-
cals are as likely to form as other longer free radicals.
Once free radicals are formed, they are extremely
reactive. They can undergo alpha and beta scissions,
hydrogen transfer, and polymerization. The scissions
of free radicals are the actual thermal cracking reac-
tions that produce smaller hydrocarbon molecules.
When a scission reaction of a free radical occurs, it
produces an olen and a new free radical. However,
one important dierence from the catalytic reactions is
that free radicals cannot undergo isomerization. Thus
the end product from thermal cracking has fewer
branched hydrocarbons. Another important dierence
is that the beta scission in the catalytic cracking reac-
tion produces hydrocarbons longer than C3, whereas
alpha and beta scissions of free radicals produce mole-
cules and radicals including C1 and C2. Thus, thermal
cracking leads to a product that is rich in C1 and C2
light gases.
Hydrogen transfer occurs when a free radical
extracts a hydrogen atom from a hydrocarbon mole-
cules, which ends the free radical and creates a new free
radical. The hydrogen-to-primary-carbon bond is the
strongest and the hydrogen-to-tertiary-carbon bond is
the weakest within a hydrocarbon chain. Furthermore,
the rate of hydrogen transfer decreases as the hydro-
carbon chain of the free radical increases. Thus a
methyl radical produced by the alpha scission tends
to extract a hydrogen atom from a neutral hydrocar-
bon molecule to form a methane molecule and a new
secondary or tertiary free radical. The same mechan-
ism applies to the formation of ethane through beta
scission followed by hydrogen transfer.
Two free radicals can recombine into a single mole-
cule and terminate the chain reactions of thermal
cracking. The olens formed initially by scission reac-
tions can also undergo polymerization and condensa-
tion to form coke. All in all, thermal cracking produces
more C1/C2 light gases as well as more coke than
catalytic cracking. The challenge in the FCC process
is to minimize thermal cracking while maximizing cat-
alytic cracking.
2.4 Heat Balance
Like every system operating under steady-state condi-
tions, an FCC unit must stay in heat balance at all
times. However, because catalyst circulation between
the regenerator and the reactor serves the dual pur-
poses of providing reaction activity and the heat
requirement to the reactor, the heat balance in the
FCC process has special purposes that other processes
do not have. Thus, heat balance is the key to better
understanding of how dierent variables interact with
one another in the FCC unit.
Most FCC units only have a few independent vari-
ables. Typically, these independent variables are the
feed rate, feed preheat temperature, reactor/riser tem-
perature, air ow rate to the regenerator, and catalyst
activity. The feed rate and air ow rate to the regen-
erator are set by ow controllers. The feed temperature
is set by the feed temperature controller. Catalyst activ-
ity is set by catalyst selection and fresh catalyst addi-
tion rate. Reactor temperature is controlled by the
regenerator slide valve that regulates the catalyst cir-
culation rate. The catalyst circulation rate is not
directly measured or controlled. Instead, the unit relies
on the heat balance to estimate the catalyst circulation
rate. Except for these independent variables, other
variables, such as regenerator temperature, degree of
conversion, and carbon-on-catalyst, etc., will vary
accordingly to keep the FCC unit in heat balance.
These variables are dependent variables.
The overall heat balance requires that the FCC unit
constantly adjust itself to produce and burn just the
right amount of coke in the regenerator to supply the
heat requirements for the entire unit. The major heat
requirements are heating and vaporizing the hydrocar-
bon feed, heat of reaction, heating air and steam to
system temperature, and heat losses of the unit.
Combustion of coke in the regenerator is the single
source of heat supply to run the FCC unit. The heat
generated in the generator is carried to the rest of the
unit by catalyst circulation, as described by the follow-
ing equation:
Heat transfer
lb of feed

catalyst
oil
catalyst heat capacity)
T
regen
T
react
The rst term on the right-hand side is the catalyst-to-

oil ratio, which has the dimension of pounds of cata-
lyst circulation per pound of feed processed. The last
term is the temperature dierence between the regen-
erator and the reactor.
The material balance requires that coke burned in
the regenerator be equal to coke formed in the reactor,
which is called the coke yield. Coke yield has the
dimension of pounds of coke formed per pound of
feed processed. Since catalyst circulation is the carrier
of the coke on catalyst, the coke yield can be expressed
as
Coke yield delta coke
catalyst
oil
The rst term on the right-hand side is the delta coke,
which is the dierence in coke concentration (lb coke/
lb catalyst) between the spent and the regenerated cat-
alyst. The second term on the right-hand side is the
catalyst-to-oil ratio.
Through these relationships of heat and material
balances, a change in one independent variable will
trigger a domino eect on other variables in an FCC
unit. For instance, if the reactor/riser temperature is
increased at a given feed temperature by opening the
regenerator slide valve, it will trigger the following
chain of responses (on a per-pound-of-feed-processed
basis):
Catalyst circulation will increase owing to the open-
ing of the regenerator slide valve.
Reaction conversion will increase because of higher
reaction rate at higher catalyst-to-oil ratio and
reactor temperature. This is the most common
reason for increasing the reactor temperature.
Coke yield will increase because of higher conver-
sion. This also provides additional heat for
higher reaction conversion and other heating
requirements.
Delta coke will decrease because there is more cat-
alyst circulating to the reactor to receive the coke
formation.
More air to the regenerator is required to burn more
coke. The increase in air requires an independent
adjustment to the ow controller because air ow
rate to the regenerator is not a dependent vari-
able.
Regenerator temperature will most likely increase
because of higher coke combustion rate.
However, the increase in regenerator temperature
is typically much less than the increase in reactor
temperature.
2.5 Pressure Balance and Catalyst Circulation
Catalyst circulation between the reactor and the regen-
erator has two critical functions in the FCC process.
One is to maintain the activity of the regenerated cat-
alyst by burning and removing coke on the spent cat-
alyst. The other is to keep the unit in heat balance by
continuously removing heat from the regenerator and
adding it to the reactor and the rest of the unit.
Since FCC catalyst is kept above minimum uidiza-
tion conditions everywhere in this catalyst circulation
loop, the uidized catalyst is free to ow from one
place to another. Thus the catalyst circulation is driven
by the overall pressure balance of the unit, and the
circulation rate is regulated by the two slide valves,
i.e., the stripper slide valve and the regenerator slide
valve. A minimum pressure drop across each slide
valve is set in the control system to guard against
ow reversal, which is a very serious safety issue. For
instance, a reverse ow of hydrocarbon vapor from the
reactor to the oxygen-rich regenerator can lead to a
sudden increase in combustion reaction and regenera-
tor temperature. In the extreme case, a catastrophic
explosion could occur. The overall pressure balance
of the unit determines the pressure drops available
for the slide valve control and hence the maximum
catalyst circulation rate.
The rst step in the overall pressure balance is to
determine the hydraulic heads in dierent parts of the
FCC unit. Typical apparent density in the riser is 4 to
6 lb=ft
3
. For side entry feed nozzle conguration, the
lower riser section prior to feed injection has a much
higher density than the remaining of the riser, which
depends on the preacceleration uidization velocity.
Typical standpipe density is 35 to 40 lb=ft
3
. An inclined
standpipe is known to have a slightly lower density
than a vertical standpipe. Typical stripper bed density
is about 35 to 40 lb=ft
3
, whereas the regenerator bed
has a lower density, in the range of 25 to 30 lb=ft
3
. The
dilute phase in both the reactor and the regenerator
has a density below 1 lb/ft
3
.
The next step is to estimate transition pressure
losses, which include riser liftpot, riser termination,
cyclone, spent catalyst distributor, and other ow tran-
sitions. By subtracting these transition pressure losses
from the hydraulic pressure, the remainder would be
the pressure drop available for slide valve control.
While other parameters in pressure balance seldom
change much at all, it is not uncommon to have the
standpipe apparent density lower than the typical
value. Because standpipes are long columns, a lower
standpipe apparent density means a substantial loss in
hydraulic heads and hence lower available pressure
drops for slide valve control, leading to a lower achiev-
able catalyst circulation rate. Commercial experience
has demonstrated that a proper design and operation
of the standpipe, particularly the standpipe inlet and
uidization, can increase the catalyst circulation rate
by up to 30 to 40% (Chen, 2001).
2.6 Hydrogen Balance
Every chemical reaction must satisfy an overall mate-
rial balance as well as individual atomic balances. It is
important to discuss the hydrogen balance in the FCC
process, because hydrogen in the feedstock is the limit-
ing component of the cracking reactions.
Typical FCC feedstock contains about 12 wt% of
hydrogen. The average hydrogen content of lighter
cracking products is about 13.5 wt% whereas the aver-
age hydrogen content of the heavier liquid cracking
products is about 9 wt%. Since typical hydrogen con-
tent in coke is in the range of 5 to 7 wt% and coke yield
is about 5 wt% on feed, a simple hydrogen balance
shows that typical conversion of an FCC unit should
be about 70%. Any deviation from these typical values
will shift the hydrogen balance accordingly. For
instance, an increase in feedstock hydrogen content
will allow the unit to have higher conversion, leading
to higher production of cracking products, coke and
less heavier liquid cracking products.
Hydrogen balance also provides other insights into
the FCC process. For example, it is known that ther-
mal cracking produces more C1/C2 dry gas. Because
these shortest hydrocarbon molecules have the highest
hydrogen contents, more coke and heavy cracked oils
are also produced by thermal cracking to satisfy the
hydrogen balance. Thus thermal cracking is undesir-
able. Dehydrogenation reaction, which is promoted by
metal contamination, has similar eects to those of
thermal cracking in increasing the production of coke
and heavy cracked oils based on the hydrogen balance.
Another example is to look at the eect of adding
residuals to the feedstock. As the percentage of resi-
dual in feed increases, the hydrogen content in the
overall feedstock will drop. Thus hydrogen balance
requires that a residual cracking unit operate at a
much lower conversion and have a higher coke yield
than a typical FCC unit.
3 MODERN FCC DESIGN
Several engineering companies such as UOP, Stone
and Webster, Kellogg Brown and Root, and ABB/
Lummus as well as major oil companies such as
Exxon/Mobil, Shell, and Total/ElfFina all have their
own FCC designs (Meyers, 1997). Although each FCC
design is dierent, the basic design philosophies are
actually quite similar. The following discussion will
use mainly the Shell FCC design, shown in Fig. 1, as
an example to elucidate major components in a typical
modern FCC process. As shown in the gure, the FCC
unit is a two-vessel, side-by-side design with the regen-
erator to the left and reactor/stripper to the right. The
two vessels are connected to each other with catalyst
transfer lines. Each major component of this FCC
design is discussed in detail in the following sections.
3.1 Feed Injection System
The feed injection system is the most critical compo-
nent of the modern FCC riser reactor design. Several
factors have made the feed injection system increas-
ingly important over the years. The most important
factor is that the FCC reaction time is getting shorter
and shorter. This was not the case for earlier FCC
units when the catalyst was not as active and the reac-
tion time was relatively long, i.e., on the order of min-
utes. Thus, mixing time and evaporation time of the
feed induced by the feed injection system were not as
critical. However, due to the development of the highly
active zeolite catalyst, typical reaction time has been
shortened to a few seconds in the riser reactor. Some of
the most recent FCC reactor designs have gone to
Figure 1 Modern shell FCC design.
below 1 second (Kau et al., 1996). Since catalytic
cracking reactions can only occur in the vapor phase
after liquid hydrocarbon feedstock is vaporized, mix-
ing and feed vaporization must take place as quickly as
possible in order for catalytic cracking reactions to
complete in a few seconds or less. Otherwise, only ther-
mal cracking can take place. Furthermore, liquid
hydrocarbon, which is not vaporized, is burned as
coke in the regenerator.
The second factor is that the regenerator tempera-
ture is getting higher, which is also a contributing fac-
tor to the shorter reaction time as discussed previously.
Because of the higher regenerator temperature, control
of thermal cracking in the reactor becomes more criti-
cal. Typical riser top temperature of modern FCC
units is controlled in the range of 950 to 1050
F. But
typical regenerated catalyst temperature is much
higher, in the range of 1250 to 1350
F. Thus both
catalytic and thermal cracking reaction rates are
much higher in the lower riser section because the cat-
alyst is fresh and active, and the temperature is high.
The feed injection system plays the key role of control-
ling thermal cracking by cooling o the lower riser
section with fast mixing and vaporization of the liquid
feed.
The third factor is that the FCC feedstock is getting
heavier. As the feed gets heavier, the boiling point
increases, which makes feed vaporization more di-
cult. At the same time, the viscosity of the feed also
increases, which makes feed atomization more dicult.
Thus better feed atomization must be achieved to dis-
perse the feed into even ner droplets in order to
vaporize the heavier feed in time for catalytic cracking
reactions to take place in the riser reactor within sec-
onds.
The feed injection system in FCC consists of two
key components: the feed nozzle itself and the preac-
celeration region, which is the lower riser transition
preparing the catalyst ow prior to feed injection.
These two design components must work in concert
in order to achieve desirable riser performance. There
are many dierent FCC feed nozzle designs in the
patent literature (Chen and Dewitz, 1998; Chen et
al., 1999; Bedaw et al., 1993; Haruch, 1995). All
these modern FCC feed nozzles belong to the same
category of twin-uid atomizers that utilize pressurized
gas to assist the atomization of liquid feed. Figure 2
shows the feed nozzle design of Chen and Dewitz
(1998). In this design, the atomization gas ows
along the center pipe, passes through multiple holes
at the end of the gas cap, and mixes with the liquid
feed from the annulus prior to exiting the nozzle slits.
Figure 3 shows ne atomization and feed distribution
of the fan spray pattern generated by the feed nozzle of
Chen and Dewitz (1998) during a pilot plant test.
The feed nozzle has three major functions: feed
atomization, feed distribution, and mixing with cata-
lyst.
1. Feed atomization. The objective of feed atomi-
zation is to generate a large surface area for
ne droplets to vaporize as quickly as possible.
A feed nozzle produces a spray with a certain
droplet size distribution, both in time and
space. The challenge of a successful feed nozzle
design is to produce the nest, most stable, and
uniform feed atomization using the least
amount of energy. The quality of feed atomi-
zation is typically measured by the Sauter
mean diameter of the droplet size distribution.
However, the mean droplet size is only one of
several important measurements. It is desirable
to produce a droplet size distribution as nar-
row as possible to limit the number of large
droplets. Another key performance parameter
of a feed nozzle is the energy eciency, which
is measured by the amount of atomization
steam used and the feed pressure required to
achieve desirable atomization. Commercial
experience has demonstrated that a successful
feed nozzle design can achieve ne atomization
with atomization steam in a range between 1 to
3 wt% on feed and with a pressure drop
through the nozzle of no more than 25 to 60
psi.
2. Feed distribution. Feed distribution is mea-
sured by the uniformity of liquid ux across
the entire spray pattern generated by the feed
nozzle, both in time and in space. The objective
is to achieve uniform coverage of feedstock
across the riser. Commercial experience has
shown that the best spray pattern to achieve
such an objective is a wide-angle fan spray
such as the one shown in Fig. 3.
3. Mixing with catalyst. The objective is to
achieve a at radial riser temperature prole
as quickly as possible. In doing so, the regen-
erated catalyst is uniformly cooled down by
the vaporization of hydrocarbon feedstock,
thus minimizing thermal cracking reactions in
the lower riser section. In addition to feed ato-
mization and feed distribution, commercial
experience has shown that the feed injection
angle, dened by the angle between the riser
Figure 2 CCU feed nozzle design. (From Chen and Dewitz, 1998.)
axis and the feed injection, plays a signicant
role in mixing the feed with catalyst and hence
achieving a desirable temperature prole in the
riser. Most FCC units have feed nozzles
installed through riser shrouds at a xed
angle. A new feed nozzle design by Chen et
al. (1999) as shown in Fig. 4, enables an FCC
unit to adjust the feed injection angle while
using the existing riser shrouds. This enables
an FCC unit to optimize the mixing of feed
with catalyst by adjusting the feed injection
angle to achieve the best performance of the
unit.
Commercial experience has conrmed that using better
feed nozzles can substantially reduce dry gas and
increase gasoline yield. These results are in line with
the expectation that thermal cracking reactions, which
are the primary source for dry gas, are reduced. As a
result, catalytic cracking reactions are maximized, and
more desirable cracking products are produced.
Depending on unit constraints and market demands,
other steps can be taken, such as increasing riser tem-
perature, increasing catalyst activity, or adding ZSM-5
catalyst, in order to take full advantage of the dry gas
reduction.
The second component of the feed injection system
is the design of the low riser transition prior to feed
injection. There are two common congurations of
the lower riser to prepare for feed injection; one is
bottom entry and the other is side entry congura-
tion. Most of the modern FCC units have the side
entry conguration, as shown in Fig. 1, in which
regenerated catalyst is lifted from the riser bottom
to the feed injection point by uidizing gas, normally
steam. The section between the riser bottom and the
feed injection point is called the catalyst preaccelera-
tion zone. A typical lift steam velocity for the preac-
celeration is in the range between 3.0 to 10.0 ft/s,
which is in the turbulent to fast uidization regime.
The length for the preacceleration zone is in the range
from 6 to 20 feet. At the end of preacceleration, mul-
tiple feed nozzles located on the circumference of the
riser are used to inject the feedstock. With proper
designs of the preacceleration and feed nozzles, back-
mixing of catalyst in the lower riser is minimized with
the side entry conguration.
Most of the older FCC units have the bottom entry
conguration. In this case, a single or multiple nozzles
located at the bottom of the riser are used to introduce
the feedstock directly into the region where hot regen-
erated catalyst enters the riser. The catalyst condition
in the riser bottom is denser, chaotic, and highly ero-
sive. Earlier feed nozzles with the bottom entry cong-
uration emphasize mechanical robustness, but with
primitive feed atomization and poor performance.
Because of this disadvantage many bottom entry
FCC units have been converted to side entry units in
recent years. However, recent commercial experience
has demonstrated that, with proper feed nozzle design
(Chen, 2002), a bottom entry conguration can achieve
feed contacting and riser performance similar to the
Figure 3 Atomization and spray pattern of the feed nozzle shown in Fig. 2. (From Chen and Dewitz, 1998.)
side entry conguration. In some cases, the bottom
entry conguration can even be the preferred option
because it does not require the preacceleration zone,
which can take up substantial riser volume. In addi-
tion, bottom entry conguration has lower pressure
drop in the riser, which has a positive impact on cata-
lyst circulation.
3.2 Riser/Reactor
Modern FCC units have vertical risers, which provide
the proper residence time for the hydrocarbon feedstock
to vaporize, react, and crack in presence of catalyst.
Due to the highly active modern FCC catalyst, the
cracking reactions are essentially completed at the end
Figure 4 CCU feed nozzle design. (From Chen et al., 1999.)
of the vertical riser. Thus the reactor vessel in the mod-
ern FCC unit downstream of the riser is actually a
containing vessel for spent catalyst/product vapor
separation. In some older FCC units, however, the
reactor vessel is designed to provide additional resi-
dence time for the cracking reactions to complete down-
stream of the riser while using a less active catalyst.
Due to the cracking reactions, the molar ow rate
and the volumetric ow rate of the hydrocarbon vapor
increase as it moves up the riser. In a typical FCC unit,
the volumetric expansion is in the range of 3 to 4 times
of the vaporized feedstock. Thus the riser diameter
may be increased once or twice after feed injection to
keep the vapor velocity within the range of 40 to 80
feet/s. Some modern FCC risers could further include
boundary layer strip rings, e.g., Gwyn (1993), to
reduce catalyst backmixing along the riser wall and
to achieve performance closer to an ideal plug ow
reactor.
A modern FCC riser design also includes a separa-
tion device at the end of the riser. The objective is to
achieve a quick and clean separation of spent catalyst
and hydrocarbon vapor in order to minimize post riser
cracking. Post riser cracking is undesirable because the
riser is designed to achieve optimum conversion, and
additional cracking reactions after the riser will push
the conversion outside the optimum range. One riser
termination design, called the prestripping cyclone,
uses a dead end tee at the riser top followed by a special
cyclone, as described in Parker et al. (1987). Figure 5
shows further details of this design in which the upper
part is a rough-cut cyclone for catalyst/vapor separa-
tion and the lower part is a prestripper bed where steam
is injected to remove entrained vapor. Catalyst not cap-
tured in the prestripping cyclone is separated in down-
stream secondary cyclones. In most modern FCC
designs, the secondary cyclones are close-coupled with
the primary cyclones, such as examples shown in Wilson
(1997), to minimize post riser cracking. The reactor
cyclones can be designed to operate either inside or out-
side the reactor vessel, as shown in Fig. 1.
3.3 Stripper
After the spent catalyst/product vapor separation
from a riser termination system, it is inevitable that
some hydrocarbon will be entrained by the spent cat-
alyst from the exit of the riser termination system.
Part of the entrained hydrocarbon resides in the
vapor space between catalyst particles, and the
remainder is physically adsorbed on the surface and
inside the pores of the catalyst. The primary objective
of a stripper is to recover the entrained and adsorbed
hydrocarbon as much as possible with a limited
amount of steam. The other objective is to reduce
hydrocarbon carryunder into the regenerator, which
will be burned as coke.
Most FCC strippers are designed with a counter-
current contacting of descending spent catalyst and
ascending stripping steam, operating in the bubble
ow regime. Any hydrocarbon not stripped by steam
will enter the regenerator and burned as coke. Thus the
overall coke entering the regenerator includes the coke
on the catalyst and the entrained hydrocarbon. Since
the hydrocarbon vapor removable by stripping is rela-
tively rich in hydrogen compared to coke deposition on
the spent catalyst, a direct measurement of the stripper
performance is the level of hydrogen content in the
overall coke entering the regenerator. Another mea-
surement of the stripper performance is the energy e-
ciency expressed in terms of the amount of stripping
steam used. A properly designed stripper should be
able to reach a hydrogen-in-coke level of 5 to 7 wt%
using 2 to 4 pounds of steam per 1000 pounds of cat-
alyst circulation.
It is important to note that the hydrocarbon vapor
continues to react and crack under typical FCC strip-
per conditions. Thus early dissociation of the hydro-
carbon vapor from catalyst is critical to minimize
overcracking of high-value hydrocarbon products
into low-value light gas and coke. This can be achieved
by a proper design of two-stage stripping. In the rst
stage, a relatively high ow of stripping steam is used
to displace quickly the easily desorbed and entrained
hydrocarbon vapor. This is typically done by placing a
steam ring in the upper part of the stripper or by using
a prestripping cyclone, as shown in Fig. 5. In second
stage stripping, a moderate ow rate of the stripping
steam combined with an adequate residence time
removes crackable fragments of the more strongly
adsorbed heavy hydrocarbons.
3.4 Regenerator
The primary objective of the regenerator is to burn o
coke deposition on the spent catalyst to restore catalyst
activity. However, other important aspects should also
be taken into consideration as well. The challenge of a
successful regenerator design is to achieve the follow-
ing objectives all at the same time, to
Achieve low coke on regenerated catalyst (CRC) to
restore catalyst activity.
Burn more coke at a given amount of blower air.
Figure 5 CCU stripper cyclone. (From Parker et al., 1987.)
Minimize catalyst deactivation.
Minimize afterburn, which will be discussed later in
this section.
There are two types of regenerators in FCC opera-
tions; one operates in the partial combustion mode
and the other in the total combustion mode. In
partial combustion mode, a less-than-theoretical, or
stoichiometric, amount of air is provided to the
regenerator. Only part of the carbon in coke is
reacted to carbon dioxide, and the remainder of the
carbon is reacted to carbon monoxide. Ideally, all
oxygen should be consumed and no oxygen should
be present in the ue gas.
One important parameter to monitor in partial
combustion regeneration is the CO=CO
2
ratio in the
ue gas, which is typically operated in the range from
0.5 to 2.0.
In the total combustion mode, excess air is provided
to the regenerator. Ideally, all the carbon component
in the coke should be reacted to carbon dioxide, and
no carbon monoxide should be present in the ue gas.
One important parameter to monitor in total com-
bustion regeneration is the oxygen content in the ue
gas, which is typically operated in the range from 1.0 to
3.0 percent on a dry basis.
Coke on regenerated catalyst (CRC) is a key perfor-
mance measurement for both partial and total combus-
tion regenerators. Units with total combustion
regenerators can achieve a typical CRC level of
about 0.05 wt% or lower. Units with partial combus-
tion regenerators have higher CRC levels, typically
about 0.1 wt% or higher.
A partial combustion regenerator is typically de-
signed to operate in a single-stage countercurrent
ow with uidizing air supplied close to the bottom
of the vessel and spent catalyst distributed close to
the top, as shown in Fig. 1. Typical uidization velo-
city is in the range of 2 to 9 feet/s in turbulent to fast
uidization regimes. Partial combustion regenerators
have several advantages over total combustion regen-
erators. One is that more coke can be burned at a given
amount of blower air because partial combustion of
carbon requires less than the stoichiometric amount
of air. Another benet is that catalyst deactivation is
minimized by maintaining moderate regenerator tem-
peratures. A potential drawback of the partial combus-
tion regenerator is higher coke on regenerated catalyst
(CRC).
The phenomenon of afterburn in the partial com-
bustion regenerator is due mostly to the escape of oxy-
gen through the regenerator bed. This is likely to
happen when the spent catalyst is poorly distributed
and some areas of the regenerator have little coke to
burn. As a result, oxygen in the uidizing air in that
particular area is not completely consumed in the
regenerator bed. When oxygen reaches the freeboard
and reacts with carbon monoxide from other parts of
the regenerator to produce carbon dioxide, a large
amount of heat combustion is released. Since only a
little catalyst is present in the freeboard, the heat capa-
city is low and the temperature escalates quickly, which
is called afterburn. Severe afterburn, sometimes as high
as 150
F, can cause severe mechanical damage to the

regenerator cyclone system.
Both high CRC and afterburn in partial combustion
regenerators can be overcome by proper design of the
spent catalyst distributor and the air grid. Figure 6
shows one example of a modern spent catalyst distri-
butor designed by Khouw et al. (1994), which distri-
butes spent catalyst laterally through several
horizontal arms.
The most signicant advantage of total combus-
tion regenerators is that the CRC is low and catalyst
Figure 6 CCU spent catalyst distributor design. (From
Khouw et al., 1994.)
activity is higher. Potential drawbacks of total com-
bustion regenerators include higher catalyst deactiva-
tion owing to higher regenerator temperature, lower
coke burning capacity owing to the requirement of
excess air, and higher heat release owing to total com-
bustion reaction. Catalyst deactivation can be reduced
by using two-stage regeneration; see, e.g., Herbst et
al. (1989). In the two-stage regeneration, the rst
stage is operated at a moderate temperature to burn
o the hydrogen component on the coke, which has a
higher reaction rate, and some of the carbon compo-
nent. The second stage uses excess air to burn o the
carbon component on the coke at higher temperature
to carbon dioxide. Because of the absence of water
vapor in the second stage regenerator, catalyst deac-
tivation at high temperature can be minimized. Two-
stage regeneration is more expensive and more com-
plicated to operate. The coke burning issue is typi-
cally addressed by having a large air blower, which
adds to higher capital and operational costs. The
third issue in total combustion regeneration is the
heat balance, because more heat is generated in the
regenerator at the same coke burning rate when car-
bon is reacted to carbon dioxide instead of carbon
monoxide. If more heat is generated in the regenera-
tor than what is required for the reactor, a common
solution is to add a catalyst cooler to the regenerator
to remove the excess heat, as shown in Fig. 1. The
catalyst cooler is a heat exchanger that produces
steam while removing heat from regenerator. This
also adds to capital costs.
The phenomenon of afterburn also occurs in total
combustion regenerators. This is most likely to hap-
pen when the spent catalyst is poorly distributed and
a particular area of the regenerator has too much
coke. Although, on average, air is always in excess
in the total combustion mode, local areas could
have insucient air to burn all the carbon to carbon
dioxide if spent catalyst is not well distributed. As a
result, some carbon monoxide will escape through the
regenerator bed from these areas. Excess oxygen is
always present in the freeboard to react with carbon
monoxide and to produce carbon dioxide, causing an
afterburn problem. Afterburn in total combustion
regenerators can be overcome by using higher excess
air, promoter, or a better design of the spent catalyst
distributor, e.g., Fig. 6.
Residual cracking poses another challenge to regen-
erator design and operation. For a given conversion,
coke yield is much higher in residual cracking com-
pared to the conventional FCC because of heavier
feedstocks and higher metal contents. Catalyst coolers
are required, regardless whether the regenerator oper-
ates in partial or total combustion, in order to keep the
unit in heat balance. In addition, metal contamination
on equilibrium catalyst is high, which has detrimental
eects on the catalytic reactions. There are several
ways of dealing with this problem. These include selec-
tion of catalyst with metal trapping capability, o-line
removal of metal contamination, and management of
equilibrium catalyst to control the maximum concen-
tration of metals.
3.5 Third-Stage Separator
A modern FCC unit may further include a power
recovery system to recover energy from the regenerator
ue gas, which is a high volumetric gas ow at an
elevated temperature and a moderate pressure. It
becomes critical to control particulates in the ue gas
in order to protect the blades of an expander in the
power recovery system. However, most regenerator
cyclone systems have a limited capability of removing
particles around 10 microns.
A special design called the third-stage separator
(TSS) is shown in Fig. 7 (Dries et al., 2000) and
can be used for this purpose. It uses swirl tubes to
generate a very high centrifugal force and removes
practically all particles of 10 microns and larger
from the hot ue gas. In fact, the TSS design has
recently been improved so that it has a compatible
capability for catalyst removal as an electro static
precipitator (ESP) from the emission control view-
point. A TSS can be installed either inside or outside
the regenerator vessel.
3.6 Standpipe and Standpipe Inlet
Standpipe ow is the main driving force behind cata-
lyst circulation between the regenerator and the reac-
tor, which drives the entire FCC process. The objective
of a standpipe is rather straightforward. It is to build a
hydraulic head by holding a column of uidized cata-
lyst. This enables the catalyst to move in the direction
against pressure gradient, from low pressure at the top
to high pressure at the bottom of the standpipe. When
standpipes do not function properly, the FCC unit
cannot circulate catalyst at the design rate and the
unit is forced to cut feed rate. Even with such an
important role in the FCC process, standpipe ow
remains poorly understood. The existing theories on
standpipe ow, e.g., Leung (1977) and Chen et al.
(1984), are overly simplied.
Designing a standpipe is simple from mechanical
viewpoint, and yet very complex from hydrodynamic
viewpoint. A standpipe is typically a vertical, or nearly
vertical, column with a constant diameter, which is the
simplest mechanical structure of the entire FCC unit.
There are only two design elements that inuence the
standpipe ow besides the slide valve, which controls
the outlet end of the standpipe. The rst design ele-
ment is aeration along the standpipe. As pressure
increases down the standpipe, the gas phase within
the uidized catalyst is compressed and must be
made up by additional uidizing gas. Otherwise, a
long standpipe can become deuidized at the lower
end and restrict the catalyst ow. The complex nature
of the aeration design is that an overaerated standpipe
has a hydrodynamic behavior similar to that of an
underaerated standpipe, i.e., it cannot build proper
pressure, and the catalyst ow is unstable. In order
to distinguish the two cases, one must look beyond
the supercial symptoms. In the overaerated case the
instability is caused by the formation and release of
large bubbles, whereas in the underaerated case it is
caused by deuidization leading to a stick-slip ow.
The second design element is the standpipe inlet.
Because both stripper and regenerator uidized beds
are operated at relatively high gas ow rates, excessive
gas can be drawn into the standpipe through the inlet,
which is highly undesirable. To address the gas entrain-
ment issue, the conventional design is to add an inlet
hopper at the top of the standpipe. The typical inlet
hopper size is about 2 to 2.5 times the standpipe dia-
meter. The concept of the inlet hopper is to provide
enough residence time for small entrained bubbles to
coalesce into larger ones. Since larger bubbles have a
higher rising velocity, these large bubbles have a better
chance to escape from the inlet hopper, thus minimiz-
ing gas entrainment into the standpipe. The complex
nature of the design is that a poorly designed inlet
hopper can either entrain too much gas or cause deui-
dization. The overall hydrodynamic behaviors of the
two cases are again very similar, i.e., the standpipe
cannot build proper pressure and the catalyst ow is
unstable. If catalyst deuidizes in the inlet hopper, put-
ting a local aeration into the hopper can solve the
problem (Chen, 1986). However, if the hopper draws
too much gas into the standpipe, any additional aera-
tion will only worsen the situation. One particular pro-
blem with the conventional inlet hopper is that the
amount of gas entrainment cannot be controlled.
Thus a standpipe can suddenly become unstable as
the catalyst circulation rate is increased or decreased
beyond a certain rate.
A new standpipe inlet design has recently been
developed by Chen and Brosten (2001), as shown in
Fig. 8. Instead of using an inlet hopper, this new inlet
design uses a disk positioned directly below the stand-
pipe inlet. The concept is to trap uidizing gas from
below, causing a local partial deuidization above the
disk and forming a dense bed region near the proxi-
mity of the standpipe inlet. A small amount of uidi-
zation gas can be introduced above the disk to control
the uidization condition of the standpipe inlet region
independently of regenerator and stripper uidization
conditions, which are set by process requirements.
Commercial experience has demonstrated that both
catalyst circulation rate and stability can be signi-
cantly improved by simply replacing the conventional
standpipe inlet with the better inlet design of Fig. 8.
Figure 7 Third stage separator, blow-up of swirl tube, emissions improvements.
REFERENCES
Bedaw RE. Atomizing spray nozzle for mixing a liquid with a
gas. U.S. Patent 5,240,183, 1993.
Chen Y. A theoretical investigation of an aerated hopper
ow. J Chinese Inst Chem Eng 17:195, 1986.
Chen Y. Feed nozzle. U.S. Patent 6,387,247, 2002.
Chen Y, Brosten D. Standpipe inlet. U.S. Patent, 6,228,328,
2001.
Chen Y, Dewitz TS. Feed nozzle. U.S. Patent 5,794,857,
1998.
Chen Y, Rangachari R, Jackson R. Theoretical and experi-
mental investigation of uid and particle ow in a vertical
standpipe. Ind Eng Chem Fundam 23:354, 1984.
Chen Y, Brosten D, Nielson JW. Feed nozzle. U.S. Patent
5,979,799, 1999.
de Kruij GT, van Hattem A. Apparatus for solidsuid
separation. U.S. Patent 5,1998,005, 1993.
Dries H, Patel M, van Dijk N. New advances in third-stage
separators. World Rening 3034, Oct. 2000.
Geldart D. Types of gas uidization. Power Technol
7:285292, 1973.
Gwyn JE. Entrance, exit and wall eects on gas/particulate
solids ow regimes. Proceedings of 4th International Conf
on Circulating Fluid Beds, Somerset, Pennsylvania, 1993,
pp 679684.
Harauch J. Nozzle for use in uidized catalytic cracking.
European Patent EP 0-717-095-A2, 1995.
Herbst JA. Fluid catalytic cracking process and apparatus
for more eective regeneration of zeolite catalyst. U.S.
Patent 4,814,068, 1989.
Kau DA, Bartholic CA, Steves CA, Keim MR. Successful
application of the MSCC process. Paper AM-96-27,
NPRA meeting, 1996.
Figure 8 CCU standpipe inlet design. (From Chen and Brosten, 2001.)
Khouw FHH, van Poelje WM, van der Honing G. Process
and apparatus for distributing uids in a container. E.P.
Patent 0622116, 1994.
Leung LS. Design of uidized gassolids ow in standpipes.
Meyers RA. Handbook of Petroleum Rening Processes. 2d
ed. 1997, pp 3.33.99.
Parker WA, Dewitz TS, Hinds GP, Gwyn JE, Bilgic AH,
Hardesty DE. Apparatus for the separation of uid crack-
ing catalyst particles from gaseous hydrocarbons. U.S.
Patent 4,692,311, 1987.
Sadeghbeigi R. Fluid Catalytic Cracking Handbook. 1995.
Wilson JW. Fluid catalytic crackingtechnology and opera-
tion. 1997.
15
Applications for Gasiers and Combustors
Richard A. Newby
1 INTRODUCTION
Carbonaceous fuels are used throughout the world as
major energy sources. They exist primarily in solid
forms, or heavy liquid forms, containing signicant
process and environmental contaminants compared
to cleaner fuels such as natural gas. Coals and biomass
fuels are two of the many carbonaceous fuel types that
can be eectively processed in uidized bed reactors,
and they are currently of great industrial interest for
uidized bed gasication and combustion.
Fluidized bed gasiers and uidized bed combustors
for carbonaceous fuels are relatively simple chemical
reactors that can be utilized for several industrial appli-
cations that include chemical synthesis, process heat
supply, steam generation, and power generation.
Only relatively recently have carbonaceous fuel ui-
dized bed combustors and gasiers reached or
approached commercial status, with uidized bed com-
bustors gaining commercial acceptance earlier than
uidized bed gasiers.
These uidized bed technologies compete directly
with commercially mature entrained-particle and mov-
ing-bed gasiers and combustors (pulverized-fuel-red
boilers, and stoker-red boilers) not only with respect
to plant cost and eciency but also with respect to
plant availability. The uidized bed combustor and
gasier competitive positions relative to these more
mature combustor and gasier technologies are still
being established and depend, to a great extent, on
the specic application and the properties of the car-
bonaceous fuel being processed.
Fluidized bed gasiers dier from uidized bed
combustors in their basic reaction mechanisms, func-
tions, features, and operating conditions. They have in
common, though, several characteristics. Fluidized bed
gasiers and combustors operate at relatively low tem-
peratures, below the fuel ash agglomeration tem-
perature, and they have relatively uniform
temperature throughout their volume. Their designs
are tailored to the properties and behavior of the car-
bonaceous fuels to be processed. Finally, they must
overcome operational and performance diculties
associated with the phenomena of nonuniform ui-
dized bed gas mixing and bypassing, nonuniform fuel
feeding and mixing, bed overtemperature zones, bed
agglomeration, ash deposition, particle erosion, mate-
rials corrosion, and residual carbon conversion.
The many variations in design congurations and
features of the key types of uidized bed gasiers and
combustors have been developed from uidization
principles, chemical reactor principles, test observa-
tions, and empirical rules, coupled with conventional
engineering practice. Coals and biomass fuels represent
two contrasting extremes in fuel properties and beha-
vior, as well as in industrial application capacity, and
have thus evolved uidized bed processing technolo-
gies having somewhat diering congurations and fea-
tures. Other carbonaceous fuels, having their own
unique set of properties and behavior, have preferred
uidized bed processing congurations and features
that, to some extent, may be drawn from these two
carbonaceous fuel types by analogy.
2 FLUIDIZED BED GASIFIERS
Fluidized bed gasiers for coals and biomass fuels can,
in principle, generate reducing gas, liquid, and char
products for a variety of applications. Several uidized
bed gasier congurations have been proposed that
apply well-known principles of bubbling bed and cir-
culating bed uidization technologies. The uidized
bed gasier congurations nearing, or at, commercial
demonstration primarily generate a fuel gas product
and operate in uidization regimes and with maximum
feasible fuel capacities that dier considerably. The
design considerations applied to these uidized bed
gasier congurations reect these dierences, as well
as dierences in their design fuel properties.
2.1 Fluidized Bed Gasier Principles
The principles of design and operation of uidized bed
gasiers for carbonaceous fuels are relatively simple.
The challenge is to design the uidized bed gasier
for reliable operation, meeting its performance require-
ments while avoiding the consequences of the undesir-
able phenomena that typify uidized bed gasication.
2.1.1 Gasier Products and Applications
Fluidized bed gasication might be applied to carbo-
naceous fuels to produce steam or process heat, che-
micals, or electric power, or combinations of these.
The uidized bed gasier combines processed carbo-
naceous fuel (dried and sized) with oxidant, steam,
and/or an external heat source to produce a raw redu-
cing gas stream, a raw ash stream, or a raw char
stream, and possibly a raw liquid products stream.
Inert gases, such as nitrogen and recycled product gas
will also be fed to the gasier, functioning as solids
transport, aeration, and purge gases. These inert
gases can contribute signicantly to the makeup of
the raw reducing gas. Sorbent particles or inert bed
makeup particles may also be fed to the gasier.
The uidized bed gasier raw reducing gas stream
may be processed further to generate a syngas used
to produce chemicals, or a fuel gas used to generate
process steam, process heat, or electric power (Ghate
and Longanbach, 1988; Schobert et al., 1998). In some
gasication processes, liquid products (tars and oils)
might also be extracted from the raw reducing gas as
another chemical synthesis feed stock. The gasier raw
char might be applied for the synthesis of chemicals, or
combusted for steam, process heat, or electric power
generation. Raw ash may be processed further to
extract remaining fuel energy content (e.g., residual
carbon) or sensible heat, and to prepare it for disposal
or byproduct use.
Gasication is used in this chapter as a generic
term representing several alternative methods for con-
verting a carbonaceous fuel into a primarily gaseous
form having signicant heating value (steam reform-
ing, hydrogasication, hydropyrolysis, partial oxida-
tion, carbonization, thermal pyrolysis). Specic
gasication alternatives considered in this chapter are
designated as partial oxidation, carbonization,
and pyrolysis. Fuel gasication based on combining
substoichiometric oxidant with carbonaceous fuel to
combust partially the char and volatiles is termed
either partial oxidation or carbonization, depend-
ing on the extent of xed-carbon conversion that
occurs. Gasication under substoichiometric oxidation
conditions, intended to produce essentially no char, is
partial oxidation, under conditions intended to pro-
duce char as a product it is designated carbonation.
2.1.2 Functions and Requirements
The functional requirements of the uidized bed gasi-
er are to convert eciently and reliably the carbonac-
eous fuel into a raw reducing gas stream and a raw
char, or raw ash stream meeting all of the product
stream compositional, temperature, pressure, and
ow rate specications that are imposed on the gasier.
The uidized bed gasier is designed to promote a
reaction environment having good gasparticle con-
tacting, good particleparticle mixing, and relatively
uniform temperature conditions, and to avoid opera-
tional diculties resulting from the agglomeration,
deposition, erosion, and corrosion characteristics asso-
ciated with the gasication of carbonaceous fuels. The
design engineer is challenged to select fuel and sorbent/
inert particle feed sizes, uidization conditions, tem-
perature environment, and gasier reactant feed
stream proportions that avoid the nonidealities com-
mon to uidization (for example, deuidization, gas
bypassing, excessive particle attriation and elutriation,
particle segregation, generation of hot spots in the
dense bed and freeboard, and nozzle and drain plug-
ging).
The eciency of the gasier conversion might be
dened in a variety of ways, based on the energy con-
tent of the raw reducing gas, raw liquid products, raw
char, or raw ash streams, and the energy content of the
carbonaceous fuel feed stream. It is most common to
determine the cold gas eciency, dened as the ratio
of the energy content of the raw reducing gas at 25
C
to the carbonaceous fuel input energy and the frac-
tional fuel carbon conversion as measures of the gasi-
er performance. In some cases, there may also be a
specication on the maximum acceptable carbon con-
tent of the raw ash stream.
Availability is a measure of the time that the gasier
is able to operate within performance specications
relative to the time that the overall process plant
desires to operate. It is a critical factor in the commer-
cial feasibility of the uidized bed gasier technology.
2.1.3 Reaction Environment
Fluidized bed carbonaceous fuel gasication features
relatively low-temperature, generally less than 1050
C,
nonslagging gasication. It is also characterized by
nearly uniform gasier temperature throughout.
While uidized bed gasication can operate eectively
with a wide range of carbonaceous fuels, the gasier
operating conditions and design features must be spe-
cically selected to optimize performance with each
carbonaceous fuel type.
The simplied, overall reaction steps (some nonstoi-
chiometric relationships) are
Partial combustion of carbonaceous fuel char and
volatiles
Carbonaceous fuel {C, H, O, N, S,. . ., mineral
matterg oxidant {O
2
; N
2
g ! raw reducing
gas {CO; CO
2
; H
2
; H
2
O; CH
4
; N
2
, higher
hydrocarbonsg raw reducing gas contami-
nants {nitrogen compounds (NH
3
; HCN), sul-
fur compounds (H
2
S; COS), halogen
compounds (HCl), alkali metal compounds
(NaCl; NaOH; KCl; KOHg liquids {oils
and tarsg ash/char {mineral matter,
carbon; . . .g
Heterogeneous char reforming
Char fC; H; O; N; S; . . . ; mineral matterg
H
2
O=CO
2
! CO; H
2
; H
2
S; . . . ; ash
Homogeneous gas-phase reactions
CO H
2
O ! CO
2
H
2
C 2H
2
! CH
4
Boudouard reaction
2CO ! Csolid CO
2
Steam may be fed to the gasier in some cases to
provide a reaction environment that will promote
xed-carbon reforming and minimize the production
of ne carbon particles by the Boudouard reaction.
The exothermic partial combustion reactions produce
the reaction heat required to heat the input reactant
streams and perform the endothermic volatilization
and char reforming reactions. The extent of conver-
sion of char into raw reducing gas depends on the
operating conditions in the uidized bed (fuel-to-oxi-
dant feed ratio, fuel-to-steam feed ratio, bed tempera-
ture and pressure), and the char particle residence
time in the uidized bed. It has been observed that
limestone particles, added as sulfur sorbents, may also
catalyze char reforming reactions as well as induce tar
cracking (Simell et al., 1992; Agrawal and Haldipur,
1988).
Alternatively, uidized bed gasiers may be oper-
ated without oxidant supply, having only injected
heat and steam to pyrolyze the carbonaceous
fuel, driving o its volatile content and partially
reforming the char. The simplied overall reactions
are in this case,
Pyrolysis of the carbonaceous fuel
Carbonaceous fuel fC; H; O; N; S; . . . ; mineral
matterg heat! volatiles tars=oils char
Heterogeneous char reforming reactions
Char fC; H; O; N; S; . . . ; mineral matterg H
2
O
and H
2
! CO; H
2
; CH
4
; H
2
S; and ash
Homogeneous gas-phase reactions
Volatiles fC; H; N; S; . . . H
2
O and CO
2
! CO,
H
2
; . . . ; CO H
2
O ! CO
2
H
2
Boudouard reaction
2CO ! Csolid CO
2
The heat input may be direct heat input, for exam-
ple, injected hot solids or hot recycled product fuel gas;
or indirect heat input, as by high-temperature heat
transfer surfaces located in the uid bed.
With respect to uidization phenomena, the ui-
dized bed will consist of a distribution of char and
ash particles, with their relative amounts depending
on the carbonaceous fuel reactivity and the gasier
operating conditions. Sorbent or inert particles, if fed
to the gasier, may occupy the greatest portion of the
bed. The size distribution of particles in the gasier
bed, and the bed density prole in the gasier, depend
on a number of factors, including the carbonaceous
fuel and sorbent/inert particle feed size and density
distribution, the attrition properties of the char and
sorbent/inert particles after exposure to the gasier
environment, the gasier axial velocity prole, the exis-
tence of high attrition zones (e.g., high-velocity jets) in
the bed, and the use of cyclones to recycle elutriated
particles.
The gasication reaction phenomena in uidized
beds involve fuel devolatilization followed by the par-
tial combustion of volatiles and char separately with
oxygen, and char reforming by H
2
O and CO
2
in both
the bubble phase and the emulsion phase of the ui-
dized bed. This multiphase mass transfer and reaction
aspect of uidized bed gasication is an inherent fea-
ture of all uidized bed reactor systems. Carbonaceous
fuel devolatilization and combustion have similar,
characteristically high initial reaction rates and are
thus strongly inuenced by rate-limiting uidized bed
mixing and heat transfer phenomena. Char reforming,
though, is characteristically orders of magnitude
slower than devolatilization and combustion, and it
is a controlling reaction step in ecient uidized bed
gasication (Wen and Tonen, 1978).
The uidized bed may be designed and operated to
behave as a bubbling bed with a level of turbulence
depending on the uidization velocity and the particle
size and density distribution in the bed, or as a circu-
lating bed having fast uidization behavior.
Recirculation of nes collected by cyclones will gener-
ally be used even with bubbling bed gasiers.
2.2 Status of Fluidized Bed Gasier Technology
Gasication of carbonaceous fuels is a relatively old
technology. During the twentieth century, many ui-
dized bed gasiers have been proposed to gasify coals,
biomass fuels, petroleum products, and other carbo-
naceous fuels for the purpose of producing liquid
fuels, synthetic natural gas, syngases, chars, and fuel
gases. Some low-pressure uidized bed biomass fuel
gasiers are considered commercial.
Hebden and Stroud (1981) have reviewed many of
the uidized bed coal gasiers under development dur-
ing the 1970s. Similarly, Seglin and Bresler (1981)
review uidized bed low-temperature coal pyrolysis
technology of the 1970s. Low-temperature pyrolysis
in uidized bed reactors for the purpose of tar/liquid
and coke generation has not achieved commercial sta-
tus, and only limited activities are currently progres-
sing (Ghate and Longanbach, 1988).
The coal-fueled uidized bed gasier types nearing
demonstration and commercialization, and their appli-
cations, are characterized in Table 1. These represent
both operating facilities and projects in design or con-
struction. The three types of gasication are repre-
sented: partial oxidation, carbonization, and
pyrolysis, and bubbling bed uidization is the preva-
lent regime.
Table 2 presents a similar characterization of bio-
mass-fueled gasiers nearing demonstration and com-
mercialization. Partial oxidation at both elevated
pressure and low pressure are included, and both bub-
bling bed and circulating bed uidization are feasible
regimes. As with coal, the biomass gasiers are primar-
ily fuel gas applications for steam or combustion tur-
bine power.
Figure 1 illustrates the general schemes representa-
tive of the uidized bed coal gasication processes of
signicance today that are listed above, and Fig. 2
represents the key biomass fuel gasication processes.
The process schemes shown are oversimplied and
show only the major streams, gasier components,
and process interfaces relating to the primary uidized
bed conversions. Gasier stream interfaces are iden-
tied in the gures to emphasize the critical need for
satisfying all interface requirements. The gures also
indicate optional streams that are used depending
on the specic fuel properties and the application
needs. Nonmechanical valve locations for solids
recycle streams are identied, without aeration gas
being shown. In general, feed streams carrying solids
also require an inert transport gas (for example,
nitrogen or recycled product gas), and these transport
gases are not included in either Fig. 2 or 3. Inert gases
are also used at various places within the gasiers for
purging, aeration, local uidization, and stripping, and
these also are not shown in the gures.
Three schemes are shown in Fig. 1: (A) pressurized
bubbling bed partial oxidation of coal; (B) pressurized
bubbling bed carbonization of coal; and (C ) low-pres-
sure bubbling bed pyrolysis of coal. Figure 1A is repre-
sentative of pressurized bubbling bed partial oxidation
of coal. The sketch indicates that the bubbling bed
uidized bed gasier vessel cross-section is expanded
in the freeboard to minimize particle elutriation, and
this is generally true of all bubbling bed coal gasiers.
Elutriated nes recycling is used in general, with non-
mechanical valves to seal the recycle system and to
control the reinjection of nes into the uidized bed
gasier. The raw fuel gas will have a low heating value
of about 3.35.6 MJ/Nm
3
(LHV) for air-blown opera-
tion, and a medium heating value of about 8.211.2
MJ/Nm
3
(LHV) for oxygen-blown operation. The ves-
sel will drain a low-carbon ash product.
The gasier must interface with the pressurized oxi-
dant delivery system (air or oxygen-blown gasiers
being feasible for coals); with the pressurized coal
and sorbent preparation and feed systems; with the
raw fuel gas pressurized processing system; and with
the pressurized ash processing system. Coals may be
fed in either dry or paste form; the paste form is
claimed to improve feed reliability and cost. If the
application is to provide a pressurized fuel gas to re
a combustion turbine, the raw fuel gas must be well
cleaned of particulate and alkali species to protect the
combustion turbine. If the pressurized fuel gas is used
to re a process heater, such high levels of cleaning
may not be needed. Coals will generally result in a
raw fuel gas that needs to be desulfurized, and the
fuel gas ammonia content may be a concern for stack
gas NOx emissions.
Figure 1B is representative of pressurized bubbling
bed carbonization of coal. Here, cyclone nes recycling
is not used, and the carbonizer vessel and cyclone will
drain the raw char product. The gasier must interface
with the pressurized air delivery system; with the pres-
surized coal and sorbent preparation and feed systems;
with the raw fuel gas pressurized processing system;
and with the pressurized char processing system.
Coals may be fed in either dry or paste forms. The
raw fuel gas will have a low heating value of about
3.35.6 MJ/Nm
3
(LHV), similar to the raw fuel gas
issued from the pressurized air blown bubbling bed
partial oxidation gasier.
The third scheme shown, Fig. 1C, is a low-pressure
bubbling bed coal pyrolysis process based on a parallel
circulating bed char combustor that provides hot cir-
culating solids to the gasier. The bubbling bed pyro-
Table 1 Leading Coal Fueled Gasier Systems
Gasier type/Fluidization type/Pressure mode: Partial oxidation/Bubbling bed/Elevated
Developer Applications Description References
Kellogg
KRW gasier
Fuel gas for combustion
turbine power
20,000 MJ/h development unit testing
completed on wide range of coals;
Sierra Pacic Power
Co., 1994;
800,000 MJ/h coal demonstration plant
startup initiated 1998;
Demuth and Smith,
1998
air-blow operation;
21002400 kPa operating pressure
IGT
U-gas gasier
Coke oven fuel gas 105,000 MJ/h commercial plant
operational;
Bryan and Hoppe,
1998
air-blow operation;
160220 kPa operating pressure
Rheinbraum AG
HTW gasier
turbine power;
Lignite operation only;
450,000 MJ/h demonstration project
Renzenbrink et al.
1998
syngas for methanol completed;
air and oxygen-blown operations
Gasier type/Fluidization type/Pressure mode: Carbonization/Bubbling bed/Elevated
Foster Wheeler
Development Corp.
Fuel gas and char for
combustion turbine
10,000 MJ/h development unit testing
completed on range of coals;
Robertson, 1995
power 50,000 MJ/h pilot plant construction
completed 1998;
air-blow operation
British Coal Fuel gas for turbine
power;
40,000 MJ/h pilot plant operation
completed 1992;
Dawes et al., 1997
char for steam generation air-blow operation;
integrated gas cleaning and turbine
Gasier type/Fluidization type/Pressure mode: Pyrolyzer/Bubbling bed/Low
Tsinghua
University
Fuel gas for town gas;
char for pyrolysis heat
5,000 MJ/h development unit operated
on Chinese coals;
Lu et al., 1996
and steam air-blown operation;
100,000 MJ/h demonstration plant in
design
lyzer is uidized by steam, and its design and operation
must be closely integrated with the circulating bed
combustor. The pyrolysis gasier must also interface
with low-pressure coal and sorbent preparation and
feed systems. The low-pressure raw fuel gas can be
used for process heat or for steam generation, and it
must be processed to meet the requirements of those
applications. The raw fuel gas may also be cooled,
scrubbed and compressed to act as a combustion tur-
bine fuel gas. The product fuel gas, after cooling and
steam condensation, will have a medium heating value
of about 1216 MJ/Nm
3
(LHV). The uidized bed
Table 2 Leading Biomass-fueled Gasier Systems
Gasier type/Fluidization type/Pressure mode: Partial oxidation/Bubbling bed/Elevated
Developer Applications Description References
Carbona Corp
(IGT Renugas gasier)
turbine power
100,000 MJ/h pilot unit operated on
range of biomass fuels;
DeLong et al.,
1995
integrated gas cleaning;
750,000 MJ/h alfalfa-stem plantation
unit in design
Cratech gasier Fuel gas for small
combustion turbines
10,000 MJ/h pilot unit operating on
cotton gin mill waste;
Craig and Purvis,
1998
integrated gas cleaning;
100 kW turbine operation planned
Gasier type/Fluidization type/Pressure mode: Partial oxidation/Circulating bed/Elevated
Foster Wheeler
Energia Oy
turbine power
70,000 MJ/h demonstration testing
ongoing since 1995;
Stahl and Neergaard,
1998
Bioow gasier integrated with 6 MWe combustion
turbine operation;
wood chip fuel
Gasier type/Fluidization type/Pressure mode: Partial oxidation/Circulating bed/Low
Lurgi gasier Fuel gas for process
heat;
350,000 MJ/h unit for cement kiln fuel
gas in operation;
Hirschfelder and
Vierrath, 1998;
fuel gas for combustion
turbine power
160,000 MJ/h unit for combustion turbine
in design;
DeLange and Barbucci,
1998
air-blown operation
Foster Wheeler
Energia Oy
Fuel gas for process
heat and steam
55,000230,000 MJ/h commercial
units in operation;
Nieminen and Kivila,
1998
lime kiln process heat and steam;
air-blown operation
Termiska Processor
AB (TPS) gasier
Fuel gas for process
heat and steam;
Two, 50,000 MJ/h units in commercial
operation for steam generation;
Barducci et al., 1995;
Pitcher et al., 1998
turbine power
300,000 MJ/h eucalyptus unit designed
for combustion turbine plant in Brazil;
100,000 MJ/h wood chip unit for
combustion turbine in UK in
construction
Gasier type/Fluidization type/Pressure mode: Pyrolysis/Circulating bed/Low
Battelle gasier Fuel gas for process
heat and steam;
145,000 MJ/h wood chip in
operation for steam generation;
Farris et al., 1998
turbine power
future conversion for combustion
turbine operation planned
combustor produces substantial steam for export, and
a separate coal stream may be fed to the uidized bed
combustor to raise additional steam.
Figure 2 shows representations of biomass fuel ui-
dized bed gasiers. Four types are illustrated: (A) pres-
surized bubbling bed partial oxidation of biomass
fuels; (B) pressurized circulating bed partial oxidation
of biomass fuels; (C) low-pressure, circulating bed par-
tial oxidation of biomass fuels; and (D) low-pressure
circulating bed pyrolysis. For all of these, proper bio-
mass fuel drying and sizing is critical to their successful
operation.
Figure 2A is representative of pressurized bubbling
bed air blown partial oxidation of biomass fuel. The
sketch indicates that nes recycling might be used, with
nonmechanical valves to control the reinjection of nes
into the uidized bed gasier. The vessel will drain a
low-carbon ash product. The sketch also suggests that
overbed air injection might be used for fuel gas partial
oxidation as a means for fuel gas tar destruction. The
gasier must interface with the pressurized air delivery
system (only air being considered feasible with biomass
fuel gasication applications); with the pressurized fuel
and inert feed systems; with the raw fuel gas pressur-
ized processing system; and with the pressurized ash
processing system. Biomass fuels will be fed only in
dry form.
The application is to provide a pressurized fuel gas
to re a combustion turbine, and the raw fuel gas must
be well cleaned of particulate and alkali metal species
to protect the combustion turbine. Here, it is expected
that tar generation will be small, and special provision
for tar cracking or removal will not be needed. The fuel
gas must be cooled to the extent needed to achieve this
cleaning, probably to a maximum temperature of
650
C. Fuel gas ammonia content may be a particular

concern for NOx stack emissions when using some
biomass fuels. Biomass fuels generally will not require
fuel gas desulfurization, but some agricultural planta-
tion crops do have signicant sulfur content and may
Figure 1 Fluidized bed, coal gasication schemes nearing demonstration.
require gas desulfurization. The raw fuel gas will have
a low heating value similar to those raw fuel gases
produced in pressurized air blown coal gasiers.
Figure 2B is a pressurized circulating bed air blown
partial oxidation of biomass fuel. Here biomass fuel is
fed as a dry solid into a circulating bed of inert parti-
cles. While circulating bed gasication may not be
practical for most coals, the high reactivity of biomass
fuels makes circulating bed gasication an eective
process. The raw fuel gas must be processed so that
it can serve as a combustion turbine fuel gas. Again, it
is expected that tar generation will be small, and spe-
cial provision for tar cracking or removal will not be
needed. Large bed circulation rates are used to main-
tain vigorous gasparticle mixing and nearly uniform
temperatures throughout the gasier. The raw fuel gas
heating value and carbon utilization will be in the same
range as the other air blown gasiers.
The third scheme, Fig. 2C, is low-pressure (near
atmospheric pressure) circulating bed air blown partial
oxidation of biomass fuel. Here low-pressure biomass
fuel is fed as a dry solid into a circulating bed of inert
particles uidized by low-pressure air and steam. The
process challenges of pressurizing and feeding biomass
fuels at low pressure are much less than in the two
previous high-pressure schemes. The low-pressure
raw fuel gas may be cooled to reduce hot gas piping
expense and used to re a conventional boilersteam
generator with little gas cleaning. Alternatively, the
raw fuel gas might be cooled and scrubbed so that it
can be compressed to provide a combustion turbine
fuel gas. It is expected again that tar generation will
be small in most cases.
The fourth scheme, Fig. 2D, is a low-pressure bio-
mass pyrolysis process based on parallel circulating
beds, one a pyrolyzer and the other a char combustor.
The high reactivity of biomass fuels make this process
feasible in relatively compact vessels. The raw fuel gas
will have a medium heating value of about 13.016.7
MJ/Nm
3
(LHV). The low-pressure raw fuel gas can be
used to re a conventional boiler after cooling, with
little fuel gas cleaning; or it may be cooled, scrubbed,
and compressed to act as a combustion turbine fuel
gas. The concentrated pyrolysis fuel gas will generally
contain relatively high tar content, and tar cracking, or
recovery, will need to be incorporated into the raw fuel
Figure 2 Fluidized bed, biomass fuel gasication schemes nearing demonstration.
gas processing system for the combustion turbine
application. Fuel gas tar content may also be a process
concern for the boiler fuel gas application.
New techniques for pyrolysis of biomass fuels by
indirect heat input are currently at various stages of
development and demonstration for small-capacity
fuel gas supply applications: the BrightStar
Technologies low-pressure steam uidized pyrolyzer
using fuel ring to heat the windbox (Chau and
Rovner, 1995); the MTCI low-pressure steam uidized
pulse combustion pyrolyzer with immersed heat trans-
fer surfaces in the bed (Monsour et al., 1995); and the
Iowa State University cyclic (gasicationoxidation),
steam uidized pyrolyzer with high-temperature
phase change materials contained in tubes immersed
in the bed (Stobbe et al., 1996). These new pyrolyzer
congurations all have signicant materials, scale-up,
and process integration challenges to overcome.
Generally, the bubbling bed gasiers have mean bed
particles characteristic of Geldart group B classica-
tion (Geldart, 1973), with no overlap into the A or D
groups. The pressurized bubbling bed gasier appears
to approach the turbulent uidization regime much
more closely than does low-pressure bubbling bed gasi-
cation, indicating the potential for more vigorous ui-
dization, but also more extensive particle elutriation.
The low-pressure and pressurized circulating bed gasi-
ers operate well above the terminal velocity, although
the low-pressure operating region indicates that low-
pressure circulating bed gasiers may operate very near
the transition between turbulent and fast uidization.
The single gasier module maximum fuel capacity is
sometimes applied as an indicator of the feasibility of
competing congurations. A key factor inuencing the
single gasier module fuel capacity of uidized bed
gasication technologies is the cross-sectional area of
Figure 3 Comparable uidized bed pressurized partial oxidizer and carbonizer.
the uid bed gasier. The major parameters controlling
the cross-sectional area of the uid bed gasier are
The oxidant type (air or oxygen), or pyrolysis heat
source type
Operating pressure: low-pressure or pressurized
Gas velocity in the freeboard
Fuel properties
Fluidized bed gasiers are generally cylindrical vessels,
even for low-pressure operation. A 5 m vessel diameter
is about the limit for shop fabrication and road/rail
shipment of pressure vessels. The uidized bed gasier
vessels require either eld fabrication or the use of
multiple parallel gasiers, as their fuel thermal input
capacity exceeds limiting capacity.
The pressurized circulating bed biomass fuel partial
oxidation gasier is the most compact of the gasiers;
and the low-pressure circulating bed biomass fuel pyr-
olysis gasier is relatively compact despite its low oper-
ating pressure. The pressurized, bubbling bed air
blown coal partial oxidation gasiers are fairly com-
pact, with the oxygen blown gasier having very large
capacity in a single module. The low-pressure bubbling
bed coal pyrolysis gasier is very large in diameter and
will have only limited coal capacities.
2.3 Fluidized Bed Gasier Design Considerations
The design of uidized bed gasiers requires that trans-
port models be developed that apply the conservation
principles for mass, energy, and momentum to some
degree of fundamental detail. The mass and energy
balances are closely coupled, and their solutions,
applying empirical reaction kinetic and multiple
phase mixing models, provide estimates of solids and
gas composition and temperature proles and input
and output stream conditions. The momentum bal-
ances provide the gasier pressure prole and total
pressure drop when applied with appropriate uidized
bed phase density models. These empirically based
models provide the means needed to size the gasier
equipment and scale pilot plant data.
The design of uidized bed gasiers involves the
selection of several interrelating design, operating,
and performance parameters and requires the consid-
eration of performance and cost trade-os for any spe-
cic application. The type of uidized bed gasier, the
fuel properties, and the application requirements
strongly inuence the uidization parameter values
selected for the design. To make such designs properly,
for all of the types of gasiers, fuels, and applications
described in this chapter, it is currently necessary to
apply highly empirical uidized bed reactor models
that are closely calibrated with pilot plant test data,
thermodynamic estimates, and cold ow modeling
observations (Yang et al., 1995). Iterative design
codes are needed to consider all of the design para-
meters, fuel properties, process performance require-
ments, and trade-os.
2.3.1 Gasication of Coal
Figure 3 represents a pressurized coal fueled bubbling
bed partial oxidizer and a carbonizer of comparable
fuel thermal input capacity, showing their major com-
ponents and features, and relating to the respective
process schemes shown in Figs. A and B. The gasier
vessels, like all uidized bed gasier vessels, are
uncooled refractory lined pressure vessels having
appropriately located nozzles for gas and solids inlet
and outlet ows and manways for inspection and
maintenance.
The bubbling bed partial oxidizer is separated axi-
ally into several perceived functional zones: the com-
bustion zone, the gasication zone, the ash/char
separation zone, and the freeboard zone. Each of
these zones is provided with a diameter resulting in
acceptable uidization and mixing at the local ow
conditions, and a height sucient to perform the
zone functions eectively.
When gasifying caking coals, the coal is introduced
into the combustion zone via a vertical jet, operating at
2550 m/s nozzle velocity, that intensely mixes the feed
streams of oxidant, coal, recycled fuel gas, and steam
with internally circulating char, ash, and sorbent par-
ticles entrained into the jet. This results in the complete
and rapid consumption of oxygen, the pyrolysis and
partial combustion of char and volatiles, and the gen-
eration of a hot zone in the jet that allows ash agglom-
eration, if it is desirable with the coal being gasied.
Noncaking coals may be fed more simply by mechan-
ical or pneumatic means at the vessel wall, since coal
agglomeration is not a concern in this case, but high
reactivity coals might still be fed through a central jet
to avoid nonuniform volatiles release in the bed. The
oxidant may also be injected as jets from nozzles dis-
tributed around the vessel wall but must always be
injected with vigorous mixing and solids entrainment
to avoid local hot spots and excessively large ash
agglomerates or deposits.
In the ash/char separation zone, if it is desirable to
incorporate such a zone, controlled uidization results
in ecient segregation of the larger ash agglomerates
from the char particles, acting as a mechanism to
maintain a high char content in the bed but a low
carbon content in the drained ash. An alternative
ashchar separation method is pneumatic stripping of
char from the draining ash agglomerates. Bottom bed
drainage is performed with countercurrent cooling of
the draining ash and heating of injected gases. Bed
drainage must be controlled to provide and maintain
sucient gasier bed depth. The feeding of limestone
to the gasier bed catalyzes the steam/carbon dioxide
char reforming reactions and makes charash separa-
tion less important, with the carbon content in the bed
being lowered. Coals having high reactivity, such as
many subbituminous coals and lignites, may not
require ash agglomeration or special ashchar separa-
tion features to be used eciently, and may be gasied
at lower bed temperatures.
The gasication zone is an expanded lower velocity
zone above the combustion zone where the slower
charsteam/carbon dioxide reforming reactions occur
that result in ecient utilization of the coal char. The
reaction conversion of this zone is greatly inuenced by
the large bubbles issued from the combustion zone jet.
The freeboard is both a disengaging zone for the very
large bubbles that break the bubbling bed surface and
a zone for homogeneous fuel gas conversions to occur.
Tar formation is generally relatively small, but various
ash-related deposits can form within the gasier vessel.
The recycle cyclone, if this is used to reduce carbon
losses, must be placed at a sucient elevation above
the relatively dense uidized bed to provide a su-
ciently long standleg to balance the circulation loop
pressure prole. The bed elutriation will tend to result
in recycled ne particles having low bulk density com-
pared to the dense uidized bed. Fine recycled particles
must be injected into the combustion zone eectively
to provide additional carbon conversion.
Figure 3 shows that the equivalent pressurized air
blown bubbling bed coal carbonizer vessel might be
about half the total height of, and slightly smaller in
diameter than, the coal partial oxidation vessel due to
its primary objective of generating a char product
rather than a low-carbon ash. Similar central jet feed-
ing of coal may be used into a combustion zone, but
segregated charash draining is not required. Simpler
top bed overow drainage can be used to control the
bed height. Generally there is also no need to recycle
overhead nes back into the carbonizer vessel, this
stream being primarily product char. The uidized
bed carbonizer may operate at a lower temperature
than the uidized bed partial oxidizer. Thus the bub-
bling bed coal carbonizer design is signicantly simpler
than the bubbling bed coal partial oxidizer design.
Several design decisions and estimates that must be
addressed to generate a reliable and ecient pressurized
bubbling bed coal partial oxidizer or carbonizer are
Fluidized bed temperaturethis represents a trade-
o between carbon utilization, oxidant and
steam consumption, raw fuel gas heating value,
and the undesirable possibilities of deposits,
agglomerates, nozzle and drain plugging, and
excessive tar and ammonia generation.
Coal and sorbent feed size distributions.
Oxidant/coal, steam/coal, recycle fuel gas/coal feed
ratios
Gasication and combustion zone uidization velo-
city
Freeboard zone velocity and height to minimize
particle carryovera trade-o inuencing equip-
ment cost, particle recycle system capacity, and
gasier performance.
Oxidant injection design.
Steam and recycle fuel gas distribution for local
uidization to avoid dead zones in the bed and
to minimize local hot spots resulting in uncon-
trolled ash agglomeration.
Coal feed location(s) and transport method.
Recycled ne particles feed location and transport
method.
Combustion zone height and gasication zone
height required for eective partial oxidation,
carbonization, and desulfurization.
Special features for limiting bed drainage carbon
losses, if needed to meet performance require-
ments.
Overall gasier and recycle loop pressure drop pro-
les and heat losses.
Eective turndown and control methods.
Design to accommodate multiple fuel types.
Matching gasier input and output conditions and
requirements with compatible capabilities at
interfacing equipment and systems, and achiev-
ing overall gasierprocess integration.
Low-pressure bubbling bed pyrolysis of coal, as shown
schematically in Fig. 1C, must apply many of the same
types of design considerations as those listed above for
pressurized bubbling bed coal partial oxidation, but it
diers in its characteristics, performance, and behavior
expectations in several ways.
The performance of the low-pressure bubbling bed
pyrolysis gasier will be extremely sensitive to the
volatile content of the coal and its charsteam
reforming reactivity.
The coal feed size distribution may need to be
restricted in its top and bottom sizes, for
example, limited to 0.53.2 mm, to avoid ne
particle losses and to obtain good pyrolysis con-
version.
No oxidant supply is required, and uidization is
provided by steam, as well as by the evolution
of fuel gas from the pyrolyzing coal particles
a major trade-o exists between factors relating
to steam consumption and uidization velocity.
The pyrolysis gasier temperature (greater than
800
C for most coals) is controlled by the char

fueled circulating bed combustor temperature
(greater than 900
C) and the solids circulation

rate, and the gasier bed temperature may be
limited to lower than optimum values by ash
agglomeration limitations in the circulating bed
combustor.
The circulating bed media may be coal ash particles
or limestone particles, depending on the process
desulfurization needsthe limestone reaction
conversions in this cyclic oxidizingreducing
environment may have reaction heat eects that
inuence the process performance.
The low-pressure bubbling bed pyrolysis unit will be
much shallower than the pressurized bubbling
bed partial oxidation gasier to meet its pyrolysis
conversion and pressure drop constraints.
Rapid mixing of hot circulating solids, fed above the
pyrolysis bed, with the coal feed, is required, as
well as good distribution of coal to the pyrolysis
bed so that nonuniform volatile release to the
freeboard does not occur.
Greater tar and ammonia content in the bubbling
bed pyrolysis fuel gas is expected than in the
pressurized bubbling bed partial oxidation coal
gasier, although this may be moderated by the
large steam content.
The oxidizingreducing interfaces between the gasi-
er and combustor must be properly sealed and
purged.
2.3.2 Gasication of Biomass Fuels
A pressurized bubbling bed partial oxidizer for bio-
mass fuel, shown schematically in Fig. 2A, will be
designed much like the pressurized bubbling bed coal
partial oxidizer in Fig. 3. The higher reactivity of the
biomass fuel relative to coal, and the probable elimina-
tion of in-bed desulfurization with biomass fuels,
might result in shorter combustion and char gasica-
tion zones if the gasier operating temperature is not
limited to low values due to bed agglomeration poten-
tial. The high reactivity of biomass fuels makes rela-
tively uniform fuel feed distribution even more
important than it is with coals. Reactive biomass
fuels may not require the use of ne particle recycling
or charash separation features. Secondary air might
be injected into the freeboard of the biomass fuel gasi-
er to promote tar destruction through partial fuel gas
combustion. The biomass fuel bubbling bed would
consist largely of inert sandlike particles containing
small amounts of biomass fuel char and ash.
The same design considerations must be applied
with biomass fuels as with coals. Biomass fuel gasica-
tion is relatively simple due to the generally highly
reactive nature of biomass fuels. The greatest risk in
biomass fuel gasication is associated with ash, or ash-
inert particle agglomeration within the bed that may
lead to deuidization and plugging of nozzles and
drains. With the tendency for biomass fuel bed
agglomeration, relatively low partial oxidation tem-
peratures may need to be selected that result in rela-
tively high carbon losses. The tendency for greater
formation of tars in the raw fuel gas, and increased
gasier drain material carbon content, as the tempera-
ture is decreased, must be weighed against the tendency
for greater bed agglomeration, and possibly greater
alkali vapor release, as the gasier temperature is
increased. The reliable, controllable preparation, pres-
surization, and feeding of biomass fuels are signicant
challenges to be fullled to make the technology a
commercial success.
Pressurized circulating bed air blown partial oxida-
tion of biomass fuels, shown schematically in Fig. 2B,
is feasible because of the high reactively of biomass
fuels. Proper biomass fuel drying and fuel and inert
bed material particle size control (size reduction or
pelletization) are important to the gasier perfor-
mance. The biomass fuel is fed several meters above
the air distributor to a relatively low-density zone of
the fast uidized bed. The initial pyrolysis of the bio-
mass fuel, with some partial combustion of the vola-
tiles and char, will occur rapidly near the fuel injection
point, and the vessel diameter may be expanded at this
point owing to the large evolution of biomass fuel
volatiles. Steam reforming of char particles will con-
tinue throughout the fast bed riser, with the conversion
becoming less eective at the lower density higher ele-
vations of the fast bed. The recirculating solids, con-
taining unconverted char and sand, are injected into
the turbulent, dense uidized bed at the base of the
vessel, near the air distributor, so that unconverted
char can be burned.
Low-pressure circulating bed partial oxidation of
biomass fuels, shown schematically in Fig. 2C, is con-
ceptually very similar in design to the pressurized cir-
culating bed partial oxidizer. It faces similar
performance constraints and design considerations to
those for the pressurized circulating bed partial oxidi-
zer. The low-pressure partial oxidizer uidization and
reaction kinetics are greatly inuenced by its lower
pressure operation, but it can operate eectively with
fuel and inert material feed particle sizes, velocities,
temperatures, and bed depths that are selected to pro-
vide acceptable performance.
Low-pressure circulating bed pyrolysis of biomass
fuels, shown schematically in Fig. 2D, diers signi-
cantly from coal pyrolysis in that the volumetric evolu-
tion of fuel gas from the pyrolyzing biomass fuel
particles is sucient to uidize the inert bed media in
the gasier vessel. Only local uidization with steam,
or nitrogen, is needed to avoid deuidized, nonmixing
locations in the vessel. The circulating bed pyrolyzer
and the char fueled circulating bed combustor are both
adiabatic reactors. A balance is required between the
carbon conversion in the pyrolyzer at its selected oper-
ating temperature, the carbon feed rate to the circulat-
ing bed combustor and its operating temperature, and
the solids circulation rate between the vessels. The rela-
tive complexity of this thermal and chemical balance
scheme, and its high potential for raw fuel gas, are
design issues. Proper sealing and stripping at the redu-
cingoxidizing interfaces is required.
3 FLUIDIZED BED COMBUSTORS
Fluidized bed combustors for coals and biomass fuels
are applied for steam generation, process heat supply,
and electric power generation. They apply the princi-
ples of bubbling uidized beds, and circulating ui-
dized beds, and they operate at both atmospheric
pressure (AFBC) and elevated pressures (PFBC).
AFBC congurations, both bubbling bed and circu-
lating bed, have achieved commercial status, and
PFBC is in an early commercial scale-up phase. The
design considerations and the characteristics of ui-
dized bed combustors for coals and biomass fuels
reect the signicant dierences between these two
types of fuels.
3.1 Fluidized Bed Combustor Principles
Fluidized bed combustors burn carbonaceous fuels to
produce steam, process heat, or electric power. The
major distinction of uidized bed combustors com-
pared to other combustor types is that carbonaceous
fuel combustion occurs at a relatively low temperature
of 760930
C, within an almost uniform temperature

reactor environment. Carbonaceous fuel is fed into a
uidized bed consisting primarily of sulfur-sorbent
(limestone) particles, or inert particles, and containing
smaller portions of fuel char and ash particles. The bed
is uidized by air, fed in excess of that needed for
complete fuel combustion, to perform combustion
within the relatively well-mixed, relatively uniform
temperature of the uidized bed. The uidized bed
combustor may operate at near-atmospheric pressure
for steam generation, process heat production, or
Rankine cycle electric power generation; or at elevated
pressure for combined cycle power generation.
The uidized bed temperature, high enough to per-
form ecient combustion and sulfur removal, but low
enough to avoid bed agglomeration, is maintained by
rapidly removing heat to generate steam through heat
transfer surfaces immersed in the uidized bed. A raw
combustion gas is produced that is processed, primar-
ily removing entrained particles, so that the product
combustion gas can have further energy extracted
from it. This gas processing is considerably less com-
plex than the gas processing associated with uidized
bed gasication. Bed drain solids, a mixture of the
carbonaceous fuel ash and sulfur-sorbent products,
are produced that may require further processing
(heat removal, chemical/physical processing) before
disposal or by-product use.
The functional requirements for uidized bed com-
bustors are generally more complex than those for ui-
dized bed gasiers: they must eciently and reliably
combust the carbonaceous fuel, achieve specied sulfur
removal performance with economic consumption of
sorbent, generate a raw combustion gas meeting ow,
composition (SO
2
; NO
x
; CO; O
2
, particulate), and
temperature specications, generate steam meeting
ow, pressure, temperature, and quality requirements,
and generate a raw ash stream meeting its ow, com-
position, and temperature constraints. The uidized
bed combustors overall thermal eciency is the ratio
of the recoverable thermal energy content of the out-
put streams (steam, raw combustion gas, raw ash) to
the carbonaceous fuel input energy. This thermal e-
ciency is strongly inuenced by the carbonaceous fuel
properties and the nature of the process application
and is sensitive to unburned carbon losses. The com-
bustion ineciency is the fuel value of the waste solids
streams (primarily based on their carbon content) over
the carbonaceous fuel input.
While uidized bed combustion may be performed
in a staged-combustion arrangement, having both
reducing and oxidizing zones, the overall reaction
steps are representative of oxidizing conditions (non-
stoichiometric relationships):
Combustion of carbonaceous fuel
Carbonaceous fuel {C; H; O; N; S; . . . ; mineral
matterg air fO
2
; N
2
g ! raw Gas fCO
2
; H
2
O;
N
2
, excess O
2
g raw gas contaminants {CO,
SO
2
; SO
3
; NO
x
, halogen compounds [HCl],
alkali metal compounds NaCl; Na
2
SO
4
; KCl,
K
2
SO
4
g ash {mineral matter, residual carbon}
Limestone sulfation
Limestone fCaCO
3
; MgCO
3
, mineral matter g
SO
2
O
2
! CaSO
4
; CaO; CaCO
3
; MgO; CO
2
The uidized bed combustion phenomenon involves
carbonaceous fuel devolatilization followed by the
combustion of volatiles and char separately with oxy-
gen in the dense bed bubble and emulsion phases, and
within the dilute phases of the uidized bed (Leckner,
1998). In general, both reducing and oxidizing regions
exist within the uidized bed combustor even with vig-
orous mixing, especially in the vicinity of carbonaceous
fuel feed points. Similarly, limestone sulfation involves
calcination and contacting of limestone particles with
sulfur oxides and oxygen in the bed emulsion and bub-
ble phases.
Fluidized bed combustors may be designed and
operated as either bubbling uidized beds or as circu-
lating uidized beds. They may also be operated near
atmospheric pressure (AFBC) or at elevated pressures
(PFBC) where the product combustion gas is expanded
through a turbine to generate electric power.
Bubbling bed combustors normally operate with
Geldard group B particles that provide generally
slow bubble behavior with good bubble-to-emulsion
phase gas mixing. Circulating bed combustors, with
ner particle sizes and higher velocities, operate in
the fast uidization regime, providing excellent gas
particle mixing due to the large slip velocity between
gas and particle clusters characteristic of fast uidiza-
tion.
The pressurized bubbling bed combustors (bubbling
PFBC) appear to approach the turbulent uidization
regime much more closely than do the low-pressure
bubbling bed combustors (bubbling AFBC). The
low-pressure and pressurized circulating bed combus-
tors (circulating AFBC and circulating PFBC) operate
well above the terminal velocity, although circulating
AFBC may operate just above the transition between
turbulent and fast uidization.
The low-pressure and pressurized uidized bed
combustors operate in similar regimes, but with
dierent uidization characteristics with regard to
mixing and particle elutriation and entrainment.
Changes in operating pressure also result in signicant
changes in combustion and sulfation reaction kinetics.
Devolatilization and combustion, being kinetically
rapid even at the relatively low temperature of uidized
bed combustors, are generally limited by large-scale
mixing phenomena in the dense bed and dilute regions
(gasgas, and gasparticle mixing). Combustion of CO
and residual carbon represent the limiting reaction
steps.
Figure 4 illustrates the bubbling bed and circulating
bed congurations and is representative of both AFBC
and PFBC. The process schemes shown are oversim-
plied and show only the process steps relating to the
uidized bed conversions. Bubbling bed combustors
are depicted in the gure to be larger in cross section
but shorter in height than circulating bed combustors.
This results from several factors: bubbling bed com-
bustors operate at much lower uidization velocities,
and with much denser beds than circulating bed com-
bustors; bubbling bed combustors normally pass all of
the combustion air through the bubbling bed, while
circulating bed combustors use staged combustion,
where only about half of the combustion air is used
as primary air in a primary turbulent bubbling bed,
and the remaining air is injected into the fast bed
zone as secondary air.
Bubbling bed combustors conduct almost all of
their combustion and, simultaneously, considerable
heat transfer within the relatively dense bubbling
bed. The freeboard is primarily used to complete
CO and volatiles combustion and to disengage solids
from the raw combustion gas before the gas reaches
the convective heat transfer surface. Elutriated parti-
cles that are captured by cyclones may be recycled, at
relatively low temperature, to the uidized bed com-
bustor.
In contrast, circulating bed combustors perform
partial combustion and devolatilization in the primary,
reducing bed zone, and complete combustion in the
relatively dilute, but well mixed, fast uidized bed
zone. Heat transfer surface is distributed throughout
the secondary fast uidized bed combustor and the
solids recycle leg (water-cooled cyclone and external
heat exchanger), requiring relatively large furnace
heights similar to those of pulverized fuelred com-
bustors. Solids circulation rates are very large, so that
little temperature dierence occurs across the full
height of the circulating bed.
In AFBC applications, the convective heat transfer
surface is arranged so that maximum steam genera-
tion is accomplished by cooling the product combus-
tion gas down to the lowest permissible stack
temperature. In PFBC operation, the product com-
bustion gas is maintained at its highest permissible
temperature, that of the uidized bed combustor, so
that maximum turbine power and eciency can be
obtained. PFBC operates at bed temperatures
(815870
C) much lower than conventional combus-

tion turbine inlet temperatures (11501430
C), so the
turbine performance in PFBC is relatively poor, but it
still contributes signicantly to the power plant gen-
erating output and eciency.
Another form of PFBC, topped-PFBC, has been
devised to maximize the gas turbine performance. In
topped-PFBC, a uidized bed carbonizer (see Sec. 3),
operated at 815930
C, generates a lowheating value

fuel gas and a char product. The char is combusted in a
pressurized uidized bed combustor at 870930
C,
producing a hot combustion product gas stream con-
taining excess oxygen. Both uid bed reactors use in-
situ sulfur removal with limestone. The carbonizer fuel
gas and the uidized bed combustor product gas com-
bine in a low-NO
x
topping combustor, resulting in an
expansion gas temperature characteristic of modern
heavy-duty gas turbines. The two parallel gas streams
from the carbonizer and combustor are cleaned at, or
near, their generation temperatures so that the com-
bined streams meet both the gas turbine protection
requirements and the power plant environmental emis-
sion standards. The hot gas cleaning system performs
Figure 4 Fluidized bed combustion process schemes.
particle removal and alkali removal functions, if alkali
metal content is a concern.
3.2 Status of Fluidized Bed Combustors
AFBC is a commercial technology that is widespread
across the world. Coal-fueled AFBC competes with
conventional stoker-red boilers, with practical capa-
city of 20,000 to 300,000 MJ/h fuel energy input, all the
way up to the lower end of base-load pulverized fuel
red boilers, having practical capacities of 250,000 to
9,000,000 MJ/h. Among the many claimed practical
advantages of AFBC are the following:
Feed coarse particle sizes of crushed carbonac-
eous fuel compared to pulverized fuelred com-
bustors, thus not requiring expensive, power-
consuming fuel pulverization.
Feed broad particle size ranges of carbonaceous
fuels, thus not requiring the elimination of ne
particles as set by stokers.
Operate at low temperatures, eliminating ash slag-
ging, reducing ash deposition, and allowing a
greater variety of carbonaceous fuels to be eec-
tively combusted.
Can accomplish in-combustor desulfurization with
limestone, generating a dry solid waste product
and produce low NOx emissions without special
controls.
Characterized by high uidized bed-to-surface heat
transfer coecients that result in compact heat
exchanger surface even with the lower tempera-
ture driving forces relative to stokers and pulver-
ized fuelred combustors.
Combustion eciency higher than stokers and com-
parable to pulverized fuelred combustors.
Biomass fueled AFBC is commercial at small capaci-
ties commensurate with biomass fuel supplies but is
not as widespread as coal-red AFBC. PFBC with
coals is a relatively new commercial power generation
technology still being scaled up to base-load electrical
utility capacities. No activity on PFBC with biomass
fuels is underway, as it is not appropriate for the lim-
ited biomass fuel supplies available, and given the
practical problems of pressurized biomass fuel opera-
tion (fuel feeding, limited bed temperature to avoid
agglomeration, high alkali metal vapor potential).
The key technology suppliers for large-capacity ui-
dized bed combustors are primarily conventional sto-
ker and PC boiler manufacturers (for example, Asea
Brown Boveri, Babcock and Wilcox, Foster Wheeler
Pyropower, Kvaerner Pulping), while a variety of ven-
dors are involved in smaller capacity AFBC. A key
factor inuencing the single uidized bed combustor
commercial capacity of these technologies is the
cross-sectional area of the combustor vessel. The
major parameters inuencing the cross-sectional area
of the combustor vessel are
Fluidization velocity (characteristic of circulating
and bubbling bed operations)
Operating pressure: atmospheric pressure or pres-
surized
Excess air level
Fuel properties
AFBC combustor vessels are normally of rectangular
cross section, with at, water-walled, shop- or eld-
fabricated construction. PFBC combustor vessels
may be cylindrical pressure vessels having refractory
lining and wall-walled construction. Alternatively, the
PFBC combustor vessel may have rectangular water-
walled construction, similar to AFBC, and must then
be contained within a larger external pressure vessel.
The two key parameter values are uidization velo-
city and pressure. Clearly, the scale of some of the
uidized bed combustor vessels, particularly for
AFBC, has grown to enormous levels. PFBC vessels
having relatively large capacities can be shop fabri-
cated pressure vessels:
500,000800,000 MJ/h for bubbling bed PFBC
2,400,0003,5000,000 MJ/h for circulating bed
PFBC
Outstanding engineering innovations in the area of
uidized bed design to achieve economical vessel
arrangements while maintaining acceptable uidiza-
tion, distribution and mixing of fuel and air, steam
generation, and turndown capability has been required
for the uidized bed combustor technologies to reach
large commercial scales.
3.3 Fluidized Bed Combustor Design Considerations
The design of uidized bed combustors, as with the
design of uidized bed gasiers, requires that transport
models be developed that apply the conservation prin-
ciples for mass, energy, and momentum to some degree
of fundamental detail. The uidized bed combustor
mass and energy balances are closely coupled, and
their solutions, applying empirical reaction kinetic
and multiple phase mixing models, provide estimates
of solids and gas composition and temperature proles
and input and output stream conditions. The momen-
tum balances provide uidized bed combustor pressure
proles when applied with appropriate uidized bed
phase density models.
These empirically based models provide the means
needed to size the combustor equipment and scale pilot
plant data. A signicant aspect of uidized bed com-
bustor modeling and design is associated with heat
transfer from the uidized bed to immersed tube sur-
faces for steam generation (LaFanechere et al., 1998).
The general arrangement of uidized bed combustion
systems, especially the mechanical arrangements of
immersed steam generating heat transfer surfaces and
convective passages, have been strongly inuenced by
commercial stoker-red boiler and pulverized fuel
red boiler designs.
The design of uidized bed combustors involves the
selection of several interrelating design, operating, and
performance parameters and requires the considera-
tion of performance and cost trade-os for any specic
application. The type of uidized bed combustor and
the application requirements strongly inuence the
uidization parameter values selected for the design.
The fuel properties (heating value, moisture content,
ash content, volatile/xed carbon ratio, sulfur content,
nitrogen content, chlorine content, ash alkalinity,
alkali metal content, and ash fusibility) represent the
major parameters that the designer must apply, and
these properties may limit the potential performance
(thermal eciency, environmental, operating cost,
availability) that can be achieved by the uidized com-
bustor (Makansi, 1990).
The design of a uidized bed combustor must result
in acceptable unit availability, accounting for the most
signicant operating problems that arise (Makansi,
1997; Jones, 1995):
Tube failures
Refractory damage
Plugging and erosion of nozzles and drains
Deposits and blockages of circulating solids leg seal-
valve
Wear and plugging of coal and limestone feed lines
Coal handing and feeding system
Ash handling system
Many of these phenomena are common to almost all
uidized bed reactors, and uidized bed combustor
design improvements are evolving as experience
increases.
The nature of the uidized bed combustor applica-
tion and the fuel properties lead to unique design con-
gurations and features in coal fueled AFBC, biomass
fueled AFBC, and coal fueled PFBC.
3.3.1 Coal Fueled AFBC
The features of bubbling bed and circulating bed
AFBC relating to uidization technology are illu-
strated in Fig. 5. The AFBC conguration and operat-
ing conditions are inuenced by the coal properties, the
steam conditions (superheated temperature and pres-
sure, feed water temperature, need for reheating), and
the unit capacity and turndown requirements
(Bernstein et al., 1995). A requirement to operate
with several coals may also exist, which means the
design must be based on the worst coal properties.
Bubbling bed AFBC with coals has been determined
most suitable for smaller capacity steam generators
because of its low uidization velocity (1.54 m/s),
resulting in a large uidized bed cross-sectional area
compared to circulating bed AFBC (Gaglia and Hall,
1987). Nonetheless, several large-capacity bubbling
bed AFBC units have been constructed and are suc-
cessfully operating (Anderson et al., 1997; Takahashi
et al., 1995). Larger scale bubbling bed AFBC units
(greater than 50,000 MJ/h thermal input) generally
must be laid out in modularized uid bed compart-
ments, either as a single level, ranch style design,
or as a stacked compartment design, with diering
steam generation functions placed in separate combus-
tion compartments (Manaker et al., 1982).
Figure 5 shows bubbling bed AFBC with two ui-
dized compartments on a single level. The air inlet
plenum, or windbox, is compartmentalized so that
individual bed sections can be deuidized for control,
and these bed sections may have physical boundaries
between them. The key features are the dense bubbling
bed with its splash zone and freeboard zone, the air
distributor, the coal and limestone feed arrangements,
the heat transfer surface arrangements, and the nes
recycling.
Bubbling bed AFBC may use underbed feeding of
coal or overbed feeding, and coal is fed in dry form.
Underbed feeding requires a multitude of pneumatic
feed points, as many as one per 12 m
2
of bed cross
section to achieve acceptable coal combustion e-
ciency (9698%). Overbed feeding can use a more eco-
nomical spreader feed system accommodating
relatively large surface areas of bed. Overbed feed,
though, requires a coarser coal top size (< 25 mm
with removal of coal nes) than underbed feeding
(< 12 mm 0 to 6 mm 0) and generally achieves a
combustion eciency about 1% lower than underbed
feeding. Fluidized bed particle and gas mixing, particle
segregation phenomena, splash-zone behavior, and
particle attrition and elutriation phenomena control
the performance of both the underbed and the overbed
feed systems. Nonuniform charbed mixing and vola-
tilesoxygen mixing may result in bed material agglom-
eration, high carbon losses, volatiles breakthrough
with above-bed burning, reduced sulfur removal, and
other undesirable results.
Typical bubbling bed temperatures are 815870
C,
and excess air levels range from 20 to 25%, with
lower reactivity coals requiring higher excess air,
higher temperatures, and fewer coal feed points than
higher reactivity coals. Bubbling bed AFBC places
horizontal heat transfer surface within the dense ui-
dized bed, achieving high heat transfer coecients
(250400 W=m
2
C) but also facing the possibility of
steam tube erosion. Heat transfer surface is also
placed above the dilute freeboard zone where convec-
tive steam generation occurs. Bubbling bed combus-
tor dense beds are generally about 11.5 m deep,
allowing them to achieve acceptable pressure drops
(1525 kPa), while the splash zone and freeboard
height is about 10 m from the top of the dense bed.
The combustion gas velocity above the transport dis-
Figure 5 Coal fueled AFBC congurations.
engaging height is usually accelerated up to normal
furnace convective pass velocities.
Smaller scale bubbling bed AFBC units can apply
innovations not feasible at large capacities, such as
internal uidized bed recirculation schemes.
Turndown can be achieved by a number of methods,
including shutting down bed compartments, reducing
uidization velocity, reducing uidized bed depth, and
using ue gas recirculation.
Circulating bed AFBC uses a more complex com-
bustor design than bubbling bed AFBC, as is illu-
strated in Fig. 5. It uses no heat transfer in the
dense, reducing, primary bubbling bed zone, and it
places all heat transfer surfaces exposed to high bed
velocity in a vertical orientation to minimize the pos-
sibility of erosion. Refractory covering is also used to
protect the metal heat transfer surfaces in some loca-
tions. The dense, primary, bubbling bed zone is usually
refractory lined to minimize heat transfer surface ero-
sion. The dense bed is a turbulent bed uidized by
substoichiometric primary air (about half of the total
air) at a uidization velocity of about 24 m/s. The
dense bed is supported by an air distributor whose
primary function is to distribute air uniformly to
induce particle mixing between injected coal and
recycle solids, and carry out partial coal oxidation.
The distributor must operate within specied pressure
drop limits, without plugging, erosion damage, or
buckling. A variety of designs are in use, including
refractory insulated bubble nozzles, water-cooled bub-
ble nozzles, and pipe distributors. Allowance must also
be made for bed material drainage from the primary
bed, with the drainage ports integrated with the air
distributor, but most drainage is taken from the low-
carbon solids in the recycle leg.
Coal is fed to the dense bed by a small number of
simple slide chutes fed by rotary valves and located
along one wall of the unit. A single coal feed point
may service 930 m
2
of bed cross section (Brereton,
1997). Alternatively, coal may be fed at a single point
into the solids recirculation leg. Limestone (less than 6
mm in diameter) is normally fed separately above the
bed.
Recirculating bed media are fed hot into the pri-
mary bed zone at a ratio of 50 to 100 times the coal
feed rate. Various types of reinjection, nonmechanical
valves and seals are used, such as seal pots, L-valves,
J-valves, and others. These are simple refractory lined
ducts having solids holding volumes that provide a
loop seal, and appropriate aeration nozzles that
induce and control solids ow. Cyclones are normally
used to separate recirculating solids from the over-
head gas stream, with solid uxes on the order of 10
kg=m
2
s (Brereton, 1997). Alternative devices may
also be used to remove entrained particulates for
recycling, such as impactors (Belin and Flynn,
1991). Nonmechanical valves are discussed in
Chapter 21, Standpipes and Nonmechanical
Valves. Cyclones are discussed in Chapter 22,
Cyclone Separators.
Secondary air is injected from nozzles located at the
vessel wall at an elevation close to the top of the dense
bed zone, resulting in increased uidization velocity to
about 49 m/s with high particle entrainment from the
dense bed, completion of char and volatile combus-
tion, and vigorous circulating bed, gasparticle mixing
due to the high slip velocity and solids backow that
exists. Additional higher levels of secondary air injec-
tion may also be used. Circulating bed AFBC can
operate with excess air of 10-20%, slightly lower than
that of bubbling bed AFBC. The secondary air jets
must penetrate suciently far into the fast uid bed
to promote eective mixing, and this jet penetration
length limits the allowable furnace depth to about 9
11 m (Makansi, 1993). Even though coal is not uni-
formly mixed in the dense primary bed zone, the vig-
orous circulating bed furnace provides ample mixing
and residence time to achieve high levels of char and
volatiles combustion and high sulfur removal e-
ciency.
The density of the circulating bed falls rapidly with
increased elevation above the dense bed and reaches
density levels as small as about 1040 kg=m
3
near the
top of the circulating bed furnace. The bed density
prole may be inuenced by the design of the solids
gas exit region (Horio, 1997). The furnace height must
be sucient to provide residence time for ecient com-
bustion and sulfur removal, and must also accommo-
date the required heat transfer surface for steam
generation while resulting in acceptable gas pressure
drop. Typically, the furnace height is limited to
about 40 m.
The bed-to-surface heat transfer coecient (about
85280 W=m
2
C) by solids convection and radiation is
lower in circulating bed AFBC than in bubbling bed
AFBC due to the lower bed density and vertical heat
transfer surface orientation, and the heat transfer coef-
cient decreases with increased height. The dilute zone
of the furnace is water-walled, with the heat transfer
surfaces placed at the perimeter of the rectangular ves-
sel enclosure. Additional vertical heat transfer surface
walls or wing walls may be hung within the vessel to
increase its heat removal capacity. Another means to
increase the heat transfer surface is to place an exter-
nal heat exchanger in the circulating solids loop (see
Fig. 4). This external heat exchanger is a relatively low-
velocity (1.53 m/s) bubbling uidized bed containing
horizontal heat transfer surface; it is uidized by sec-
ondary air and has no in-bed reducing zones.
Scaling to larger capacity single circulating bed
AFBC units requires proper engineering of the geome-
try of the rectangular furnace and the use of the several
heat transfer surface conguration options, and it
should provide for acceptable coal feed distribution
and secondary air injection (Lee, 1997). Models of
the integrated uidized bed combustor and solids cir-
culation system are required to design the equipment
and control scheme properly. Circulating bed AFBC
turndown techniques dier signicantly from those uti-
lized by bubbling bed AFBC. These have been
reviewed by Brereton (1997).
3.3.2 Biomass Fueled AFBC
The relatively small capacity of biomass fueled AFBC
applications (generally less than 300,000 MJ/h fuel
thermal input), and the highly reactive nature of bio-
mass fuels make bubbling bed AFBC a highly compact
and eective combustion technology. Many aspects of
biomass fueled bubbling bed AFBC are similar to coal
fueled bubbling bed AFBC:
The uidization velocity is about 3 m/s or less to
minimize bed material elutriation.
The bed temperature is 760870
C depending on the
biomass fuel ash agglomeration tendencies.
The total excess air is about 2030%.
The turndown is achieved by compartmentalizing
the bed and by ue gas recirculation.
The bubbling bed depth is about 1 m with in-bed
heat transfer surface (nned tubes).
The dierences between coal-red bubbling bed AFBC
and biomass fueled bubbling bed AFBC, though, are
signicant (Douglas and Morrison, 1997; Hanson,
1991).
Coarse size biomass fuel feeding is used (100% less
than 90150 mm maximum particle dimension).
An inert bed of sand or limestone particles is used.
Above-bed biomass fuel feeding through chutes is
used.
The dense bed is operated substoichiometrically,
with primary air ranging from 40 to 80% of the
total air.
At the top of the freeboard, the gas is accelerated
(velocity as high as 5 m/s) and secondary air is
injected to complete volatiles combustion, result-
ing in a combustion gas temperature as much as
200
C above the dense bed temperature.

Special features may be used to provide online
removal of oversized tramp materials brought
in with the biomass fuel.
Agglomeration in the uidized bed, due to alkali metal
(Na, K) reactions with the inert bed sand particles is
common and may require periodic renewal of the bed
material. Convective surface fouling can also be exten-
sive and may require the use of screens before the
superheater surfaces.
In larger installations, biomass fueled circulating
bed AFBC can also be used, with designs very similar
to those used in coal fueled, circulating bed AFBC
(Skoglund, 1997). Circulating bed AFBC with biomass
fuels has little or no advantage over bubbling bed
AFBC with biomass fuels.
3.3.3 Coal Fueled PFBC
As with bubbling bed and circulating bed AFBC tech-
nology evolution, bubbling bed PFBC has led the
development of PFBC over the still-to-be-demon-
strated circulating bed PFBC. Bubbling bed and circu-
lating bed PFBC both appear to be attractive options
for advanced high-eciency power generation systems
for large electric generating applications.
The basic features of bubbling bed and circulating
bed PFBC are analogous to those of bubbling bed and
circulating bed AFBC, except for the following points:
The elevated pressure of PFBC (10001500 kPa)
means that even the bubbling bed combustor
cross-sectional area is reasonable for shop fabri-
cation at relatively large capacities.
The elevated pressure of PFBC means that the dense
bubbling bed can be relatively deep (35 m) and
still meet pressure drop limits, so that the distri-
bution of coal feed points is not as demanding as
it is in bubbling bed AFBC.
The bubbling bed PFBC combustor vessel is typi-
cally tapered near the base to provide higher ui-
dization velocities and more intense mixing in the
regions where coal is fed.
Coal may be fed by either dry lock hopper methods
or by more reliable waterpaste methods.
The PFBC vessel, cyclones, solids circulation equip-
ment, and the external heat exchanger may be con-
structed as independent cylindrical pressure vessels.
Alternatively, at water-walled construction, like that
used in AFBC may be placed, along with cyclones and
many other pieces of process equipment designed for
low-pressure operation, within the pressurized environ-
ment of a single large external pressure vessel, allowing
the combustor to retain many of its AFBC features.
Figure 6 illustrates the PFBC power plant cong-
uration, showing the main power plant components.
The gure is representative of bubbling bed PFBC
with the currently commercial process conguration:
multiple stages of cyclones are used to remove particu-
late from the combustion gas, and the combustion gas
is expanded through a ruggedized turbine expander, an
expander designed to tolerate large particulate content
in the gas. The stack gas must be further cleaned of
particulates in commercial equipment (electrostatic
precipitator or fabric bag lter) to meet environmental
standards. The PFBC unit and several other compo-
nents are located in a large external pressure vessel that
is pressurized and cooled by the warm compressor air.
This arrangement separates the thermal boundary
from the pressure boundary and provides several prac-
tical equipment and process advantages over the use of
separate pressure vessels for each component
(McDonald et al., 1987; Huryn et al., 1987).
Bubbling PFBC uses dense bed depth adjustment as
a means for load control, exposing normally immersed
heat transfer surfaces to the freeboard zone; a vessel
for storage and transfer of bed ash is used for this
purpose and is housed in the external pressure vessel.
The external heat exchanger used with circulating
PFBC represents a relatively large uidized bed that
is also housed in this pressure vessel (Walter et al.,
1997). In scaling up PFBC, the size of the external
pressure vessel is a major consideration. For example,
the Cottbus PFBC plant has a fuel thermal input (lig-
nite fed dry) of about 700,000 MJ/h and operates at a
pressure ratio of about 13 : 1. Its external pressure ves-
sel is about 32 m high and 11 m in diameter (Walter et
al., 1997). The Karita PFBC unit has about ve times
the fuel thermal input as the Cottbus and feeds coal as
a water paste (Jes, 1997). With its higher operating
pressure, the Karita external pressure vessel might be
about 50 m high, with a diameter of about 1516 m, a
formidable vessel to fabricate, ship, and install.
Topped PFBC applies both uidized bed gasica-
tion and combustion technologies in concert. Topped
Figure 6 PFBC power plant schematic.
PFBC will operate with conventional heavy-duty com-
bustion turbines having higher ring temperatures and
higher pressure ratios than those used in PFBC. Hot
gas lters must be used to remove particulate from
both the combustion gas and the carbonizer fuel gas
to protect the turbine expander from erosion and
deposition damage. The additional challenge of topped
PFBC is to integrate and control the combined opera-
tion of the carbonizer and combustor and their inter-
facing components.
The design of the uidized bed carbonizer and com-
bustor unit for topped PFBC change little from those
design factors already reviewed. Because a higher tur-
bine inlet temperature is used than in PFBC, the oper-
ating pressure of the uidized bed combustor would be
higher, resulting in smaller bed diameter and smaller
external pressure vessel diameter than for a compar-
able PFBC. The uidized bed combustor is fed warm
char from the uidized bed carbonizer at appropriate
char feed distribution points, and the combustor must
be sized to burn eciently the low-volatile char and
convert the CaS from the uidized bed carbonizer
into CaSO
4
. Topped PFBC also has the option of feed-
ing coal to the uidized bed combustor as well as to the
uidized bed carbonizer, making much greater power
plant exibility possible.
4 THE FUTURE OF FLUIDIZED BED
GASIFIERS AND COMBUSTORS
The recent practical advancements of uidized bed
gasiers and combustors have occurred with little
dependence on the fundamental understanding of ui-
dization phenomena and associated reaction phenom-
ena. Fluidized bed gasiers and combustors are
designed following highly empirical principles based
on the extrapolation of pilot plant test data. The like-
lihood of further improvements in the competitiveness
of uidized bed gasiers and combustors depends in
part on the health of the commercial markets for
their applications and the relative economic and envir-
onmental pressures on the carbonaceous fuels that they
utilize (coals, biomass fuels, and others).
Improved understanding of uidization fundamen-
tals related to gasication and combustion phenomena
may eventually guide commercial equipment design
and selection of special features. More immediate
impacts may result, though, from engineering innova-
tions based on operating plant observations that
improve performance, reliability, operability, and
cost. Experimentation (laboratory scale conceptual
testing, cold ow simulations, pilot plant parametric
testing, demonstration plant experience) will continue
to be an important avenue for such engineering inno-
vations.
REFERENCES
Agrawal RK, Haldipur GB. Advanced gasier-desulfurizer
process development for SNG application. Final Report
by KRW Energy Systems, Inc., to Gas Research Institute,
GRI-89/0011, 1988.
Anderson KD, Manaker AM, Stephans EA Jr. Operating
experience of the Tennessee Valley Authoritys 160-MW
atmospheric uidized bed combustion demonstration
unit. Proceedings of the 14th International Conference
on Fluidized Bed Combustion, ASME, 1997, pp 3945.
Barducci G, Daddi P, Polzinetti G, Ulivieri P. Thermic and
electric power production and use from gasication of
biomass and RDF: Experience at CFBG plant at Greve
in Chianti. Proceedings of the Second Biomass
Conference of the Americas, Portland, 1995, NREL/CP-
200-8098, pp 565575.
Belin F, Flynn TJ. Circulating uidized bed boiler solids
system with in-furnace particle separator. Proceedings of
the 1991 Fluidized Bed Combustion Conference, ASME,
1991, pp 287294.
Bernstein N, Goidich S, Li S, Phalen J. Inuence of fuel type
and steam cycle on CFB boiler conguration. Proceedings
of the 13th Fluidized Bed Combustion Conference,
ASME, 1995, pp 2737.
Brereton C. Combustion performance. In: Grace JR, Avidan
AA, Knowlton T, eds. Circulating Fluidized Beds. New
York: Blackie, 1997, pp 376416.
Bryan BG, Hoppe JA. Status of the Shanghai Coking and
Chemical Companys U-gas coal gasication plant.
Proceedings of the Fifteenth Annual International
Pittsburgh Coal Conference, Pittsburgh (CD-ROM),
1998.
Chau LL, Rovner JM. Wood gasication for combined cycle
power applications. Paper presented at the AIChE
Summer National Meeting, July, 1995.
Craig JD, Purvis CR. A small scale biomass fueled gas tur-
bine engine. Paper presented at the ASME, International
Gas Turbine and Aeroengine Congress and Exhibition,
Stockholm, Sweden, June 1998.
Dawes SG, Mordecai M, Welford GB. Recent design and
cost studies for air blown gasication. Proceedings of
the 14th International Conference on Fluidized Bed
Combustion, ASME, 1997, pp 347358.
DeLong MM, Oelke EA, Onischak M, Schmid MR, Wiant
BC. Sustainable biomass energy production and rural
economic development using alfalfa as feedstock.
Proceedings of the Second Biomass Conference of the
Americas, Portland, 1995, NREL/CP-200-8098, pp
15821591.
Demuth JE, Smith HG. Pinon Pine Project gasier startup.
Presented at the Advanced Coal-Based Power and
Environmental Systems 98 Conference, Federal Energy
Technology Center, Morgantown, 1998.
Douglas MA, Morrison SA. The design of an advanced BFB
steam generator for biomass. Proceedings of the 14th
ASME, 1997, pp 151157.
Farris M, Paisley MA, Irving J, Overend RP. The biomass
gasication process by Battelle/FERCO: design, engineer-
ing, construction, and startup. Proceedings of the EPRI
1998 Gasication Technologies Conference, San
Francisco, 1998.
Gaglia BN, Hall A. Comparison of bubbling and circulating
uidized bed industrial steam generation. Proceedings of
the 1987 International Conference on Fluidized Bed
Geldart D. Powder Technology 7:285, 1973.
Ghate M, Longanbach J. Coal gasicationtechnology sta-
tus report. DOE/METC-88/0262, January 1988.
Hanson JL. Agricultural waste red uid bed combustor,
Delano, California. Proceedings of the 1991
ASME, 1991, pp 11051109.
Hebden D, Stroud HJF. Coal gasication processes. In:
Elliott MA, ed. Chemistry of Coal Utilization. Second
supplementary volume. New York: John Wiley, 1981,
pp 15991752.
Hirschfelder H, Vierrath H. CFB biomass gasication for
energy and industry operational results. Proceedings of
the EPRI 1998 Gasication Technologies Conference,
San Francisco, 1998.
Huryn JB, Wickstrom B, Rederstor B. Design concept for a
PFBC commercial plant. Proceedings of the 1987
ASME, 1987, pp 272277.
Jes E. Karita: a quantum leap for PFBC. Turbomachinery
International, March 1997, pp 1617.
Jones C. O&M experience underscores maturity of CFB tech-
nology. Power, May 1995, pp 4655.
LaFanechere L, Basu B, Jestin L. Use of an expert system to
study the eect of steam parameters on the size and con-
guration of circulating uidized bed boilers. Journal of
Engineering for Gas Turbines and PowerTransactions
of the ASME 120(4):813819, 1998.
Leckner B. Fluidized bed combustion: mixing and pollutant
limitation. Progress in Energy and Combustion Science
24(1):3161, 1998.
Lee YY. Design considerations for CFB boilers. In: Grace
JR, Avidan AA, Knowlton T, eds. Circulating Fluidized
Beds. New York: Blackie, 1997, pp 417440.
Lu Z, Li D, Shen Y, Xu X, Cao B. An industrial demonstra-
tion project on tri-generation process for gas, heat and
power using CFB technology. Presented at the China
Korea Technology Cooperation Conf. On Coal
Applications, Tsinghua University, China, 1996.
Makansi J. Fuel type, preparation emerge as critical to FBC
design. Power, January 1990, pp 4144.
Makansi J. Can uid bed take on P-C units in the 250400
MW range? Power, September 1993, pp 4550.
Makansi J. Designers, users air concerns on CFBs. Power,
March/April 1997, pp 45.
Manaker AM, Dunn GC, Cruikshank KA, Wyatt JN. TVAs
technical assessment and cost comparison of nal concep-
tual designs by Babcock Contractors, Inc. (BCI), Babcock
and Wilcox Company (B&W), and Combustion
Engineering, Inc. (CE), of a 200-MW Atmospheric
Fluidized Bed Combustion (AFBC) Demonstration
Plant. Proceedings of the 7th Conference on Fluidized
Bed Combustion, 1982, pp 691700.
McDonald DK, Weitzel PS, Thornblad P. PFBC design inte-
grationASEA Babcock PFBCs approach. Proceedings
of the 1987 International Conference on Fluidized Bed
Monsour N, Durai-Swamy K, Voelker G. MTCI/thermo-
chem steam reforming process for biomass. Proceedings
of the Second Biomass Conference of the Americas,
Portland, Oregon, 1995, NREL/CP-200-8098, pp 543
552.
Nieminen J, Kivela M. Biomass CFG gasier connected to a
350 MW
th
steam boiler red with coal and natural gas
thermie demonstration project in Lahti in Finland.
Biomass and Bioenergy 15(3):251257, 1998.
Pitcher K, Hilton B, Lundberg H. The ARBRE project: pro-
gress achieved. Biomass and Bioenergy 15(3):213218,
1998.
Renzenbrink W, Wischnewski R, Engelhard J, Mittelstadt A.
High temperature Winkler (HTW) coal gasicationa
fully developed process for methanol and electricity pro-
duction. Proceedings of the EPRI 1998 Gasication
Technologies Conference, San Francisco, 1998.
Robertson A. Second generation PFBC systems research and
developmentpilot plant becomes demonstration plant
design. Presented at the Advanced Coal-Fired Power
Systems 95 Review Meeting, Federal Energy
Technology Center, Morgantown, 1995.
Seglin L, Bresler SA. Low-temperature pyrolysis technology.
In: Elliott MA, ed. Chemistry of Coal Utilization. Second
Supplementary Volume. New York: John Wiley, 1981, pp
785846.
Schobert HH, Rusinko F, Mathews JP. W(h)ither the coal
industry?the long-term view. Proceedings of the
Fifteenth Annual International Pittsburgh Coal
Conference, Pittsburgh (CD-ROM), 1998.
Sierra Pacic Power Company. Tracy Power StationUnit
No. 4, Pinon Pine Power Project. Public Design Report,
1994, NTIS, DOE/MC/293094056.
Simell P, Kurkela E, Stahlberg P. Formation and catalytic
decomposition of tars from uidized-bed gasication.
Presentation at the International Conference on
Advances in Thermochemical Biomass Conversion,
Interlaken, Switzerland, 1992.
Skoglund B. Six years of experience with Swedens largest
CFB boiler. Proceedings of the 14th International
Conference on Fluidized Bed Combustion, ASME,
1997, pp 4756.
Stahl K, Neergaard M. IGCC power plant for biomass uti-
lization. Biomass and Bioenergy 15(3):205211, 1998.
Stobbe S, Oatley J, Brown RC. Indirectly-heated biomass
gasication using latent-heat ballasting of a uidized reac-
tor. Proceedings of the IECEC 96 Conference, Institute of
Electrical and Electronics Engineers, 1996. pp 17241729.
Takahashi M, Nakabayashi Y, Kimura N. The 350 MWe
Takehara plant. VGB Kraftwerkstechnik 75:427434,
1995.
Walter E, Krartz HT, Almhen P. Lignite red combined
cycle heat and power plant using pressurized uidized
bed combustion. Proceedings of the 14th International
Conference on Fluidized Bed Combustion, ASME,
1997, pp 359369.
Wen CY, Tone S. Coal conversion reaction engineering. In:
Chemical Reaction Engineering Reviews, Houston (Luss
D, Weekman VW, eds.). ACS Symposium Series 72, 1978,
pp 85109.
Yang WC, Newby RA, Keairns DL. Large-scale uidized
Technology 82:333381, 1995.
16
Applications for Chemical Production and Processing
Behzad Jazayeri
Fluor Daniel, Inc., Aliso Viejo, California, U.S.A.
This chapter provides an overview of some of the more
important commercial applications of uidized bed
technology not covered in the other sections. Recent
developments with high potentials of commercializa-
tion in the near future are also discussed.
1 CHEMICAL SYNTHESIS
1.1 Acrylonitrile
The Sohio Company (now BP Chemicals) introduced
in 1950s what is now considered to be the most suc-
cessful application of uidized beds for chemical synth-
esis (Fig. 1). In this application, air, propane, and
ammonia are reacted together in a turbulent uid
bed of Geldart group A (Chapter 3) catalyst to pro-
duce acrylonitrile by the reaction:
C
3
H
8
NH
3

3
2
O
2
! C
2
H
3
CN 3H
2
O 1
The features of the original process as reported by
Kunii and Levenspiel (1991a) and some recent devel-
opments follow. Reactor operating conditions are 400
to 500
C at about 1.7 atmosphere, with a reactor con-

tact time of 5 to 20 seconds. In addition to acryloni-
trile, carbon monoxide, carbon dioxide and water,
small quantities of hydrogen cyanide, acetonitrile,
and acrolein are also produced. The original plants
were designed to operate at around 52 cm/s supercial
velocity. Modern plants are now operating with nes
content (10 to 45 micron) of around 2545% at about
66 cm/s (Pell and Jordan, 1987), and possibly as high
as 100 cm/s. BP has introduced several new catalysts
with increasing selectivity to acrylonitrile. With these
changes, propylene conversion is now approaching
100%, and per pass reactor yield to propylene is
about 85%. The reaction is highly exothermic, releas-
ing about 670750 kJ/mol. The heat of reaction is
removed by direct generation of high pressure steam
in serpentine coils located inside the reactor. The reac-
tor hydraulic diameter (Volk et al., 1962) is reported as
100 to 150 cm.
Sohio initially tested the process in both a small
bench scale unit and a circa 61 cm pilot unit before
building the rst commercial plant (circa 300 cm). Over
51 units have been built, representing over 95% of the
worlds acrylonitrile capacity. The largest units now
utilize reactors with estimated diameters of 800 to pos-
sibly 1000 cm.
In the process, air is introduced uniformly into the
bottom of the reactor via a distributor plate. The pro-
pylene and ammonia are introduced into the uidized
bed above the distributor plate via a separate sparger.
The design of these gas distribution systems has
evolved over the years improving the reactor yield
(Ohta and Yokura, 1996.)
The original units used several parallel sets of three-
stage cyclones for catalyst recovery. Recently, one
cyclone vendor has announced the use of multiple
cyclones made of a single stage followed by two second
stages.
The Monsanto Company has also commercialized a
uidized bed for the production of acrylonitrile for its
own use. The Monsanto reactor is expected to have
similar features to that of BP but uses a dierent cat-
alyst. In early 1990s, Monsanto decided to license this
process and has licensed one unit in Korea.
BP Chemicals recently announced the piloting of a
propane-based uid bed process for acrylonitrile pro-
duction (1997 annual report). Mitsubishi has also
received patents on a similar propane-based process.
Use of propane may oer production cost benets as
propane is typically cheaper than propylene. The pro-
pane-based catalyst is believed to be dierent from the
propylene based catalyst, ruling out the possibility of a
swing operation depending on the market prices (a
single catalyst producing acrylonitrile using either pro-
pane or propylene as feedstock.) Based on a review of
literature and patents issued to BP and Mitsubishi, it is
the authors opinion that the propane-based process
will operate with a low per pass propane conversion.
If so, such a process is likely to use recycle, which may
increase the process complexity and cost. Use of
recycle may also introduce additional safety issues as
the recycle gas will contain unconverted propane and
possibly some oxygen. Additionally, it is the authors
opinion that the capacity of a given reactor will be
lower with the propane-based catalyst. In such a situa-
tion, revamping an existing propylene based plant to
propane can result in loss of production. Since in a
revamp situation, the reactor volumetric throughput
cannot be increased beyond the capacity of the down-
stream recovery equipment, one possible solution to
this issue may be to use enriched air, or possibly oxy-
gen. Use of enriched air or oxygen is likely to increase
the operating cost. It remains to be seen how these
challenges will be resolved and how, in the ultimate
analysis, the economics of the propane-based process
compare with the more mature propylene-based pro-
cess. In late 1998, BP announced that it has successfully
completed the pilot plant program and will move ahead
with a demonstration scale plant. It therefore appears
that BP has resolved these issues and that this new
process may indeed oer an economic advantage.
1.2 Maleic Anhydride
Production of maleic anhydride by the vapor phase
partial oxidation of C4s is an extremely exothermic
reaction (1410 KJ/mol) and is therefore well suited to
uidized bed application. The rst commercial ui-
dized bed application was by Mitsubishi Chemical
Industries in 1970. This process converted a mixture
of butadiene and butene to maleic anhydride in a ui-
dized bed of 6 m diameter using a vanadium phosphor-
ous oxide (VPO) catalyst (Kunii and Levenspiel,
1991a.) In the early 1980s, three new uid bed pro-
cesses were developed for conversion of n-butane to
maleic anhydride. The rst, by Badger, was taken
through the demonstration stage but was never com-
mercialized because of the global depression of the
early 1980s. The second process was developed by BP
Chemicals and has been licensed once. The third pro-
cess was commercialized by Lummus at about the
same time as BPs. Nine units have been licensed to
date. The Lummus process is now licensed by Lonza.
The process developed by these four companies uses
dierent formulations of VPO catalyst in a conven-
tional uidized bed similar in design to the Sohio acry-
Figure 1 Sohio acrylonitrile reactor.
lonitrile reactor (Fig. 1). Air, owing through a grid
plate, is used to uidize the catalyst. Vaporized C4s is
sparged into the uidized bed above the grid plate. The
reaction heat is removed by direct generation of steam
inside serpentine cooling coils placed in the uidized
bed. The uidized bed reactor can operate within the
ammability envelope with much higher concentra-
tions of n-butane in the total feed compared with the
xed bed process. Reactor operating conditions are
410 to 420
C at about 1 to 5 atmospheres. Main reac-

tion by-products are acetic and acrylic acids. The pri-
mary reaction is
C
4
H
10
O
2
! C
4
H
2
O
3
! H
2
O CO CO
2
2
An evaluation of patents and literature in this eld has
led this author to conclude the following:
Fixed bed reactors are tubular in design and have
features similar to that of a shell and tube heat
exchanger with xed tube sheets. Molten salt
owing on the shell side is used to remove the
heat of reaction. Mechanical considerations
limit the number of tubes in such reactors. As a
result, the author estimates the xed bed reactor
capacity to be limited to circa 15,00020,000 MT/
year. Multiple reactors are therefore required for
higher capacities. The uidized bed, on the other
hand, can be designed as a single reactor unit
with capacities of circa 70,000 MT/year and pos-
sibly more.
Fixed bed reactors are limited by ammability con-
siderations to about 1.2 to 1.8 mol% butane in
the feed, as the butane and air are premixed out-
side the reactor. The uidized bed can operate
with up to possibly 5 mol% butane in the feed
because butane is sparged separately into the bed
and the mass of catalyst acts as a heat sink. The
higher butane-to-air feed ratio reduces the size of
the downstream gas handling section of the pro-
cess, reducing the capital cost.
The sensible heat of the purge gas vented to atmo-
sphere from the uidized bed process can be as
low as 30% of that in the xed bed process
because of lower gas volumetric ow. Therefore
for a given reactor capacity more of the reaction
heat is recovered in generating steam.
The uidized bed maleic anhydride process, how-
ever, has also several apparent disadvantages:
The catalyst volume required in the uidized process
is several times higher than that of the xed bed
process. Since VPO catalyst is very expensive, the
additional initial cost of the catalyst may negate
any savings in the downstream equipment.
Reaction selectivity to maleic anhydride tends to be
lower in the uidized bed process.
These competing characteristics tend to favor the ui-
dized bed when a larger capacity unit is required and
there is a high demand for steam. The author estimates
a rough break point to be around 10,00020,000 MT/
year capacity.
The maleic anhydride uidized bed reactor also
oers several design and operational problems that
must be carefully addressed:
The VPO catalyst used is somewhat dense.
Therefore the particle size distribution of the cat-
alyst, the design of the reactor internals, and the
operating velocity must be carefully selected.
Failure to do so will result in gas bypassing,
which will aect reactor conversion and
increased oxygen slip.
The VPO catalyst is susceptible to loss of selectivity
when exposed to reducing environment. To pre-
vent catalyst overreduction, a threshold minimum
oxygen concentration is needed in the reactor o-
gas. Additionally, the reactor must operate at
higher velocities to enhance gas/solid contact.
To maintain suitable oxidation environment for the
catalyst, butane conversions must be kept within
a certain range.
Meeting the above constraints results in operation
with a reactor ogas that is close to autoignition
and is therefore susceptible to afterburn.
Engineering solutions have been developed to
address this and permit the operation of these
units in a safe manner.
Little information has been published on the above
four processes, but some educated guesses can be
made. The catalyst is Geldart group A with possibly
25 to 55% nes (10 to 45 micron) to counter its higher
density. Operation within the turbulent regime is
needed to enhance gas/solid contact and promote uni-
formity between the emulsion and void gas and there-
fore minimize catalyst overreduction. Operating
velocities of about 40 and possibly as much as 100
cm/s are therefore required.
In an eort to resolve the design issues discussed
above, DuPont has developed and commercialized a
circulating uidized bed for conversion of butane to
maleic anhydride (Contractor, 1985). The process is
depicted conceptually in Fig. 2. The VPO-based cata-
lyst is structurally dierent from the conventional ui-
dized bed catalyst discussed above. Details of the cat-
alyst structure are provided by Contractor et al. (1978).
A signicant departure from conventional uidized
bed design is the use of the catalyst as oxygen carrier.
The catalyst is oxidized in a separate regeneration ves-
sel by contact with air. It is then circulated to a riser
reactor where it is reduced by reaction with butane.
The reactor operates at high velocities with a relatively
short contact time. Per pass conversion of butane is
therefore low, requiring recycling of unconverted
butane. Butane feed concentration can be high, possi-
bly over 20 mol%. To avoid catalyst overreduction in
Figure 2 DuPont maleic anhydride reactor.
such a harsh environment, the reactor is designed to
minimize catalyst residence time and operates at rela-
tively low temperatures, possibly as low as 360
C. This
mode of operation helps maintain reaction selectively
even though butane concentrations are very high.
The reduced catalyst is separated from the reactor
ogas, stripped to minimize butane and maleic anhy-
dride losses, regenerated by contact with air in the
regenerator, and then returned back to the reactor.
DuPont has indicated that a further stripping step
may be used between the regenerator and reactor.
DuPont claims that by carrying out the oxidation
and reduction cycles in two separate zones, each zone
can be designed to operate close to its optimal point,
thus increasing reaction selectivity and catalyst pro-
ductivity. Afterburn of the reactor ogas is not
expected, as the gas contains little or no oxygen.
DuPont bench tested this process in mid-1980s. A
15 cm diameter by 27.4 m high riser was operated for
over one year to prove the process concept (Contractor
et al., 1993). The table shows the conditions for the
demonstration unit.
Parameter Range tested
Reaction temperature,

C 360420
Reaction pressure, atmosphere < 5
Catalyst ux, kg=cm
2
s 2501100
Gas velocity, m/s 410
Butane concentration, mol% < 25
The riser solid volume concentration approached 0.2.
The rst commercial facility was started in the mid-
1990s.
The DuPont process incorporates design features of
uid catalytic cracking (Chapter 14) in addition to
many features that are unique to chemical synthesis
reactors.
The Mitsubishi process described above has since
been converted to n-butane feedstock. Mitsubishi just
recently announced piloting an enhanced process with
butane recycle and catalyst regeneration using a new
non-VPO catalyst formulation, which is claimed to
reduce cost.
1.3 Oxychlorination of Ethylene
Ethylene is reacted with dry hydrogen chloride (HCl)
and either air or pure oxygen over a supported cupric
chloride catalyst in a uidized bed to produce ethylene
dichloride (EDC) and water by the vapor phase reac-
tion
C
2
H
4
2HCl
1
2
O
2
! C
2
H
4
Cl
2
H
2
O 3
The oxychlorination reactor is part of a balanced vinyl
chloride monomer complex (Fig. 3). The EDC is the
intermediate product and is thermally cracked to VCM
and HCl. Operating conditions of the oxychlorination
reactor are 220 to 245
C at 2.5 to 6 atmospheres. For

the air-based process, ethylene and air are fed in slight
excess of stoichiometry to ensure high conversions of
hydrogen chloride and minimize losses of excess ethy-
lene in the vent gas. Under these conditions, 94 to 99%
conversion of ethylene and 98 to 99.5% conversion of
hydrogen chloride are obtained at a reported selectivity
to EDC of 94 to 97% (Kirk-Othmer, 1997.) The oxy-
gen-based process operates at lower temperatures and
has a higher yield and provides a drastic reduction in
the volume of vent gas. It is operated with a much
higher excess of ethylene. BF Goodrich (now Geon)
is the dominant licensor of the uidized bed process.
The rst commercial plant was started in 1960s. A total
of thirty-six plants have been licensed. Other licensors
are listed by Naworski and Velez, 1983.
The cupric chloride catalyst used in the uidized bed
process is a group A powder with a nes content (10
45 micron) of about 30 wt% (Naworski and Velez,
1983.) The catalyst is subject to stickiness under certain
(usually upset) conditions that promote formation of
dendritic growths of cupric chloride on the catalyst
surface. Particle agglomeration can occur with loss of
uidity, and, in severe cases, collapse of the bed.
The oxychlorination reactor is a vertical cylindrical
shell made of carbon steel with a support grid/air spar-
ger system and internal cooling coils. Internal or exter-
nal cyclones are used to minimize catalyst carryover.
The reactor internal parts are made from corrosion-
resistant alloy. The reactor has many design features
depicted in Fig. 1.
1.4 Phthalic Anhydride
Coal tar naphthalene is oxidized with air over a vana-
dia catalyst in a uidized bed according to the follow-
ing highly exothermic partial oxidation reaction:
C
10
H
8
4SO
2
! C
8
H
4
O
3
2H
2
O 2CO
2
4
Reaction yields are about 100 kg of phthalic anhydride
per 100 kg of naphthalene with desulfurized feed.
Desulfurized molten naphthalene is introduced into
the bed above the distributor plate through multiple
spray nozzles, while air is admitted from below.
Contact of reactants occurs inside the uidized bed,
permitting reactor operation within the ammability
envelope (Graham, 1970). The reactor operates at
around 2.7 atmosphere and between 345 and 385
C,
with a velocity of between 30 and 60 cm/s and a con-
tact time of 10 to 20 seconds (Graham and Way, 1962).
The catalyst is Geldart group A with nes content (10
to 45 micron) of 28 % (Johnsson et al., 1987). The
reactor is believed to operate within the turbulent
regime. Besides carbon monoxide, carbon dioxide,
and water, small quantities of maleic anhydride,
naphthoquinone, and benzoic acid are also formed as
by-products.
The uidized bed process was commercialized by
Sherwin Williams in the mid-1940s but is now known
as the Badger phthalic anhydride process. Over 20
reactors have been built, with the largest having a
capacity of 100,000 metric tons per year. The last reac-
tor was built in China in the mid-1980s. The process is
now considered noncompetitive with o-xylene-based
xed bed plants due to limited availability of coal tar
naphthalene. Attempts to develop an o-xylene-based
uid bed process have been hampered by the unavail-
ability of a suitable attrition resistance catalyst support
(Bolthrunis, 1989).
Excellent accounts of the development, operation,
and safety aspect of this process have been provided
elsewhere (Miseralis et al., 1991; Graham, 1970) and
are summarized here.
The original Sherwin Williams rst-generation reac-
tor used an external shell-and-tube catalyst cooler and
external ceramic lter elements (Fig. 4a). Reaction heat
was removed by circulating catalyst through the exter-
nal cooler via a standpipe. The cooled catalyst was
returned back to the reactor through the distributor
plate. This operation was never satisfactory and suf-
fered from both mechanical and operational problems.
Reactor throughput was often limited due to insu-
cient heat removal capability. In the mid-1950s, Badger
Figure 3 Balanced oxychlorination process.
suggested experimenting with internal serpentine cool-
ing coils as a way to increase the heat removal capabil-
ity of the reactor. This experiment proved to be very
successful. Heat removal was increased so much that
the external cooler was no longer needed. The behavior
of the uidized bed was also markedly improved with
the addition of the coils. The second generation reac-
tors were designed with internal cooling coils only.
Medium to high pressure steam was generated directly
in the coils.
Entrained catalyst leaving the bed was originally
removed by externally mounted ceramic lter elements.
The hot reactor ogas is susceptible to afterburn
(Graham and Way, 1962). Long runs of pipes between
the reactor and lter housings were used to cool the gas
to about 260
C before the catalyst nes were removed.

The recovered dust was blown back to the reactor with
heated air to prevent thermal shock of the lter ele-
ments. Catalyst was purged periodically from the lters
to prevent buildup of very ne dust (below 10 micron)
in the reactor. The ne dust is produced by attrition,
and at high enough levels it can aect the uidity of the
catalyst. Ceramic lter elements are susceptible to
cracking from thermal shock and mechanical stress.
They were eventually replaced by lter elements con-
structed of glass-wool-wound steel pipe cores and
nally micrometalic lters with marked improvement
in operating reliability. With the improvement in
cyclone design and operation, the lters were even-
tually replaced by internal multistage cyclones.
Figure 4b shows a third-generation phthalic anhydride
reactor. Two uidized beds are stacked one on top of
the other. The bottom reactor is used for reaction. The
top reactor is a quench zone. This quench zone is
necessary to prevent afterburn of the reactor ogas
after the entrained catalyst is removed by the cyclones.
Figure 4 Badger phthalic anhydride reactors. (From Miseralis et al., 1991.)
A venturi scrubber is also provided to remove the small
quantity of catalyst carryover from the cyclones and to
protect downstream switch condensers in the event
that the cyclones malfunction.
2 GAS TO LIQUIDS
2.1 Sasol
Synthesis of hydrocarbons from H
2
and CO gases by
the FischerTropsch reaction is strongly exothermic
and occurs at about 200 or 340
C and 20 to 50 atmo-
sphere by the reaction:
nCO 2nH
2
! CH
2

n
nH
2
O 5
The higher temperature operation produces gasoline
range products and is used exclusively with uidized
beds. The lower temperature operation is used for wax
production in either xed bed tubular or slurry bed
reactors. Sasol of South Africa commercialized the
rst uidized bed FischerTropsch process in 1955.
The synthol uidized bed process produces hydrocar-
bons in the gasoline boiling range using an iron cata-
lyst. The scale-up and reactor design development are
depicted in Fig. 5 (Jones, 1991). Sasol 1A and B used a
circulating uidized bed design based on the original
10 cm pilot plant work of the WM Kellogg company in
the 1940s. These rst commercial reactors have a dia-
meter of 2.3 m and operate at about 2 m/s. They repre-
sent a 500-fold increase in gas throughput compared to
the pilot plant. Catalyst is circulated between the
upow leg of the reactor and the downow leg of the
disengaging hopper and standpipe. The reactors were
originally built with two banks of xed tube sheets for
heat removal. The reacting gas and catalyst mixture
owed through the tubes and was cooled by oil circu-
lating on the outside. Operating problems plagued the
startup and successful operation of the Sasol 1A and B
reactors (Shingles and Jones, 1986). The disengaging
hopper cone was too shallow. Funnel ow occurred,
causing bridging of the catalyst. The standpipe was
found to operate in stickslip ow (Chapter 21) and
was subject to bridging and loss of circulation. The
Figure 5 Sasol Synthol reactor development. (From Jones, 1991.)
reactors were designed originally to operate with con-
tinuous catalyst addition and withdrawal to maintain
catalyst activity. This was never practiced in these rst
generation reactors as the valve technology capable of
reliable operation was not available at that time. The
most serious problem was however with the tube
sheets. Localized pressure uctuations resulted in pre-
ferential gas ow through some tubes, leaving the
others with a dense solid pocket that quickly plugged
due to formation of waxy compounds. The increased
gas ow through the open tubes then resulted in tube
erosion.
It took Sasol two years to learn to operate these rst
reactors with acceptable run times between shutdowns.
In 1960s, a third reactor, Sasol 1C, was added; it used
longer tubes for increased heat removal and a taller
standpipe to improve the pressure recovery. In 1974,
Sasol, reacting to the world oil crisis, decided to build
the much larger Sasol 2 complex. A cooperative pro-
gram was started with Badger to review the Synthol
reactor operation and determine the best options to
eliminate the operating problems. Excellent accounts
of this cooperative development have been provided
by others (Shingles and Jones, 1986; Jones, 1991). A
brief review follows.
Sasol 2 was built using the Synthol concept and
was deemed a major success. The disengaging hopper
cone angle was increased to eect mass ow, eliminat-
ing previous bridging problems. The standpipes were
designed with smaller diameters, eliminating stickslip
ow. However, the most signicant design improve-
ment was the replacement of the xed tube sheets
with serpentine cooling coils. By allowing the gas/
solid mixture to ow on the outside of the coils, the
instability problems associated with previous xed
tube sheet design were eliminated, allowing reliable
continuous operation. According to Jones (1991), a
test coil bank was installed in Sasol 1C to demon-
strate feasibility and measure erosion. The test proved
successful. Very high heat transfer coecients were
measured, and the mechanical integrity of the design
was proven. The 10 Synthol reactors of Sasol 2
(Jones, 1991) have now operated for over 20 years
with no reported problems with the coils. In the
early 1980s, the Sasol 3 complex, essentially a copy
of Sasol 2, was added. The second generation Synthol
technology was licensed for use in the Mosselbay pro-
ject that started in 1991. The Sasol 2 reactors are
apparently still operated batchwise. As the catalyst
ages, the density is reduced, and the mass ow
through the reactor must be increased to maintain
constant holdup (Shingles and McDonald, 1988).
The fractional solid holdup within the reactor ranges
from 0.12 at the start of the run, to 0.17 at the end of
the run.
In 1974, as part of the original SasolBadger devel-
opment program, Badger tested and conrmed the
possibility of using a conventional xed uidized bed
(FFB) with the iron catalyst. At that time, this new
reactor concept was given a lower priority due to sche-
dule constraints of the Sasol 2 project and the need for
demonstrating the new concept. The new reactor con-
cept was bench tested in 1983 by Sasol in a 100 bbl/day
unit. In 1989, Sasol returned to this idea. By the early
1990s, a 3500 bbl/day demonstration unit, designed by
Badger, was brought on-line and proved to be an
instant success. Following this, an 11000 bbl/day unit
was started in 1995. Sasol has announced plans to
replace all of the Synthol reactors with the third-gen-
eration FFB reactors. Figure 6 shows the comparison
of the Synthol and FFB reactors. Some details of the
FFB design and development have been published else-
where (Silverman et al., 1986; Jaeger et al., 1990). The
FFB reactor is believed to operate in the turbulent
regime.
In the early 1980s, Sasol bench tested and then
piloted in a 5 cm unit, a new slurry phase reactor
concept for conversion of syngas to waxy liquids
using an iron catalyst. The wax is upgraded by con-
ventional technology to naphtha and diesel. In 1990,
Sasol, in cooperation with Badger, demonstrated this
new concept in a 100 bbl/day, 1 m diameter unit. A
commercial unit of 2500 bbl/day capacity (circa 5 m
reactor) was started in 1993. The slurry reactor is
reported to operate at 240
C and about 20 atmos-

pheres and will eventually replace the older xed
bed Arge reactors used at Sasol since 1955. The slurry
reactor is shown in Fig. 7. Syngas is bubbled into the
reactor via a support grid/distributor plate and rises
through the bed. Reaction between the syngas and the
liquid occurs on the surface of the suspended catalyst.
Gaseous products and unreacted syngas exit the bed
via the freeboard, while liquid product is withdrawn
from the bed. Internal serpentine cooling coils are
used to remove the heat of reaction by direct steam
generation.
The slurry phase reactor provides much better and
more exible temperature control compared to the
Arge reactor. It can operate at higher temperatures
and with more active catalyst without formation of
coke or catalyst breakup. The key operating issues
are catalyst/liquid separation and catalyst attrition.
Sasol has apparently resolved these design issues
successfully.
2.2 Exxon
Exxon (now ExxonMobil) has developed a two-step
process for conversion of natural gas to liquid hydro-
carbons. The process, referred to as AGC-21, utilizes a
uidized bed for catalytic partial oxidation/steam
reforming of natural gas to syn gas, and a slurry bed
reactor for conversion of the syn gas to hydrocarbons.
The hydrocarbons are processed further to produce
liquid products that are used as renery or chemical
plant feedstock.
The feed gas to the syn gas reactor is rst treated for
removal of H
2
S and sulfur compounds. The treated
gas, together with oxygen and steam, is then converted
to syn gas in a high-temperature high-pressure ui-
dized bed reactor (Goetsch et al., 1989). An alpha
alumina (95 wt% minimum) supported catalyst con-
taining 0.5 to 2.5 wt% nickel and substantially no silica
is used. The catalyst density is very high, between 2.4
and 3.9 g=cm
3
. Exxon claims that catalyst nes leaving
the reactor cyclone system can lay down in the over-
head piping and will promote the reverse reaction of
syn gas to methane, increasing methane slip. The above
catalyst characteristics are considered essential by
Exxon for minimizing nes generation and thus achiev-
ing high methane conversion. The preferred reactor
temperature is between 980 and 1000
C. Operating
pressure can be as high as 40 atmosphere, but the pre-
ferred range is 20 to 30 atmospheres. The uidized bed
syngas generator is shown in Fig. 8. Natural gas/steam
and oxygen enter the bed via separate spargers. The
reactor ogas is separated from entrained solids using
Figure 6 Sasol Synthol and FFB reactors. (From Jones, 1991.)
external cyclones and is then rapidly quenched to pre-
vent the reverse reaction to methane and carbon for-
mation by the Boudard reaction. Exxon claims that a
minimum cooling rate of 150
C and preferably 200
C
is required. A circa 1.5 meter pilot plant was operated
in early 1990s to prove the syngas reactor concept.
Conversion of syngas to hydrocarbons is by the
FischerTropsch synthesis. A multiphase slurry bed
reactor is used. Information on the slurry reactor is
limited, but the general design is expected to be simi-
lar to that in Fig. 7. Exxon claims that the addition of
an inert solid enhances bed expansion and reactor
performance.
2.3 Syntroleum
The Syntroleum company has announced the develop-
ment of an air-based gas conversion process. Syngas is
produced using an air-blown authothermal reformer.
The syngas and diluent nitrogen is then passed on a
once-through basis through the conversion reactor.
Syntroleum is developing xed, uidized, and slurry
bed conversion reactors using cobalt catalyst. The
company claims favorable economics at capacities as
low as 5,000 barrels per day and has joined up with
ARCO and Texaco to develop slurry bed conversion
reactors.
Figure 7 Sasol slurry bed reactor.
3 POLYMERIZATION OF OLEFINS
The UNIPOL gas phase uidized bed reactor for the
production of polymers was commercialized by
Union Carbide (now Dow Chemical) in 1968. This
reactor produced high-density polyethylene (HDPE).
The UNIPOL process was extended in 1975 to the
production of linear low-density polyethylene
(LLDPE) and in 1985 to polypropylene production.
In the late 1980s, BP Chemicals began licensing its
own gas phase Innovene uidized bed process in
competition with Union Carbide. The UNIPOL pro-
cess currently holds the lions share of the market
with over 120 reaction lines sold or under construc-
tion. The reactor design is similar for all types of
polymers and is shown conceptually in Fig. 9. A
similar reactor is used by BP Chemicals, which has
licensed 27 units.
Monomer, comonomer, hydrogen, and uncon-
verted monomer are fed through a distributor plate
into the uidized bed, while catalyst and cocatalyst
are fed into the bed above the distributor plate.
Polymerization occurs on the catalyst surface creating
particles in the 250 to 1000 micron range. The reac-
tion is highly exothermic and has a low per pass
conversion of the monomer. Chromium, Ziegler, or
the new generation of metallocene catalysts are used
in the process. Reactor operating conditions range
from 75 to 105
C and from 20 to 25 atmosphere

depending on the type of polymer being produced
and the catalyst. The bed aspect ratio has been
reported at 2.7 to 4.7.
Figure 8 Exxon syn gas reactor.
Key reactor design issues are catalyst addition
point, polymer withdrawal point, distributor plate
design, aspect ratio, operating velocity, and tempera-
ture control. Temperature uniformity is critical as the
reactor operates close to the polymer melting point.
Maldistribution can result in polymer sheet formation
or bed collapse; under severe conditions, there can be
complete solidication of the reactor.
The upper section of the reactor is expanded to
reduce velocity and minimize polymer entrainment.
In the UNIPOL process, the gas and the small amount
of entrained polymer are cooled and recycled back to
the process. In the BP design, a cyclone is used rst to
recover the polymer before the gas is cooled and
recycled.
Reactor capacity is limited by the system heat
removal capability. Internal cooling coils cannot be
used, as these will be quickly rendered useless by poly-
mer lm formation. Temperature control in the rst
generation reactors was accomplished by cooling of
the recycle gas. Capacity of the newer plants have
been increased by as much as 200% by operating in
the condensing mode. The condensing technology is
oered by Union Carbide, BP, and Exxon. The recycle
gas is cooled below its dew point to form some liquid.
The latent heat of the liquid is then used to increase the
reactor capacity. Union Carbide and Exxon have
joined forces to oer supercondensing operation with
as much as 50% liquid in the total feed. BP claims
operation with up to circa 40% liquid. In the Union
Carbide and Exxon process, the gas and liquid are fed
together through the distributor plate. In the BP pro-
cess, the gas and liquid are separated, and the liquid is
fed above the distributor plate (Newton et al., 1995).
Swing reactor operation, where the operating con-
ditions are changed to make dierent polymers, have
been considered theoretically but do not appear to
have been used in practice. Dual reactors in series
have been used for impact polypropylene copolymer
production (Burdett, 1992).
The polymer product and unreacted catalyst is with-
drawn from the reactor and degassed with nitrogen in
Figure 9 Unipol process.
a dual lock hopper system to drive o interstitial
hydrocarbon gases (Burdett, 1992). The amount of
unreacted catalyst withdrawn with the polymer is
extremely small owing to the very high productivity
of the catalyst. As a result, the unreacted catalyst is
left in the polymer.
4 HEAVY OIL UPGRADING
In addition to gasication (Chapter 15), several ui-
dized bed processes are available for upgrading of
heavy oil. These upgrading processes fall into two
groups. The rst group upgrades heavy oil by carbon
rejection. These include resid cracking (see Chapter
14), uid coking, and exicoking. The second group
upgrades heavy oil by hydrogen addition. These
include H-Oil, LC Fining, Combi-Cracking, HDH,
and U-can.
4.1 Fluid Coking
This continuous uidized bed process was developed
by Exxon (now ExxonMobil) in the 1950s. The process
was piloted in a 100 BPD unit. The rst commercial
plant was built in 1955 with a capacity of 3,800 BPD,
representing a 38-fold scale-up. A total of thirteen
units have been built, the largest having a capacity of
72,000 BPD. The process is shown in Fig. 10.
Heavy oil is sprayed into a reactor of ne coke par-
ticles with an average diameter of 150 to 200 micron,
uidized by steam. The reactor operates at 500 to
600
C. The reaction products are coke, light ends,

and distillates. The vapor products leave the reactor
Figure 10 Exxon Fluid Coking process.
as gas bubbles, pass through cyclones and into a scrub-
ber, where they are quenched before being fractio-
nated. The coke particles from the reactor are rst
steam stripped for product recovery and then ow to
the burner vessel, where a portion of the coke is burnt
with air. Hot coke is then returned back to the reactor
to supply the heat of reaction.
4.2 Flexicoking
This process uses the same conguration as the uid
coker but has an additional coke gasication step (Fig.
11), where excess coke is gasied to renery fuel. The
process is commerciallyprovenandis licensedbyExxon.
4.3 H-Oil and LC Fining
These processes use an ebullated reactor. The H-Oil
process is licensed by HRI (now Axens). The LC
ning is licensed by Lummus. Both processes have
been proven commercially. The H-Oil reactor is
shown in Fig. 12. Oil and hydrogen are fed into
the bottom of the reactor, containing an ebullated
bed of catalyst, where the oil is hydrogenated in
the liquid phase. The reactor is operated in back-
mixed mode by recycling the liquid. Catalyst activity
is maintained by on-line addition and withdrawal.
Five H-Oil units have been built ranging in capacity
from 2,500 to 35,000 BPSD. The rst unit was com-
mercialized in 1963.
4.4 Combi-Cracking
The heavy oil is rst hydrocracked in a liquid phase
uidized bed and is then hydrotreated in a xed bed
reactor. The upgraded product is very low in sulfur.
The process was developed by Veba. The liquid phase
reactor has been demonstrated in a 3,000 BPD unit.
Figure 11 Exxon Flexicoking process.
Figure 12 H-Oil ebullated bed reactor.
4.5 HDH
This process, developed by Intevep, has been only
piloted. It is very similar to the rst step in Combi-
Cracking. A slurry bed is used to hydrocrack the
heavy oil.
4.6 U-CAN
A bubble slurry phase reactor is used to upgrade the
heavy oil. The process was developed by SCN-Lavalin.
The reactor was demonstrated in a 5,000 BPD unit.
5 SEMICONDUCTOR SILICON
MEMC Electronic Materials uses a uidized bed reac-
tor originally developed by the Ethyl Corporation to
produce high-purity silicon for use in the semiconduc-
tor industry. The process was piloted (Ibrahim and
Johnston, 1990) using several pilot plant scale units
of increasing size before being demonstrated in a 125
tons per year market development unit. A commercial
plant with 1250 tons per year capacity was started in
1987. The high purity silicon is produced in the ui-
dized bed reactor by decomposition of silane in a
hydrogen atmosphere by the reaction
SiH
4
! Si 2H
2
6
This reaction is exothermic, but additional heat must
be provided to balance thermal losses. Recycled puri-
ed hydrogen and silane are introduced continuously
in the bottom of the reactor (Fig. 13), while high-purity
silicon seed particles are added and the product with-
drawn on a periodic basis. The product is cooled and
then collected in storage hoppers. An inert gas is used
to displace any entrained process gas. The product
silicon is classied to remove nes before being trans-
ferred to shipping containers. The hydrogen-rich reac-
tor ogas is cooled, cleaned of nes, compressed,
puried, and returned to the reactor
To maintain high purity, equipment and piping
made of high-purity silicon are used to transfer and
handle product and seed silicon. Storage and shipping
containers are stainless steel with uorocarbon lining.
The reactor is constructed with an internal lining
coated with silicon. Additional silicon deposition dur-
ing the course of normal operation of the reactor pro-
vides the ultimate high-purity containment of the
reactor. The authors indicate it is undesirable for silane
to decompose in the reactor above the uidized bed, as
this produces nes that deposit on the upper parts of
the reactor. Use of a quench zone in this section of the
reactor is further suggested as a way to prevent this
problem.
The chemistry of the process is complex.
Heterogeneous decomposition results in particle size
growth by direct deposition of silicon onto the particles
as the silane decomposes. Homogenous decomposition
occurs as the silane decomposes in the bubble or gas
stream and results in silicon dust ranging from submi-
cron size to 10 micron. These dust particles are elu-
triated from the reactor and represent an economic
Figure 13 MEMC high-purity silicon process.
loss. Dust generation increases with higher silane con-
centration, increased reactor temperature, a larger
reactor bubble holdup, higher gas velocity, and low-
ered bed height. However, without adequate bubble
fraction, solid mixing is reduced and the particles can
agglomerate when silane is decomposing on the sur-
face. In extreme cases, the whole bed can collapse.
Operating conditions of the reactor are reported as
600 to 800
C and 5 to 15 psig. Average particle size

is 700 to 1100 micron. Operating velocity is 1.2 to 3.5
minimum uidization. Under these conditions, the
reactor operates in a slugging mode.
Ibrahim and Johnston suggest a two-cycle reactor
operation. In the initial cycle, the reactor is operated in
a high productivity condition with silane concentra-
tions as high as possible. This operating condition
results in appreciable homogeneous decomposition
and nes generation. In the nal cycle, the silane con-
centration is reduced for a short time. This condition
promotes heterogeneous decomposition. A layer of
silicon of possibly no more than 1 micron is deposited
on the particles, which cements the loosely adhering
silicon dust formed in the initial cycle. It is presumed
that product withdrawal occurs immediately after the
nal cycle.
The silicon produced in this process typically con-
tains 10 to 44 ppmw of hydrogen. This hydrogen level
can be further reduced to 2 ppmw by heat treatment in
a second uidized bed reactor.
6 METALLURGICAL PROCESSING
6.1 Aluminum Flouride Synthesis
In 1948, Montedison and Lurgi piloted a uidized bed
for synthesis of AlF
3
from aluminum hydroxide
AlOH
3
and hydrouoric acid by the reaction
Al OH
3
3HF ! AlF
3
3H
2
O 7
The pilot plant produced 28 ton/day of AlF
3
with a
purity of 92 to 94% (Reh, 1971). The process has since
been commercialized successfully. Moisture-free alumi-
num hydroxide of about 45 micron average size is fed
into the expanded upper section of a two-zone uidized
bed (Fig. 14.) The hydroxide contacts superheated
hydrouoric acid in a zone of increased solid concen-
tration at the bottom of the lower reactor. Hot ue gas
is introduced immediately below the expanded section
to provide the endothermic heat of reaction and
increase operating velocity to around 140 cm/s, aect-
ing circulation of the solids through the cyclone to the
reactor bottom. In this manner the ue gas provides
the necessary heat of reaction without reducing the
concentration of the reactants. Reaction temperature
is around 530
C. Operating pressure is close to ambi-

ent.
6.2 Alumina Calcination
This process represents the rst large-scale application
of circulating uidized bed (CFB) reactors in the
metallurgical industry. The process was piloted by
VAW and Lurgi in a 24 ton/day pilot plant in the
early 1960s. A 500 ton/day unit was started in 1970.
Over 30 units are currently in operation with capacities
of up to 2000 ton/day representing over 50% of
worlds alumina Al
2
O
3
production. The process is
depicted in Fig. 15. Wet aluminium hydroxide is
dried in a cyclone dryer with hot ue gas from the
calciner. After recovery in an electrostatic precipitator,
the dry hydroxide is air lifted into a venturi preheater
where it contacts hot gas from the calciner. The pre-
heated hydroxide is mixed with circulating solids from
the calciner in the recycle cyclone and is fed to the
bottom of the calciner via the seal loop. Endothermic
heat of reaction is provided by ue gas from a burner.
The calciner is operated in a staged combustion mode
for NO
X
control. Operating temperature can be
adjusted between 800 and 1200
C depending on the
properties required for the product. Operating pressure
is near ambient. The product is cooled by heat
exchange with cold air and water in the cooler.
6.3 Ore Roasting
Fluidized beds have been used for the roasting of ores
since 1947. Over 550 units have been built to date.
Both KTI/DorrOliver and Lurgi oer conventional
single or double stage uidized bed roasters.
Roasting applications include iron pyrite and pyrrho-
tite for sulfuric acid, roasts for arsenic removal and
roasts of metal suldes for recovery of zinc, copper,
cobalt, nickel, gold, tin, and molybdenum. Both slurry
feed and dry feed systems are used. Refer to Kunii and
Levenspiel (1991b) and Bunk (1990) for additional
information.
In 1989, Lurgi commercialized a 575 ton/day wet
feed slurry circulating uidized bed refractory gold
roaster operating at about 640
C and near ambient

pressure (Peinemann, 1990). A second unit of identical
design was added in 1990. A third unit with dry feed
handling and a capacity of about 2000 ton/day gold
containing refractory ore was started in 1991. KTI/
DorrOliver also has a circulating unit in operation
7 OTHER PROCESSES
7.1 UOP/Hydro Methanol to Olens
UOP and Norsk Hydro have jointly developed and
piloted a uid bed process for the conversion of metha-
nol to olens (Fig. 16.) The process uses Union
Carbides (now part of UOP) SAPO-34 catalyst. A
two uidized bed reactor/regenerator system is used,
between which catalyst is circulated. Methanol conver-
sion is essentially 100%. The reaction produces a small
quantity of coke that will slowly deactivate the cata-
lyst. To maintain activity, catalyst is circulated to the
regenerator where the coke is burnt o with air. The
heat of reaction is removed by steam generation. As
the reactor severity is increased, the ethylene/propylene
ratio and the coke yield increase, while the overall ole-
n yield drops (Vora et al., 1996). A 0.5 MT/day
demonstration unit was operated in the mid-1990s.
UOP has announced the successful completion of the
demonstration unit and the availability of this process
for licensing.
Figure 14 Lurgi aluminum uoride synthesis.
7.2 MobilBadger Technologies
Three uidized bed processes are available for license
from MobilBadger (now ExxonMobilBadger).
These are MBR (Mobil benzene reduction), MOG
(Mobil olens to gasoline), and MOI (Mobil olen
interconversion). All three use zeolite catalyst and a
dual dense uidized bed reactorregenerator system.
The MOI process is shown in Fig. 17. The other two
processes are conceptually very similar.
The MBR process reduces the benzene content of
light reformate, FCC gasoline, or pyrolysis gasoline to
below 1 vol% while boosting pool octane up to one
point. The zeolite catalyst alkylates benzene with light
olens to form higher octane C8 to C10 aromatics.
Single pass benzene conversion is 60 to 70%.
However, overall benzene conversion can be increased
to 90% by recycling a portion of the MBR reactor
euent. Once-through olen conversion is greater
than 90%.
The MOG process converts olens, including light
olens contained in dilute streams, to C5+ gasoline
pool components with yields of 61 weight percent at
high severity and 75 weight percent at low severity.
Olen conversion is over 90%.
The MOI reactor operates at a lower pressure and
shorter contact time than the MOG reactor to maxi-
mize light olen production. Operating temperature is
Figure 15 Lurgi alumina calcining.
about 538
C. By recycling, up to 40 weight percent of

the total feed can be converted to propylene and iso-
butylene. The reactor product slate can be changed by
adjusting its operating temperature.
7.3 Catalytic Oxidation of Chlorinated Byproducts
The Catoxid uidized bed process was developed by
BF Goodrich (now Geon) in collaboration with
Badger. The rst unit was started in 1974. At least 4
units have been built.
Chlorinated hydrocarbon liquid streams from the
oxychlorination process (this chapter), containing
nely divided carbonaceous solids and metallic corro-
sion products in solution and suspension are pumped
into the lower portion of the uidized bed. Air is used
to uidize the bed and to oxidize the feed. The reaction
occurs at below 540
C. Combustion of the feed is

essentially complete with no signicant breakthrough
of elemental chlorine and only minor breakthrough of
chlorinated hydrocarbons (Benson, 1979). The heat of
combustion is recovered by generating medium to high
pressure steam inside coils placed in the bed. The pro-
Figure 16 UOP/Hydro methanol to olen reactor system.
cess requires auxiliary fuel only when the feed contains
signicantly more than 70% chlorine. The auxiliary
fuel is required to sustain the correct bed temperature
and satisfy the hydrogenchlorine ratio for hydrogen
chloride production.
The combustion gas leaving the Catoxid process
contains hydrogen chloride, water, nitrogen, oxygen,
and carbon oxides and can be recycled directly
back to the oxychlorination reactor, as shown in
Fig. 3.
The Catoxid catalyst is robust and is compatible
with the oxychlorination reaction. Catalyst makeup is
required only for attrition losses. The catoxid catalyst
dust moves through the oxychlorination uid bed
reactor and leaves the system with the oxy catalyst
nes.
7.4 Isophthalonitrile
GB Biosciences (now part of Zenneca) has been oper-
ating a dual uidized bed process to produce isophtha-
lonitrile by ammoxidation of m-xylene, by the reaction
C
8
H
10
2NH
3
3O
2
! C
8
H
4
N
2
6H
2
O 8
The process was commercialized in 1976. A vanadium
catalyst with Geldard group A/C characteristics is
used. The catalyst is circulated between a reactor oper-
ating at 400
C and a reoxidizer operating at 427
C.
The reactor operating pressure is not reported but is
expected to be 1 to 2 atmosphere. Catalyst losses are
kept to a minimum by the use of sintered metal lters
(Fig. 18).
Figure 17 MobilBadger MOI reactor system.
REFERENCES
Benson JS. Catoxid for chlorinated byproducts.
Hydrocarbon Processing, pp 107108, October 1979.
Bolthrunis CO. An industrial perspective on uid bed reactor
models. Chem Eng Progress, pp 5154, May 1989.
Bunk S. Fluid-bed roasting advances. Randol Gold Forum,
pp 173174, September 1990.
Burdett ID. A continuing success: the UNIPOL process.
Chemtech, pp 616623, October 1992.
Contractor RM. US Patent 4,666,802 to E I DuPont de
Nemours and Company, May 26, 1985.
Contractor RM, Bergna HE, Horowitz HS, Blackstone CM,
Malone B, Torardi CC, Griths B, Chowdry U, Sleight
AW. Butane oxidation to maleic anhydride over vana-
dium phosphate catalysts. Catalysis Today 1:4958, 1978.
Contractor RM, Pateince GS, Garnett DI, Horowitz HS,
Sisler GM, Bergna HE. A new process for n-butane oxi-
dation to maleic anhydride using a circulating uidized
bed reactor. Proceedings of the Fourth International
Fluidized Beds Conference, Hidden Valley, 1993, pp
387391.
Goetsch AD, Say GR, Vargas JM, Eberly PE. US Patent
4,888,131 to Exxon Research and Engineering, 1989.
Graham JJ. Safety features of uid bed process. Chem Eng
Progress 66(9):54, September 1970.
Graham JJ, Way PF. Phthalic anhydride by uid bed pro-
cess. Chem Eng Progress 58(1):96100, 1962.
Ibrahim J, Johnson SW. High purity polysilicon production
from silane by uidized bed process. Proceedings 2nd
Symposium on Fluidized Bed Processing of Powder
Materials. Soc Chem Eng Japan 1990, pp 2636.
Figure 18 GB Biosciences isophthalonitrile reactor system.
Jaeger B, Dry ME, Shingles T, Stynberg A. Experience with a
new type of reactor for FischerTropsch synthesis.
AIChE Spring National Meeting, Orlando, March 1990.
Johnsson JE, Grace JR, Graham JJ. Fluidized-bed reactor
model verication on a reactor of industrial scale. AIChE
J 33(4):619627, April 1987.
Jones DH. Development of Sasols FischerTropsch technol-
ogy. AIChE Spring National Meeting, Houston, Texas,
April 1991.
KirkOthmer. Encyclopedia of Chemical Technology. 4th
ed. Vol. 24. New York: John Wiley, 1996, pp 860865.
Kunii D, Levenspiel O. Fluidization Engineering. 2nd ed.
Boston ButterworthHeinemann, 1991a, p 32.
Kunii D, Levenspiel O. Fluidization Engineering. 2nd ed.
Boston Butterworth-Heinemann, 1991b, p 51.
Miseralis CD, Way PF, Jones DH. The evolution of ui-
dized-bed reactors in the phthalic anhydride production.
AIChE Spring National Meeting, Houston, Texas, April
1991.
Naworski JS, Velez ES. Oxychlorination of ethylene. In:
Leach BE, ed. Applied Industrial Catalysis. Vol. 1.
Newton D, Chin JC, Power M. Advances in uidized bed
gas-phase technology. HTI Quarterly, pp 8185, Autumn
1995.
Ohta M, Yokura M. Fluidized-bed reactor and reaction pro-
cess using the same. International Patent Application
PCT/JP96/00171 to Asahi Kasei Kogyo KKK, January
30, 1996.
Peinemann B. Lurgis circulating uid bed roasting process
improves gold recovery. Randol Gold Forum, September
1990, pp 169172.
Pell M, Jordan SP. Eects of nes and velocity on uid bed
reactor performance. AIChE Symp Ser. Vol. 84, No. 262,
pp 6872, 1987.
Reh L. Fluidized bed processing. Chem Eng Prog 167:5863,
1971.
Shingles T, Jones DH. The development of synthol circulat-
ing uidized bed reactors. Chem SA, pp 179182, August
1986.
synthol CFB reactors. Circulating Fluidized Bed
Technology II, Compiegne, 1988, pp 4350.
Silverman RW, Thompson AH, Stynberg A, Yukawa Y,
Shingles T. Development of a dense phase uidized bed
FischerTropsch reactor. Fluidization V. New York:
Volk WC, Johnson CA, Stotler HH. Eect of reactor inter-
nals on quality of uidization. Chem Eng Prog 58:4447,
1962.
Vora BV, Lentz A, Marker TL. UOP/Hydro MTO process.
World Petrochemical Conference CMAI, Houston,
March 1996.
17
Applications for Coating and Granulation
Gabriel I. Tardos
The City College of the City University of New York, New York, U.S.A.
Paul R. Mort
Procter & Gamble, Cincinnati, Ohio, U.S.A.
1 SUMMARY
We present, in this chapter, applications of uidized
beds to coating and granulation of powders. The pro-
cess, widely used in industry, is designed to agglomer-
ate ne powders into larger granules and/or to coat
large particles or granules with nes. To achieve aggre-
gation and growth and to ensure sticking together of
powdery particles, a binder as a solution or a melt is
used. The goal of this work is to present a general
practical and theoretical framework of binder granula-
tion that takes an agglomeration process from binder
selection and testing, to granule formation, growth,
and consolidation, and nally to granule deformation
and breakup.
For agglomeration and granule growth to take
place, a certain amount of binder has to be introduced
into the uid bed granulator: special instrumentation
and procedures for binder selection are presented in
this chapter. In such granulation processes as deter-
gents and pharmaceutical products, both the powders
to be agglomerated and the binders are dened by the
formulation, and little liberty is given to alter the
chemistry. Binder selection in this case is practically
reduced to adjusting the properties of the binder using
small amounts of additives (mostly surfactants) and
tailoring the binder to exhibit specic behavior. This
allows ne-tuning of binder properties that include
surface wetting, spreading, adsorption, binder
strengthening, and solid bridge strength.
The bulk of the presentation is dedicated to the
theory of growth kinetics during granulation and
the prediction of critical sizes that delimit dierent
regimes of granulation. Several dimensionless para-
meters based on energy dissipation principles are pre-
sented and examples given about how these
parameters, and the critical sizes they dene, can be
used to predict the outcome of granulation and the
scale-up of the process. The above theoretical frame-
work is then tested with experimental data from the
literature and with granulation results obtained by the
present authors.
A section is dedicated to the interpretation of gran-
ulation experiments performed at dierent scales and
how these can be used in scale-up of the process from
laboratory and pilot size to full-scale industrial equip-
ment. It is only very recently that criteria for such
scale-up have become available mainly from theoreti-
cal considerations of granule growth and breakup at
the micro-scale, and this new knowledge is briey
presented. Finally, some recent computations are pre-
sented in which the process of both agglomeration and
growth and deformation and breakup are simulated
using numerical techniques. These results are signi-
cant because they shed light on the details of the
micro-scale process and aid in scale-up.
1 INTRODUCTION
Granulation is a size enlargement operation by which a
ne powder is agglomerated into larger granules to
generate a specic size and shape, improve owability
and appearance, and, in general, produce a powder
with specic properties such as dissolution rates, gran-
ule strength, and apparent bulk density. Granulation
of powders is practiced at an enormous scale in indus-
tries including detergents, foods, agricultural chemi-
cals, and pharmaceutical products. During binder
granulation, a liquid binder such as a solution or a
melt is pumped, poured, or atomized onto an agitated
bed of dierent powders contained in a mixer whose
main role is to provide shearing forces in the powder
mass. As solvent evaporates from the binder (liquid) or
the melt thickens, powder particles stick together and,
as interparticle bridges strengthen, larger granules of
the original powder are formed. These granules are
further consolidated by forces in the mixer and, upon
nal solidication of the binder or melt, strong
agglomerates are left behind.
There are several key product transformations that
occur in granulation processes, including binder ato-
mization, uidization, wetting and spreading of bin-
ders on powder surfaces, agglomeration (including
nucleation, coalescence, and layering), consolidation,
binder solidication, drying, and attrition. In addition,
a reactive binder can be used to create solid bridges of
the reaction product between particles. In this later
case, the kinetics of the chemical reaction is an addi-
tional transformation that is key to the overall process
result. The interdependence of the many transforma-
tions involved in a typical granulation further adds
complexity to the process. Given this degree of com-
plexity, one is compelled to identify the key individual
transformations and then reduce them to a fundamen-
tal basis, i.e., to micro-scale interactions. Fundamental
understanding on the micro-scale level provides then a
foundation for predicting the interactions on the
macro-scale and the eects of process parameters and
key material properties.
Binder granulation can be achieved in dierent types
of mixers ranging from rotating drums and pans to
high shear mixers and uidized and spouted beds.
During the present analysis, the accent is put on uid
bed agglomeration and layering or coating, although
combined mixers that include both uidization and
mechanical agitation are also considered. Fluidized
bed granulation is in some ways dierent from other
types of mixer granulation in that the gas supplied to
produce powder agitation through uidization also
causes binder evaporation and cooling (or heating) of
the powder. In addition, particle size increase in a ui-
dized bed is associated with many changes in uidiza-
tion characteristics, the most important of which are
the mixing properties of the bed. These interacting
phenomena make uid bed granulation by far the
most complex while at the same time the most versa-
tile, allowing drying and cooling transformations to be
carried out simultaneously with size increase (agglom-
eration). On the other hand, simultaneous transforma-
tions pose dicult challenges to scale-up of industrial
applications.
Several excellent reviews of uid bed granulation
were published by Nienow and coworkers (Nienow,
1983; Nienow and Rowe, 1985; Nienow, 1994), in
which the operation is described in detail. Heat and
mass balances on the system are presented and a gran-
ule growth model is proposed. The advantages and
disadvantages of the operation are discussed. An over-
all review of all types of granulation was also presented
by Sommer (1988). The present contribution diers
from the above works by looking at the micro-scale
processes of granule formation, growth, and breakage
and by studying the behavior of liquid bridges between
agglomerating particles. The micro-scale view is
powerful in that it generates an overall picture of bin-
der granulation from such specic phenomena as par-
ticle coalescence, layering of nes, deformation and
breakage of wet granules, and nal product consolida-
tion.
2. THE FLUID BED AS A MIXER/
GRANULATOR
Some typical uid bed granulation equipment and a
few of their main characteristics are given in Fig. 1.
The conventional uid bed granulator (Fig. 1a) can
be operated in either top (as shown) or bottom
spray mode. The bed is usually tapered, as shown, to
improve solid circulation. The spouted bed granulator
(Fig. 1b) is somewhat dierent in that it has no distri-
butor and that there is a calibrated gap between the
nozzle and the bottom of the spout that allows dis-
charge of granulated material. It is common industrial
practice to combine the two uid beds presented so far
Figure 1 Schematic representation of some industrial granulators. (a) Conventional uid bed granulator. (b) Conventional
spouted bed granulator. (c) Spouted bed with draft tube. (d) High shear rotor granulator.
in large units with the spouted bed forming units,
which when combined yield large granulation areas
each with individual spray nozzles. The uid bed
with the draft tube (Fig. 1c) is mainly used for coat-
ing operations. The tube is used to create strong solid
circulation in the bed so that a leaner phase is created
inside as particles are blown upward in the tube and a
more dense phase at the periphery where drying takes
place.
The tendency in present industrial granulation is
oriented toward the use of combination equipment
such as the high shearmechanical uidized bed
shown in Fig. 1d. The device has a rotating distributor
(that can be at or made of several intertting cones)
through which air is pumped. Nozzles are placed at the
periphery, and a strong solid circulation is achieved
due to wall friction on the distributor and on the
tapered section of the housing. These devices are very
versatile and allow, besides very eective particle
growth and coating, also drying, heating, and grinding
operations in a very short operation cycle. Several
other constructions of combination devices are
known, and these are exhaustively reviewed by
Ormos (1994).
The main purpose of the uid bed in granulation is
to act as a mixer for the granulating powder. This is
achieved by creating shear layers in the mix, character-
ized by an average shear rate, ,
, where particles move

relative to each other. The main characteristic of the
uid bed is the relative velocity imparted to the parti-
cles, U
0
, that is a strong function of the size of the
particles and the gas velocity in the bed and was
shown to be given by (Ennis et al.,1991)
U
0
- a,
= 18
U
B
a
D
B
o
2
(1)
where a is the average particle size, U
B
is the bubble
velocity, D
B
is the bubble diameter, and o is the dimen-
sionless bubble spacing. The rst expression on the
RHS applies to combination mixers in general while
the last expression on the RHS of Eq. (1) applies to
uidized beds with no rotating parts where shear is
induced by the motion of bubbles only.
To obtain certain size granules from a uid bed
granulator, several other unit operations have to be
used. These are combined in so called granulation
circuits where the uid bed is the central unit with
other peripheral equipment also included. An example
of such a circuit is given in Fig. 2 (after Dencs and
Ormos, 1993). The uid bed granulator shown is of a
very special kind and, as seen, is equipped with grind-
ing rollers activated by a motor; this unit performs
granulation and grinding operations simultaneously.
In general, however, these two operations are per-
formed separately in dierent pieces of equipment
and therefore a conventional granulation circuit also
includes a grinder and some sieving equipment for
granule size control. Other peripheral equipment
includes tanks and mixers for the preparation and
feeding of the binder (Binder solution in Fig. 2),
Figure 2 Schematic representation of a uid bed granulation circuit. Shown in the circuit is a uid bed grindergranulator. 1.
Fluid bed houseing, 2. Fluidized bed, 3. Grinding roller, 4. Drive, 5. Spray head, 6. Binder solution, 7. Pump, 8. Fluidization air
fan, 9. Air heater, 10. Air control valve, 11. Cyclone, 12 and 13. Rotary valves, 14. Filter bag-house. (After Dencs and Ormos,
1993.)
pumps, heaters for air supply, gas cleaning equipment
such as cyclones and lters for the separation of nes
and granule discharge, and transport devices. More
complex circuits have several uid bed granulators in
series in which dierent operations such as granula-
tion, layering, wetting, drying, heating, and cooling
are also performed. For details on such specialty
devices, the reader is referred to the review of Ormos
(1994) and references within.
3 MICROSCOPIC PHENOMENA
The role of the spray nozzle shown in Figs. 1 and 2 is
to introduce binder (liquid) into the mixing powder
mass in as uniform a layer as possible. Consequently,
we encounter, in the shearing powder mix, individual
particles covered by a liquid layer. The process by
which these ne powder particles, partially or totally
covered by a liquid layer, stick together in a shearing
mass of powder to form large granules that possess, in
the end, enough strength to survive in the granulator is
quite complex (Ennis et al., 1990, 1991; Ennis, 1990;
Tardos et al., 1985, 1993, 1997).
Figure 3 depicts some, more common, ways in
which agglomerate growth may occur. It must be
emphasized that these micro-scale mechanisms depend
on the degree of binder dispersion in the powder.
Typically, dispersion varies with binder atomization,
addition rate, and state of uidization or shear in the
mixer. Nucleation is dened as the sticking together of
primary particles owing to the presence of a liquid
binder on the solid powder surface (Fig. 3a).
Coalescence, on the other hand, is the process by
which two larger agglomerates combine to form a
granule (Fig. 3b). In coalescence, the porous granule
surface is saturated with binder, and the colliding gran-
ules are suciently malleable to allow for deformation
and bonding (Kristensen et al., 1985; Kristensen, 1991;
Kristensen and Schaefer, 1987; Schaefer et al., 1990;
Schaefer, 1988; Tardos et al., 1997). Other modes of
granule growth are layering of a binder-coated granule
by small ne particles (Fig. 3c) and the capture of nes
by a partially lled binder droplet (Fig. 3d).
The growth mechanisms shown in Fig. 3 depend on
the collision and bonding of particles. One can assume
that shear forces in a mixer will cause particles to col-
lide at some point along their trajectory. For bonding
of particles to occur, it is essential that some binder be
present at the point of contact, as depicted in Fig. 4.
This can be in the form of pure binder, as shown in the
gure, or the solid surface of the particle at the contact
point can exhibit some softness, i.e., malleable char-
acter, which would allow surfaces to bond upon con-
tact. From the simplied picture of Fig. 4, all
mechanisms of Fig. 3 can be easily reconstructed if
one allows for variation in size and nature (solid or
liquid) of the colliding particles.
Figure 3 Granule growth mechanisms (a) Agglomerate for-
mation by nucleation of particles. (b) Agglomerate growth by
coalescence. (c) Layering of a binder-coated granule. (d)
Layering of partially lled binder droplet.
Figure 4 Coalescence of two binder covered particles
(granules)
The study of the kinetics of granule growth and con-
solidation in binder-granulation can be undertaken
using the picture in Fig. 4 that encompasses, in a sim-
plied way, all mechanisms depicted in Fig. 3. The
relative velocity, U
0
, between binder-coated particles
is generated by the mixer, i.e., the uidized bed, in
this case. The magnitude of the relative velocity is
mixer dependent and is predicted approximately by
Eq. (1). The kinetics of the process essentially reect
the conditions under which two colliding particles will
stick together or rebound in the shear eld: sticking
will yield growth while rebounding will not. The rate
of growth will depend on the relative sizes of the indi-
vidual particles involved. The mechanisms of coales-
cence (Fig. 3b) and nucleation (Fig. 3a) involve
collisions of similar-sized particles and will yield higher
growth rates than layering (Fig. 3c) and engulfment
(Fig. 3d), which incorporate nes particles into a larger
granule. Once formed and grown to a certain size,
granules have to survive in the shearing mass; special
conditions between the magnitude of shear in the outer
shearing mass and the inner strength of the granule
have to be satised in order for this to happen.
Growth is also inuenced by granule consolidation
that is closely related to the viscosity of the binder
after granule formation and the time allowed for defor-
mation of the formed bridges.
Several essential conditions have to be fullled a
priori in order for granule growth and consolidation
to occur. These conditions pertain to the presence of
appropriate amounts of binder on the granular surface at
the points of contact and the binders physical and ow
properties in the bridge; these aspects of granulation
will be examined rst. Following this, granule growth
kinetics and considerations regarding granule consoli-
dation will be presented.
4 CONDITIONS FOR GRANULATION
It is intuitively obvious from the schematic representa-
tion in Fig. 4 that a certain amount of binder is
required in the powder mass before enough will be
present on the surface to ensure stickiness. This criti-
cal, minimum amount of binder is an important char-
acteristic of the system and must be determined
beforehand. Equally important are (1) the initial dis-
tribution of the binder within the bulk powder and (2)
the time span over which the binder either spreads on
the granular surface and/or penetrates into the pores of
the powder. Also critical is the time over which the
binder bond increases in strength through evaporation
of solvent, reactive transformation, cooling, or other
mechanism of solidication. The critical binder/pow-
der ratio, the characteristic spreading/penetration time,
and the binder strengthening rate will be discussed
before the theory of growth kinetics is presented.
4.1 Critical Binder (Liquid)/Powder Ratio
Mixer torque rheometry is a useful method of deter-
mining critical binder/powder ratios. There has been a
large amount of recent work on this subject relative to
scale-up of high-shear mixer-agglomerators (Landin et
al., 1996a,b; Rowe and Parker, 1994; Hancock et al.,
1994). The instrument consists essentially of a mixer in
which the powder is slowly agitated and the binder is
continuously introduced by a metering pump. The
instrument allows the torque of the mixer to be mon-
itored continuously. An important condition to be ful-
lled during this procedure is to allow adequate time
during binder addition for the liquid to spread and
adsorb into the interstices before new binder is intro-
duced. In this way one ensures that the nite spreading
and adsorption rate of the liquid does not inuence the
results.
To illustrate the above procedure, the use of a
mixer-torque device to measure torque versus binder
content (di-butyl-phthalate, DBT) for a sodium carbo-
nate mixture is shown in Fig. 5. A slight increase in
measured torque occurs at 25% by volume while a
sharp increase occurs at 65% by volume DBT.
These two points correspond to the saturation of the
capillary-loaded granules and in particular to the
Figure 5 Experimental torque vs. binder/solid ratio for a
porous sodium carbonate powder.
crossover to the funicular and the capillary states,
respectively. Granulation of the powder shown in
Fig. 5 can be performed at the lower binder content
of 2530% binder. Since binder is only lling some
voids, growth will occur mostly by nucleation with
very little layering and no coalescence because insu-
cient binder is present on the granule surface. At liquid
loading corresponding to the drastic increase in torque
(at 65% binder), all mechanisms of growth occur
including coalescence: this value is considered for this
powder to be the critical binder/solid ratio.
One has to note that the rate of binder addition and
the shear rate in the mixer also aect the binder/solid
ratio actually achieved during industrial granulation.
Decreasing the shear rate and increasing the binder
addition rate will result in excessive growth and lump-
ing not because the critical binder/powder ratio was
exceeded but because not enough time was given for
the liquid to spread and adsorb into pores. The above
considerations are presented here as an example of
how to determine binder/powder (L/S) ratios. The
instrumentation used is not unique, and in general
any small size mixer of the medium or high shear
type can be used (see also work by Tardos et al.,
1993). Alternatively, the binder/powder (solid) ratio
can be determined in actual granulation runs by per-
forming granulation with increasing amounts of binder
until a wet mass is obtained. It is essential, however,
that this determination be done properly, since too
high or too low amounts of binder yield undergranu-
lated and dusty or overgranulated and sticky products.
4.2 Binder and Interparticle Bridge Properties
Under normal conditions of granulation, the binder is
introduced into the powder bed as a spray of small
droplets. These droplets impact on the solid powder
surface and deposit on it: if solidliquid contact angles
are such that surface spreading can take place, the
binder droplet will atten and cover an ever-increasing
area of the surface. At the same time, however, liquid
will penetrate into the surface pores of the granule.
Both surface wetting and spreading are necessary,
since a nonwetting liquid will either not stick to the
surface or be present on or cover a very small area,
thereby restricting the number of collisions that yield
growth. Penetration into the granule surface pores is
also required to give the granules the malleable, plastic
property needed for coalescence (Fig. 3b).
To characterize spreading/penetration characteris-
tics of binder/powder systems, sessile drop experiments
are performed using commercial goniometers. Binders
are spread onto tablets pressed from the powder to be
granulated and the process is observed under a high-
magnication microscope. In a more sophisticated ver-
sion of the instrument, the solidliquid contact line is
digitized by a computer and the liquid volume and
spreading area is computed as a function of time (see
for details Mazzone et al., 1987; Ennis et al., 1990;
Tardos and Gupta, 1996). If spreading, penetration,
and liquid bridge formation are found inadequate for
a given binder, dierent binders are chosen or, if this is
not an option, binder properties are altered by using
additives, surface active agents, etc.
A further condition of successful granulation is the
production of granules that gain sucient strength so
that they do not fall apart once they are formed. This is
achieved in granulation by choosing an appropriate
binder that generates bridges that strengthen in time
owing either to solvent evaporation or drying or to
cooling of the melt. Strengthening is a result of
increased binder viscosity and depends on the binder
chemistry and on the heat and mass transfer rates in
the granulator. Extensive research showed that, in gen-
eral, rates of heat and mass transfer in uid bed gran-
ulators are quite high, and so the choice of the
appropriate binder chemistry is crucial.
A special instrument was developed (Ennis, 1990;
Ennis et al., 1990; Tardos and Gupta, 1996; Tardos
et al., 1997) to test binders and measure their strength-
ening rates. The instrument is essentially a force trans-
ducer and an oscillating mechanism activated by an
electric motor mounted in the view of a projection
microscope. Two powder granules or tablets pressed
from the ungranulated material are mounted (glued)
onto the transducer and the oscillating arm. A droplet
of binder is inserted between the two particles (tablets),
and a pendular bridge is formed. The oscillator is put
in motion and the bridge strength is recorded by the
transducer as a function of time. Both the amplitude
and the frequency of oscillation and the bridge tem-
perature can be changed at will. At the same time,
pictures of the deforming bridge can be taken, and
imperfections in wetting and spreading can also be
observed.
The measured force due to the bridge (Gupta and
Tardos, 1996; Tardos et al., 1997) increases from an
initial value that can be calculated from the knowledge
of the liquid surface tension, to a maximum value after
which the mostly solidied bridge ruptures. Both the
maximum force and the critical bridge rupture time are
important for granulation, but it was found that the
critical strengthening time had to be within very nar-
rowly dened limits for the granulation to be success-
ful. These limits depend on the granule properties to be
achieved: for example, to produce open, high-porosity,
easily soluble granules, the binder strengthening time
should not exceed one-third to one-half of the total
granulation time, while in coating or layering opera-
tions it must be very short, of the order of several
seconds or a fraction of a minute. The above ndings
are quite signicant, since they show a clear way to
design solvent evaporation in the bed when the binder
is a solution, or the cooling time when the binder is a
melt. One has to stress, however, that the above esti-
mates are probably not very general and may require
corrections for dierent pairs of powders and binders.
Since the experimental procedure and the instrument
are quite simple and straightforward to use, such mea-
surements should be easy to perform for each specic
case.
5 GRANULE GROWTH KINETICS
It will be assumed for the present considerations that
sucient binder is present in the granulator as deter-
mined by the binder/powder ratio, and that the binder
is appropriately spread on enough granular surfaces to
ensure that most random collisions between particles
will occur on binder-covered areas. It will also be
assumed that the particles are mostly spherical having
a characteristic dimension, a. The liquid is character-
ized by its surface tension , and its viscosity j. In those
cases where the surface of the colliding particles is itself
soft and malleable, the viscosity j must be taken as
that of the surface layer. The relative velocity U
0
is
taken to be only the normal component between par-
ticles, as seen in Fig. 3, while the tangential component
is neglected.
The system of the two approaching particles is
shown schematically in Fig. 6. The particles are
depicted after the liquid layers on their surfaces have
already touched and a liquid bridge has formed. While
there may be some additional binder on the free sur-
face of the particles, this is not shown in the gure and
is irrelevant for the collision process at hand. As the
two particles approach with the relative velocity U
0
,
the liquid will be squeezed out from the space between
them to the point where the two solid surfaces will
touch. A solid rebound will occur based on the elasti-
city of the surface characterized be a coecient of res-
titution, e, and the particles will start to move apart.
Liquid binder will now be sucked into the interparticle
gap up to the point where a liquid bridge will form;
upon further movement the bridge will elongate and
nally rupture as the particles separate entirely. When
the particles are deformable (malleable) throughout as
in the process depicted in Fig. 3b, rebound will not
occur, and coalescence will depend on the deformabil-
ity of the granulebinder system.
It was shown by many researchers (a summary of
pertinent work in this area is given in Ennis et al.,
1990) that under fairly general conditions the total
force, F, induced by a liquid bridge between two
solid particles can be calculated from the summation
of two eects: a surface tension contribution propor-
tional to the bridge volume or the lling angle, , and
the surface tension of the uid, ,, and a viscous con-
tribution dominated by the relative velocity and the
viscosity of the liquid, j. The superposition gives accu-
rate results and can be expressed analytically by
F = F
vis
F
cap
=
3jU
0
a
2
4h
,a sin
2
C
0
2 ( ) [ ]
(2)
where C
0
is the LaplaceYoung pressure deciency due
to the curvature of the free surface of the liquid and h
is the thickness of the liquid layer (see Fig. 6).
5.1 Conditions of Coalescence
The outcome of the collision of two binder-covered
particles is determined by the ratio of the initial kinetic
energy of the system and the energy dissipated in the
liquid bridge and in the particles (Ennis et al., 1991;
Tardos et al., 1997). This can be expressed analytically
by the denition of a so-called Stokes number related
to particle coalescence, St
coal
:
Figure 6 Liquid bridge formed between two moving, sphe-
rical particles covered with a liquid layer.
St
coal
=
initial kinetic energy
dissipated energy in the bridge
St
coal
=
2m
p
U
2
0
2F
vis
h
=
8,
p
U
0
a
9j
(3)
where ,
p
is the particle density, and m
p
is the mass of
the particle, F
vis
is the viscous force given in Eq. (2),
and j is the surface viscosity. This dimensionless num-
ber increases in value as particle (granule) size
increases during granulation. Simple energy considera-
tions show that if the Stokes number dened above is
smaller than a critical value St
coal
- St
+
coal
, collisions
between particles are eective and coalescence occurs,
while if St
coal
> St
+
coal
, particles rebound because the
total incoming energy is larger than the one dissipated
during collision. One has to note that j in the above
equation can be taken to be the binder viscosity or an
equivalent viscosity of the granular surface.
An important observation is related to the charac-
teristic particle size, a, which is well dened for two
equal particles but, in a more general case, becomes
an equivalent size calculated from
2
a
=
1
a
1
1
a
2
(4)
where a
1
and a
2
are the sizes of the colliding particles
(granules). It is immediately apparent from this equa-
tion that during collisions between small and large par-
ticles when a
1
a
2
, the equivalent size is that of the
smaller particle, a a
2
, and so for this case the size of
the ne particle is the only relevant dimension to be used
in the calculation of the Stokes number. The implication
is that for small particles the dimensionless Stokes num-
ber is always small, always less then the critical value,
St
+
coal
, and therefore small particles will be preferentially
captured by larger ones if some binder is present on the
surface in a process called layered growth (Fig. 3c).
Trajectory calculations by Tardos and coworkers
(Ennis et al., 1991) yield analytical expressions for the
critical Stokes number, St
+
coal
, for simplied cases when
either the viscous force or the capillary force in Eq. (2)
is dominant. These values only apply for the case when
the colliding particles have an internal, solid core to
which a restitution coecient, e, can be assigned. In
the more general case, when there is no solid core or
when particles are deformable, critical Stokes numbers
cannot be calculated analytically and have to be esti-
mated from numerical integration or measured experi-
mentally. Further, in a real process, binder surface
tension and viscosity will both act together to dissipate
energy and ensure sticking and coalescence, but no sim-
ple analytical solution exists for this case. It was also
shown by Ennis and Tardos (Ennis et al., 1991) that
conditions based on viscous dissipation are more
restrictive then those based on capillary forces, and so
the discussion of critical sizes is limited, in this section,
to the former. In a subsequent section, computer simu-
lations are presented in which all eects, both capillary
and viscous, are taken into account to determine critical
values of the Stokes number.
5.2 Prediction of Critical Sizes
To use the above model for actual predictions, it is
necessary to assign values to the relative velocity, U
0
,
between moving particles. Prediction of U
0
at the pre-
sent level of knowledge is an extremely dicult task
since the powder ow eld in uid bed mixers is very
complicated and not amenable to simple solutions.
This is complicated by the presence of xed and mov-
ing walls such as, for example, draft tubes and rotating
propellers. A rough estimate of the relative velocity
between particles was given in Eq. (1). Using the
above velocities, one can calculate limiting (critical)
granule sizes at the point when the process crosses
over between granulation regimes associated with coa-
lescence where St
coal
= St
+
coal
:
a
coal
cr
=
9jSt
+
coal
8,
p
,
_ _
1,2
A
[,
]
1,2
(5)
Below the critical size, a
cr
, granule growth is rapid,
while above it it slows considerably; this behavior
occurs because the coalescence of larger particles
ceases at the critical point where St
coal
= St
+
coal
. Small
particles for which the critical Stokes number is still
small (since the characteristic dimension is that of the
small particle) will continue to stick to large particles
or granules in a layering type of mechanism and thus
lead to some slow growth. This slow growth process
will continue while nes are present and large
granules surfaces are covered by binder.
The proportionality condition in Eq. (5), the last
expression on the RHS, yields an approximate depen-
dence of critical size on the inverse of the square root
of the average shear rate and, since the actual value of
the critical Stokes number is not available for most
granulations, this is the most information that can be
extracted from the above model. To be able to use the
model as a predictive tool for detailed calculations, a
value of the critical Stokes number has to be estimated
from theoretical considerations (see the section on
computer simulations) or obtained by direct measure-
ments.
5.3 Wet Granule Deformation and Breakup
It would appear from the above analysis that once
granules reach the critical size characteristic of the
layering (coating) regime, any further increase in the
Stokes number, relative velocity, or shear rate will
maintain the size of the granules. This is certainly
true for the case in which binder drying and solidica-
tion accompanies growth, and granule strength
increases appropriately as large granules are formed.
For this to happen, however, one has to tailor both the
binder characteristics and the heating (or cooling) rate
in the agglomerator very carefully to achieve the
required properties in the optimal time frame.
We present, in this section, a simplied model that
accounts for the behavior of green, i.e., wet gran-
ules. It is assumed that these green agglomerates pos-
sess, upon formation, only the strength imparted to
them by the liquid bridges that assured coalescence
in the rst place. They have had, however, no oppor-
tunity to strengthen signicantly owing to either the
lack of time or to the fact that the binder did not
become more viscous. Such cases of granulation are
very common when, for example, oily binders are
used that do not evaporate easily or do not solidify,
or when the bridge is formed by a slow chemical reac-
tion between the liquid and the powder, i.e., hydration
of a powder. One is then left, within the shearing pow-
der mass, with deformable granules that can grow by
layering but can also deform and break.
To formalize mathematically the above concepts of
granule deformation and breakage, it is useful to dene
a new dimensionless Stokes number that relates initial
kinetic energy in the shearing mass to internal energy
resisting deformation, in the form:
St
def
=
Externally applied kinetic energy
Energy required for deformation
St
def
=
m
p
U
2
0
2V
p
t(,
)
(6)
The notations in this equation are similar to those in
Eq. (3). In addition, V
p
is the particle (granule)
volume, m
p
= V
p
,
p
is the granule mass, and t(,
) is
some characteristic strength of the granule. In the most
general case, this stress can be taken according to the
HerschelBulkley model
t(,
) = t
y
k,
n
(7)
where t
y
is the yield strength, k is an apparent viscos-
ity, and n is the ow index. A green unsolidied
granule is, from a rheological point of view, a complex
system that usually exhibits both a yield strength and
some non-Newtonian behavior as shown in Eq. (7).
Assuming that the granule is a very concentrated slurry
of the binder and the original particles, one can take,
as a rst approximation, the apparent viscosity being
negligible compared with the yield strength. So
t(,
) = t
y
. Under these conditions, and again taking
U
0
- a,
as before, one nds

St
def
=
,
p
a
2
,
2
2t
y
(8)
The Stokes number dened above increases with
increasing particle size, a, and reaches at some point
during granulation a critical value, St
+
def
, above which
granules start to deform and break. At the critical
point, when St
def
= St
+
def
, a limiting granule size is
dened as
a
def
cr
=
(2t
y
St
+
def
),,
p
_ _
1,2
,

B
0
,
(9)
This equation is similar in structure to Eq. (5) but
predicts an inverse linear dependence of the critical
size on the average shear rate.
5.4 Summary of Growth and Deformation Kinetics
The above considerations are summarized in Fig. 7,
where both the growth limit characterized by St
coal
=
St
+
coal
and the deformation (breakage) limit character-
ized by St
def
= St
+
def
are presented. The ratio of Stokes
numbers is presented on the abscissa, while the critical
sizes associated with equilibrium points are presented
on the ordinate. Both scenarios depicted in the gure
are possible: one in which the growth limit is below the
breakage limit and another in which the growth limit is
above the breakage limit. Arrows in the gure indicate
the direction of the drift in the process, while heavy
dots represent equilibrium points.
Take for example the situation in Fig. 7a: particles
will grow by nucleation and coalescence until they
reach the growth limit, a
coal
cr
. Since this limit is calcu-
lated strictly for equal size granules or particles,
unequal particles can still grow by layering to yield
granules somewhat larger than a
coal
cr
. Granules that
grow beyond a
def
cr
become unstable, deform, and even-
tually break. The process will stabilize at some point
between the two limiting sizes and will be characterized
by a size distribution that will shift slowly toward lar-
ger granules as long as binder and nes are present.
The situation depicted in Fig. 7b is somewhat sim-
pler in that there is only one stable point at the lower
critical size. Particles will grow to a size a
def
cr
and will
start to deform and break if they grow further.
Depending on the balance of growth and breakage
forces, this process may yield a narrow size distribution
around the critical value. Alternatively, the product
distribution may contain excess nes if coalescence
and breakage outweigh layering.
Since the critical sizes a
cr
depend on the average
shear rate in dierent ways [see Eqs. (5) and (9)], it is
possible to choose a shear rate that will bring the two
curves in Fig. 7 close to each other or even make them
overlap. This will yield a granulation with a very nar-
row size distribution, since small granules will grow
and large granules will break and thus in the end
yield the critical size. Such an overlap will also generate
a situation when St
+
def
= St
+
coal
; this is important since
under these conditions the actual value of the critical
Stokes number can be easily measured experimentally
(see the section on constant shear experiments).
To put the above considerations in perspective, the
simplied granulation model as developed above is
presented schematically in Fig. 8. The granulator is
assumed to impart a certain known, constant shear to
the granular media, as indicated by the arrows in the
gure. The granules are taken to be of the same size at
any given moment, while growth and breakup keep the
system in equilibrium as depicted in the gure. The
coalescence criterion developed above determines the
conditions under which two identical (smaller) gran-
ules in the schematic representation to the left coalesce
and grow to yield the larger size depicted on the right.
Furthermore, the deformationbreakup criterion
denes the conditions under which a larger granule
shown in the schematic at the right will deform and
break to yield the size on the left. This is quite an
unrealistic picture of the process, since in reality a
wide range of sizes result from both coalescence and
breakup. The simplied picture, however, yields some
basic understanding of the process, and two semiana-
lytical solutions that can be checked experimentally,
and therefore is quite powerful.
Further simplications, also shown in Fig. 8, are
related to the way formed granules are idealized: this
is depicted on the right lower part of the gure. Instead
of assuming granules to be made of individual smaller
particles held together by the binder, we assume instead
that granules are spherical and surrounded by liquid as
depicted. The premise here is that once granules are
Figure 7 Equilibrium Stokes numbers vs. critical granule
size. (A) Coalescence limit above deformation limit. (B)
Deformation limit above coalescence limit. Figure 8 Details of granule growth and breakage.
formed, particles are moved toward the center to form a
Acore, and binder is squeezed toward the surface to
insure further sticking. The simplication introduced
by this assumption can be relieved somewhat by taking,
instead of actual binder properties, the surface tension
and viscosity of a binderparticle surface layer.
6 GRANULE CONSOLIDATION
We use the above simplied model to gain some insight
into the process by which granules become more dense
after formation. This process called consolidation
can only take place when some bridges inside a formed
granule remain viscous and deformable or if the gran-
ule itself retains some of its plastic, malleable proper-
ties for some time. Under these conditions, external
compressive and shear forces will tend to push particles
together and hence reduce apparent density. It is clear
from the above that the granule has to be in a regime
somewhere between the growth limit and the deforma-
tion and breakup limit, as depicted in Fig. 7a.
We will now assume that the system subjected to
consolidation forces is the one shown in Figs. 4 and
6. The equation describing the motion of the two par-
ticles can be written as (Ennis et al., 1991)
U = U
0
1 St
1
ln
h
x
_ _ _ _
(10)
Where U is the instantaneous particle relative velocity,
St is the Stokes number (characteristic of either the
coalescence or the deformation limit), 2h is the initial
distance between particles, and 2x is the actual separa-
tion distance. Solving this equation between two suc-
cessive consolidation positions during which the
particles get closer by a distance x, one gets the
expression
c
=
x
2h
= 1 exp(St) [ ] (11)
where
c
is the degree of consolidation. Since the
Stokes number is inversely proportional to both binder
(granule surface) viscosity and/or the binder yield
strength [see Eqs. (3) and (6)], the above expression
immediately predicts that if either or both are very
large, no consolidation occurs since x
0. Ennis
(1990) extended the above correlation to a linear
chain of n identical particles that undergo i step-wise
consolidations and obtained
c
=
1
(n 1)
n1
i=1
1 exp
(n i)St
n
_ _ _ _
(12)
Both Eq. (11) and Eq. (12) reduce to the consolidation,
c
, being proportional to the Stokes number (
c
St)
for small values of this parameter (St _ 1). Assuming
that the consolidation of a granule takes place homo-
geneously in three spatial directions and that in each
direction the movement is described by,
c
St, we
nd a correlation between the bulk density of the gran-
ule before, ,
g
and after, ,
a
, consolidation as
,
a
,
g
= (1
c
)
3
- (1 3St) (13)
Here we assumed that both the Stokes number and the
degree of consolidation are small (St.
c
_ 1). The
above equation gives a quantitative relation between
the density change during consolidation and the shear
rate and other parameters of the system through the
Stokes number St.
7 EXPERIMENTAL VERIFICATION
This section contains several examples, one from the
authors own experiments, where critical values for the
Stokes number were measured experimentally, and
others, in which we use the above theoretical frame-
work to match experimental granulation data from the
literature.
7.1 Granulation in a Constant Shear Fluidized Bed
This experiment was devised to generate a set of con-
ditions under which the deformation (breakage) limit
overlaps the coalescence limit, St
+
coal
= St
+
def
, as shown
in Fig. 7. We use this condition to measure critical
values of the Stokes number. This is facilitated by
the fact that under these conditions formed granules
deform and break immediately after reaching equili-
brium; therefore this point can be easily detected
experimentally, as mentioned above.
To insure that the shear eld in the granulator is
constant and uniform, a uidized bed Couette device
was used in which a bed of particles at minimum ui-
dization conditions was sheared between two con-
centric rotating rough-walled cylinders (Fig. 9).
Fluidized air is supplied in the gap between the sta-
tionary outer and the rotating inner cylinder, thereby
creating a shear zone in which particles are suspended
in the upward moving gas. It was shown by both
numerical and analytical calculations (Campbell,
1990; Campbell and Brennen, 1985) that a granular
material moving in the rapid granular ow regime,
i.e., where inertial interactions between particles over-
whelm frictional eects, indeed exhibits a linear velo-
city prole, provided that conditions close to nonslip
can be generated at the walls. The shear rate for the
system can thus be calculated from
,
=
V
wall
z
(14)
where V
wall
is the velocity at the wall and z, is the gap
width.
To simplify the granule growth process in the pre-
sent experiment, agglomerates were generated by intro-
ducing a very viscous binder droplet into a uidized
bed of unsheared particles and by allowing enough
time for the granule to form by solids being imbedded
into the droplet (i.e., Fig. 3d). The initial granule size
could then be calculated from
D
g
= D
d
(c
0
)
1,3
(15)
where D
d
is the diameter of the binder droplet and c
0
,
is the void between the solids; it is assumed in the
above equation that the uid lls all voids. It was
found experimentally that this was indeed the case
and that due to surface tension, the resulting granule
was mostly spherical.
Experiments were performed with the uidized bed
Couette device described above (see also Khan and
Tardos, 1997) using glass particles of a mean size of
300 micrometers so that 80% of the particles lay
within the range of 350 to 250 jm. The density of par-
ticles was within the range of 2.42 to 2.5 g/cc; under
Geldarts classication scheme (Kunii and Levenspiel,
1991), these particles classify as class B. The minimum
uidization velocity of the particles was found ex-
perimentally to be about 8 cm/s. The volume fraction
or the porosity e of the bed at minimum uidization
conditions was taken as 0.4 (Kunii and Levenspiel,
1991).
Granules of the above powder using dierent visc-
osity Carbowax (PVP) solutions were formed, sheared
in the device, and subsequently were let to solidify in a
shear-free environment (uid bed at minimum uidiza-
tion conditions). Stokes numbers, St, were calculated
using Eq. (3) and taking U
0
- a,
. A granule deforma-
tion parameter dened as D = (L B),(L B), where
L and B are the major and minor axes of a deformed
agglomerate (D = 0 for a sphere), was also determined.
The correlation with the Stokes number is given in Fig.
10, where we also superimposed the picture of the
granule formed at the given conditions. It was found
that for dierent values of the Stokes number, but
otherwise identical conditions, formed granules
acquire dierent shapes. For low values of the Stokes
number (high binder viscosity), spherical agglomerates
are generated. Increasing the Stokes number (decreas-
ing the binder viscosity) yields regions in which only
slightly deformed granules are found and regions in
which the granules are totally deformed. A further
increase in the value of the Stokes number yields no
Figure 9 Schematic representation of the uidized bed
Couette device.
Figure 10 Experimental results. Elongation parameter (D)
vs. Stokes number for granules produced in a constant shear
Couette device.
agglomerates; binder-covered particles are dispersed
into the shearing powder. The important nding is
that beyond a critical value of the Stokes number of
about St
+
= 0.03, formed granules are so elongated
and eroded (see the gure) that their integrity is no
longer maintained. This is quite signicant, since one
would expect a value of the order of unity, O(1), for
this critical parameter, and therefore the experiment
shows that, at least as a rst approximation, the theo-
retical approach described above is realistic.
Figure 10a shows granules formed in the constant
shear granulator from a ne (average size 65 microns)
starch powder granulated with water. The pictures
were taken with a TV camera placed above the bed
and show the upper surface of the granulating powder
at six consecutive time steps as depicted in the gure.
The powder bed is relatively shallow in the direction
perpendicular to the picture (about 3 cm), while the
shear is applied in the plane of the picture. One can
clearly see signicant granule growth during water
(binder) addition in pictures AE; it is also apparent,
however, that those granules that form are almost at a
constant nal size. Subsequent granulation (see pic-
tures D and E) is overwhelmingly controlled by layer-
ing of small particles onto large granules. This process
continues during wet-massing (while no further binder
is added) in picture F until practically no nes are left.
This sequence shows quite convincingly that, in a gran-
ulator where the deformationbreakage limit is
reached, the process follows closely the steps predicted
by the theoretical model in Fig. 7. The coalescence
growth phase of granule formation mostly overlaps
wetting and spreading of the binder in the shearing
mass of powder, while growth by layering immediately
follows. The deformationbreakage regime clearly lim-
its growth to yield the nal, mostly uniform size dis-
tribution observed in Fig. 10a.
7.2 Experimental Data by Watano et al., 1995
Recently reported experiments by the above authors are
especially relevant to the present analysis of agglomer-
ate growth and consolidation. These authors published
a series of papers on granulation in a Roto-uidized bed
with a top-spray (i.e., a hybrid between devices shown in
Figs. 1c and 1d), where shear is induced into the bed by
both bubble motion (controlled by the uidization air
velocity) and the motion of the propeller (agitator),
which creates shear forces in a plane perpendicular to
the gas ow. They measured granule size, size distribu-
tion, and bulk density as a function of gas velocity and
rotational speed of the propeller (agitator) using three
scales of granulation equipment. In all cases, a pharma-
ceutical formulation of an initial size of 45 microns was
granulated using a polymeric binder. The following sec-
tions analyze the data in terms of binder dispersion,
growth, and consolidation.
7.2.1 Binder Dispersion
As discussed in the background on microscopic phe-
nomena, the prerequisite for growth by binder granu-
lation is the adequate presence of binder at contact
points. Therefore it is important to consider the initial
step of binder dispersion and its eect on growth.
Inadequate binder dispersion in the agglomeration
device can result in product heterogeneity. For exam-
ple, some particles with ample binder will grow large in
size while others remain unagglomerated owing to
local scarcity of binder. Binder delivery is measured
in terms of ux of binder per surface wetted area of
powder. The ux is dened by the binder ow rate
relative to the spray coverage area, and the degree of
powder uidization or mixedness that provides fresh
powder surface to the spray.
Watano et al. (1995) report three scales of agitated
uid beds that dier in binder spray area relative to
batch size (Table 1). The batch size was increased in
proportion to the mixer volume, maintaining a xed
powder height/diameter ratio for all scales. In addi-
tion, the area of spray coverage was kept in proportion
to the cross-sectional area of the vessel. However, the
ratio of the spray area to the product bulk volume
decreased with scale. This resulted in poorer binder
dispersion and broader granule size distribution at lar-
ger scales. In addition, the uidization air velocity was
also varied. At higher air velocity, the turnover of pow-
der in the spray zone increased, leading to improved
dispersion and narrower size distribution. The com-
bined eect of both variables is shown in Fig. 11
where the geometric standard deviation, (GSD) o
g
is
given as a function of relative spray area and gas velo-
city. As seen, the data is continuously arranged
throughout the eld of the gure: improved dispersion
of binder caused by increase in air ow velocity,
increase in spray surface area, or decrease in mass in
the mixer yields narrower product size distribution.
The data is also continuous in mixer scale as long as
the above conditions are equal.
7.2.2 Agglomerate Growth
To compare the results of Watano et al. to the current
analysis, we rewrite the result given in Eqs. (7) and (9)
but with the assumption that the yield strength, t
y
in
Eq. (7) is negligibly small and that the granules
internal strength is given by t(,
) = k,
n
.
Equation (9) can then be rewritten as
a
2
cr
,
2
= C,
n
(17)
where C is a constant coecient that incorporates all
parameters in Eq. (9) except the diameter and shear rate.
Taking the logarithm of the above expression yields
log(a
cr
) = 1
n
2
_ _
log(,
) log C = mlog(,
) c
(18)
Figure 10a Time sequence of powder granulation in the constant shear granulator.
This expression shows that on a loglog plot, the slope
of the curve of granule diameter versus shear rate gives
the shear index, n, of the binder through the coecient
m, while the intercept gives a measure of the critical
Stokes number through the value of c.
The experimental data shown in Fig. 12 include
median granule size as a function of varying tip
speed at three independent gas uidization velocities
(U
f
= 0.4, 0.6, 0.8 m/s) at three machine scales.
Equation (18) predicts reduced growth at high propel-
ler tip speed. This eect, however, is counteracted
somewhat by increased gas uidization velocity that
cushions particles from the impact of the propeller.
Data regression shows that the eect of air uidization
velocity (U
f
) relative to the propeller tip speed is
approximately to the 2 power. The full set of data
in Fig. 12 t the loglog trend predicted by
log(a
cr
) = mlog
U
i
U
2
f
_ _
c (19)
Here we assumed that the shear rate, ,
, is propor-
tional to the impeller tip speed, U
i
, and therefore the
quantity U
i
,U
2
f
is some modied shear rate. The aver-
age value of the slope for all granulators, large and
small, is approximately m = 0.25, which gives a rheo-
logical ow index of n = 1.5. Such values are common
for concentrated slurries of viscous liquids and small
solid particles and characterize a shear thickening
behavior. The most important conclusion from the
above considerations is that both the slope, m, and
the coecient, c, take approximately the same value
for all size granulators operated under dierent condi-
tions, especially at the high-shear range of the impeller
where the power-law term of the viscosity dominates.
This implies that the binder viscosity has the same role
in both small and large units and that the Stokes num-
ber, based on deformation, is the relevant dimension-
less parameter. It also implies that the Stokes number
may be the appropriate scaling factor when such
devices are scaled up.
7.2.3 Agglomerate Consolidation
Experimental support for the expression derived for
granule consolidation [Eq. (13)] is found again in the
work of Watano et al. (1995). Their data is shown in
Fig. 13 where the relative density ratio is given as a
function of the impeller tip speed (U
i
) for all three
Table 1 Scale-Up of an Agitated Fluid Bed Agglomerator
Model
+
: NQ-125 NQ-230 NQ-500
Vessel diameter (cm) 12.5 23 50
Cross-sectional area (cm
2
) 122.7 415.5 1963.5
Spray area (cm
2
) 38.5 132.7 594.0
Vessel volume ratio 1 6.23 64
Powder feed wt. (kg) 0.36 2.23 22.9
Spray area (cm
2
)/feed (kg) 106.9 59.5 25.9
Source: Watano et al., 1995.
Figure 11 Geometric standard deviation (GSD) of particle
size distribution (PSD) as a function of binder dispersion and
air ow velocity. (Watano et. al., (1995.)
Figure 12 Loglog plot of data by Watano et al. (1995);
median particle size (D
50
) versus impeller tip speed (U
1
)
and uidization velocity (U
f
). (D
0
is the D
50
intercept at
x = 1.)
granulators used at gas uidization velocities of 0.4,
0.6, and 0.8 m/s. Data regression shows that the eect
of air uidization velocity (U
f
) relative to the impeller
tip speed is approximately to the 0.5 power. The
relative density ratio, (,
a
,
g
),,
g
, was calculated
from the experimental data and by extrapolation for
the rst point in the gure, assuming that the two
densities are equal at zero granulator tip speed. As
seen, the bulk density increases with increasing tip
speed, i.e., increasing rotation rate or increasing
impeller diameter. To provide a comparison with pre-
dictions, we combine Eqs. (13), (6), and (7) with t
y
=
0 and n = 1.5 (as found from experiments, above) to
get
,
a
,
g
,
g
= K(,
)
1,2
= K
/
(U
i
)
1,2
(20)
with K being a parameter that incorporates all vari-
ables except the shear rate and where we again assume
the propeller tip speed to be proportional to the shear
rate.
Watano et al. (1995) found that uidization gas
velocity also has an inuence on consolidation, i.e.,
higher consolidation is obtained at lower gas velocities
as shown in the regression in Fig. 13, but this is not
captured directly in Eq. (20). A trend, however, can be
obtained by combining Eqs. (13), (2), and (4) to yield
,
a
,
g
,
g
= 48
,
p
U
B
a
2
D
B
o
2
j
= 34
,
p
a
2
g
1,2
D
1,2
B
o
2
j
(21)
where we used U
B
= 0.71 (g D
B
)
2
. It is likely that ui-
dization conditions impose a smaller bubble size, D
B
,
at lower gas velocities, and that would result in larger
consolidation; this would explain the result as found by
the above authors. Although it is not clear how the
dimensionless bubble spacing, o, changes under these
conditions, it appears that a combination of the above
two relationships would correctly reect the trend
obtained experimentally,
,
a
,
g
,
g
= K
//
U
i
U
f
_ _
2
(22)
The regression line in Fig. 13 represents the above
correlation with K
//
= 0.14. It is important to note
that correlations (22) and (19) were obtained for all
three scales of granulation but that growth (granule
size) and consolidation (densication) scale dierently
with both impeller tip speed and uidization velocity
posing dicult problems in scale-up.
7.3 Experimental Data by Dencs and Ormos (1993,
1994)
These authors, with a long history of eminent granula-
tion research starting as early as 1962, published
recently a set of papers reporting on granulations
aimed to recover solids from a concentrated aqueous
solution. They used a uidized bed in which shear
forces are controlled by bubble motion and by a rotat-
ing insert. The insert has two cylindrical rollers
mounted on the ends of a rotating arm that keeps
the rollers adjacent to the wall of the uidized bed,
as shown in Fig. 2 (a similar construction has the roll-
ers moving parallel to the distributor plate). The gap
between the rollers and the wall (distributor) can be
adjusted, and by controlling the speed of the rotating
arm, strong shear forces can be generated. The device
is called a uid bed grinder-granulator by the
authors to emphasize the grinding action of the rollers
combined with the granulation function of the uid
bed.
The unit can be run in a semicontinuous mode in
which the bed is started with seed particles from a
previous run and solution is introduced continuously
while solids are discharged through a tube close to the
distributor. The granulator is operated at some ele-
vated temperature above 100
C so that water from

the solution is evaporated and the residual solid mate-
rial is layered onto the existing solids in the bed. A
constant agglomerate diameter is obtained after some
time of operation, and from there on a steady state sets
Figure 13 Granule consolidation data by Watano et. al.
(1995); relative density increase [(,
a
,
g
),,
g
], versus impel-
ler tip speed (U
1
) and gas uidization velocity (U
f
).
in. The data reported in this section is this steady state
agglomerate size as a function of dierent operating
variables such as rotation rates of the arm, gap dis-
tance between the roller and the wall, gas uidization
velocity, and uidized bed diameter. Two materials
were used: a sucrose solution (700 g/liter) obtained
from a carrot processing unit and a zinc sulfate solu-
tion (500 g/liter). The data was taken from Dencs and
Ormos, (1993, 1994), and an additional large part
directly from the authors (Dencs, 1998).
To extract general conclusions from the data, we
employ here the same model prediction as used in
Eq. (19): log(a
cr
) = mlog(U
i
,U
2
f
) c, where U
i
is the
impeller tip speed and U
f
is the uidization velocity. In
this case however, shear is determined, in addition to
the impeller tip speed, U
i
, also by the gap distance, r,
between the roller and the wall. We rewrite the above
relationship to read
log(a
cr
) = mlog
U
i
rU
2
f
_ _
c (23)
where again m = (1 n,2), and c are material con-
stants that determine the non-Newtonian behavior of
the solidwater solution and the characteristic Stokes
number, respectively, and n is the binders ow index.
The experimental data is shown in Fig. 14 as the
average agglomerate diameter, d
agg
, as a function of
the modied shear rate ,
U
2
f
= U
i
,rU
2
f
, as given in
Eq. (23). An important conclusion from the gure is
that data t optimization gave the same dependence on
the variables as found for the data by Watano et al.
The tted curves to the data obtained for sucrose and
zinc sulfate solutions yield a coecient m = 0.183
and m = 0.15 (or values of the index n = 1.63 and
n = 1.79, respectively) and c = 0.37 and c = 0.35,
respectively. The values for the shear index, n, are
somewhat higher than those obtained during granula-
tion of a pharmaceutical powder by Watano et al. (see
Fig. 12, where n = 1.5). These values again point to a
ow index characteristic of a shear thickening solution
in agreement with the general nature of the material.
It is interesting to note that the data show a leveling
o as the shear rate is increased, suggesting that at
higher shear, characteristic of some additional para-
meters of the system, the inuence of shear on particle
size is reduced. This kind of behavior has been
observed in previous granulations by several authors
and is attributed here to wall slip. In other words,
depending on the stickiness (roughness) of the particle
(granule) surface, a shear rate is reached by either
Figure 14 Experimental data by Dencs et al. (1993) and Dencs (1998); agglomerate size versus modied shear rate and
uidization velocity. Data for sucrose and zinc sulfate solutions in small and large uid beds.
increasing the tip speed of the rotating arm or reducing
the gap distance, when particles start to slip signi-
cantly. From this point on, further increases in shear
rate are not transmitted to the powder, and the
agglomerate size is not signicantly inuenced by
further increases in shearing. Further considerations
of how the above experimental and theoretical ndings
are used to scale up industrial granulators can be
found in Mort and Tardos (1999).
8 COMPUTER SIMULATION OF GRANULE
GROWTH AND BREAKUP
We describe in this section several attempts to simulate
granulation in a shearing mass of powder in which a
sticky binder is present on a signicant fraction of
particles. We assume that the shear is constant, that
particles are of uniform size and shape (spherical), and
that the binder is present on the particle surface in a
constant, uniform layer as depicted in Fig. 8. It is
further assumed that particles, while spherical, can
only move in a two-dimensional space (2D simulation)
and that gravity is negligibly small. We follow the con-
cept of growth by coalescence if particles are small and
of deformation and breakup when granules become
excessively large, as predicted by the theoretical
model presented earlier.
The simulation is based on the discrete element
approach widely used in the literature to predict
ows of solid-particulate materials. This method in
turn is based on the analogy between the character of
granular ows with that of the ow of gas molecules
and is a generalization of the so called molecular
dynamics simulation or MDS, employed in physics.
Instead of point particles and force potentials, how-
ever, rapid granular ow simulations or RGFS use
actual expressions for the magnitude of forces to
describe interactions between nite size particles.
During this kind of simulation (RGFG), Newtons
equation (F
ext
= mdU,dt) is solved for pairwise inter-
actions between a large number of particles starting
from an initial condition, until steady state is achieved.
Several of these kinds of calculations have been per-
formed lately, a summary of some results being pre-
sented in Khan and Tardos (1997). Interactions
between particles in these works are restricted to fric-
tion and elastic and sometimes plastic deformations of
the surface. It is this restriction that limits the applic-
ability of these models to dry powder ows.
During a recent eort (Khan and Tardos, 1997), the
above RGFS were extended to include viscous interac-
tions between solid particles that also include the inu-
ence of the liquid surface tension; details of the
procedure are given in the above paper. The most
important addition to the original RGFS simulation
is the inclusion of viscous and capillary forces [as
given in Eq. (2)] into the force balance used to solve
Newtons law (F
ext
= mdU,dt). These forces are added
vectorially to supplement friction, elastic, and some-
times plastic forces used in previous simulation.
While the addition of viscous forces looks like a simple
extension of previous work, it is, in fact, a very serious
complication: this is due to the extensive (long) range
action of viscous forces and to the fact that they
become very large when particles come into close
proximity (and become unbounded when particles
touch). In addition, simple solutions only apply to
very slow movement (low Reynolds number, inertialess
ows) of the uid between the particles.
The simulation was successfully implemented for a
maximum of 40,000 particles, of which up to 30%
carried the sticky binder; a parallel computer system
at the University of Illinois at UrbanaChampaign and
a Sun computer cluster at CCNY were used. To limit
the complexity of the code and to keep computer run-
ning time to acceptable levels (1214 hours), it was
necessary to assume that particles covered with binder
(uid) at the onset of binder introduction were the only
ones carrying the binder for the duration of computa-
tions. In other words, sticky binder-covered particles
always remained sticky with a constant uid layer
on them, while dry (unsticky) particles remained
always dry, and no binder transfer between particles
was allowed during a particular run. Additional binder
could be introduced at set intervals but only to make
more (or less) particles sticky which then remained
binder covered for the duration of the run.
8.1 Simulation of Granule Growth by Coalescence
The simulation domain with 40,000 particles is pre-
sented in Fig. 15. The shear is introduced at a constant
rate through the upper and lower cell boundaries as
shown. Mirror images of the simulation domain ensure
that the inuence of solid boundaries is neglected (see
Khan and Tardos, 1997). The Stokes [St, dened in Eq.
(3)] and capillary (Ca, dened in Fig. 6) numbers are
the dimensionless parameters of the problem that have
to be perceived as dimensionless binder viscosity or
shear rate and uid surface tension, respectively.
Increasing binder viscosity and decreasing shear rate
result in lower Stokes numbers, while increasing the
liquid surface tension results in lower capillary num-
bers. The other important parameter of the system is
the ratio of sticky/dry particles in the domain (r
B
),
which is a measure of the total amount of binder pre-
sent. [It was found during many runs that the binder
layer thickness is a secondary parameter if the layer is
taken as less then about 1520% of the particle radius.
In this case, the only important consideration is
whether the binder is present or the particle is sticky,
while the further inuence of the layer thickness can be
neglected (for such shallow layers). In practice, creat-
ing layers that are more than about 510% of the par-
ticle radius is quite dicult, and therefore this is not a
serious limitation.]
The simulation domain on the left in Fig. 15 depicts
the moment in dimensionless time when the uid layer
is introduced onto the particles at t
+
= 0; binder-cov-
ered, sticky particles are shown in black, while dry
particles are depicted in gray. Introduction of binder is
achieved by randomly assigning stickiness to a percen-
tage of particles (20% or 8,000 in this case) in a
domain of dry (unsticky) particles that achieved steady
state after some previous running time. The picture at
the right in Fig. 15 shows the domain at steady state
after a dimensionless time of t
+
= 200. Lighter areas
are devoid of particles, while denser areas depict clus-
ters. The intense black in the picture shows agglomer-
ates, usually surrounded by clusters of dry powder. A
pattern recognition routine on a PC was used to gen-
erate granule size distributions from pictures similar to
the one shown in Fig. 15.
Granule size distributions for values of the Stokes
number, St = 1, and capillary number, Ca = 1, and for
a ratio of sticky/dry particles (r
B
) from 10% to 60%,
are shown in Fig. 16. The ratio of binder thickness, h,
to the particle diameter, d
p
, was taken as h,d
p
= 0.1.
Several other results of this kind, with a more detailed
description of the numerical scheme used during com-
putation, are given in Talu et al. (1999). Most of these
results were correlated as the average [Sauter mean,
d
(3.2)
] granule size (in multiples of initial particle dia-
meters) with the parameters of the system using a gen-
eral equation of the form
Figure 15 Shear ow of dry and binder covered particles;
the 2D simulation domain at dimensionless time t
+
= 0 and
at steady state, t
+
= 200. Total number of particles,
N = 40,000 of which 20% (8,000) are covered with binder.
Stokes number St = 0.1 and capillary number Ca = 1.0. Dry
particles are shown in gray while binder covered particles
(sticky) are depicted in black.
Figure 16 Simulated granulation with increasing amounts of binder. Cumulative particle size distribution vs. relative agglom-
erate diameter. Stokes number St = 0.1 and capillary number Ca = 1.0.
d
agg
d
p
=
St
+
St
_ _
o
(24)
where the exponent was found to be o = 0.4. Values
for the critical Stokes number, St
+
, were extracted
from the computer simulation results. These contain
all other parameters of the problem that are not
directly incorporated into the Stokes number. They
are given for the case of growth by coalescence by
St
+
=
44r
1.65
B
Ca
0.32
_ _
2.5
for St - 1 (25a)
St
+
=
92r
1.65
B
h
d
p
_ _
1,4
Ca
0.32
_
_
_
_
_
2.5
for St > 1 (25b)
It can be easily seen from the above results and from
the exponents of dierent dimensionless quantities in
Eqs. (25), that the inuence on granule size of binder
surface tension (through the Ca number) and binder
viscosity (through the St number) are approximately of
the same order, while the inuence of the amount of
binder present (through the binder-covered-to-dry par-
ticle ratio, r
B
) is overwhelming. These conclusions are
amply supported by industrial practice where it is well
known that only after the amount of binder in the
granulator is xed does the inuence of other para-
meters, such as binder properties, begin to have an
inuence on granule size.
8.2 Simulation of Granule Deformation and Breakup
A second set of simulations was performed (see also
Talu et al., 1999) with all binder covered, sticky
particles contained in one initial spherical granule as
shown in Fig. 17. This was done to simulate the defor-
mation and eventual breakup of a large agglomerate
within the shearing mass of dry powder. Figure 17
depicts the simulation domain at steady state after
the initially round (spherical) granule was kept in the
shear eld for a long time (dimensionless time
t
+
= 100). As seen, at a Stokes number of St = 0.01
(and Ca = 10) the agglomerate does not deform and
remains spherical throughout. Increasing the value of
this parameter to St = 0.02 results in a deformed
agglomerate that, however, does not break apart. At
larger Stokes numbers (lower binder viscosity or higher
shear), the agglomerate breaks into two (St = 0.03)
and into three parts (St = 0.04). The critical value of
the Stokes number, when the agglomerate can still be
considered intact, is somewhere between St = 0.01 and
0.02.
Critical values of the Stokes number were calcu-
lated for several dierent initial granule sizes and dif-
ferent values of the capillary number (Ca), and these
are shown in Fig. 18. In this gure, the relative
agglomerate size d
agg
,d
p
is given versus the Stokes
number for three values of the capillary number Ca =
10 (small capillary eects), 1, and 0.5 (large capillary
eects). The experimental value for the critical Stokes
number, as obtained from the constant shear Couette
device (see Fig. 10), is also shown; the t is quite
remarkable even though the value of the capillary
number in the experiment is not known (it is assumed
to be large, Ca = 10 or larger, since the binder is very
viscous).
The lines in Fig. 18 were tted by functions of the
general form given in Eq. (24) with an exponent, a = 1.
Critical Stokes numbers were extracted that, for gran-
ule breakup, have the form
St
+
=
2.25
Ca
for Ca - 10 (26)
Figure 17 The inuence of the Stokes number on the nal
shape of a granule surrounded by dry powder. Binder cov-
ered agglomerates in a shearing medium at steady state,
t
+
= 100. Total number of particles in the domain,
N = 10,000; capillary number Ca = 10.0.
This last equation yields [with Eq. (24) and a = 1] the
largest possible agglomerate that will survive in the
granulator under given conditions.
CONCLUSIONS
A comprehensive model of granulation was presented,
one that takes an agglomeration process from binder
selection and testing to granule growth, deformation,
and consolidation. An instrument and procedures were
described to measure the critical amount of binder that
ensures that coalescence and particle growth will
occur. It was also shown how the pendular bridge
apparatus (PBA), developed earlier by the present
authors, can be used to test the bridge-forming char-
acteristics of the binder while at the same time deter-
mine binder spreading and penetration rates and the
critical time of binder strengthening. These operations
can be performed before any granulation runs are car-
ried out, and binder characteristics can be modied to
achieve proper wetting, bridge formation characteris-
tics, and binder strengthening (viscosity increase).
A theoretical model of particle/granule growth
based on simple principles of energy dissipation in
the liquid (binder) bridge between two colliding parti-
cles was developed. When the initial energy of the par-
ticles is smaller than the dissipated energy, particle
coalesce and growth will take place. This concept, ori-
ginally formalized in Ennis et al. (1991) is extended in
the present paper to totally deformable particles where
the binder viscosity is replaced by an equivalent visc-
osity of the malleable surface. Furthermore, the con-
cept of energy dissipation is also applied to determine a
criterion of green particle deformation and breakup
and to granule consolidation. A set of dimensionless
numbers that characterize granule growth and the
deformation process are dened. It is shown that
these numbers (Stokes numbers) delimit dierent gran-
ulation regimes, and critical values are found that can
be used to calculate the critical granule sizes that are
characteristic at the transition from one regime to the
other. Several experimental results from the literature
are used to illustrate the usefulness of the theoretical
approach. A set of very careful experiments performed
by the present authors in a specially constructed gran-
ulator with constant shear are used to measure critical
Stokes numbers. Employing these values, it is possible
to predict the outcome of a granulation process pro-
vided that the shear in the device is essentially constant
and that binder bridge formation is not superimposed
on granule growth.
Finally, it was shown that rapid granular ow simu-
lations (RGFS) into which the eect of binder viscosity
and surface tension are introduced, readily predict the
outcome of granulation. These simulations support the
simplied theoretical models of granule growth by coa-
lescence and granule deformation and breakup and
Figure 18 Values of the critical Stokes number, St
+
, during granule deformation and break-up. Dimensionless agglomerate size
vs. Stokes number for values of the capillary number of Ca = 10, 1.0 and 0.5. Experimental result reproduced from Fig. 10.
give more realistic values for the critical parameters
that dene dierent regimes of granulation.
Equations were obtained for granule sizes generated
in a granulation process as a function of the dimen-
sionless parameters of the problem such as the capil-
lary number, Ca, the Stokes Number, St, and the
amount of binder present in the system as dened by
the fraction of binder-covered particles in the domain.
Throughout this work, it has been shown that
micromechanical analysis is a powerful tool in predict-
ing fundamental transformations in the granulation
process, including binder dispersion, growth, and
deformation. Micromechanical analysis shows the
interdependence between key process parameters,
such as shear rate, and material properties, such as
binder viscosity and agglomerate yield stress. Scale-
up of agglomeration in industrial practice is greatly
aided by an understanding of the key process transfor-
mations and their controlling factors on the microscale.
ACKNOWLEDGMENTS
The authors would like to thank Dr. Bela Dencs from
the Research Institute for Chemical Engineering at the
Pannon University of Agricultural Sciences in
Veszprem, Hungary, for the information in Fig. 2
and the data in Fig. 14 and for fruitful discussions
on the topic of granulation. The help of Dr. I. M.
Khan and Dr. I. Talu from CCNY in preparing the
simulation data, and for performing some experiments,
is also greatly appreciated
NOTATIONS
a. a
1
. a
2
= particle, granule radius
B = width of deformed granule
Ca = Capillary number (Fig. 6)
A. B
0
. C. K = constants
C
0
= constant in Eq. (2)
C = log C = constant in Eq. (18)
d
(3.2)
= Sauter mean diameter
d
agg
= agglomerate diameter
d
p
= particle diameter
D
50
. d
50
= granule median diameter at
50% of size distribution
D
0
= D
50
[at x = 1] = reference granule diameter (in
Fig. 12)
D
g
= granule diameter
D
d
= binder droplet diameter
D = (L B),(L B) = deformation parameter
D
B
= bubble diameter in uidized
bed
e = solid core restitution coecient
F = force exerted by liquid bridge
g = acceleration due to gravity
h = binder layer thickness
k = apparent viscosity
L = length of deformed granule
m
p
= V
p
,
p
= mass of particle
m = (1 n,2) = constant in Eq. (18)
n = ow index [Eq. (7)]
N = number of particles in
simulation domain
r = coordinate, gap width in
experimental uid bed
r
B
= ratio of binder covered to dry
particles in the simulation
domain
St = Stokes number (dimensionless)
t = time
t
+
= dimensionless time
U = particle instantaneous velocity
U
0
= relative particle velocity
U
B
= bubble velocity in uidized bed
U
f
= gas uidization velocity
U
i
= impeller tip speed
V
p
= volume of particle
V
wall
= granular velocity at the wall of
the Couette device
x = distance between colliding
particles
z = coordinate
Greek Letters
o = exponent in Eq. (24)
j = binder or granule surface
viscosity
, = binder surface tension
,8 = shear rate
= lling angle
c
0
= void volume in granular bed
,
a
= apparent density of granule
,
g
= granule density before
consolidation
,
p
= particle (solid) density
o = bubble spacing in uidized bed
t = shear stress
t
y
= granule yield strength
z = gap width in the Couette
device
c
= degree of consolidation
o
g
= geometric standard deviation
(GSD)
Subscripts and Superscripts
coal = coalescence
def = deformation
cr = critical
+ = dimensionless, limiting value
vis = due to viscous eects
cap = due to capillary eects
wall = at the wall
ext = exterior
REFERENCES
Adams MJ, Briscoe BJ, Kamjab M. The deformation and
ow of highly concentrated dispersions. Adv in Colloid
and Interface Sci 44:143, 1993.
Campbell CS. Rapid granular ows. Ann Rev Fluid Mech
22:57, 1990.
Campbell CS, Brennen, CE. Computer simulation of granu-
lar shear ows. J Fluid Mech 151:167, 1985.
Dencs B, Ormos Z. A particle size control in a uidized bed
spray-dryer and granulator during the recovery of solids
from liquids. Hungarian J Ind Chem Veszprem
21:225231, 1993.
Dencs B, Ormos Z. A Orlo-gorgos uidizacios-porlasztos es
granulalo berendezes fojadekok szarazanyag-tartalmanak
kinyeresere. J Hungarian Chem 7, 1994 (in Hungarian).
Dencs B. Personal communication, April 1998.
Ennis BJ, Tardos GI, Pfeer R. A micro-level-based classi-
cation of granulation regimes and mechanisms.
Proceedings of the Second World Congress on Powder
Technology, Kyoto, Japan, 1990, p. 409.
Ennis BJ. On the mechanisms of granulation. Ph.D. disserta-
tion, City University of New York, 1990.
Ennis BJ, Li J, Tardos GI, Pfeer R. The inuence of visc-
osity on the strength of an axially strained pendular liquid
bridge. Chem Eng Sci 45:30713087, 1990.
Ennis BJ, Tardos GI, Pfeer R. A micro-level-based charac-
terization of granulation phenomena. Powder Technol
65:257272, 1991.
Hancock BC, York P, Rowe RC. An assessment of sub-
stratebinder interactions in model wet masses. 1: Mixer
torque rheometry. Int J Pharm 102:167176, 1994.
Khan MI, Tardos GI. Stability of wet agglomerates in gran-
ular shear ows. J Fluid Mech 347:347368, 1997.
Kristensen HG, Holm P, Schaefer T. Mechanical properties
of moist agglomerates in relation to granulation mechan-
isms. Part I: Deformability of moist densied agglomer-
ates. Powder Technol 44:227237, 1985.
Kristensen HG, Holm P, Schaefer T. Mechanical properties
of moist agglomerates in relation to granulation mechan-
isms. Part II: Eect of particle size distribution. Powder
Technol. 44:239247, 1985.
Kristensen HG, Schaefer T. A review on pharmaceutical wet
granulation. Drug Develop Ind Pharm. 13(4,5):803, 1987.
Kristensen HG. Agglomeration of powders. Acta Pharm Soc
25:187204, 1988.
Landin M, York P, Cli MJ, Row RC, Wigmore AJ, Scale-
up of a pharmaceutical granulation in a xed bowl mixer
granulator. Int J Pharm 133:127131, 1996a.
Landin M, York P, Cli MJ, Row RC, Wigmore AJ. The
eect of batch size on scale-up of a pharmaceutical gran-
ulation in a xed bowl mixergranulator. Int J Pharm
134:243246, 1996b.
Mazzone DN, Tardos GI, Pfeer R. The behavior of liquid
bridges between relatively moving particles. Powder
Technol 51:71, 1987.
Mort PR, Tardos GI. Scale-up of agglomeration processes
using transformations. Kona 17, 1999.
Nienow AW. Fluidized bed granulation. In: Stanley-Wood
NG, ed. Enlargement and Composition of Particulate
Solids. London: Butterworths, 1983.
Nienow AW, Rowe PM. Particle growth and coating in gas-
uidized beds. In: Davidson JF, Clift R, Harrison D, eds.
Fluidization. 2d ed. London: Academic Press, 1985.
Nienow AW. Fluidized bed granulation and coating:
application to materials, agriculture, and biotechnology.
Proceedings of the First International Particle Technology
Forum, Denver, Co., August 1719, 1994.
Ormos ZD. Granulation and coating. In: Chulia D, Deleuil
M, Pourcelot Y, eds. Powder Technology and
Pharmaceutical Processes, 1994, pp. 359376.
Rowe RC, Parker MD. Mixer torque rheometry: an update.
Pharm Technol, March, 1994, pp 6982.
Schaefer T. Equipment for wet granulation. Acta Pharm Soc
25:205228, 1988.
Schaefer T, Holm P, Kristensen HG. Wet granulation in a
laboratory scale high shear mixer. Pharm Ind 52(9):1147,
1990.
Sommer K. Size enlargement. In: Ullmans Encyclopedia
of Industrial Chemistry. 5th ed. Vol. B2. VCH Verlag,
1988.
Talu I, Tardos GI, Khan MI. Use of rapid granular ow
simulations to predict granulation of ne powders.
Powder Technol, in press, 1999.
Tardos GI, Mazzone D, Pfeer R. Destabilization of ui-
dized beds due to agglomeration, Part I and Part II.
Can J Chem Eng 63:377383, 384, 1985.
Tardos GI, Garcia J, Ahart R, Ratuiste F. How and why
some ow aids promote bulk powder ow. AIChE
Symposium Series. Weimer, AW, ed. 89(296), 1993.
Tardos GI, Gupta R. Forces generated in solidifying liquid
bridges between two small particles. Powder Technol
87:175180, 1996.
Tardos GI, Khan MI, Mort PR. Critical parameters and
limiting conditions in binder granulation of ne powders.
Watano S, Sato Y, Miyanami K, Murakami T. Scale-up of
agitation uidized bed granulation. Chem Pharm Bull
43(7): Part I: 1212, Part II: 1217, Part III: 1224, Part
IV: 1227, 1995.
18
Applications for Fluidized Bed Drying
Arun S. Mujumdar
National University of Singapore, Singapore
Sakamon Devahastin
King Mongkuts University of Technology Thonburi, Bangkok, Thailand
1 INTRODUCTION
Fluidized bed dryers (FBDs) have found widespread
applications for the drying of particulate or granular
solids in the chemical, food, ceramic, pharmaceutical,
agriculture, polymer, and waste management indus-
tries. More recently, they have also found special appli-
cations in the drying of slurries. Suspensions,
solutions, dilute pastes, or sludges are atomized into
a uidized bed of inert particles and the dry powder is
separated from the exhaust gases.
For drying of powders in the 50 to 2000 jm range,
uidized beds compete successfully with other more
traditional dryer types, e.g., rotary, tunnel, conveyor,
continuous tray. Among the advantages of the ui-
dized bed dryers one may cite
High drying rates due to excellent gasparticle con-
tact leading to high heat and mass transfer rates
Smaller ow area
Higher thermal eciency, especially if part of the
thermal energy for drying is supplied by internal
heat exchangers
Lower capital and maintenance costs, compared to
rotary dryers
Ease of control
However, they also suer from some limitations such
as
High power consumption due to the need to sus-
pend the entire bed in gas phase leading to high
pressure drop
Increased gas handling requirement due to extensive
recirculation of exhaust gas for high thermal e-
ciency operation, especially when drying pro-
ducts with extensive internal moisture
High potential of attrition; in some cases granula-
tion or agglomeration
Low exibility and potential of deuidization if the
feed is too wet
Generally not recommended when organic solvents
need to be removed during drying.
Table 1 compares the key characteristics of the ui-
dized bed dryer with those of some of the competing
dryers. Note that the ash dryer is applicable only for
the removal of surface moisture from smaller size par-
ticles, while the uidized bed is recommended for
particles requiring longer drying times, e.g., 1060 min-
utes compared to only 1030 seconds for ash drying.
Some of the characteristics listed in Table 1 are subject
to modication by changes in the design of the hard-
ware as well as the operating conditions. It is impor-
tant to point out that for removing organic solvents or
handling toxic or ammable solids, the conventional
hot air uidized bed dryer is not a good choice, since
there is danger of re or explosion if ammability lim-
its are exceeded. Also, owing to the inherent need to
handle large volumes of air due to the need to achieve
uidization, complete collection of the toxic material is
not economically feasible.
2 CLASSIFICATION AND SELECTION
CRITERIA
The simplest and most common uidized bed dryer is a
circular cross section vessel in which the particulate
material is dried convectively either in the batch
mode (for smaller quantities or if the feed is produced
in batches) or continuously. Rectangular and other
noncircular cross section vessels may be employed if
needed, however. For most operations, uidized beds
operate at slight negative pressure to avoid leakage of
hot and humid gas into the ambient; it is generally not
cost-eective to fabricate uidized bed dryers as pres-
sure vessels or totally leak-proof.
Figure 1 provides a coarse classication scheme that
covers most (but not all) uidized bed dryers available
commercially. Some special variants of the conven-
tional uidized bed may be needed for special applica-
tions. It should be noted at the outset that the well-
mixed dryer type (Fig. 2) has the advantage of being
able to handle higher moisture content feed which may
not be uidizable under normal conditions. As the wet
feed enters a well-mixed uidized bed of a lower aver-
age moisture content, the danger of deuidization or
formation of large chunks (or agglomerates) is
Table 1 Comparison of Fluidized Bed Dryer with Competing Dryers
Criterion Rotary Flash Conveyor Fluid bed
Particle size Large range Fine particles 500 jm10 mm 1002000 jm
Particle size distribution Flexible Limited size distribution Flexible Limited size distribution
Drying time Up to 60 min 1030 s Up to 120 min Up to 60 min
Floor area Large Large length Large Small
Turndown ratio Large Small Small Small
Attrition High High Low High
Power consumption High Low Low Medium
Maintenance High Medium Medium Medium
Energy eciency Medium Medium High High
Ease of control Low Medium High High
Capacity High Medium Medium Medium
Figure 1 A coarse classication scheme for uidized bed dryer.
reduced. On the negative side, a well-mixed continuous
dryer will necessarily lead to a distribution in the
moisture content of individual particles; equilibration
of moisture during storage yields a product of uniform
moisture content. When processing wet feeds that do
not uidize readily, one may use mechanical assists
(e.g., agitation or vibration of the bed) to separate
the particles or employ solids backmixing. In the latter
case, the wet feed is blended in appropriate proportion
with dried product so that the mixed feed is uidizable.
This adds to both the capital and operating costs of the
drying operation, however.
Further, most uidized bed dryers are nearly adia-
batic contactors, e.g., they are well insulated, and the
heat and mass transfer between the drying particles
and the hot gas is by convection only. This implies
that the product surface temperature (when surface
moisture is being removed) will attain the wet-bulb
temperature corresponding to the inlet gas temperature
and humidity, while the state of the exit gas follows the
adiabatic saturation line. When the internal moisture is
being removed, the product temperature is above its
wet-bulb temperature, and in the limit it attains the
uidizing gas temperature at long drying times.
To improve the thermal eciency of the dryer by
reducing the gas consumption, it is often a good idea
to supply a part of the heat for drying by immersing
heat exchanger panels or tubes in the bed. Thus lower
gas temperatures (and ow rates) can be deployed with-
out extending the drying time excessively, since a part of
the heat (3060 percent) is supplied indirectly by con-
duction. This is especially important for heat-sensitive
materials, which limit the uidizing gas temperature to
low values. For certain polymer pellet applications, ui-
dized bed dryers with immersed heat exchange panels
(nonadiabatic) are the only cost-eective means for dry-
ing. It should be noted, however, that heat transfer
considerations make this option most attractive only
for ne particles of the order of 100 jm in size.
Table 2 summarizes the diverse variants of the ui-
dized bed technique; not all the types noted are
commercially available, however.
Figure 3 shows the basic steps involved in the
selection of a dryer. It is beyond the scope of this pre-
sentation to discuss the selection of dryers from a very
large pool of possible dryers. The interested reader is
asked to refer to Mujumdar and Menon (1995) and
Baker (1997) for a more general discussion. Note
that while over four hundred dryer types have been
reported in the literature, over one hundred are sold
commercially. Careful selection of the dryer type is
crucial to the success of the nal design and optimiza-
tion of the drying system.
For any dryer, the steps shown in Fig. 3 apply. Unless
reliable data exist for the same product it is strongly
recommended that laboratory scale tests be carried
out to verify that the material can be processed in a
uidized bed. The conditions for uidization (e.g., mini-
mum uidization velocity) may be estimated using pub-
lished correlations, but serious errors may accrue due to
the surface wetness of the particulate solid, which may
generally behave entirely dierently from a surface-dry
particle. Even trace amounts of surface wetness may
render the material unuidizable without mechanical
energy input. In the case of a uidized bed dryer with
immersed heat exchangers, it is obvious that the mate-
rial must not stick to the heat exchange surfaces.
While the convective heat and mass transfer rates in
wet particle uidization may be estimated reasonably
well with dry particle correlations available readily in
the literature, particle-to-immersed-surface heat trans-
fer rates are very sensitive to the presence of surface
moisture. Evaporation of the surface moisture may
lead to a severalfold increase in the heat transfer rate
until the particle is surface-dry, e.g., it reaches its cri-
tical moisture content or attains its equilibrium moist-
ure content at the surface.
It is suggested that the laboratory scale tests be car-
ried out in a chamber of at least 100 mm in diameter so
that the wall eects are not severe. When possible, pilot
scale tests are also recommended owing to the highly
nonlinear nature of the hydrodynamics. Especially
noteworthy is that the product quality (e.g., physical
size, size distributions, structure, texture, color) may
vary with the scale of operation depending on the pro-
Figure 2 Well-mixed continuous uidized bed dryer.
duct. Attrition and agglomeration rates may dier as
well. When the quality of the product is of paramount
importance, it is essential to carry out pilot scale tests.
Table 3 summarizes the main criteria that must be
specied in the design of a uidized bed dryer. Detailed
design or analysis procedures for the diverse uidized
bed dryer types are beyond the scope of this presenta-
tion.
The nal selection of the uidized bed drying system
includes the selection of the feed system(including back-
mixer, if needed) as well as gas cleaning (using cyclones,
bag lters, etc.) and partial exhaust gas recycle for
improved thermal eciency. A typical ow diagram of
the whole uidized bed drying system is shown schema-
tically in Fig. 4. A number of computer codes have been
developed for the design of a uidized bed dryer and its
essential ancillaries, e.g., cyclones, fans. An example of
such a code is one that is developed by Jumah and
Mujumdar (1993). This code also permits estimation
of the energy consumption as well as the capital and
operating costs of uidized bed dryers. More sophisti-
cated codes for design and analysis of uidized bed
dryers are available from commercial sources.
Table 2 Classication of Fluidized Bed Dryers
Criterion Type of dryer (application)
Operating pressure Low-pressure (e.g., for heat-sensitive products)
Near atmospheric (most common)
High-pressure (5 bars, steam dryers)
Particulate ow regime Well-mixed
Plug ow
Hybrid (well-mixed followed by plug ow)
Processing mode Batch
Continuous
Fluidizing gas ow Continuous
Pulsed
Fluidizing gas temperature Constant
Time-dependent
Heat supply Convection or convection/conduction
Continuous/pulsed
Fluidization action By gas ow (pneumatic) only
Downward set ow (jet-zone)
With mechanical assist, e.g., vibration or agitation for sticky
or polydispersed solids
Fluidized material Particulate solid
Paste/slurry sprayed onto a bed of inert particles
Slurry sprayed onto absorbent particles (e.g., silica gel, biomass, etc.)
Fluidizing medium Air/ue gases/direct combustion products
Superheated steam (or vapor)
Number of stages Single
Multiple
Figure 3 Basic steps in dryer selection.
Note that when removing an organic solvent it is
normally necessary to use an inert gas (e.g., nitrogen);
the solvent is recovered by condensation and the inert
gas reheated indirectly and returned. A small bleed
is necessary to avoid the buildup of impurities in the
circulating gas stream. When drying solids that can
undergo combustion, care must be taken to avoid con-
ditions susceptible to re or explosion. Suitably
designed rupture disks should be installed and the oxy-
gen level in the drying system (including cyclones and
baghouses) must be controlled to be outside the explo-
sion limits. All internal metal parts must be grounded
to avoid the potential of buildup of electrostatic
charges by the ne dry dust.
Control of uidized bed dryers is discussed by
Jumah et al. (1995). More recently, Liptak (1998) has
discussed advanced control strategies for continuous
uidized bed dryers. Attempts have also been reported
in the literature on the use of fuzzy logic and articial
neural nets to control uidized bed drying and granu-
lation.
Finally, some remarks on capital costs of uidized
bed dryers are appropriate. Typically, the equipment
cost rises as the 0.60 to 0.70 power of the capacity.
The cost of the ancillary equipment, e.g., feeders,
blowers, gas cleaning equipment, can exceed the
cost of the dryer itself several times, depending on
specic applications. It is therefore important to
cost the entire drying system rather than just the
dryer.
3 BASICS OF DRYING KINETICS
Figure 5 represents a typical textbook drying rate
curve obtained from a batch of wet particles uidized
with hot air of xed temperature, humidity, and ow
rate. After an initial transient period (generally a
heatup period) if the particles are very wet, i.e., cov-
ered with a liquid lm, the drying rate depends entirely
on the external heat and mass transfer rates, e.g., a rise
in air temperature or air ow rate or a reduction in air
Table 3 Design Criteria for a Fluidized Bed Dryer
Criterion/parameter Main consideration
Feed moisture content, feed rate,
product moisture content
Design specications
Bed area Calculate in design; given in analysis problem
Maximum air temperature Depends on heat sensitivity of product
Maximum product temperature must be specied
Outlet gas temperature and
humidity
Exit gas in equilibrium with bed
Equilibrium moisture content of product must be known
Maximum carryover Related to attrition
Quality of product aected
Figure 4 A schematic ow diagram of the whole uidized bed drying system.
humidity will lead to an increase in the drying rate.
This rate continues unchanged as long as the liquid
lm covers particle surfaces fully, since the drying con-
ditions are held constant. Thus the so-called constant
rate drying period exists as long as the drying condi-
tions remain unchanged. Once the particle surfaces
become partially dried, the rate of drying is reduced
since in the area not covered with liquid lm the rate of
drying will depend also on the rate at which the inter-
nal liquid arrives at the surface prior to vaporization.
The moisture content at which this drop in drying rate
rst occurs is called the critical moisture content.
Unfortunately, this critical value is not a material
property in that it is also a function of the drying
conditions for a given material. It must be determined
experimentally. Once the surface lm disappears, the
surface attains its equilibrium moisture content corre-
sponding to the temperature and humidity conditions
it is exposed to. [See Mujumdar and Menon (1995) for
details of the denitions and terminology in psychro-
metry and drying.] Approximate critical moisture con-
tent values for some selected materials are given in
Table 4.
For a purely convective adiabatic dryer, neglecting
sensible heat eects, the surface temperature attained
by the drying particles in the constant rate period is the
so-called wet-bulb temperature corresponding to the
gas conditions they are exposed to. Below the critical
moisture content the particle temperature will rise
above the wet-bulb temperature. The gas temperature
follows the adiabatic saturation line (in both constant
and falling rate periods).
It is useful to summarize the key results of numer-
ous experimental studies on uidized bed drying
kinetics. The following generalizations apply mainly
to the conventional hot gas uidized adiabatic dryer.
3.1 Effect of Bed Height
For materials with high mobility of internal moisture
(e.g., iron ore, ion exchange resins, silica gel), most
drying takes place close to the distributor plate.
Increasing bed height beyond a particular value has
Figure 5 Typical textbook batch drying rate curve under constant drying conditions.
Table 4 Approximate Critical Moisture Contents for
Various Materials
Material
Critical moisture content
(kg water/kg dry solid)
Salt crystals, rock salt,
sand, wool
0.050.10
Brick clay, kaolin, crushed
sand
0.100.20
Pigments, paper, soil, worsted
wool fabric
0.200.40
Several foods, copper
carbonate, sludges
0.400.80
Chrome leather, vegetables,
fruits, gelatin, gels
> 0.80
no eect on drying rate for the bed. For materials with
main resistance to drying within the material, e.g.,
grains, increase of bed height will decrease the drying
rate.
3.2 Effect of Particle Size
The time required to remove a given amount of moist-
ure increases as the square of the particle diameter for
a Geldarts type B particle, all other conditions being
equal. For type A particles the eect is much smaller.
Geldarts classication of powders is discussed in
Chapter 3, Bubbling Fluidized Beds.
3.3 Effect of Gas Velocity
In removing surface moisture, gas velocity has a domi-
nant eect. For particles with high internal resistance
to moisture transfer, gas velocity has marginal to no
eect.
3.4 Effect of Bed Temperature
The eect is complex and depends on the relative sig-
nicance of external and internal resistances to moist-
ure transfer. Higher external heat uxes can lead to
increased bed temperature, which in turn leads to
higher moisture diusivities and hence drying rates
(see Reay and Baker, 1985).
4 FEATURES OF SPECIFIC TYPES OF
FLUIDIZED BED DRYERS
Here we summarize the key features of the more com-
monly used uidized bed dryers. Some of the more
innovative and specialized uidized bed dryer designs
are discussed in the nal part of this chapter.
4.1 Batch Fluidized Bed Dryers
Batch uidized bed dryers are used for low throughput
(normally - 50 kg h
1
and good for - 1000 kg h
1
),
multiproduct applications. Drying air is heated directly
or indirectly usually to a xed temperature. The drying
gas ow rate is also usually xed. However, it is pos-
sible to start drying at a higher inlet gas temperature
(and ow rate) and lower it once the product moisture
content falls below the critical value. Batch uidized
bed dryers are available commercially that can adjust
the drying gas conditions automatically to maintain
the bed temperature constant throughout the drying
process. Mechanical agitators or vibration may be
needed if the material is dicult to uidize.
4.2 Well-Mixed, Continuous Fluidized Bed Dryers
(WMFBDs)
In this type of dryer (Fig. 2), the bed temperature is
uniform and is equal to the product and exhaust gas
temperatures. However, due to inherent product resi-
dence time distribution, product moisture content will
span the range from inlet moisture content to a lower
value. One advantage of the perfect mixing dryer is
that the feed falls into a bed of relatively dry material
and so is easy to uidize.
4.3 Plug-Flow Fluidized Bed Dryers (PFFBDs)
In plug-ow uidized bed dryers the bed usually has a
length-to-width ratio in the range 5 : 1 to 30 : 1; the
solids ow continuously as a plug through the channel
from the inlet to the exit. This ensures approximately
equal residence time for all particles, regardless of their
size. For nearly monodisperse particles this ensures
uniformity of product moisture content. The main
operational problems occur at the feed end where
wet feedstock must be uidized directly rather than
mixed with drier material as in a well-mixed unit. To
handle this problem, several alternative strategies may
be employed, e.g.,
Use agitator in feed region
Use backmixing of solids
Use a ash dryer to remove surface moisture prior
to uidized bed drying
At the tail end of drying, thermal eciency can be poor
as little drying takes place while the gas ow rate
remains high to maintain uidization. Zoning of the
plenum (so that the drying gas temperature, and velo-
city to lesser extent, is progressively reduced as the
material dries) is helpful to enhance eciency as well
as to reduce thermal degradation of product. A sche-
matic sketch of the plug-ow uidized bed dryer is
shown in Fig. 6.
4.4 Vibrated Fluidized Bed Dryers (VFBDs)
For beds of particles that are dicult to uidize due to
strong polydispersity, particle size, or particle-to-parti-
cle adhesive forces (stickiness), it is worth considering a
batch or continuous vibrated uidized bed dryer. An
application of nearly vertical sinusoidal mechanical
vibration (half-amplitude 35 mm; frequency 1050
Hz) allows pseudo-uidization of the bed with
rather low air ow rates. In this case, the requirements
of hydrodynamics and heat/mass transfer are eec-
tively decoupled. The gas velocity can be chosen to
t the needs of the drying kinetics; it must be signi-
cantly less than the minimum uidization velocity,
since above such a velocity the bed is no longer in
continuous contact with the vibrating distributor
plate. Furthermore, since the vibrational energy trans-
mitted is attenuated with distance from the plate, the
bed depth is limited to about 1015 cm. Vibrated bed
dryers can also be used to reduce attrition by gentle
processing. A schematic sketch of this dryer is shown
in Fig. 7. Most vibrated uidized bed dryers are con-
tinuous units; the vibration vector is applied at a small
angle to the vertical to assist with conveying of the
material from the feed end to the exit weir.
4.5 Mechanically Agitated Fluidized Bed Dryers
Several versions of such dryers are in use. For drying
of pastes or sludges, one variant uses a cylindrical ves-
sel with a fast spinning agitator at the bottom onto
which the feed drops by gravity for dispersion into
an upward spiral of hot drying gas. Other versions
use a high rpm chopper that disperses the feed into
hot air. More commonly, slowly rotating agitators
(or rakes) are used to facilitate uidization in the
feed zone where very wet feed is fed into a continuous
plug-ow dryer. Often, the rst stage of a plug-ow
dryer may be a well-mixed unit designed primarily to
act as an ecient solids backmixer to facilitate uidi-
zation in the plug-ow stage.
4.6 Centrifugal Fluidized Bed Dryers
To intensify heat and mass transfer rates for rapid
drying of surface-wet particles, a centrifuge-type device
may be used so that the drag force due to the uidizing
gas can be balanced with an articial gravity gener-
ated by rotating the bed on a vertical axis. The rotating
uidized bed equipment is complex, and the decrease
in drying times for most materials is normally not high
enough or essential enough to justify the cost and com-
plexity.
4.7 Spouted Bed Dryers (SBDs)
Spouted bed dryers are found suitable for drying of
Geldarts type D particles, which are too coarse and
dense to uidize well without channeling. Unlike ui-
dized beds where the particle motion is random, the
movement of particles in spouted beds is a regular
recirculatory motion. Both batch and continuous
modes of operation are possible. Owing to their limited
processing capability per unit oor area and their high
power consumption, spouted beds have not yet found
major commercial applications. They are more com-
monly used for roasting (e.g., coee/cocoa beans,
Figure 6 Plug-ow uidized bed dryer.
Figure 7 Vibrated uidized bed dryer.
soya beans). Also, they have found limited applications
for small-scale drying of slurries and suspensions
sprayed into a spouted bed of inert particles (e.g.,
Teon, polyethylene pellets). Another application
involving drying is that of the coating of pharmaceu-
tical tablets.
We have not discussed multistage uidized bed
dryers for lack of space. However, in some applications
they yield major advantages. A dryer may be a part of
a multistage FBD operated at dierent conditions or
as a part of other dryer types, e.g., a ash dryer or
spray dryer followed by a FBD.
Table 5 summarizes a selection guide for various
uidized bed dryer types based on their operating char-
acteristics. Suggested dryers for drying of pasty feed-
stocks are listed in Table 6. For drying granular
materials, a suitable dryer may be chosen with the
aid of a decision tree illustrated in Fig. 8.
5 DESIGN CONSIDERATIONS
5.1 A Simple Calculation Method for Batch Drying
in a Fluidized Bed Dryer
If the amount of moisture to be removed from the
particles in the nal stages is very small, then the latent
heat requirements can be ignored relative to the sensi-
Table 5 Fluidized Bed Dryer Selection Guide (Based on Operating
Characteristics)
Selection criterion WMFBD PFFBD VFBD SBD
Feed Inert Inert Inert
Liquid/suspension/slurry particle bed No particle bed particle bed
Wet particles
Free-owing Yes Yes Yes Yes
Cohesive No No No No
Particle size
Small Yes Yes Yes No
Medium Yes Yes Yes Yes
Large No No Yes No
Polydisperse No No Yes No
Moisture
Surface Yes Yes Yes Yes
Internal Yes Yes Yes Yes
Product specication
Uniform moisture No Yes Yes No
Low moisture No Yes Yes No
Fragile No No No No
Heat-sensitive Yes No No No
Drying time
510 min Yes No No No
1060 min Yes Yes Yes Yes
60 min No Yes No Yes
Throughput
Low Yes Yes Yes Yes
Medium Yes Yes Yes Yes
High Yes Yes Yes No
Immersed heaters Yes No No No
Temperature zoning No Yes Yes No
Flexibility in operation No Yes Yes No
ble heat required to bring the product temperature
from its wet-bulb value (at the end of the surface dry-
ing period in a purely convective dryer) to the nal
temperature that (from laboratory experiments) the
product must reach at the end of drying to achieve
the desired moisture content. In such cases, the batch
drying time in the nal drying period can be estimated
reasonably accurately by a simple transient energy bal-
ance to determine the time required to heat up the
particles from the wet-bulb to the nal desired tem-
perature. In most cases, it is also necessary to cool
the product down to a lower temperature for storage
or packaging. Indeed, the same design equation can be
used to estimate the cool down time. The governing
equation for the energy balance is (vant Land, 1991)
dT
p
dt

G
g
C
pg
T
in
T
p
t
m
s
C
ps
1
Upon integration, this yields the heat-up time, which is
the dryer residence time needed for the falling rate
period drying:
t
m
s
C
ps
G
g
C
pg
log
T
in
T
p
t 0
T
in
T
p
t

2
In the case when the amount of moisture to be
removed is large and the eect of latent heat cannot
be neglected, dierent procedures must be employed.
Kunii and Levenspiel (1991) have presented simplied
procedures for analysis and design of batch and con-
tinuous uidized bed dryers. As expected, dierent cal-
culation procedures are required to calculate the
drying time for the constant rate and falling rate per-
iods. When the moisture diusivity is high, the particle
size small, and the drying conditions mild, much of the
drying will occur in the constant rate period. In prac-
tice, uidized bed dryers are used when a signicant
portion of the moisture is internal, in which case a
simple diusion model can be used to estimate the
drying time for the falling rate period. Interested read-
ers are referred to the design procedures and the exam-
ple design problems presented by Kunii and Levenspiel
(1991).
Table 6 Suggested Fluidized Bed Dryers for Pasty Feedstocks
Feedstock consistency Dryer suggested
Pumpable slurry, suspension Fluid bed, spouted bed of inert particles
Soft pastes/sludges WMFBD/backmix with dry product
Hard pastes FBD or SBD
Figure 8 Decision tree for selection of uidized bed dryers for granular feedstock. Note: Other dryer types are also suited for
several of the above applications. If the feedstock contains a solvent, uidized bed dryers are not recommended in general;
indirect or vacuum dryers are preferred instead.
5.2 Predicting the Performance of a Continuous
Fluidized Bed Dryer from Batch Drying Data
Provided that a batch drying curve Xt is available,
the performance of a continuously well-mixed uidized
bed dryer can be predicted, at least in principle, by a
so-called integral model of drying rst proposed by
Vanecek et al. (1964). The model is based on the fol-
lowing equation for calculating the average outlet
moisture content, X
0
, from a continuous dryer:
X
0

1
0
EtXt dt 3
where Et is a function describing the residence time
distribution of the material in the dryer. Xt is a dry-
ing curve function describing how the moisture content
varies as the material passes through the dryer.
In a continuously well-mixed dryer, where the resi-
dence time distribution is that of a perfectly mixed
vessel, Eq. (3) becomes
X
0

1
t
m
1
0
Xt exp
t
t
m

dt 4
where t
m
is the mean residence time of particles in the
bed.
Reay and Allen (1982) discussed the diculty asso-
ciated with the use of this equation. The batch drying
curve Xt must be obtained at a constant bed tem-
perature corresponding to the temperature selected
for the continuous operation. This value is normally
not known at the start of the development process, and
in any case is extremely dicult to obtain experimen-
tally. The experiment has to be performed with a high
initial inlet air temperature, which must then be
decreased at the later stage of drying in order to keep
the bed temperature constant at the desired value. This
is very dicult to achieve even with a sophisticated
automatic control system.
The problem was overcome, however, by Reay and
Allen (1982) following their derivation of a normaliza-
tion rule for the eect of bed temperature on batch
drying curves for both type A and B materials. They
proposed, and veried experimentally, the following
steps for constructing a constant bed temperature
batch drying curve at any desired bed temperature
from a batch drying curve measured at any convenient
constant inlet air temperature.
1. Obtain a record of the changing bed tempera-
ture T
ii
during constant inlet air temperature
run.
2. Divide the constant inlet air temperature batch
drying curve Xt into increments of length
X. For each increment note the time t
ii
required to accomplish that amount of drying.
3. Calculate the time t
ib
required to accomplish
the same increment of drying at constant bed
temperature by the use of the equation:
t
ib
t
ii
p
s
p
i
X X
e

T
ii
p
s
p
i
X X
e

T
ib
5
4. Build up the constant bed temperature batch
drying curve by increments.
By solving Eq. (4) for a range of mean residence
times, a so-called design curve can be constructed
and used to extract the mean residence time required
to achieve a specied outlet moisture content. The bed
area can then be calculated for a given throughput S
and bed mass per unit area m
b
(Bahu, 1994):
A
St
m
m
b
6
The choice of the bed mass per unit area is dictated by
the need to break up the wet feed and to ensure rapid
dispersion away from the feedpoint.
For a plug-ow uidized bed dryer, the residence
time will deviate from idealized plug-ow because of
backmixing. This can be accounted for by employing
the axial dispersion number, B Dt
m
,L
2
, where D is
the particle diusivity given by Reay (1978):
D
3.71 10
4
U U
mf
U
1,3
mf
7
and L is the bed length. For small deviations from
plug-ow, i.e., for small values of B- 0.1, the resi-
dence time function is
Et
1
2
B
p exp
I t,t
m
2
4B
" #
8
It should be noted that the validity of Reays corre-
lation for particle diusivity has only been conrmed
for bed depths up to 0.10 m. There is some evidence in
the literature that D may be an order of magnitude
larger in much deeper beds (Reay and Baker, 1985).
The shallowest practicable beds are thus recommended
if the objective is a close approach to plug-ow beha-
vior of solid particles in the bed.
A summary of the basic scale-up of a uidized bed
dryer is illustrated in Fig. 9. For more accurate analy-
sis or optimization, more advanced mathematical
models can be found in the literature. Indeed, numer-
ous mathematical models have appeared in the techni-
cal literature on modeling of the heat and mass transfer
processes in uidized bed drying of particles as well as
uidized bed drying of slurries using beds of inert par-
ticles. Borde et al. (1997) have presented a simple cal-
culation procedure for the design of a vibrated bed
dryer. The review presented by Pakowski et al. (1984)
is also very useful to obtain a clear idea about the ow
and heat/mass transfer characteristics of vibrated ui-
dized beds. For the design and analysis of conventional
spouted beds as well as certain modied spouted beds
the reader is referred to the work of Passos et al.
(1987). For the special case of intermittently spouted
beds for grain drying, Jumah et al. (1996) present a
simple diusion-based model for the estimation of dry-
ing times. Their model may be used only if the entire
drying occurs in the falling rate period and the diu-
sion model adequately describes the falling rate period.
Unfortunately, no degree of sophistication in mod-
eling can predict product quality parameters. Thus
pilot testing is essential prior to scale-up to full-scale
dryer design.
6 NOVEL FLUIDIZED BED DRYERS
Numerous innovative designs and operational modi-
cations have been proposed in the literature while
fewer are readily available commercially.
Tables 7, 8, and 9 summarize the key features of the
innovative dryer concepts relative to the more con-
ventional FBDs. Many of the ideas presented are new
and have not been investigated even at the laboratory
stage. It must be noted that not all the novel ideas are
necessarily better, but many are worth exploring, since
they appear to oer some distinct benets over the
more conventional ones.
6.1 Modied Fluidized Bed Dryers
Depending on special requirements it is possible to
develop modied versions of the traditional uidized
bed dryers. Two-stage uidized bed dryers consisting
of several dryers are now in commercial use. Some
examples of these combinations are given in Table 10.
Figure 9 Basic steps in scale-up of uidized bed dryer.
Table 7 Fluidized Bed Dryers: Conventional Versus Innovative Concepts
Conventional Innovative
Convective heat transfer Convection conduction (immersed heaters in bed)
Steady gas ow Pulsed gas ow
Constant gas temperature Variable gas temperature
Pneumatic uidization Mechanically assisted uidization (vibration/agitation)
Used for drying of particles Drying pastes, slurries using inert media
Air/combustion gas as drying medium Superheated steam for uidization/drying
Air drag resisted by gravity Centrifugal uid beds (articial gravity generated by rotation)
Single stage/multistage uid beds Multistage with dierent dryer types
Simultaneous uidization of entire bed Moving uidization zone (pulsating uidized bed)
Pulsing the uidizing gas ow has been demon-
strated to enhance drying performance in some cases
at the expense of added capital costs. Periodic local
uidization of the bed by moving the distributor
zone at regular intervals (so-called pulsated uidized
beds) has the advantage of reduced energy and air
consumption when batch drying particulate materials
with most of the heat and mass transfer resistances
within the particles (Gawrzynski and Glaser, 1996).
For larger particles (e.g., Geldarts type D), intermit-
tent spouting of the bed with a rotating spouting jet
has been shown by Jumah et al. (1996) to reduce
energy consumption at only a marginal increase in dry-
ing time for batch drying. Indeed, in this case, it is also
possible to introduce another type of intermittency,
e.g., periodic heating of the spouting air when drying
in the batch mode.
To conserve energy while reducing the drying time
in batch uidized bed drying one may remove the sur-
face moisture using a higher gas temperature initially;
the gas temperature can be reduced progressively as the
material dries so that it does not exceed its maximum
permissible temperature (Devahastin and Mujumdar,
1999). This can be achieved with a fuzzy logic control-
ler or with a model-based control scheme. In plug-ow
dryers this idea can be extended by temperature-zoning
along the length of the dryer. The nal section may
generally act as a uid bed cooler to reduce the product
temperature to a safe value for storage or packaging
(Bahu, 1997). If the product is highly hygroscopic, and
low nal moisture contents are desired, the cooling
stage may require dehumidied air.
In the following we will review only a few innovative
FBD techniques. For further details, see Kudra and
Mujumdar (1995).
6.2 Superheated Steam Drying in Fluidized Beds
Although the idea was rst published as early as 1898,
the potential for superheated steam as a drying med-
ium was not exploited industrially for at least six dec-
ades thereafter. Superheated steam drying oers one or
more of the following advantages (Mujumdar, 1995):
No re or explosion hazards
No oxidative damage
Ability to operate at vacuum or high-pressure oper-
ating conditions
Ease of recovery of latent heat supplied for evapora-
tion; eectively a multiple eect operation is fea-
sible
Table 8 Spouted Bed Dryers: Conventional Versus Innovative Concepts
Pneumatic spouting Mechanical spouting (screw, vibration)
Single spout Multiple spouts
Constant gas ow/continuous spouting Variable gas ow/pulsed gas ow
Drying particles Drying pastes, slurries using inert media
Spatially xed spout Moving spout (rotation, oscillation)
Convective drying Combined convection and conduction
Axisymmetric Two-dimensional, annular, hexagonal, etc.
Table 9 Vibrated Bed Dryers: Conventional Versus Innovative
Concepts
Constant gas ow Variable gas ow
Aerated Nonaerated
Convective drying Combined conduction/radiation/microwave
Near atmospheric pressure Vacuum operation
Horizontal trough Vertical spiral trough
Fixed frequency/amplitude Variable frequency/amplitude
A higher drying rate above a critical inversion tem-
perature when removing surface moisture
Better quality product under certain conditions
Closed system operation to minimize air pollution
Osetting some of these advantages are some limita-
tions, e.g.,
Higher product temperature
Higher capital costs compared to hot air drying
Possibility of air inltration making heat recovery
from exhaust steam dicult by compression or
condensation
Among industrial scale applications of superheated
steam uidized bed drying one may cite
Drying of pulverized coal (Faber et al., 1986)
Drying of high-moisture lignite using a uidized bed
with internal heat exchangers to improve energy
eciency and reduce dryer size (Potter et al.,
1986)
Drying of pulps, bagasse, sludges, hog fuel, spent
grain from breweries, etc., in a pressurized super-
heated steam dryer (35 bars) manufactured by
Niro A/S, Denmark (Jensen, 1992)
As may be expected, the net energy consumption of
such dryers can be very small. Up to 90 percent fuel
saving over conventional dryers has been reported for
drying of beet pulp in a high-pressure uidized bed
dryer (Bosse and Valentin, 1988).
6.3 Low-Pressure Fluidized Bed Drying
Owing to the high potential of thermal degradation
and re or explosion hazards if the wet solids contain
an organic solventcommon occurrences in the ne
chemical and pharmaceutical industrieslow-pressure
operation oers several advantages (Mujumdar, 1995).
Similar problems may arise in coating operations. To
alleviate this problem, low-pressure uidization may
be used to lower the operating temperature, thus redu-
cing the probability of thermal degradation; it may
allow operation in conditions out of the ammability
limits. Arnaldos et al. (1997) have presented experi-
mental results on the hydrodynamic behavior of
vacuum uidized bed dryers and their applications
for drying granular solids and coating of particles.
Note that the minimum uidization velocity increases
as the pressure decreases. The ow behavior at lower
pressures is quite dierent from that at atmospheric or
high-pressure operation. The uidization is progres-
sive, and the concept of minimum uidization velocity
loses some of its signicance. Experimental measure-
ments of the velocity for complete uidization are
needed for design purposes.
Little work is published on vacuum uidized bed
drying. Arnaldos et al. (1997) report vacuum uidized
bed drying data at 200, 400, and 500 mbar pressures.
Silica d
p
975 jm, ,
s
1650 kg,m
3
;
s
0.7) and
millet (d
p
1800 jm, ,
s
1600 kg,m
3
;
s
0.90)
were used as test materials. Interestingly, the highly
porous silica particles dried faster as the pressure was
lowered, while the denser millet particles dried slower
since much of its resistance to drying is internal. Thus
the eect of lower pressure depends on the internal
heat and moisture resistance of the particles. Because
of their inherently higher capital and operating costs,
vacuum uidized bed dryers have found application
mainly in the pharmaceutical industry.
CLOSING REMARKS
Fluidized bed dryers, both conventional and innova-
tive, will continue to nd increasing applications in
various industries. For further details, the reader is
asked to refer to Vanecek et al. (1965), Gupta and
Mujumdar (1983), Pakowski et al. (1984), Reay and
Baker (1985), Hovmand (1995), Keey (1992), and
Mujumdar (1995), which contains an extensive anno-
Table 10 Two-Stage Fluidized Bed Dryers
Type Remarks
Similar FBD stages stacked one below the other Reduces oor area/bed depth of each stage; low product
moisture content possible
Flash dryer stage preceding FBD stage Fast removal of surface moisture; reduced stickiness leads
to easy uidization/smaller uidized bed dryer
Spray dryer stage followed by FBD stage Signicantly reduced spray dryer size
Well-mixed FBD followed by plug-ow FBD
stage underneath in same vessel
Ease of uidization for high moisture content feed; uniform/
low nal product moisture content possible
tated bibliography of the major literature sources on
drying. Numerous papers dealing with mathematical
models for conventional and modied uidized bed
dryers appear regularly in Drying TechnologyAn
International Journal (Marcel Dekker, NY).
NOTATIONS
A = bed cross-sectional area, m
2
B = axial dispersion number
C
pg
= specic heat of gas, kJ kg
1
K
1
C
ps
= specic heat of solid, kJ kg
1
K
1
D = particle diusivity, m
2
s
1
d
p
G
g
= gas ow rate, kg s
1
L = bed length, m
m
b
= bed mass per unit area, kg m
2
m
s
= bed mass, kg
p
i
= partial pressure of vapor in gas entering bed,
kPa
p
s
= saturation partial pressure of vapor at the bed
temperature, kPa
S = throughput, kg s
1
T
ib
= bed temperature in a batch drying curve with
constant bed temperature, K
T
ii
= bed temperature in a batch drying curve with
constant inlet temperature, K
T
in
= inlet gas temperature, K
T
p
= particle temperature, K
t = time, s
X = moisture content, kg water/kg dry solid
X
e
= equilibrium moisture content, kg water/kg dry
solid
X
0
= mean product moisture content, kg water/kg dry
solid
U = supercial gas velocity, m s
1
U
mf
= supercial minimum uidization gas velocity, m
s
1
Subscripts
b = bed
e = equilibrium
g = gas
p = particle
s = solid
Greek Symbols
s
= particle sphericity
,
s
= particle density, kg m
3
t
m
= mean particle residence time, s
REFERENCES
Arnaldos J, Martra J, Casal J, Kozanoglu B. Vacuum uidi-
zation processes: drying and coating. Powder Handling
and Processing 9:315319, 1997.
Bahu RE. Fluidised bed dryer scale-up. Drying
TechnologyAn International Journal 12:329339,
1994.
Bahu RE. Fluidized bed dryers. In: Baker CGJ, ed. Industrial
Drying of Foods. London: Blackie Academic and
Professional, 1997, pp. 6589.
Baker CGJ. Dryer selection. In: Baker CGJ, ed. Industrial
Drying of Foods. London: Blackie Academic and
Borde I, Dukhovny M, Elperin T. Heat and mass transfer in
a moving vibrouidized granular bed. Powder Handling
Bosse D, Valentin P. The thermal dehydration of pulp in a
large scale steam dryer. Proceedings of the Sixth
International Drying Symposium, Versailles, 1988, pp.
337343.
Devahastin S, Mujumdar AS. Batch drying of grains in a
well-mixed dryer-Eect of continuous and stepwise
change in drying air temperature. Transactions of the
ASAE 42:421425, 1999.
Faber EF, Heydenrych MD, Seppa RUI, Hicks RE. A
techno-economic comparison of air and steam drying.
In Mujumdar AS, ed. Drying86, Vol. 2. Washington:
Hemisphere, 1986, pp. 588594.
Gawrzynski Z, Glaser R. Drying in a pulsed-uid bed with
relocated gas stream. Drying Technology-An
International Journal 14:11211172, 1996.
Gupta R, Mujumdar AS. Recent developments in uidized
bed drying. In Mujumdar AS, ed. Advances in Drying.
Vol. 2. Washington: Hemisphere, 1983, pp. 155192.
Hovmand S. Fluidized bed drying. In: Mujumdar AS, ed.
Handbook of Industrial Drying. 2d ed. New York:
Marcel Dekker, 1995, pp. 195248.
Jensen AS. Pressurized drying in a uid bed with steam.
Proceedings of the Eighth International Drying
Symposium, Montreal, 1992, pp. 15931601.
Jumah RY, Mujumdar AS. A PC program for preliminary
design of a continuous well-mixed uid bed dryer.
Drying TechnologyAn International Journal 11:831
846, 1993.
Jumah RY, Mujumdar AS, Raghavan GSV. Control of
industrial dryers. In: Mujumdar AS, ed. Handbook of
Industrial Drying. 2d ed. New York: Marcel Dekker,
1995, pp. 13431368.
Jumah RY, Mujumdar AS, Raghavan GSV. A mathemati-
cal model for constant and intermittent batch drying of
grains in a novel rotating jet spouted bed. Drying
TechnologyAn International Journal 14:765802,
1996.
Keey RB. Drying of Loose and Particulate Materials.
Washington: Hemisphere, 1992.
Kudra T, Mujumdar AS. Special drying techniques and
novel dryers. In: Mujumdar AS, ed. Handbook of
Industrial Drying. 2d ed. New York: Marcel Dekker,
1995, pp. 10871149.
Liptak B. Optimizing dryer performance through better con-
trol. Chemical Engineering 105:96104, 1998.
Mujumdar AS. Superheated steam drying. In: Mujumdar
AS, ed. Handbook of Industrial Drying. 2d ed. New
York: Marcel Dekker, 1995, pp. 10711086.
Mujumdar AS, Menon AS. Drying of solids. In: Mujumdar
AS, ed. Handbook of Industrial Drying. 2d ed. New
York: Marcel Dekker, 1995, pp. 146.
Mujumdar AS, Alterman DS. Drying in the pharmaceutical
and biotech elds. In Goldberg E, ed. Handbook of
Downstream Processing. London: Blackie Academic and
Pakowski Z, Mujumdar AS, Strumillo C. Theory and appli-
cation of vibrated beds and vibrated uid beds for drying
processes. In Mujumdar AS, ed. Advances in Drying. Vol.
3. Washington: Hemisphere, 1984, pp. 245306.
Passos ML, Mujumdar AS, Raghavan GSV. Spouted bed for
drying: principles and design considerations. In
Mujumdar AS, ed. Advances in Drying. Vol. 4.
Washington: Hemisphere, 1987, pp. 359398.
Potter OE, Beeby C. Modeling tube to bed heat transfer in
uidized bed steam drying. In Mujumdar AS, ed.
Drying86. Vol. 2. Washington: Hemisphere, 1986,
pp. 595603.
Reay D. Particle residence time distributions in shallow rec-
tangular uidised beds. Proceedings of the First
International Drying Symposium, Montreal, 1978, pp.
136144.
Reay D, Allen RWK. Predicting the performance of a
continuous well-mixed uid bed dryer from batch tests.
Proceedings of the Third International Drying
Symposium, Birmingham, 1982, pp. 130140.
Reay D, Baker CGJ. Drying. In: Davidson JF, Clift R,
Harrison D, eds. Fluidization 2d ed. London: Academic
Press, 1985, pp. 529562.
Vanecek V, Picka J, Najmr S. Some basic information on the
drying of granulated NPK fertilizers. International
Chemical Engineering 4:9399, 1964.
Vanecek V, Markvart M, Drbohlav R. Fluidized Bed
Drying. London: Leonard Hill, 1965.
vant Land CM Industrial Drying Equipment: Selection and
Application. New York: Marcel Dekker, 1991.
19
Circulating Fluidized Beds
John R. Grace and Hsiaotao Bi
University of British Columbia, Vancouver, British Columbia, Canada
Mohammad Golriz
Umea University, Umea, Sweden
1 INTRODUCTION
The term circulating uidized bed, commonly abbre-
viated CFB, has been in common usage since the mid-
1970s, although the origins of the technology date back
to the 1940s for catalytic processes (Squires, 1994) and
to the 1960s for gassolid processes (Reh, 1971). The
term implies two complementary characteristics for
gassolid systems:
1. A conguration where particles, entrained at a
considerable ux from a tall main reactor or riser,
are separated eciently from the carrying uid, usually
external to the reactor, and returned to the bottom of
the riser, forming a recirculation loop for the particles.
Individual particles circulate around this loop many
times before leaving the system, whereas the uid passes
through only once. A typical setup is shown schemati-
cally in Fig. 1. As described in Sec. 4 below, many dif-
ferent geometric congurations are utilized in practice.
2. Operation at high supercial gas velocity (typi-
cally 212 m/s) and high particle ux (typically 101000
kg/m
2
s) so that there is no distinct interface in the riser
between a dense bed and a dilute region above.
Contacting is therefore carried out at gas velocities
beyond the bubbling, slugging and turbulent uidiza-
tion ow regimes, residing instead in a higher velocity
owregimefast uidization, dense suspensionupow,
or dilute pneumatic conveying (see Sec. 3 below).
In recent years, the term circulating uidized bed
has also been used for liquidsolid systems and gas
liquidsolid (three-phase) systems. These are treated in
Chapters 26 and 27. However, the predominant inter-
est in CFB systems continues to be for gassolid (two-
phase) systems, and we restrict this chapter to this case.
Some mention is also made in the literature of intern-
ally circulating uidized beds where the particles circu-
late around one or more loops within a main reactor
space or vessel. This chapter deals principally with gas
solid systems where solids recirculate through an exter-
nal solids ow system, usually involving one or more
cyclones, a standpipe, and a valve or seal, either non-
mechanical (e.g., L-valve or loop seal; see Chapter 21)
or mechanical (e.g., slide valve).
The key operating variables for commercial opera-
tion usually fall into the following ranges:
Supercial gas velocity: 212 m/s
Net solids ux through the riser: 101000 kg/m
2
s
Temperature: 20950
C
Pressure: 1002000 kPa
Mean particle diameter: 50500 jm
Overall riser height: 1540 m
Historical accounts of the development of circulat-
ing uidized beds have been prepared by Reh (1986)
and Squires (1994). Advantages of circulating uidized
beds relative to bubbling beds and other types of gas
solid contactors such as packed beds and rotary kilns
frequently include
High gas throughputs
Limited backmixing of gas
Long and controllable residence time of particles
Temperature uniformity, without hot spots
Flexibility in handling particles of widely diering
sizes, densities, and shapes
Eective contacting between gas and particles
Lack of bypassing of gas with minimal mass trans-
fer limitations
Opportunity for separate and complementary
operation (e.g., catalyst regeneration or particle
cooling) in the return loop
Disadvantages often include:
Need for very tall vessel: small scale CFB processes
are therefore seldom viable
Substantial backmixing of solid particles
Internals (e.g., baes, heat transfer surfaces) not
viable because of wear/attrition
Wall wastage sometimes a serious problem
Suspension-to-surface heat transfer less favorable
than for low-velocity uidization
Lateral gradients can be considerable
Losses of particles due to entrainment.
Extensive reviews of all aspects of circulating ui-
dized beds up to about 1996 are provided by the books
edited by Kwauk (1994) and Grace et al. (1997). More
limited reviews have been provided by Yoshida and
Mineo (1989) and by Berruti et al. (1995). This chapter
summarizes the most important and useful ndings
and approaches to circulating uidized beds. This is
an active area for research and development with
new results and models emerging each year.
International conferences devoted explicitly to circulat-
ing uidized beds are held every three years, each lead-
ing to published proceedings, cited frequently in this
chapter. Many CFB papers also appear in journals and
proceedings of other conferences, e.g., the triannual
Fluidization Conferences under the auspices of the
Engineering Foundation and the biannual
International Fluidized Bed Combustion conferences.
2 APPLICATIONS
Major commercial applications of circulating uidized
beds are listed in Table 1, together with key references.
These applications include solid-catalyzed gas reac-
tions and gassolid reactions, as well as physical opera-
tions. While catalytic cracking and solid-fuel
combustion are the predominant catalytic and gas
solid reaction applications, respectively, there are also
a number of other processes where the unique charac-
teristics of circulating uidized beds are being
exploited. Further details on the most important appli-
cations are presented in Chapters 14 through 16.
3 FLOW REGIMES
3.1 Onset of Fast Fluidization
A circulating uidized bed (CFB) is operated in the
transport mode, with solids carried over from the top
of the riser separated and returned to the bottom of the
riser via a standpipe and feeding or control device. The
transition from low-velocity uidization to transport
operation occurs when signicant solids entrainment
commences with increasing supercial gas velocity.
At least seven methods have been proposed (Bi et al.,
2000) to quantify the transition. The criteria can be
divided into two groups, one based on solids entrain-
ment and the other on solids concentration proles.
3.1.1 Transport Velocity, U
tr
, based on Phase
Diagrams
A critical solid circulation rate may exist when a sharp
change in pressure drop over the lower part of a riser
takes place when varying the solid circulation ux at a
Figure 1 Schematic of typical circulating uidized bed
system.
given gas velocity, U
g
(see Fig. 2). As U increases
beyond a certain point, the sharp change in the pressure
gradient disappears. The gas velocity at this critical
point, dened as the transport velocity U
tr
, marks the
onset of fast uidization (Yerushalmi and Cankurt,
1979). An examination of pressure gradient proles
(Bi, 2002) reveals that U
tr
varies with height. U
tr
may
indicate a transition of axial voidage proles in the riser.
Below this velocity, a distinct interface exists between
the top-dilute and bottom-dense regions. Beyond this
velocity, the interface becomes relatively diuse.
To predict U
tr
, several correlations have been devel-
oped, most of the form
Re
tr
= kAr
n
(1)
Typical k and n values are given in Bi et al. (2000), with
k = 2.28 and n = 0.419 from Bi and Fan (1992). In
view of the diculty of determining U
tr
experimen-
tally, and the variation of U
tr
with measurement loca-
tion and riser geometry, caution needs to be exercised
when Eq. (1) is used to predict U
tr
.
3.1.2 Critical Velocity, U
se
, Based on Solids
Entrainment
In a gassolids transport system, solids ux and gas
velocity are related by
G
s
= ,
p
(1 c)
U
g
c
U
slip
_ _
(2)
A linear relationship between G
s
and U
g
in the high-
velocity range with e constant in Fig. 3 suggests that
U
slip
in Eq. (2) approaches a constant value, deter-
mined from the intercept and slope of the linear part
of the curve. This critical velocity, designated U
se
(Bi et
Table 1 Major Commercial Applications of Circulating Fluidized Beds
Application Key References Comments
Fluid catalytic cracking Avidan, 1997; Sec. 8.1 of this handbook Hundreds of units worldwide; mainstay of
petroleum rening
FischerTropsch synthesis Shingles & McDonald, 1988; Steynberg
et al., 1991; Matsen, 1997
Applied for many years as Synthol process in
South Africa
Maleic anhydride Matsen, 1997; Contractor, 1999 One commercial reactor in Spain
Combustion of coal,
biomass, wastes, o-gases
Li and Zhang, 1994; Brereton, 1997;
Lee, 1997; Basu, 1999; Plass, 2001
Widespread usage for power generation and
boilers in Europe, North America, and Asia
Gasication Hirschfelder & Vierrath, 1999; Plass, 2001 Commercial units gaining a foothold, especially
in Europe
Calcination (e.g., of
aluminium trihydrate
and carbonates)
Reh, 1971, 1986; Schmidt, 1999 Lurgi units used widely
Catalyst regeneration Chen et al., 1994 Applied in China
Roasting of ores Dry & Beeby, 1997; Pienermann et al.,
1992
Applied in Australia
Reduction of iron ore Dry & Beeby, 1997; Husain et al., 1999;
Plass, 2001
Lurgi plant in Trinidad
Smelter o-gas treatment Hiltunen & Moyo ha nen, 1992 One plant in Australia supplied by Ahlstrom
Flue gas dry scrubbing of
HF, HCl, SO
2
, dioxins,
mercury, etc.
Graf, 1999; Mayer-Schwinning &
Herden, 1999
Commercial units since the 1970s, primarily in
Europe
Figure 2 Denition of transport velocity, U
tr
, of
Yerushalmi and Cankurt (1979).
al., 1995), can be considered a hindered or apparent
terminal velocity of bed particles. U
se
can also be deter-
mined in a batch uidized bed from the emptying time
(Bi et al., 1995). A correlation for U
se
based on data
from relatively tall (H > 5 m) and large (D > 0.05 m)
columns (Bi et al., 1995) gives
Re
se
= 1.53Ar
0.5
(3)
When U
se
from Eq. (3) - the terminal velocity, u
t
, of
single particles, U
se
should be taken = u
t
.
3.2. Flow Regimes in GasSolids Circulating
Fluidized Beds
3.2.1 Transition from Pneumatic Transport to Fast
Fluidization
Fast uidization is characterized by a dense region at
the bottom of a circulating uidized bed, leading
smoothly (without a sharp interface) into a lean region
above (Li and Kwauk, 1980). In contrast, in the
dilute-phase ow regime, the pressure gradient, except
for an acceleration zone at the bottom, is nearly uni-
form. Hence the transition from fast uidization to
dilute ow can be characterized by the disappearance
of the S-shaped inection point (Li and Kwauk, 1980),
or by the disappearance of nonuniform axial density
proles (Takeuchi et al., 1986).
In conventional pneumatic transport, a minimum in
the dP,dz vs. G
s
curve at a xed gas velocity is com-
monly used to separate dense phase ow from lean
phase transport (Leung, 1980). Experimentally, inter-
nal solids circulation increases dramatically right after
the gas velocity is reduced to reach the minimum pres-
sure gradient point (Matsumoto and Harakawa, 1987).
Radial segregation into a dilute core and dense annulus
results from the balance between the force needed to
suspend the weight of the particles and wall friction,
and the transition velocity is inuenced by the unit
geometry.
Several correlations have been proposed to predict
the transition velocity, U
CA
(see Bi et al., 1993). The
correlations of Bi and Fan (1991),
U
CA
= 21.6
gd
p
_
G
s
,
g
U
CA
_ _
0.542
Ar
0.105
(4)
and Yang (1975),
2gD c
4.7
CA
1
_ _
U
CA
,c
CA
u
t
( )
2
= 6.81 10
5
,
g
,
p
_ _
2.2
(5)
show the best agreement with literature data.
Improved correlations are needed.
3.2.2 Transition from Fast Fluidization to Dense
Suspension Upow
Stable operation of gassolids upward transport lines
becomes impossible when the blower is unable to pro-
vide sucient pressure head. This condition is referred
to as blower-induced instability (Bi et al., 1993).
Another type of instability can occur when an
upow riser is directly coupled with a downcomer
that returns entrained particles to the bottom of the
riser. A pressure balance between the riser and the
downcomer is required to maintain steady operation.
If the gas velocity is decreased at a given solids circula-
tion rate, a critical state may be reached at which
steady operation at a given solids ux is impossible;
instability occurs because solids cannot be fed to the
riser at the prescribed rate. Such an instability, referred
to as standpipe-induced (Bi et al., 1993), occurs at a
lower critical velocity for a higher solids holdup in the
riser. The point of instability can be predicted based on
an analysis of the pressure balance in the riserdown-
comer loop (Bi and Zhu, 1993). To circumvent stand-
pipe-induced instability, the solids inventory in the
standpipe needs to be suciently high or, alternatively,
the riser needs to be uncoupled from the downcomer,
e.g., by employing screw feeders.
The third instability, classical choking, occurs on
reducing the supercial gas velocity in the riser when
slugs start to form right after a dense bed is created in
the bottom section of a riser. Stable operation of gas
solids upward slug/plug ow may still be possible in
Figure 3 Denition of critical velocity, U
se
, of Bi et al. (1995).
some systems (Konrad, 1986). However, when slug
ow develops to such an extent that severe pressure
uctuations cause the whole system to uctuate dra-
matically, stable operation is no longer possible.
Classical choking only occurs in systems which are
capable of slugging. Criteria for classifying systems
as slugging or nonslugging have been proposed based
on instability analysis of uniform suspension ow
(Yous and Gau, 1974), stability of slugs (Yang,
1975), and propagation of continuity waves (Smith,
1978). Slug ow is not encountered when the maxi-
mum stable bubble size is signicantly smaller than
the column diameter. The onset of classical choking
can be estimated from Yous and Gau (1974):
U
CC
= 32
gd
p
_
G
s
,
g
U
CC
_ _
0.28
Re
0.06
t
(6)
When a riser system has been designed so that blower-
and standpipe-induced instabilities and classical chok-
ing are all avoided, the ow pattern passes from fast
uidization to dense suspension upow (Bi and Grace,
1999) with increasing G
s
, as shown in Fig. 4. Based on
recent experimental observations that solids no longer
fall downward near the riser wall when the riser is oper-
ated at relatively high G
s
and U
g
(e.g., Issangya et al.,
1998; Karri and Knowlton, 1998) and that gas back-
mixing decreases after reaching a maximum at high G
s
,
Grace et al. (1999) proposed that a transition from fast
uidization to dense suspension upow occurs when
the net solids ux near the riser wall changes from
downward to upward with increasing G
s
at a given
U
g
. A rst correlation (Grace et al., 1999) based on
limited experimental data within the range of
500 _Ar _ 6, (1 c
av
) _ 0.07, 100 _G
s
,(,
g
U
g
_ 7,
0.305 _ D_0.051 m, 27.4 _ H _ 6.1 m, with air at
ambient conditions as the gas, gives
U
DSU
= 0.0113G
1.192
s
,
1.064
g
j
g
g ,
,
,
g
_ _ _ _
0.064
(7)
3.3 Flow Regime and Operating Diagrams
A number of ow regime diagrams have been pro-
posed, beginning with Zenz (1949), who plotted the
pressure gradient (dP,dz) vs. supercial gas velocity
(Fig. 5). The cocurrent gassolids ow region spans
the ow regimes encountered in the circulating ui-
dized beds, with the lower limit set by the choking
velocity. There is an inoperable region between the
low-velocity uidization and cocurrent upow,
because the transport line is choked as discussed
above. Such a regime diagram has been further
extended to incorporate more subregions within cocur-
rent ow (e.g., Drahos et al., 1988). The pressure gra-
dient (dP,dz) is almost proportional to the solids
fraction, since friction and acceleration/deceleration
terms can usually be neglected. One can alternatively
plot solids fraction or bed voidage (e.g., Li and
Kwauk, 1980) vs. supercial gas velocity (U
g
), normal-
ized supercial gas velocity (U
g
,u
t
), or slip velocity.
Figure 4 Transitions between dense suspension upow, fast
uidization, and pneumatic transport with increasing solids
circulation ux at a constant gas velocity.
Figure 5 Flow regime diagram based on pressure gradient
and supercial gas velocity coordinates. (Adapted from Zenz,
1949.)
Another set of diagrams, referred to as operating
diagrams (e.g., Takeuchi et al., 1986; Yang, 2001)
plot G
s
or loading ratio [G
s
,(,
g
U
g
)] versus U
g
or nor-
malized supercial gas velocity. A typical diagram of
this type is shown in Fig. 6, with the ow patterns
divided into homogeneous dilute-phase ow, core
annular ow, fast uidization, and dense suspension
upow, with blower and standpipe limitations absent.
Bi and Grace (1995) plotted the dimensionless relative
velocity between the gas and particles vs. a dimension-
less particle diameter based on the Grace (1986a) phase
diagram.
It should be noted that severe slugging, blower, and
standpipe limitations, as discussed above, are not
reected in almost all of these phase diagrams.
4 HYDRODYNAMICS OF THE FAST
FLUIDIZATION FLOW REGIME
4.1 Voidage, Solids Concentration, and Pressure
Proles
4.1.1 Denitions and Measurement Techniques
The voidage is the fraction of volume occupied by the
gas in a gassolid suspension. Voidage is a key variable
in circulating uidized bed risers. Several dierent voi-
dages are in common usage:
Instantaneous local voidage in a measuring volume
that is small with respect to the equipment but
large with respect to individual particles. This is
designated c, or c(x. y. z. t) to indicate variation
with position and time.
Local time-average voidage cc, cc(r. z), or cc(x. y. z).
Cross-sectional average voidage at a given height,
denoted here by c
av
or c
av
(z).
Overall average voidage for the riser, [c], i.e., frac-
tion of entire riser volume occupied by gas.
The overbar signies a time average, whereas the sub-
script av and square brackets denote spatial
averages. Since we are mostly interested in steady
ows, time-averaging is often also carried out. For
steady axisymmetric ow in a cylindrical riser of radius
R and overall height H,
cc(r. z) = lim
to
1
t
_
t
0
c(r. z) dt (8)
c
av
(z) =
1
R
2
_
R
0
c(r. z)2r dr (9)
In some cases we are interested in the uctuations
themselves, as well as in the time-averaged quantities;
we can then write, for example,
[c] =
1
H
_
H
0
c
av
(z) (10)
c(r. z) = cc(r. z) c
/
(r. z) (11)
where the prime denotes a uctuating component, i.e.,
the instantaneous deviation from the time mean. It is
also common to report or discuss the volumetric con-
centration of solid particles, often called the solids
holdup. This terminology is used throughout this
chapter and denoted by c
s
, i.e., c
s
= 1 c. Still another
way of designating the concentration is to refer to a
suspension density, i.e., ,
susp
= ,
p
(1 c) ,
g
c.
Clearly one can obtain local time-averages and spa-
tial-averages values of c
s
and ,
susp
in an analogous
manner to obtaining these quantities for c.
Several techniques have been found to be useful for
determining local voidages/solids holdups:
Optical Probes for the measurement of voidages and
solids holdups in CFB risers were pioneered by Qin
490 Grace et al.
Figure 6 Flow regime diagram based on solids circulation
rate and supercial gas velocity. (Adapted from Bi and
Grace, 1996.)
and Liu (1982). In these probes a single ber or bundle
of bers projects light onto a local region. The solids
concentration is inferred from the intensity of either
the reected or the transmitted light. The transmission
method is restricted to low solids volume fractions
(Werther et al., 1993), while reective probes can be
used over a wide range of particle concentrations after
calibration. Advantages of optical probes include
insensitivity to electrical and magnetic elds, low signal
loss, small size, exibility of geometry, high sensitivity,
rapid response, and low cost. Like any probe they may
disturb the ow, but the extent of interference can be
minimal if the probe is suciently small. Optical ber
probes have also been extended (Johnsson and
Johnsson, 2001) to high-temperature systems with
optical ltering to separate reected light from back-
ground radiation.
If the particle diameter is larger than the ber core
diameter (type I), single particles are detected. The out-
put signals from the light receiver are then converted to
pulses. Averaging over extended periods is required to
measure the particle concentration, with only average
concentrations determined. On the other hand, if the
particle diameter is much smaller than the probe dia-
meter (type II), the reected light at any instant comes
from many particles.
Type II systems are most common. For a type II
system, penetration of light and the size of the measur-
ing volume depend on the local volumetric solids con-
centration (Lischer and Louge, 1992). Zhou (1995)
found that penetration length was 7 mm for c
approaching unity, decreasing to 4 mm for a
c = 0.6, and even less near the wall. Reh and Li
(1991) and Krol and de Lasa (1998) reduced the mea-
surement volume by having two ber bundles conver-
ging at an angle of 3035
, while Tanner et al. (1994)

used a small lens to achieve convergence with a single
ber. For very low solids holdups (0.2 to 0.7%),
Sobocinski et al. (1995) used a probe with a rounded
(hemispherical) tip.
The accuracy of the measurements relies also on the
precision of the calibration. Matsuno et al. (1983) cali-
brated their probe in a stream of glass beads falling
from a vibrating sieve, assuming the velocity to equal
the single-particle terminal falling velocity. This
method is limited to dilute systems. Matsuno et al.
(1983) allowed particles to fall through a stagnant
liquid, assuming them to attain their single particle
terminal velocity. Hindered settling likely inuenced
the calibration. Zhang et al. (1991) used the cross-sec-
tional average voidage inferred from static pressure
drop measurements. Hartge et al. (1986a) obtained a
linear relationship for quartz particles in a water-ui-
dized bed for c
s
_ 0.5; they then assumed linearity and
calibrated their probe by comparing the optical signals
to two concentrations obtained by ,-ray absorption.
Linearity has also been assumed by others (e.g., Kato
et al., 1991; Zhou et al., 1994). Werther et al. (1993),
using solidsliquid suspensions with FCC and sand
particles, suggested an empirical t of the form I =
k(1 c)
n
and found that n was independent of the uid
properties. Nieuwland et al. (1996) obtained similar
results. Qin and Liu (1982) derived a power law rela-
tionship between the light intensity and the solids
holdup for uniform spheres by assuming that the
light is reected from the rst two layers of particles
only. Lischer and Louge (1992) found a similar power
law from in situ calibration using a capacitance probe.
A simulation of the output of a single ber immersed
in a suspension of spheres showed the output to be
inuenced by the particle size and the refractive
index of the suspending medium. Therefore calibration
with liquids may cause errors. Nonlinear calibration
curves were also obtained by Herbert et al. (1994)
and Issangya (1998), who placed an optical ber
ush with the inner wall of a tube and recorded the
signal as FCC particles fell uniformly.
An iterative procedure for obtaining the true (non-
linear) calibration curve based on a particle dropping
test was proposed by Zhang et al. (1998). Nonlinearity
and errors can be caused by particles traveling through
a blind region at the tip of the probe. This can be
avoided by adding a thin transparent window (Cui et
al., 2001; Liu, 2001).
Capacitance Probes are based on the principle that
solids inside a sensing volume alter the effective di-
electric constant (permittivity) of the gas/solids mix-
ture. Since uidized bed particles and gas are
typically nonconducting, capacitance is usually easier
to measure than electrical resistance or conductivity.
Various probe congurations have been used, ranging
from parallel plates to needle shapes. The instrumen-
tation is simple, inexpensive, and easy to construct.
However, calibration is required for each bed materi-
al. Problems include signal drift, e.g. that due to hu-
midity or temperature variations, powder buildup on
the probe, and difculty in delineating the exact mea-
suring volume. Brereton and Grace (1993) found the
relationship between the voidage and the signal inten-
sity of a needle probe to be nearly linear. Hage and
Werther (1997) applied a water-cooled guarded capa-
citance probe inside a high-temperature CFB com-
bustor.
X-Ray and Gamma Ray Densitometers have a
radioactive source that emits a beam of x- or
gamma- rays across the column through its wall
to a detector. Photons produce charge pulses (ioni-
zation type detector), or light pulses (scintillation
type detector) which are then transformed into DC
signals. Film exposures can also be taken to give
snapshots of the solids distribution (e.g., Weinstein
et al., 1986). With calibration, the concentration of
particles along a chord can then be determined
from the beam attenuation. Weimer et al. (1985)
noted that the technique is too slow for time-de-
pendent measurements in uidized suspensions. The
technique gives length-averaged time-averaged den-
sities, rather than localized instantaneous measure-
ments.
Tomographic images and 3-D density maps can
be obtained if the radiation source and detector
are moved or if several sources and detectors are
employed simultaneously to scan the riser along a
number of chords (e.g., Galtier et al., 1989; Mar-
tin et al., 1992; Simons et al., 1993). Other studies
have been intrusive (e.g., Schuurmans, 1980; Wirth
et al., 1991), with the radiation source traversing
the riser and an external detector. X- and gamma-
ray measurements require a large space around the
riser and precautions to avoid radiation exposure.
In addition, the equipment is expensive. An alter-
native technique is capacitance tomography (e.g.,
Malcus and Pugsley, 2001). A system consists of
capacitance sensors, a data collection system, and
an image reconstruction computer. The capacitance
values measured between all pairs of electrodes are
fed to a computer, and a cross-sectional image of
the component distribution is reconstructed using a
suitable algorithm.
4.1.2 Axial Voidage/Solids Holdup Proles
Capacitance probes and optical ber probes can also
be traversed through risers to obtain local voidages,
then Eqs. (8) and (10) are applied to calculate time
and spatial averages. However, in practice the great
majority of measurements of spatial average voidages
have been derived from dierential pressure measure-
ments. If friction at the wall and acceleration eects are
neglected, the vertical gradient of pressure in the riser
acts like a hydrostatic pressure gradient, so that, for
gassolid systems with ,
p
,
g
, the average voidage
and solids holdup over the section in question can
then be estimated as
dP
dz
= g ,
p
(1 c
av
(z)) ,
g
c
av
(z)
_ _
(12)
c
av
(z) = 1
1
g,
,
dP
dz
or c
sav
=
1
g,
,
dP
dz
(13)
By comparing values from this approach with those
obtained by other methods, e.g., quick-closing valves
(Arena et al., 1986), ,-ray absorption (Hartge et al.,
1986a; Azzi et al., 1991), or x-ray computed tomog-
raphy (Grassler and Wirth, 1999), this approach has
been shown to give reasonably accurate measurements.
However, corrections may be needed in some cases:
The neglect of wall friction is reasonable for
large columns, but it can lead to signicant devia-
tions for D (or equivalent diameter
[= 4 Area,Perimeter] for noncircular cross
sections) < 0.1 m.
The neglect of particle acceleration eects is gener-
ally reasonable when the net solids circulation
ux, G
s
, is less than about 200 kg/m
2
s.
However, signicant errors are introduced at
the bottom, and, to a lesser extent, at the top
of the column, when the acceleration term is
ignored at higher G
s
. Approaches proposed by
Weinstein and Li (1989) and Louge and Chang
(1990) can be used to estimate the corrections
when accelerational eects are appreciable.
In order to apply Eq. (13), pressure taps should be
provided at regular intervals covering the entire height
of the riser, beginning just above the gas distributor at
the bottom of the riser, with the distance between suc-
cessive pressure taps small enough that the pressure
gradient term, dP,dz, can be estimated with little
error. The resulting axial prole of voidage (or, alter-
natively, solids holdup or pressure prole, from which
c
av
and c
sav
can be readily derived) then provides a
signature for the riser ow, indicating how the sus-
pension density and solids holdup vary with height.
Some representative studies reporting measurements
of this nature are listed in Table 2. Most reported
measurements have been from small units with air at
atmospheric temperature and pressure as the uidizing
medium, but they have covered broad ranges of parti-
cle properties, U
g
, and G
s
.
Typical proles showing how the solids holdup var-
ies with the net solids circulation rate for constant
supercial gas velocity in a column with a smooth
top exit appear in Fig. 7. Similar proles are seen for
very dierent particles (e.g., Jiang et al., 1994). The
Table 2 Representative Data Showing Axial Proles of Cross-Sectional Mean Voidage or Particle Concentration
Author Particles

dd
p
, jm U, m/s G
s
, kg,m
2
s D, m H, m
Exit
geometry Other eects studied
Li & Kwauk, 1980 iron, alumina,
FCC, pyrite
54105 0.85.6 16135 0.09 8 smooth
Rhodes & Geldart, 1986 alumina 64 4 45115 0.152 6 abrupt
Bader et al., 1988 catalyst 76 4.3, 9.1 147 0.305 12.2 smooth
Mori et al., 1992 FCC 54 1.01.5 2.429 0.050.1 1.55.5 various column geometry
Bai et al., 1994 FCC, sand,
alumina
70, 321,
633
1.53.0 7128 0.097,
0.15
3.0 smooth binary mixtures
Brereton & Grace, 1994 sand 148 3.79.2 989 0.152 9.3 various exit geometry, secondary air
Karri & Knowlton, 1997 catalyst 76 3.08.1 98586 0.3 13 smooth pressure
Mastellone & Arena, 1999a glass, sand,
FCC
67310 36 35144 0.12 5.75 smooth particle properties
Issangya et al., 1999 FCC 70 48 18425 0.076 6.1 smooth
solids holdup is highest near the bottom of the riser
and decreases monotonically with height right to the
top when there is no constriction or sharp bend at the
exit. The holdup is seen to increase at all heights with
increasing G
s
at constant U
g
. Except at very low G
s
or
very high U
g
, where particles tend to be carried right
through the riser in pneumatic conveying, the curves
tend to be sigmoidal in shape, approaching an asymp-
totic value at high heights. This shape, rst shown by
Li and Kwauk (1980), was once considered to be uni-
versal. Many risers, however, have constricted top exits
that aect the solids concentration proles signi-
cantly. An abrupt or constricted exit acts as an internal
separator, with the particles in the CFB riser being
large and heavy enough that they do not readily follow
the gas as its path varies in direction while negotiating
the exit. A signicant fraction of the particles arriving
at the top (typically 2080% depending on the exit
shape, U
g
, and G
s
) are therefore internally separated
from the gas and sent back down the riser, primarily
along the outer walls. With constricted exits, the con-
gregation of particles at the top causes an increase in
particle concentration near the exit (e.g., Jin et al.,
1988; Brereton and Grace, 1994; Pugsley et al., 1997).
The inuence of a constricted exit is shown schemati-
cally in Fig. 8. Note that as U
g
and G
s
increase in this
column with a constricted exit, the solids holdup
increases throughout, with a particularly large buildup
occurring near the top exit. The inuence on the shape
of voidage proles of a wide range of variables, includ-
Figure 7 Axial proles of solids holdup for FCC particles in riser of diameter 0.1 m and height 16 m with a smooth uncon-
stricted exit. d
p
= 67 jm, ,
p
= 1500 kg,m
3
, U
g
= 3.0 m,s. (Huang and Zhu, 2001.)
Figure 8 Axial proles of solids holdup in riser of diameter
152 mm and height 9.3 m with a constricted exit.
d
p
= 148 jm, ,
p
= 2650 kg,m
3
, air as uidizing gas.
(Brereton, 1987.)
ing gas velocity, net solids circulation ux, particle
diameter, particle density, overall riser height, inlet
and outlet geometry, and total inventory of solids,
has been well summarized by Bai et al. (1992a).
Voidage proles can also be inuenced by other
factors. For example, addition of secondary gas part
way up the riser tends to disrupt the downow of solids
along the outer wall, causing the particle concentration
to increase locally (e.g., Brereton and Grace, 1994).
Similarily, ring baes along the outer wall of the
riser cause particles to be stripped from the descending
wall layer, leading to local increases in particle concen-
tration (Jiang et al., 1991; Balasubramanian and
Srinivasakannan, 1998; Bu and Zhu, 1999).
Where it is necessary to estimate these proles in
advance or where pressure taps have not been pro-
vided, several methods are available for estimating
c
av
(z) or c
sav
(z). Bai and Kato (1999) provide excellent
summary tables showing previous correlations and
sources of data. They then provide useful new correla-
tions based on two cases, one for net circulation uxes,
G
s
, less than, and the other for them greater than, the
saturation carrying capacity ux, G
+
s
, correlated by
G
+
s
d
p
j
= 0.125Fr
1.85
d
Ar
0.63
,
p
,
g
,
g
(14)
where 4.7 - Ar = Archimedes number = ,
g
(,
p
,
g
)
gd
3
p
,j
2
g
- 1020, 41 - Fr
d
= Froude number = U
g
,
(gd
p
)
0.5
- 226, and 607 - density ratio = (,
p
,
g
),
,
g
- 3610. The solids holdups c
sd
in the bottom
dense zone and c
+
s
at the top exit were correlated by
(i) For G
s
- G
+
s
, proles are relatively at with
c
sd
c
/
s
= 1 6.14 10
3
,
p
U
g
G
s
_ _
0.23
,
p
,
g
,
g
_ _
1.21
U
g
gD
_
_ _
0.383
(15)
c
+
s
= 4.04(c
/
s
)
1.214
(16)
(ii) For G
s
_ G
+
s
, there are S-shaped proles with the
limiting holdups given by
c
sd
c
/
s
= 1 0.103
,
p
U
g
G
s
_ _
1.13
,
p
,
g
,
g
_ _
0.013
(17)
c
+
s
c
/
s
= 1 0.208
,
p
U
g
G
s
_
,
p
,
g
,
g
_ _
0.082
(18)
where c
/
s
= G
s
,[,
p
(U
g
u
t
)] is the solids holdup
expected for the ideal case where all particles travel
with a velocity equal to U
g
minus the terminal setting
velocity, u
t
. Huang and Zhu (2001) found that the
distance required to achieve fully developed voidage
or solids holdup proles in a riser of small D increases
with increasing G
s
and decreases with increasing U
g
.
Models that predict axial voidage proles in addition
to other hydrodynamic properties are covered in Sec.
4.6 below.
4.1.3 Radial/Lateral Voidage/Solids Holdup Proles
Measurements based on a variety of probes (e.g., capa-
citance and optical), as well as x-rays, ,-rays, and
tomographic techniques, have shown that particle con-
centrations near the wall tend to be much greater than
in the interior of CFB risers. Table 3 summarizes some
major sources of experimental data. An alternative
tabulation, also listing the experimental techniques
employed, was provided by Xu et al. (1999). Typical
results appear in Fig. 9. Note that in this gure and
some other measured proles, the shape is like the
letter W (or an M, if the voidage, rather than the solids
holdup, is plotted). There is slight asymmetry in this
particular prole, due to the side top exit (as discussed
below). Except in the bottom zone (Rhodes et al.,
1998) or at very high uxes where a dierent ow
Table 3 Representative Data Showing Radial Proles of Local Voidage in CFB Units
Author Particles

dd
p
, jm U
g
, m/s G
s
, kg,m
2
s D, m H, m z, m
Hartge et al., 1986b sand 56 3.9 100 0.050 3.3 0.62.2
Weinstein et al., 1986 sand 56 5.0 75 0.40 7.8 0.656.0
Tung et al., 1988 FCC 54 1.43.7 591 0.09 10 2.25
Bader et al., 1988 catalyst 76 3.7 98 0.305 12.2 4.0, 9.1
Harris & Davidson, 1994 FCC 60 4.4 2652 0.140 5.1 4
Jiang et al., 1994 polymer 325 2.4 721 0.102 6.3 0.6
Tanner et al., 1994 glass 110 2.56.5 30107 0.411 8.5 36
Wei et al., 1998 FCC 54 2.34.6 40180 0.186 8 2.36.3
Grassler & Wirth, 1999 glass 60 28 30600 0.19 15 4.411.6
regime can be achieved (see Secs. 3 and 5), particles
travel predominantly downward near the wall. The
observed pattern of particles being carried upwards
in the core of the riser and descending at the outer
wall has given rise to coreannulus models.
Nevertheless, the variation of local time-mean voidage
with radial position is continuous, not subject to a step
change, as might be implied by the coreannulus
designation.
Many groups (e.g., Zhang et al., 1991; Rhodes et al.,
1992; Beaud et al., 1996) have found that time mean
radial voidage proles are broadly self-similar. Hence
the radial variation can be expressed, at least to a rst
approximation, as functions of the cross-sectional
average voidage, which in turn can be estimated
using the experimental methods or correlations dis-
cussed above or the predictive methods covered in
Sec. 4.6 below. Of the various correlations, one that
gives good predictions over a very broad range of sus-
pension densities and net solids uxes (up to 425 kg/
m
2
s), i.e., for dense suspension upow as well as fast
uidization, is that of Issangya et al. (2001):
cc(r. z) = c
mf
c
av
(z) c
mf
[ ]c
1.52.1
3.1
5.0
8.8
av
(19)
where = r,R is the dimensionless radius and R is the
column radius. Several alternative forms of correlation
(e.g., Zhang et al., 1991; Godfroy et al., 1999b) have
also been suggested. Xu et al. (1999) gave separate
correlations for the time-mean voidage at the wall,
c
R
(z) = c
mf
4(1 c
mf
) c
av
(z) 0.75 [ ] (20)
and on the axis of the riser,
c
0
(z) = 1 0.42 1 c
av
(z) [ ] (21)
Both of these equations are only intended for
c
av
(z) _ 0.75. Results of Schuurmans (1980) conrm
that patterns of radial variation are similar (maximum
at center line, minimum at wall) in industrial equip-
ment as in the smaller units in which most experimen-
tal measurements have been obtained.
Voidage uctuations have been recorded in several
studies. While early models often assumed that the
clusters at the wall have voidages of c
mf
, experimental
ndings show a wide variety of wall voidages. Brereton
and Grace (1993) introduced an intermittency index,
dened as the ratio of the standard deviation of voi-
dage at a given point divided by that which would be
obtained if a ow having the same time mean voidage
was achieved by alternating voids and clusters of voi-
dage c
mf
. This index typically has values between 0.1
and 0.6 (Brereton and Grace, 1993; Johnsson et al.,
1997). It tends to decrease with increasing height, indi-
cating development of a coreannulus structure with
increasing height. Radially it reaches a maximum at a
small distance from the outer wall.
4.1.4 Angular Variations and Asymmetries
Flows in CFB risers of circular cross section are
usually assumed to be axisymmetric. However, particle
feeding and recirculation near the bottom of the riser,
as well as particle eux from near the top, are gener-
ally assymmetric, leading to angular variations, espe-
cially near the top and bottom of the riser. Some data
showing asymmetric distributions and/or angular var-
iations have been reported (e.g., Rhodes and
Laussmann, 1992b; Martin et al., 1992; van der Meer
et al., 1997; Srivastawa et al., 1998). The presence of
the exit on one side leads to signicant asymmetry of
particle voidage (Zhou et al., 1994), particle velocity
(Zhou et al., 1995a) and solids ux (van der Meer et al.,
1997; Brobecker et al., 2001) near the top of the risers
owing to the lateral motion of gas and particles as they
approach the exit. Similarly reentry of circulated solids
from one side can lead to marked asymmetry at the
bottom (Sun et al., 1999).
4.2 Particle Velocities
4.2.1 Measurement Techniques
It is more dicult to measure local particle velocities
than their local concentrations. Photographic and video
techniques can provide information on particle velo-
Figure 9 Lateral prole of local time mean solids holdup
along a midplane of 146 mm146 mm, 9.14 m tall riser at z
= 7.06 m for U
g
= 5.5 m/s, G
s
= 40 kg/m
2
s. (Zhou et al.,
1995a.)
city, direction of motion, and acceleration. However,
small, fast-moving particles are dicult to photograph.
Backlighting can be used for contrast in two-dimen-
sional risers (Arena et al., 1989) as well as circular or
square columns (Rhodes et al., 1991a). Photography is
dicult in regions where particle concentrations are
high, e.g., the bottom region of CFB risers. Zheng et
al. (1992) employed microcomputer-controlled multi-
color stroboscopic photography to determine particle
velocity, particle acceleration, and directions of motion
adjacent to the wall. Boroscope optical ber probes
have been applied (Takeuchi and Hirama, 1991; Li et
al., 1991) to study the ow structure in the central
region of the riser. However, the viewing area tends
to be small. Wider three-dimensional images have
been obtained using a laser sheet technique (Kuroki
and Horio, 1994), but this is only suitable for dilute
suspensions.
When two optical probes are separated by a small
distance L, the cross-correlation function of the two
signals f
1
and f
2
registered by probes 1 and 2 over time
t
tot
is given by
12
(t) =
1
t
tot
_
tot
0
f
1
(t) f
2
(t t) dt (22)
The mean particle velocity is estimated as
u
p
= L,t
max
, where t
max
is the delay time at which
the
12
(t) reaches a maximum. To determine the velo-
city, one also needs the eective separation distance
between the two probes. This generally diers from
the actual separation distance and needs to be obtained
via calibration. Dierent probes have been used to
measure local time mean particle velocities in CFB
risers. Hartge et al. (1988) used a pair of 1 mm dia-
meter bers separated by 4.4 mm. Horio et al. (1988,
1992) employed a probe with two detection bers and a
single emission ber in between. To capture individual
particle motion, Ohki and Shirai (1976) recommended
that the cross-correlation is improved if the diameter of
the optical ber is the same as the particle diameter,
with the separation distance between the receiving
bers of similar magnitude.
The delay time or time interval, t, can also be
obtained using peak detectors (Zhou et al., 1995a).
To avoid false signals, a ve-ber probe was used
by Zhou et al. (1995a). The velocity was accepted
only if the dierence between two independent
values was within a certain tolerance. This probe
eliminates errors when light received by the two
photomultipliers is from dierent particles, but
only measures velocities of particles traveling verti-
cally, rather than the vertical component of all par-
ticle velocities.
In laser Doppler anemometry, a laser beam is split
into two beams of equal intensity, which are then
focused at the measuring point, forming a fringe pat-
tern in the intersection region. When a particle passes
normal to the fringes, the intensity of its reected light
varies with a frequency proportional to the particle
velocity and inversely proportional to the fringe spa-
cing. Both forward (Arastoopour and Yang, 1992) and
backward (Yang et al., 1992) scattering can be used,
but forward scattering is only suitable under very
dilute ow conditions (e.g., c
s
- 2%), whereas back-
scattering provides data at the wall, or one can use an
intrusive probe (Wang et al., 1993; Yang et al., 1992).
Particle velocities can also be detected by tracking
the trajectory of a single tagged particle over extended
time (Godfroy et al., 1999a). However, it is currently
very dicult to track particles smaller than about 0.5
mm in diameter, so that the technique has not yet
contributed signicantly to CFB research.
4.2.2 Experimental Results
Some major studies that have reported data related to
particle velocities in CFB risers are summarized in
Table 4. Time mean particle velocities in fast uidized
beds are upwards in the core of the column, with a
magnitude similar to the supercial gas velocity at the
column axis. The average velocity then falls as the sen-
sor moves outwards toward the wall, becoming nega-
tive in a layer adjacent to the wall. The downward
particle velocity right at the wall is generally of the
order of 1 m/s. Available data for the velocity at the
wall have been correlated (Grith and Louge, 1998) by
v
pw
= 36
gd
p
_
(23)
Such coreannulus patterns have been observed fre-
quently in small-scale risers, but also in much larger
equipment, even in a 14.7 11.5 m cross section com-
bustor operated by Electricite de France (Caloz et al.,
1999).
Typical lateral proles of time mean local particle
velocity, both upward and downward, along a mid-
plane of a column of square cross section are shown
in Fig. 10. The supercial gas velocity is seen to have
little inuence. Ascending particles are dominant in the
center of the column, whereas there are more descend-
ing than ascending particles near the wall (i.e., as y,Y
approaches 1). The magnitudes of the velocities of
rising particles at the axis of the column are similar
to the supercial gas velocity, while the magnitudes of
downward velocities are signicantly lower.
There are substantial velocity uctuations, espe-
cially near the wall, with the time mean uctuating
component typically reaching a maximum near the
edge of the wall layer (Yang et al., 1992; Caloz et al.,
1999). Flows near the wall undergo reversals, i.e., the
particle velocity there is sometimes upward and some-
times downward. The fraction of individual particles
that are ascending is close to 100% toward the axis of
the column, while near the wall, typically 1040% of
particles travel upward, with the rest descending (Zhou
et al., 1995a). For rectangular columns, the magnitude
of the downwards mean particle velocity and the frac-
tion of particles that are descending when averaged
over time are greater in the corners than elsewhere
along the outer wall.
Zhou et al. (1995b) inferred horizontal components
of velocity for particles crossing the axis of a riser of
square cross section. The measured lateral components
increased to a maximum of 40% of the supercial
gas velocity, U
g
, in the middle of the riser and then
decreased toward the top.
4.3 Particle Flux
4.3.1 Net Solids Circulation Flux
Most methods of determining the net solids circulation
ux, G
s
, in CFB systems assume that the cyclones are
100% ecient so that the solids ow can be inferred
from the downward ow in the standpipe/downcomer.
If plug ow is assumed, the solids ux in the down-
comer can be estimated by timing the descent of iden-
tiable particles along the downcomer wall (Burkell et
al., 1988). If the standpipe wall is not perfectly smooth,
this method underestimates the solids ux because par-
ticles at the column surface are retarded by friction and
travel more slowly than those in the interior. Another
Table 4 Representative Data Showing Radial Proles of Particle Vertical Flux or Velocity
Author Particles

dd
p
jm U
g
, m/s G
s
, kg,m
2
s D, m H, m z, m
Monceaux et al., 1986 FCC 59 3.2 3168 0.144 8 4.8
Bader et al., 1988 catalyst 76 4.6 147 0.305 12.2 9.1
Hartge et al., 1988 FCC, ash 85, 120 2.9, 3.7 49, 30 0.4 8.4 0.94.7
Rhodes et al., 1989 alumina 70 34 3040 0.152 6 1.74.9
Azzi et al., 1991 FCC 75 6.2 150 0.190.95 11.7 3.116
Herb et al., 1992 FCC, sand 68, 276 2.46 2044 0.05, 0.15 2.7, 10.8 1.5, 2.1, 5.5
Yang et al., 1992 FCC 54 1.56.5 625 0.140 11 3.3, 6.6
deDiego et al., 1995 sand, coal 400650 45.5 2898 0.10 4 2.13.2
Wang et al., 1996
+
sand 120 2.5
+
up to 26 0.161 6.2 1.3, 4.4
Wei et al., 1997 FCC 54 1.910.1 37236 0.186 8.5 1.8, 3.9
Nicolai & Reh, 1997 glass 40300 1.59.6 NA 0.41 8.5 4.04.2
Corenella & Deng, 1998 sand 200 3.45.4 1030 0.114 2.7 0.7, 2.5
Issangya et al., 1998 FCC 70 4.57.5 38325 0.076 6.1 2.8
Mastellone & Arena,
1999a, 1999b
FCC, glass,
sand
70310 6 35, 55 0.120 5.75 1.24.2
Caloz et al., 1999
#
glass 62 for 0.41
083 m units
1.73.3,
5.5
1798 0.41, 0.83,
14.7 11.5
8.5, 11.2,
35
2.621.5
+
Includes measurements up to 550
C.
#
Includes measurements under full combustion conditions.
Figure 10 Lateral particle velocity proles based on optical
ber probe for midplane of 146 mm square column at z = 6.2
m at G
s
= 40 kg/m
2
s. (Zhou et al. 1995a.)
popular method is to close a perforated buttery valve
in the downcomer, with small holes to allow gas to
pass. The solids circulation rate is then calculated
from the time to accumulate a known volume of solids
(Herb et al., 1992), from the time to reach a given
pressure drop across the bed that collects above the
valve (Brereton, 1987; Li et al., 1991), or from the
rate of descent of the moving packed bed level in the
downcomer below the valve (Weinstein et al., 1986).
Alternatively, the solids ow can be diverted to a mea-
suring cylinder for a given period of time before being
switched back to the downcomer (Bai et al., 1987). A
drawback of these methods is that they disturb the
pressure balance in the system when particles are tem-
porarily diverted.
Patience and Chaouki (1991) correlated the solids
ux to the pressure drop over the horizontal section of
the exit pipe connecting the riser to the separator at a
given gas velocity. Ultrasound can also be employed in
the downcomer to measure the solids circulation rate
(Tallon and Davies, 1998).
None of the above methods, with the possible
exception of the exit pressure drop method (where cali-
bration may be impossible), is likely to be feasible for
large-scale commercial CFB systems. Hence industrial
CFB systems usually operate without G
s
being known.
4.3.2 Local Vertical Flux Measurement Techniques
Time mean local solids uxes in CFB risers have often
been measured using solids sampling probes. Ideally,
the suction velocity through the sampling tube matches
the local gas velocity at the sampling location to pro-
vide isokinetic sampling. However, higher suction velo-
cities may be required to prevent blockage. Therefore
nonisokinetic sampling probes have often been
employed. Monceaux et al. (1986) found that the solids
collected rst increased with suction velocity and then
stayed constant over a wide range and nally increased
again. Bader et al. (1988) and Werther et al. (1993)
observed no signicant inuence of suction gas velo-
city. Azzi et al. (1991) found that the particle ux
remained nearly constant if the suction velocity was
within 1.5 m/s of U
g
. Suction velocities equal to the
supercial gas velocity were reported (Aguillon et al.,
1996) to give satisfactory results. Herb et al. (1992)
observed that measured solids uxes in the core region
were not strongly dependent on the suction velocity
near the isokinetic point. Rhodes (1990) and Miller
and Gidaspow (1992) found that although the upward
and downward mass uxes were sensitive to the
sampling gas velocity, the net solids mass ux
(obtained by dierence) was virtually independent of
the suction velocity, provided that it was sucient to
prevent blockage. However, suction velocity has a sig-
nicant eect when sampling in the wall region. To
obtain the true ux near the wall, the suction velocity
should be just enough to keep the probe free of block-
age (Bierl et al., 1980; Harris et al., 1994). The true ux
can be obtained by extrapolating the solids ux to zero
suction velocity (Bierl et al., 1980). de Diego et al.
(1995), however, found that both the downward and
upward solids uxes in the dense outer region of the
riser were greatly aected by the suction velocity. A
sensitivity study by Herb et al. (1992) showed that
the downward component of the ux was sensitive to
the suction velocity near the wall. Liu (2001) found
that the sampling method gives reasonable results
when the local ow is predominantly in one direction,
but signicant errors for locally reversing ows. To
overcome the sampling problem at the wall, Bolton
and Davidson (1988) used 5 mm ID scoops to col-
lect particles falling near the wall. Water-cooled sam-
pling probes can be used in high temperature CFB
risers to obtain local time mean solids uxes
(Couturier et al., 1991; Johnsson et al., 1997).
Liu (2001) constructed and calibrated an optical
ber probe that can simultaneously measure local par-
ticle holdup and local particle velocity. From these pro-
ducts, multiplied by the particle density, one obtains the
local instantaneous particle ux. Integration of the
local time average uxes across the cross section led
to overall net uxes that were in good agreement with
G
s
, the net ux determined independently by trapping
recirculating solids in the return loop. The optical ber
technique conrmed that the uctuations of particle
velocity and voidage are strongly correlated, as noted
in the next section, so that it is incorrect to assume that
one can obtain the time mean ux from the product of
particle density, local time mean velocity, and local
time mean solids holdup.
4.3.3 Radial Proles of Solids Flux
Studies that report local particle ux data are summar-
ized in Table 4. For the fast uidization ow regime,
radial proles of local particle ux are generally similar
in shape to the radial proles of particle velocity
described aboveupward near the center line, drop-
ping o with increasing r, and negative near the wall.
One such experimental prole, shown in Figure 11, is
seen to be symmetrical around the axis of the column.
Note that the ux is downward in this case only in a
thin region adjacent to the outer wall of the column.
Radial proles of upward particle momentum tend to
be similar in shape to the solids ux proles (e.g., Azzi
et al., 1991).
4.3.4 Wall or Annular Layer Thickness
The concept of a coreannulus structure has led many
groups to report the thickness of the outer annular wall
layer. This thickness is then used in coreannulus reac-
tor models (Sec. 9 below) and in heat transfer models
(Sec. 7). Wall layer thicknesses have been based on
radial proles of either particle velocity or solids ux,
with the radial position at which the time mean value is
0 taken to dene the boundary. However, as shown by
Bi et al. (1996) and Fig. 12, the location where the time
mean velocity is 0 diers from that where the time
mean ux is 0. This arises because uctuations of
local instantaneous voidage are strongly correlated
with uctuations of local instantaneous particle velo-
city. The most meaningful wall layer thickness is based
on the point at which the time mean particle ux is 0.
This thickness is well correlated (Bi et al., 1996) by
o(z)
D
= 0.5 1
1.34 1.30 1 c
av
(z) [ ]
0.2
1 c
av
(z) [ ]
1.4
_ _ _
0.80 _ c
av
_ 0.9985
(24)
If the column cross section is square or rectangular
rather than circular, the wall layer tends to be distorted
as shown in Fig. 13 (Zhou et al., 1995a), with a greater
thickness in the corners.
The annular wall layer thickness may shrink or
grow with height. With a constricted exit, the thick-
ness passes through a minimum part way up the col-
umn (as in Fig. 12), indicating a net transfer of solids
inward from the wall region to the core in the upper
part of the riser, while there is a net outward transfer
in the lower part of the riser as solids descend along
the outer wall.
Several groups (e.g., Rhodes et al., 1992; Horio and
Kuroki, 1994; Wei et al., 1995a; Lim et al., 1997) have
studied the sizes, shapes, motion, and breakup of indi-
vidual clusters and streamers. Streamers tend to be U-
Figure 11 Radial prole of solids ux determined by
Bodelin et al. (1994) for U
g
= 5.4 m/s, G
s
= 43.4 kg/m
2
s, D =
144 mm, H = 10 m, d
p
= 180 jm, ,
p
= 2650 kg/m
3
com-
pared with predictions of Pugsley and Berruti (1996).
Figure 12 Wall layer thickness based on both where the
time mean solids ux is 0 and where the time mean particle
velocity is 0 versus height along the mid-plane of 146 mm
square cross-section column. U
g
= 5.5 m/s, G
s
= 40 kg/m
2
s,
d
p
= 213 jm, ,
p
= 2640 kg/m
3
.
Figure 13 Distorted wall layer boundaries at two heights for
riser of square 146 mm146 mm9.14 m tall column based
on particle velocity proles obtained by Zhou et al. (1995a)
for U
g
= 5.5 m/s and G
s
= 40 kg/m
2
s. Solid line: z = 5.1 m;
broken line: z = 6.2 m.
shaped, as shown in Fig. 14, and to become thicker as
they descend until they become unstable owing to the
stripping of particles by the rapid inner core ow.
Idealized streamer shapes assumed by Lim et al.
(1997) in modeling streamers are also indicated in the
gure.
4.3.5 Lateral Flux
Horizontal particle motion is caused by interparticle
collisions and gas turbulence. Unlike diusion pro-
cesses, the net transfer is from a more dilute region
(core) to a denser one (annulus) (Brereton and
Grace, 1993; Davidson, 2000). Particles transferred
outward toward the wall tend to be captured by the
descending wall layer, while inward transfer occurs as
particles are stripped from unstable streamers descend-
ing along the outer wall.
The lateral (horizontal) ux of particles has been
determined by sampling in a similar manner to the
vertical ux sampling method described above, except
that the opening of the sampling tube is now vertical,
directed toward the near wall to measure the inward
ux, or away to determine the outward ux. The net
cross-ow is the dierence between the inward and the
outward ux, and it should be zero if the ow is fully
developed. Net horizontal uxes have been found to be
one to two orders of magnitude less than G
s
(Zhou et
al., 1995b; Jiang and Fan, 1999). The inward and out-
ward uxes are relatively small in the center of the
riser, increase to a maximum near the wall, and
decrease again at the outside (Jiang and Fan, 1999).
The net horizontal ux has been found (Zhou et al.,
1995b) to be inward near the top of the riser and out-
ward in the lower part, consistent with the appearance
of a minimum wall layer thickness in the middle of the
riser, as in Sec. 4.3.4 above. There is some evidence
(Werther, 1994) that lateral uxes decrease as the reac-
tor scale increases.
4.4 Gas Velocities
Local gas velocities are very dicult to measure
because of the presence of the particles and high levels
of turbulence. Experimental techniques include isoki-
netic sampling (van Bruegel et al., 1969; Harris and
Davidson, 1992), injecting a tracer at one level and
then cross-correlating the signals from two nearby
downstream levels (Horio et al., 1992; Martin et al.,
1992), Pitot tubes (Yang et al., 1994), and laser
Doppler anemometry (Yang et al., 1992).
Experimental local time mean gas velocities tend to
increase monotonically as one traverses inward from
the wall of the column, reaching a maximum that is
typically 1.5 to 2 times the supercial gas velocity at
the column axis, with the proles being more nonuni-
form with increasing G
s
. Like particle uctuating velo-
cities, gas uctuating rms velocities are of order 1 m/s
and appear to reach a maximum at a small distance
inside the wall, falling to relatively low values near the
axis (Yang et al., 1992).
4.5 Geometric and Operating Variable Effects
The inuence of riser geometry on hydrodynamics was
reviewed by Grace (1997). Only a brief summary is
therefore given here.
4.5.1 Inuence of Cross Section Shape: Circular vs.
Rectangular
While most CFB hydrodynamic studies have been car-
ried out in risers of circular cross section, some have
been conducted in risers of rectangular (e.g. Saberi et
al., 1998) and square (e.g., Leckner et al., 1991; Zhou
et al., 1994) cross section, geometries that are especially
relevant to CFB combustors. Some work (e.g.,
Yerushalmi et al., 1978; Arena et al., 1992) has also
been performed in thin (two-dimensional) risers.
Phenomena in rectangular columns are qualitatively
similar to those in risers of circular cross section,
with the usual dilute upward moving core and dense
downward ow in an outer annular region.
Figure 14 Streamer shapes (a) as viewed at the wall of a
column of 146 mm square cross-section; (b) as idealized in
simple cluster model; (c) as idealized in plane normal to wall.
(Lim et al., 1997.)
Relationships developed for circular risers may then be
applied as a rst approximation, with the diameter, D,
replaced by an equivalent (hydraulic)
diameter = 4 cross sectional area perimeter.
However, the corners of rectangular units are very
prone to downow (Andersson and Leckner, 1992;
Zhou et al., 1995a; van der Meer et al., 1997), with
higher downward velocities and particle concentra-
tions there than elsewhere along the walls. Moreover,
the wall layer becomes distorted as shown in Fig. 13.
Fabre et al. (1997) found quite dierent mass ux pro-
les along two transverse horizontal sections when the
length was much greater than the width.
4.5.2 Inuence of Wall Shape and Roughness
Refractory walls in CFB reactors tend to be rough.
Studies with a roughened wall (Zhou et al., 1996a)
demonstrate that wall roughness plays a signicant
role close to the wall, the primary eect being to
increase the voidage there. Studies with membrane
walls (see Sec. 7 below) show (Leckner et al. 1991;
Zhou et al., 1996b) that particle streamers travel faster
and further in the protected troughs along the at ns
between adjacent tubes than along corresponding at
walls. Local voidages also tend to be lower along the
n than on the crests of the tubes.
4.5.3 Inuence of Bottom Conguration
Unlike low-velocity uidized beds, where the distribu-
tor is often regarded as the most critical element in
achieving a successful design, there is much greater
latitude when specifying the distributor and bottom
geometry for CFB risers. In fact, it is possible to oper-
ate without any distributor plate whatsoever, so long
as the solids can be blown over at the end of each
run to be stored in the return system (i.e., the standpipe
and storage vessel, if any). Nevertheless, the inlet con-
guration plays a signicant role, and it would be
wrong to consider the design of this region unimpor-
tant.
CFB systems have various congurations for recir-
culating and reinjecting solids. The feed rate may be
regulated by a mechanical (e.g., slide valve) or non-
mechanical valve (see Chapter 11). L-valves, H-valves,
L-valves, V-valves and loop seals are common in CFB
combustors, while J-valves and slide valves are popular
in uid catalytic cracking risers. Nonmechanical valves
tend to give good ow control for Geldart group B and
D particles, while the ow of aeratable group A parti-
cles usually needs to be regulated by a slide valve or
other mechanical valve. Some common bottom geome-
tries are illustrated in Fig. 15. Cheng et al. (1998) clas-
sify inlets according to the degree of restriction of the
solids circulation and the conguration of the open
area. A strong restriction at the entry helps stabilize
the ow, but adds to the system overall pressure drop
and hence to the operating cost. Weak restrictions
cause the riser to be directly coupled with the entire
return system and with the windbox (if any), whereas a
strong restriction (high distributor pressure drop)
allows the riser to operate somewhat independently.
Hartge and Werther (1998) provide useful information
on dierent nozzle types and their inuence on erosion.
To reduce erosion they recommend nozzles that have a
section where there is downward gas ow with a mini-
mum exit velocity.
The geometry and orientation of the entering gas
can directly aect axial voidage proles over a consid-
erable height beginning in the lower part of the riser
(Weinstein et al., 1996; Cheng et al., 1998; Kostazos et
al., 1998). The bottom zone may behave as a vigor-
ously bubbling bed or as a turbulent uidized bed. It
may be clearly demarcated from the region above,
merge smoothly into the upper part of the riser, or
show a transitional splash zone between a lower
dense bed and the riser above. Addition of secondary
gas hastens the transition from a denser bottom zone
to a more dilute upper region. For low-pressure drop
distributors, as in the Chalmers University CFB boiler,
surging may occur, with pressure oscillations in the
windbox directly aecting the lower zone of the riser,
being a source of turbulence (Sterne us et al., 1999) and
causing exploding bubbles (e.g., Johnsson et al.,
1992; Schouten et al., 1999). This eect can be cor-
Figure 15 Common CFB riser bottom congurations: (a)
with loop seal; (b) with L-valve; (c) with J-valve; (d) with
slide valve.
rected by increasing the distributor pressure drop, or
by installing horizontal tubes in the bottom zone
(Johnsson et al., 2001), thereby converting the bottom
region to the turbulent uidization ow regime.
The height of the bottom zone tends to increase
with increasing net solids circulation ux, G
s
, and to
decrease with increasing supercial gas velocity, U
g
(Schlichthaerle and Werther, 1999; Sun et al., 1999).
The overall solids holdup in the bottom zone is typi-
cally 0.2 to 0.3. The tendency for local solids holdups
to be greater near the wall than in the core begins from
the very bottom of the riser. Characteristic frequencies
tend to be lower in the entry zone than at higher eleva-
tions (Louge et al., 1990). If particles are already
ascending when they re-enter the riser, e.g., owing to
a J-valve, then the bottom zone tends to be less dense
and shorter than if the solids enter sideways (e.g., via
an L-valve) or obliquely downwards (e.g., via a loop
seal) into the lower section of the riser.
The hydrodynamics at the bottom of the riser can
also be inuenced by whether the lower section of the
column is tapered (cross-sectional area increasing with
height, Grace, 1997). Tapering leads to increased super-
cial gas velocity at the bottom that would normally be
expected to decrease the solids holdup, but this does not
always occur, probably because the tapering also
decreases the area available for entraining solids.
4.5.4 Addition of Secondary Gas
Secondary air is commonly injected as jets part way up
CFB combustion risers to assist in reducing NO
x
emis-
sions. The secondary air jets can strongly inuence
axial voidage proles, while the level at which they
are injected aects the depth of the dense region at
the bottom (e.g., Naruse et al., 1991; Brereton and
Grace, 1994; Cho et al., 1996). The jets disrupt the
downward ow of solids when they reach the level of
the secondary gas injection (Brereton and Grace, 1994;
Kim, 1999).
When the secondary air jets cause swirl (tangential
motion), the overall voidage can be reduced, creating a
more homogeneous bed (Wang and Gibbs, 1991). Jets
tend to isolate the region above them from that below,
reducing downward ux along the wall (Brereton and
Grace, 1994). Near the injection level, the solids
holdup tends to increase because of the centrifugal
eect of the swirl (Cho et al., 1996). Strong swirl was
found to inuence riser hydrodynamics for a distance
of about D upstream and 23 column diameters down-
stream of the level where the swirl gas was injected
(Ran et al., 1999).
4.5.5 Exit Effects
The ow at the top of a riser can be aected pro-
foundly by the geometry of the exit port through
which the gas and solids pass to the primary cyclone
or other separator. Some congurations employed in
practice are illustrated schematically in Fig. 16.
Important factors include
The asymmetry of the exit: An exit at one side tends
to cause asymmetry, as noted above.
The degree of constriction: A severe reduction in the
cross-sectional area leads to a higher exit pres-
sure drop, aecting the overall pressure balance,
also reducing the eux of solids from the riser
for a given gas owrate.
The supercial gas velocity: Exit eects are more
important as U
g
increases.
The tortuosity of the ow path: The exit acts like a
crude gassolid separator. Any increase in curva-
ture of the gas streamlines as the gas nds its way
into the exit makes it more dicult for the par-
ticles to remain entrained and increases internal
reuxing of solids in the riser.
For horizontal exits, solids may build up in the exit
duct and dump periodically back into the main
riser (van der Meer et al., 1997).
Given these factors, the exit design should be central
and smoothly tapered if one wishes to minimize back-
mixing of gas and solids and prevent reection from
the top. However, if the designer wishes to increase
solids holdup in the riser to provide maximum resi-
Figure 16 CFB riser top exit congurations: (a) sharp right-
angle; (b) exit at side; (c) tapered right angle; (d) long-radius
bend; (e) nozzle and low pressure drop impingement separa-
tor [gas exit not shown]; (f) top exit with constriction.
dence time, it is desirable to locate the exit on one side
and to make its area signicantly less than the cross-
sectional area of the riser.
Observed eects of exit geometry follow from the
above considerations. More constricted exits lead to
increased pressure gradients and higher solids holdup
in the top region (Jin et al., 1988; Brereton and Grace,
1994; Pugsley et al., 1997). In some cases (e.g.,
Brereton and Grace, 1994; Pugley et al., 1997), the
inuence of the exit may extend downward a consider-
able distance, whereas in other cases it appears to be
conned to the top of the riser. Empirical models for
predicting the inuence of the exit have been proposed
by Pugsley (1998) and Gupta and Berruti (2000).
4.5.6 Inuence of Temperature and Pressure
Many circulating uidized beds operate at elevated
temperatures and pressures, yet the vast majority of
experimental information has been derived from col-
umns operated at ambient conditions. Fang et al.
(1999) carried out experiments at high temperature,
but in a very small riser at modest gas velocities and
solids uxes, and found a small decrease in overall
pressure drop, due presumably to the decreased gas
density, although carbon dioxide was used at high tem-
perature and air at room temperature. Increasing tem-
perature leads to a reduction in gas turbulence because
of the increased kinematic viscosity of the gas. This is
likely to inuence particle motion and mixing for rela-
tively small particles (e.g., most catalysts), but less so
for the larger denser particles employed in CFB com-
bustors (Senior and Grace, 1998). Measurements
inside relatively large CFB combustors (e.g., Leckner
et al., 1991; Lin et al., 1999) demonstrate that key
hydrodynamic features at low temperature, like down-
owing wall layers and a dilute core, are also applic-
able at high temperatures.
A study of CFB hydrodynamics at pressures up to
690 kPa (Karri and Knowlton, 1997) showed that the
solids holdup tended to decrease with increasing sys-
tem pressure. This was attributed to decreased wall
layer thickness and increased voidage close to the
wall as the pressure increased. Wirth and Gruber
(1997) found that pressurization caused the solids to
be distributed more uniformly over the full height of a
CFB riser operated at pressures up to 50 bar.
4.5.7 Internals
Various internals have been tested in laboratory CFB
units. A venturi contraction section has been used in
some FCC units to promote particle acceleration and
mixing (Avidan, 1997a). Ring type and other similar
baes have been tested to improve gassolids contact-
ing. Some baes increase the solids holdup in the riser
(e.g., Jiang et al., 1991, van der Ham et al., 1993), while
others appear to reduce solids holdup (Gan et al.,
1990; Zheng et al., 1992).
4.5.8 Geometric Variants
Circulating uidized beds can involve multiple loops.
For example, in a Total resid cracking process with
two stages of catalyst regeneration (Dean et al.,
1982), the rst riser serves as the cracking reactor.
After stripping, catalyst particles ow downward to
the rst-stage regenerator, and then through a second
riser to the second-stage regenerator before returning
to the cracking riser. The rst regenerator acts as the
downcomer of the transport riser, while the second
serves as the downcomer for the cracking riser. A simi-
lar dual-loop circulating uidized bed has been tested
(Issangya et al., 1998; Liu, 2001) to investigate hydro-
dynamics and mixing at high solids uxes (up to 600
kg/m
2
s) and high solids holdup (see Sec. 5.2). Other
dual-loop CFB reactors include the EXXON uid cok-
ing process (Matsen, 1996), N-shaped circulating ui-
dized beds (Qi et al., 1992; Zheng et al., 1989), the
Circored iron reduction process (Plass, 2001), and
Fercos CFB gasication system (Weeks and Rohrer,
1997).
A circulating uidized bed can also be constructed
in a concentric conguration as an internally circulat-
ing uidized bed. The riser is usually located in the
interior with the downcomer as an outer annulus (e.g.,
Fusey el al., 1986). Such geometries simplify construc-
tion, facilitate direct heat exchange between the riser
and the downcomer, and simplify the design of high-
pressure systems.
Other variants of the circulating uidized bed
include the downer reactor (Zhu et al., 1995; Jin et
al., 2002), quick-contact (QC) reactor (Gartside,
1989), short-contact-time uidized bed reactor
(SCTFR) (Graham et al., 1991), and entrained ow
reactor (EFR) (Raghunathan et al., 1993). As shown
in Fig. 17, there is no uidized bed in a downer
reactor, with the gas traveling downward, cocurrent
with raining particles. The distinct features of the
downer reactor include short gassolids contact times
(- 1 s) and reduced gas and solids. Downer reactors
are thus promising for very fast reactions, with the
intermediate as the desired product, where plug ow
and short contact times are essential to prevent over-
reaction and ensure good selectivity.
4.6 Hydrodynamic Models
Models describing the hydrodynamics of CFB systems
are of three types: empirical, mechanistic, and based on
computational uid dynamics (CFD).
4.6.1 Empirical Correlations
Given the complexity of multiphase ows and the
number of contributing factors, empirical methods
are common. Several of these, e.g., the approach
adopted by Bai and Kato (1999) for predicting axial
voidage proles, are covered above. However, because
of the number of variables involved and their limited
ranges of variation in experimental studies, many cor-
relations are only useful for small columns and over
very limited ranges of conditions. More comprehensive
models are generally needed to describe a broader
range of variables over extended ranges of conditions.
4.6.2 Mechanistic Models
Mechanistic models are generally preferred to empiri-
cal approaches. Ideally they incorporate the key phy-
sical factors aecting the ow, leaving out factors of
lesser importance. Mechanistic models for CFB risers
are often useful, but many suer from weaknesses,
such as
It is frequently dicult to set the boundary condi-
tion at the bottom, i.e., at z = 0.
Most approaches assume a sigmoidal voidage pro-
le and/or perfect coreannular ow; as noted
above, these are oversimplied.
Axial symmetry is usually assumed, making it
impossible to deal adequately with asymmetries
such as side feed ports, side reentry of solids, and
asymmetric top exits.
The models do not readily allow for the introduc-
tion of secondary or tertiary air (or other gases).
However, mechanistic approaches provide rational
approaches for predicting key hydrodynamic features.
Two useful models will be outlined briey.
1. Coreannulus model of Senior and Brereton
(1992). This model assumes that above an entry region,
taken as the volume below any secondary air inlets,
particles fall as dense wall streamers of semielliptical
cross-sectional shape and voidage of 0.6 at a velocity
of 1.1 m/s; a uniform dilute suspension is conveyed
upwards in the interior. Conservation of mass and
volume are applied to the two phases. The relative velo-
city between the gas and the particles in the core is set
equal to the single particle terminal settling velocity, u
t
.
The minimum streamer thickness is taken to be 8 mm.
Exchange of particles from the streamers to the core is
assumed to occur in a manner analogous to the entrain-
ment of droplets in gasliquid annular ow. Exchange
from the core to the streamers is related to lateral par-
ticle motion and collisions, with a mass transfer coe-
cient of 0.2 m/s adopted as the best overall estimate.
The model adopts an empirically tted top reection
factor, a function of riser exit geometry and gas and
particle velocities, to represent the fraction of particles
reaching the top that are internally reected back down
the column. The cross-sectional area occupied by the
streamers and core varies with height.
With tted reection factors, this model does an
excellent job of predicting axial density proles in the
upper part of a range of dierent columns with second-
ary air addition (Senior and Brereton, 1992), not only
for a small-scale pilot plant CFB combustor but also
for a prototype boiler of cross-sectional area 0.43 m
2
operated at high temperature (845
C).
2. Coreannulus model of Pugsley and Berruti
(1996) and Gupta and Berruti (1998, 2000). This
model again assumes a coreannulus structure, a bot-
tom entrance region, and a more dilute upper region.
Flow in the upper region in this case is assumed to be
fully developed with average voidage
c
av
=
U
g
,
p
G
s
[ U
g
,
p
(25)
Figure 17 Typical downer geometry.
[ is a so-called slip factor correlated for smooth exits
and group A particles by
[ = 12.2Fr
2.01
t
7.65 Re
0.57
p
(26)
with Fr
t
= u
t
,(gD)
0.5
and Re
p
= ,
g
U
g
d
p
,j
g
. For
abrupt exits, this [ is multiplied by (1 j), where
j = 22.5N
1.49
Gs
N
1.36
Gst
N
1.32
He
N
4.06
o
(27)
with N
Gs
= G
s
,(,
p
U
g
), N
Gst
= G
s
,(,
p
u
t
), N
De
= D
e
,D,
N
He
= (H H
e
),H, and N
o
= (o 45
),45
. D
e
is the
exit duct diameter and H
e
the top extension height.
Angle o, shown in Fig. 18, is taken as 45
for a
smoothly curving exit and 90
or more for an abrupt

exit. For group B powders, Eq. (26) is replaced by a
correlation due to Patience et al. (1992):
[ = 1.0
5.6
Fr
0.47Fr
0.41
t
(28)
where Fr = U
g
,(gD)
0.5
, while Eq. (27) is replaced by
j = 0.00091N
2.43
Gs
N
2.02
Gst
N
1.54
De
N
2.80
He
N
o
(29)
Once the average voidage in the fully developed region
is estimated, the average upwards particle velocity in
the fully developed region is calculated as
v
p
=
G
s
,
p
(1 c
av
)
(30)
In the acceleration zone, Pugsley and Berruti (1996)
assumed a core/annulus structure, with the core radius
the same as in the fully developed region. A dierential
force balance was written for a single particle, with the
voidage in the annular region assumed to vary linearly
with height.
The model gives reasonable predictions for such
properties as solids holdup, particle velocity, and solids
ux for simple riser geometries over a wide range of
fast uidization conditions. Agreement with a radial
prole of local solids ux is illustrated in Fig. 11.
4.6.3 Computational Fluid Dynamic (CFD) Models
CFD codes have had great success in recent decades in
simulating single-phase ows. Similar approaches have
been attempted by a number of research groups for
multiphase ows, but success has been limited for the
dense ow conditions encountered in circulating ui-
dized bed risers. There are several possible formula-
tions of the equations and a number of possible
approaches for solving them. Extensive reviews have
been provided by Gidaspow (1994), Sinclair (1997),
and Kashiwa and Yang (2002). Recent studies include
Mathieson et al. (2000) and Zhang and VanderHeyden
(2001)
Unfortunately CFD codes, despite the extensive
eort and computational power deployed, have in gen-
eral not matched experimental data any better than the
mechanistic models in challenge problem exercises
conducted at two recent (1995 and 2001)
International Fluidization Conferences. While numer-
ical (CFD) approaches will no doubt make major
future contributions, there are currently no codes cap-
able of giving consistently reliable predictions for con-
centrated and turbulent suspension ows of the kind
found in CFB risers.
4.7 Scale-Up Considerations
Circulating uidized beds cover an enormous range of
sizes, from centimeters to tens of meters in lateral
dimension. While most data have been obtained in
laboratory units of 100 to 200 mm diameter, a num-
ber of measurements in large-scale commercial units
have been reported, as summarized in Table 5. Most
of these measurements are limited to the wall region,
owing to the diculty of extending probes into the
interior of large high-temperature reactors. The overall
ow patterns appear to be qualitatively similar in small
and large diameter vessels. For example, one still nds
a fast uidization ow regime with particles traveling
predominantly upward in a dilute central core and
downward in an outer wall layer. However, some sig-
nicant quantitative dierences arise:
Because a larger riser has less surface area per unit
volume, there is greater coverage of the walls by
downowing streamers for the same U
g
and G
s
.
The thickness of the wall layer is approximately
proportional to the riser hydraulic diameter to
Figure 18 Parameters and dimensionless groups used to
characterize riser exit in mechanistic model of Gupta and
Berruti (2000).
the power of 0.6 (Werther, 1994), not to the
power of 1.
Large-scale risers, at least those used for combus-
tion processes, commonly operate under very
dilute conditions at the top (e.g., see Werther,
1994 and Sec. 5.1 below).
Turbulence is likely to play a lesser overall role as D
increases (Senior and Grace, 1998).
Because of experimental limitations, it is uncommon in
experimental CFB studies to vary the riser diameter, or
even the H,D ratio, while keeping all other variables
xed.
There have been a number of eorts to introduce a
sucient set of dimensionless groups which, when
matched and coupled with geometric similarity, assure
dynamic similarity between small- and large-scale
units. This work is well summarized by van der Meer
et al. (1999). For full hydrodynamic scaling, they
recommend that at least ve dimensionless groups be
matched:
Fr
d
=
U
g
gd
p
_ Re
p
=
,
g
d
p
U
g
j
g
G
s
,
p
U
g
,
p
,
g
D
d
p
(31)
with d
p
being the Sauter mean particle diameter. In
place of the third group above, Kehlenbeck et al.
(2001) suggest a dimensionless mass turnover, invol-
ving the solids inventory.
For Re
p
= ,
g
d
p
U
g
,j
g
- 4, the above list may be
reduced to three groups (Glicksman et al., 1993; van
der Meer et al., 1999), the Froude number (Fr
d
) and
dimensionless solids ux G
s
/(,
p
U
g
) from the above list,
plus either U
g
,U
mf
or U
g
,u
t
, e.g.,
Fr
d
=
U
g
gd
p
_
G
s
,
p
U
g
U
g
u
t
(Re
p
- 4) (32)
For Re
p
= ,
g
d
p
U
g
,j
g
> 1000, Re
p
may be dropped
from those listed in Eq. (31) above, leaving a subset
of four groups to match. D,d
p
can be replaced in this
four-group set by U
g
,U
mf
or U
g
,u
t
(Horio et al., 1989;
Glicksman et al., 1993; van der Meer et al., 1999),
yielding, for example,
Fr
d
=
U
g
gd
p
_
G
s
,
p
U
g
,
p
,
g
U
g
u
t
(Re
p
> 1000)
(33)
A number of other possible variables are ignored in
arriving at the above lists, the most notable being the
particle shape, the particle size distribution, the coe-
cients of restitution and friction for the particles them-
selves and between the particles and the wall, and any
property variations due to temperature or concentra-
tion gradients. Advanced uid dynamic models (e.g.,
Sinclair and Jackson, 1989; Senior and Grace, 1998)
and experimental results (Chang and Louge, 1992) sug-
gest that interparticle collisions and other interactions
are of considerable importance for the conditions
Table 5 Sources of Hydrodynamic Data from Large-Scale CFB Units
Reference Cross section Height (m) Aspects covered
Couturier et al., 1991 3.15 3.15 23.8 Pressure prole, solids ux prole
Leckner et al., 1991 1.7 1.4 13.5 Solids ux near wall
Werther, 1994 various Pressure proles, solids ux, wall layer thickness
Zhang et al., 1995 1.7 1.4 13.5 Particle ux proles, wall layer thickness
Lafraneche` re & Jestin, 1995 9.65 eq. dia. 36.9 Pressure proles, solids ux proles, wall layer
thickness
Hage & Werther, 1997 5.1 5.1 28.0 Local particle concentration, local particle velocity
Zhang et al., 1997 1.7 1.4 13.5 Particle momentum, concentration and velocity
Caloz et al., 1999 14.7 11.5 35 Particle velocity, uctuations
Lin et al., 1999 4 6.2 22.8 Recirculation rate, particle velocities
Schouten et al., 1999 1.2 0.8 9.0 Bottom bed hydrodynamics
Sterne us et al., 1999 1.7 1.4 13.5 Turbulence
Johnsson & Johnsson, 2001 1.7 1.4 13.5 Local particle concentration, probability density,
intermittency index
Johnsson et al., 2002 21.1 9.9 42.5 Local particle momentum, local particle concentration
encountered in CFB risers. The omission of the coe-
cients of restitution and friction and the particle size
distribution, all of which relate to particleparticle
interactions, may explain why experimental attempts
(e.g., Glicksman et al., 1991) to verify scaling based
on any of the above lists (or similar sets of groups)
have generally produced disappointing agreement.
Nevertheless, the above sets of groups give a reason-
able starting point for experimental attempts to
achieve hydrodynamic scale modeling. For example,
this approach has been used to model the eects of
temperature and pressure on CFB hydrodynamics
(Chang and Louge, 1992; Louge et al., 1999; Younis
et al., 1999).
Scaling may also be based on hydrodynamic models
(e.g., Winter et al., 1999). For example, the mechanistic
models covered in Sec. 4.5.2 incorporate large-scale
data in empirical correlations and mechanistic models
used to predict hydrodynamics of CFB risers covering
a wide range of sizes.
5 HYDRODYNAMICS OF OTHER RELEVANT
FLOW REGIMES
5.1 Pneumatic Transport
CFB systems normally operate at high enough net
solids uxes that type A choking (Bi et al., 1993)
occurs, leading to fast uidization conditions.
However, the upper region of some commercial scale
CFB combustors (e.g., Werther, 1994) is so dilute that
the dilute pneumatic transport ow regime may be of
interest. When this occurs, there is no appreciable wall
layer and particles travel upward over the entire riser
cross section.
The vast majority of work on pneumatic transport
has been in pipes and ducts of much smaller diameter
and higher H/D ratio than the riser reactors considered
in this chapter. CFB research has usually only touched
on such dilute conditions (generally c
sav
- 1%) when
necessary, e.g., in order to be able to obtain laser sheet
images. Correlations and models developed for fast
uidization conditions are unlikely to give accurate
predictions when pneumatic transport conditions pre-
vail.
5.2 High-Density Circulating Fluidized Beds
Early studies of CFB hydrodynamics were domi-
nated by modest net particle uxes (G
s
from 10
to 100 kg/m
2
s and solids holdups in the upper part
of the riser well below 10%). While these conditions
are relevant to CFB combustors, they do not match
those encountered in FCC and other catalytic CFB
reactors, where the net solids uxes can be an order
of magnitude higher and solids holdups are com-
monly 10%. Several recent studies have been dedi-
cated to ow in high-density circulating uidized bed
(HDCFB) systems. These investigations have almost
all been performed with FCC (group A) solids,
though one study with sand (group B) particles
(Karri and Knowlton, 1998) indicates that the results
also apply, at least in broad terms, to larger denser
particles.
The most notable dierence in behaviour as G
s
is
increased beyond 200300 kg/m
2
s at relatively high
U
g
is the gradual disappearance of net solids downow
at the wall (e.g., Issangya et al., 2000; Karri and
Knowlton, 1998, 1999; Liu, 2001). This leads to a dif-
ferent ow regime (dense suspension upow, as dis-
cussed in Sec. 3) and a more homogeneous ow
structure. Typical axial solids holdup proles appear
in Fig. 19. Note that at these high values of G
s
, the
proles tend to reach relatively constant holdups of the
order of 15%, indicating an approach to fully devel-
oped ow. Pa rssinen et al. (2001) found that there were
four zones in axial proles for a much taller riser.
Despite the lack of time-mean downow at the wall
in dense suspension upow, the ow is still subject to
pronounced radial gradients (Bai et al., 1999; Grassler
and Wirth, 1999; Karri and Knowlton, 1999; Liu,
2001), with signicantly lower voidages (approaching
c
mf
on a time mean basis) at the wall, while the particle
velocity prole becomes more nonuniform with
increasing G
s
. The correlation of Issangya et al.
(2001) for the radial distribution of voidage (Eq. 19
above) also works well for dense suspension upow.
Intermittency index values are in the same range and
again show a maximum near the outer wall (Issangya
et al., 2000).
The local time mean vertical component of particle
velocity at the center line in dense suspension upow
can approach (Pa rssinen et al., 2001) or even exceed
(Liu, 2001) 2U
g
. Proles tend to be very steep, as
shown in Fig. 20. As can be seen in the same gure,
the local standard deviation of particle velocity is much
atter, with local values exceeding the local time mean
particle velocity near the outer wall. This means that
there are momentary reversals in particle direction
near the wall, even when the time mean direction is
upwards.
At the center of the column all particles travel
upward, whereas there is a substantial downward ux
toward the outer wall, partially osetting the upward
ux there. Typically the local net ux is 1.52.5G
s
at
the axis of the column (Issangya et al., 1998; Karri and
Knowlton, 1999; Liu, 2001) falling to nearly 0 at the
wall. Results from a column of diameter 0.3 m at high
U
g
in Fig. 21 indicate that the radial net ux proles
can reverse, showing a minimum at the axis of the
column.
6 MIXING
6.1 Gas Mixing
Vertical and horizontal gas mixing are of considerable
importance in predicting conversions and selectivities
in CFB reactors, especially for catalytic reactions. Gas
mixing depends on the ow regime, ow pattern, par-
ticle properties, and riser conguration.
Figure 19 Axial prole of solids holdup for a range of net circulation uxes in a riser of diameter 76 mm with a smooth
unconstricted exit. d
p
= 70 jm, ,
p
= 1600 kg,m
3
, U
g
= 8 m,s. Highest uxes correspond to dense suspension upow. Note that
the entrance region was dierent for the highest solids ux case. (Issangya et al., 1999; Liu, 2001.)
Figure 20 Local time mean particle velocity and standard deviation of particle velocity at z = 4.2 m in 76 mm diameter, 6.4 m
diameter column operated at U
g
= 6 m/s, G
s
= 417 kg/m
2
s. (Liu, 2001.)
With the whole riser treated as a closed vessel with
well-mixed tracer injection at the inlet and sampling at
the outlet, the RTD curve can be obtained from a pulse
or a step-change test (Schu gerl, 1967; Brereton et al.,
1988; Liu et al., 1999). An eective dispersion coe-
cient, D
ge
, can then be derived by tting. Both vertical
and horizontal mixing/dispersion of gases contribute
to gas dispersion. In steady state tracer experiments,
a steady ow of tracer gas is introduced into the riser at
a point, and the tracer concentration is measured either
downstream or upstream. Ideally, the injection velocity
should equal the gas velocity in the riser (Bader et al.,
1988). Both adsorbing and nonadsorbing tracers can
be used. When an adsorbing tracer is selected, the equi-
librium adsorption on the particle surface needs to be
identied (Kunii and Levenspiel, 1991). Based on the
tracer concentration measured upstream of the injec-
tion point, a one-dimensional backmixing coecient,
D
gb
, can be derived (Kunii and Levenspiel, 1991). The
radial dispersion coecient, D
gr
, is obtained by analyz-
ing radial concentration proles measured downstream
of the injection point (Bader et al., 1988). To charac-
terize the mixing in the riser fully, all three coecients
need to be studied.
6.1.1 Gas RTD and Effective Gas Dispersion
The overall mixing behavior of the riser can be char-
acterized by the residence time distribution (RTD)
curve or an eective dispersion coecient. To obtain
the RTD of the entire riser, tracer is introduced below
the gas distributor, with samples taken at the riser exit
to ensure well-mixed injection and sampling. Transient
one-dimensional dispersion can be described by
oC
ot
U
g
oC
oz
D
ge
o
2
C
oz
2
= 0 (34)
With closedclosed boundary conditions, Eq. (34) can
be solved. With relatively low dispersion, the variance
and the eective dispersion coecient are related
(Levenspiel, 1998) by
o
2
t
t
2
= 2
D
ge
c
av
U
g
H
2
D
ge
c
av
U
g
H
_ _
2
1 exp
U
g
H
D
ge
c
av
_ _ _ _
(35)
The eective dispersion coecient D
ge
was reported
to be in the range 1 to 12.7 m
2
/s for FCC (Dry and
White, 1989; Liu et al., 1999) and (group B) sand
(Brereton et al., 1988; Wang et al., 1992). D
ge
increases
with increasing G
s
at relatively low solids circulation
rates, then tends to level o (Dry and White, 1989) or
to decrease after reaching a maximum at
G
s
- 200 kg,m
2
s (Liu et al., 1999). Liu et al. (1999)
related the decrease of D
ge
for G
s
> 200 kg/m
2
s to
the transition from fast uidization to dense suspen-
sion upow (see Sec. 5).
Figure 22 shows that the Peclet number, Pe
ge
, based
on the eective dispersion coecient, decreases with
increasing solids ux, especially for G
s
- 100 kg/m
2
s.
Figure 21 Radial proles of local solids ux reported by Karri and Knowlton (1999) for G
s
= 586 kg/m
2
s, z = 3.7 m in a 0.3 m
diameter, 13 m long column. Numbers shown on gure are supercial gas velocities.
At high solids ux, however, Pe
ge
appears to be no
longer sensitive to G
s
, with its value of order 10.
Gas dispersion is inuenced by the entrance and exit
structure of the riser, as well as its length and diameter.
Brereton et al. (1988) showed that D
ge
tends to be
higher for an abrupt exit than for a smooth one at a
given U
g
and G
s
. On the other hand, D
ge
from the
smooth exit was higher than D
ge
from the abrupt exit
for a given apparent suspension density.
Eective dispersion is aected by both axial and
radial dispersion, as well as by Taylor dispersion due
to the radial prole of axial gas velocities. Schu gerl
(1967) related the eective dispersion coecient, D
ge
,
axial dispersion coecient, D
gd
, radial dispersion coef-
cient, D
gr
, and the radial velocity prole by
D
ge
= D
gd
M
U
2
g
D
2
D
gr
(36)
where M is a constant which characterizes the nonuni-
formity of the axial velocity prole.
6.1.2 Radial Gas Dispersion
A radial gas dispersion coecient can be obtained
from the steady-state tracer method by tting mea-
sured radial proles of tracer concentration down-
stream of the injection point to a two-dimensional
dierential equation (Yang et al., 1983; Bader et al.,
1988; Kruse et al., 1995),
oC
ot
= D
gd
o
2
C
oz
2
D
gr
1
cr
o
or
cr
oC
or
_ _
u
g
(r)
oC
oz
(37)
D
gd
and D
gr
are both assumed to be independent of
position, and u
g
(r) is the local interstitial gas velocity.
If the axial dispersion is much smaller than the radial
dispersion, the axial dispersion term can be dropped
(van Zoonen, 1962; Yang et al., 1983; Martin et al.,
1992; Gayan et al., 1997; Namkung and Kim, 2000).
Otherwise, one needs to t both the axial and the radial
dispersion coecients, which can involve much higher
uncertainties (Bader et al., 1988). The radial velocity
prole, u
g
(r), and local voidage, c, are required to t
the radial dispersion coecient, but most studies have
assumed a at radial velocity prole in tting Eq. (37)
to tracer concentration proles.
Most researchers have reported that the radial dis-
persion coecient decreases with increasing G
s
at a
given U
g
and relatively low G
s
, and increases with
increasing G
s
at relatively high solids uxes (Zheng et
al., 1992). Similarly, D
gr
decreases with increasing
solids concentration at low solids concentration
owing to suppression of turbulence (van Zoonen,
1962; Adams, 1988; Zheng et al., 1992; Koenigsdor
and Werther, 1995), while increasing with a further
increase in solids concentration, probably because of
solids downow in the annulus region (Zheng et al.,
1992; Win et al., 1994).
At a given solids circulation rate, D
gr
decreases with
increasing U
g
, although Adams (1988) and Zheng et al.
(1992) observed that D
gr
increases with increasing gas
velocity at relatively high G
s
. When data are plotted as
Pe
gr
vs. Re
p
, as in Fig. 23, Pe
gr
increases with increas-
ing Re
p
at given G
s
. Pe
gr
is found to decrease with
increasing G
s
except at very low solids uxes (Amos
et al., 1993), where Pe
gr
is observed to increase with
increasing G
s
.
The radial gas dispersion coecient associated with
the lean phase can be obtained separately (Werther et
al., 1992; Koenigsdor and Werther, 1995; Mastellone
and Arena, 1999b). It has been found to increase with
increasing G
s
(Mastellone and Arena, 1999b) or to be
insensitive to G
s
, and it tends to increase with increas-
ing U
g
(Werther et al., 1992; Koenigsdor and
Werther, 1995).
Zheng et al. (1992) found that D
gr
increased with
increasing ,
p
at a given gas velocity and solids ux for
group B particles. Mastellone and Arena (1999b),
using group A particles, found that D
gr
decreased
Figure 22 Axial gas Peclet number as a function of solids
circulation ux at constant supercial gas velocities. Dry and
White (1989): ~ U
g
= 2.0 m/s, U
g
= 2.5 m/s, * U
g
= 3.0
m/s, r U
g
= 4.0 m/s, ^ U
g
= 5.0 m/s, & U
g
= 8.0 m/s. Liu
et al. (1999): ~ U
g
= 2.0 m/s, ! U
g
= 2.5 m/s, & U
g
= 4.7
m/s, * U
g
= 7.3 m/s.
with increasing particle density. Gayan et al. (1997)
reported that D
gr
increased when particle size was
increased from 380 to 710 jm at given U
g
and G
s
,
whereas Mastellone and Arena (1999b) reported a
reduction in D
gr
when d
p
increased from 89 to 310 jm.
Correlations for the radial dispersion coecient are
listed in Table 6. The equation of Yang et al. (1983),
based on their own experimental data in a 0.135 m
diameter column, underestimates most data obtained
in relatively large columns. The Koenigsdor and
Werther (1995) equation can only be applied to very
dilute conditions (c
av
- 0.99). The equation of Amos et
al. (1993) predicts that Pe
gr
increases with increasing G
s
over a wide range but is inconsistent with most experi-
mental data. The equations of Gayan et al. (1997) and
Namkung and Kim (2000) are recommended until
more reliable correlations are available.
6.1.3 Gas Backmixing
Gas backmixing is obtained from steady state tracer
injection tests in which tracer is injected continuously
from one location and its concentration is monitored
upstream of the injection point. A one-dimensional
vertical dispersion process with a nonadsorbing tracer
can be represented by
D
gb
d
2
C
dz
2

U
g
c
av
dC
dz
= 0 (38)
where C is the cross-sectional average tracer concen-
tration. With boundary conditions
C = C
i
at z = 0 (39)
and
Figure 23 Radial gas Peclet number as a function of
Reynolds number. Gayan et al., 1997: &d
p
= 710 jm, G
s
=
30 kg/m
2
s; * d
p
= 380 jm, G
s
= 30 kg/m
2
s, Adams (1988):
& d
p
= 250 jm, G
s
= 30 kg/m
2
s, * d
p
= 250 jm, G
s
= 44
kg/m
2
s, Bader et al. (1988): ~ d
p
= 76 jm, G
s
= 98177 kg/
m
2
s, Yang et al. (1983): ~ d
p
= 220 jm, G
s
= 57 kg/m
2
s, !
d
p
= 220 jm, G
s
= 74 kg/m
2
s)
Table 6 Correlations for Estimating Gas Radial Dispersion Coefcient
Source Equation Comment
van Zoonen (1962)
Pe
gr
= 500 1000
For small columns only.
Yang et al. (1983)
D
gr
= 43.5
U
g
(U
g
,c
av
) (G
s
,,
p
[1 c
av
])
_ _
1 c
av
c
av
0.7
No column diameter eect.
For small columns only.
Amos et al. (1993)
Pe
gr
= 6.46c
av
(G
s
,
g
U
g
)D
j
g
_ _
0.297
That Pe
gr
increases with G
s
is inconsistent
with other experimental data.
Koenigsdor and
Werther (1995)
Pe
gr
= 150 56. 000(1 c
av
)
For dilute ow only with (1 c
av
) - 0.01.
Gayan et al. (1997)
D
gr
= 1.4 10
6
(G
s
,
g
U
g
)D
j
g
_ _
1.14
(G
s
1)
d
p
,
p
_ _
1.3
D
1.85
That D
gr
decreases with increasing G
s
at
high solids ux is inconsistent with
experimental data.
Namkung and
Kim (2000)
Ud
p
D
gr
= 153.1
d
p
D
_ _
0.96
U
g
U
sl
_ _
3.73
U
sl
=
U
g
c
av
G
s
,
p
(1 c
av
)
C = 0 at z = o (40)
the solution for equation (35) is
C
C
i
= exp
U
g
(z
i
z)
c
av
D
gb
_ _
(41)
The axial dispersion coecient is obtained by plotting
ln(C,C
i
) vs. the distance, z, from the source.
Experimentally it is dicult to generate a plane
source of tracer gas. A point source of injection has
thus been used in all experiments. When tracer is
injected at the axis, little tracer is detected upstream
(Li and Weinstein, 1989; Namkung and Kim, 1998).
On the other hand, high tracer concentration and sig-
nicant radial variation were detected when tracer was
injected into the annulus. These ndings suggest that
gas backmixing is signicant in the annulus region but
negligible in the dilute core region.
To determine a one-dimensional average back-
mixing coecient based on equation (41), an average
tracer concentration is required. With axial symmetry,
C =
2
R
2
(U
g
,c
av
)
_
R
0
C(r) u
g
(r) r dr (42)
D
gb
generally increases with increasing G
s
at given
U
g
. Li and Weinstein (1989) found that D
gb
increased
with increasing U
g
at constant G
s
, while Namkung and
Kim (1998) found the opposite trend, possibly because
their data were obtained at relatively low uxes
(G
s
- 40 kg,m
2
s), while Li and Weinstein (1989)
reached relatively high solids uxes (up to 200 kg/m
2
s).
6.1.4 Axial Gas Dispersion
Axial gas dispersion can be studied with pulse or step-
change injection (van Zoonen, 1962; Li and Wu, 1991;
Luo and Yang, 1990; Martin et al., 1992; Bai et al.,
1992b). Transient one-dimensional dispersion can be
described (Li and Wu, 1991) by
oC
ot
U
g
oC
oz
D
gd
o
2
C
oz
2
= 0 (43)
For pulse injection with open-open boundary condi-
tions, the solution of equation (43) is
C =
1
2
(c
av
D
gd
,U
g
L
_ exp
(1 )
2
4(c
av
D
gd
,U
g
L)
_ _
(44)
where L is the distance between the injection and the
sampling point for single point sampling or the dis-
tance between the two sampling locations for double-
point sampling. The variance of the RTD curve is
related to the axial gas dispersion coecient for rela-
tively low dispersion by
o
2
t
t
2
= 2
D
gd
c
av
U
g
L
8
D
gd
c
av
U
g
L
_ _
2
(45)
In all reported pulse tests, the tracer was injected on
the axis, with the sampling port downstream on the
axis. The tracer concentration then reects the local
response on the axis, not the response from the entire
cross section. The dispersion coecient, D
gd
, obtained
in this way is the same as the one-dimensional disper-
sion coecient dened by equation (37) only for risers
of small diameter because of the strong radial variation
of tracer concentrations downstream of the injection
point.
D
gd
generally increases with increasing G
s
at given
gas velocities for G
s
- 200 kg/m
2
s. For G
s
> 200 kg/
m
2
s, van Zoonen (1962) reported a decrease of D
gd
with increasing solids ux. van Zoonen (1962) and
Luo and Yang (1990) found that D
gd
increases with
increasing U
g
at constant G
s
for G
s
> 50 kg/m
2
s. Li
and Wu (1991) and Bai et al. (1992) reported the oppo-
site trend for G
s
- 50 kg/m
2
s.
Luo and Yang (1990) found that D
gd
decreases with
increasing particle size at given U
g
and G
s
. The eect of
column diameter is unclear. Li and Wu (1991) corre-
lated their data by
D
gd
= 0.195c
4.11
av
(46)
which ignores any eect of particle properties. On the
other hand, Luo and Yang (1990) correlated their D
gd
data as functions of operating parameters and gas and
particle properties as
D
gd
,
g
j
g
= 0.11
,
g
U
g
d
p
j
g
_ _
1.6
G
s
,
p
u
t
_ _
0.62
,
p
,
g
_ _
1.2
(47)
6.1.5 Interphase Mass Transfer
Brereton et al. (1988) and Namkung and Kim (1998)
considered a coreannulus structure with exchange
between the two regions. With one-dimensional ow
assumed in both regions, axial dispersion in the annu-
lus and plug ow in the core, one can write
oC
c
ot
u
gc
oC
c
oz

2K
ca
R
c
(C
c
C
a
) = 0 (48)
oC
a
ot
D
gza
o
2
C
a
oz
2

2K
ca
R
c
R
2
R
2
c
(C
c
C
a
) = 0 (49)
for the core and annulus regions, respectively. Both the
axial dispersion coecient in the annulus region, D
gza
,
and the interphase exchange coecient, K
ca
, are then
tted to concentration proles. Values of K
ca
have
been reported to range from 0.015 to 0.3 m/s and to
increase with increasing G
s
. The eect of U
g
on K
ca
,
however, is uncertain. Werther et al. (1992) found that
K
ca
increases with U
g
, but White et al. (1992) and
Namkung and Kim (1998) reported a decrease of K
ca
with increasing U
g
.
6.2 Solids Mixing
Solids mixing can inuence such factors as gassolids
contacting, heat transfer, solids conversion, and deac-
tivation of catalyst particles. Solids mixing can be mea-
sured by several experimental techniques. Particle
dispersion and residence time distributions (RTD) in
the riser have been commonly studied by injecting a
pulse of tracer particles and then sampling downstream
of the injection point or at the riser exit. The tracer
particles may be radioactive (Patience et al., 1991),
magnetic (Avidan and Yerushalmi, 1985), uorescent
(Kojima et al., 1989), heated (Westphalen and
Glicksman, 1995), phosphorescent (Wei et al.,
1995b), salt or impregnated with salt (Bader et al.,
1988). On-line detection systems can be adopted for
most of these tracers but not salt.Tracer particles of
the same size, density, and shape as the bed material
should be used wherever possible.
6.2.1 Axial Dispersion of Particles
RTD curves of solid tracer particles have been ana-
lyzed to obtain an axial dispersion coecient assuming
axially dispersed one-dimensional ow (Smolders and
Baeyens, 2000b),
oC
ot
= D
pz
o
2
C
oz
2
v
p
oC
oz
(50)
where C is the concentration of tracer particles at time
t. The ow is considered fully developed, so that par-
ticle velocity v
p
is independent of z. With openopen
boundary conditions, a solution is
E() =
1
2
,Pe
pz
_ exp
Pe
pz
(1 )
2
4
_ _
(51)
with
=
t
t
and Pe
pz
=
v
p
L
D
pz
(52)
where t is the mean residence time, E() is the dimen-
sionless RTD function, and L is the distance between
the tracer injection and detection points. The mean
residence time, t, can be estimated from the rst
moment of the normalized RTD (Patience et al.,
1991), the overall solids circulation rate (Zhang et al.,
1993), or the 50% value of the cumulative solids resi-
dence time distribution of the tracer (Rhodes et al.,
1991b). In many studies, v
p
in the denition of Pe
pz
above is replaced by U
g
(Rhodes et al., 1991b; Wei et
al., 1995b; Smolders and Baeyens, 2000b). This prac-
tice will be followed here.
Figure 24 plots the axial Peclet number as a func-
tion of solids circulation ux from several sources. Pe
pz
generally ranges from 1 to 10 for risers with abrupt
exits. For a smooth exit (Diguet, 1996), Pe
pz
was higher
due to less particle reection from the exit and thus less
downow of particles near the wall. Pe
pz
tends to
decrease with increasing solids circulation rate, consis-
tent with the nding that the solids downow in the
annulus region increases with increasing solids circula-
tion. Some researchers (Rhodes et al., 1991b; Wei et
al., 1995b; Godfroy et al., 1999a), however, report that
Pe
pz
increases with increasing G
s
. At constant G
s
, Pe
pz
tends to increase slightly (Wei et al., 1995b; Diguet,
1996; Smolders and Baeyens, 2000b) or remain con-
stant (Rhodes et al., 1991b; Patience et al., 1991)
with increasing U
g
, although the axial solids dispersion
coecient in all cases increased with increasing U
g
.
Using three types of sand of dierent mean sizes,
Patience et al. (1991) found that Pe
pz
was higher for
larger than for smaller particles at the same U
g
and G
s
.
The axial gas dispersion tends to increase with col-
umn diameter, as shown above. If both gas and solids
Figure 24 Eect of solids circulation ux on axial solids
dispersion. Data: & sand, d
p
= 90 jm (Smolder and
Baeyens, 2000b), ~ (Diguet, 1996); ^ FCC, d
p
= 54 jm
(Du et al., 1999), * FCC, d
p
= 56 jm (van Zoonen, 1962),
~ sand, d
p
= 275 jm (Patience et al., 1991).
dispersion are mainly due to downow of particles next
to the wall, axial solids dispersion is expected to follow
the same trend, i.e., to increase with increasing D.
Limited data (Werther and Hirschberg, 1997) indicate
that the axial particle dispersion coecient in the
upper fully developed dilute region tends to increase
with D. However, Rhodes et al. (1991b) found that
axial dispersion decreased with increasing D. Caution
thus needs to be exercised in scaling up laboratory
data.
Several correlations are available to predict the
solids axial dispersion. The Rhodes et al. (1991b) cor-
relation and the Wei et al. (1995b) correlation predict
that Pe
pz
increases with increasing solids ux or solids
concentration, while the Smolder and Baeyens (2000)
correlation,
Pe
pz
= 2.4
18.3
(G
s
,U
g
)
0.8
(53)
predicts the opposite trend. Equation (53) is recom-
mended, although the eect of D is unclear.
6.2.2 Radial Dispersion of Particles
Radial particle dispersion in CFB risers can be studied
by measuring radial concentration proles of tracer
particles injected at a single point upstream of the mea-
surement location (van Zoonen, 1962; Wei et al.,
1995b). A two-dimensional dispersion model for fully
developed axisymmetric ow with constant dispersion
axial and radial coecients, D
pz
and D
pr
, gives
oC
ot
= D
pz
o
2
C
oz
2

D
pr
r
o
or
r
oC
or
_ _
o(v
p
C)
oz
(54)
Analytical solutions of equation (54) for pulse injection
of solid tracers can be obtained with proper boundary
conditions by assuming that v
p
is independent of radial
and axial location (van Zoonen, 1962; Wei et al.,
1995b; Patience and Chaouki, 1995). If velocity and
solids concentration proles are nonuniform, equation
(54) must be solved numerically (Koenigsdor and
Werther, 1995), and the axial and radial solids disper-
sion coecients are then obtained by tting.
Wei et al. (1995b) and Koenigsdor and Werther
(1995) reported that the radial dispersion coecient
decreased with increasing solids concentration at
given gas velocities (see Fig. 25), and that the radial
dispersion Peclet number is generally higher than for
single-phase turbulent ow. van Zoonen (1962)
observed that radial dispersion is more pronounced
in the lower entrance region and decreases with height.
This is probably associated with developing ow in the
entrance region. At constant solids concentration or
constant solids ux, D
pr
increased with increasing U
g
,
probably due to increased local turbulence (Wei et al.,
1995b; Koenigsdor and Werther, 1995).
The correlations of Wei et al. (1995b) and
Koenigsdor and Werther (1995), respectively,
Pe
pr
= 225.7(1 c
av
)
0.29
Re
0.3
p
(55)
and
Pe
pr
= 150 5.6 10
4
(1 c
av
) (56)
can be used to estimate radial solids dispersions until
more accurate correlations are available.
6.2.3 Interphase Solids Exchange Between Core and
Annulus
Although unimodal solids RTD curves have usually
been reported (Chesonis et al., 1990; Rhodes et al.,
1991b; Bai et al., 1992b; Smolders and Baeyens,
2000b; Harris, 2002), bimodal curves have been
found for relatively dense conditions (Helmrich et al.,
1986; Kojima et al., 1989; Ambler et al., 1990; Wei et
al., 1995b). The bimodal distribution is likely asso-
ciated with the internal circulation of particles and
cannot be predicted by a one-dimensional dispersion
model. Coreannulus models can account for the
bimodal solids RTD distributions, with a core/annulus
solids exchange coecient characterizing cross-ow of
particles. An unsteady solids tracer balance relative to
Figure 25 Radial solids dispersion coecient as a function
of solids concentration and gas velocity in CFB risers. Data:
* FCC, d
p
= 54 jm, U
g
= 5.7 m/s (Wei et al., 1995b); *
SiC, d
p
= 60 jm, U
g
= 4.0 m/s (Koenigsdor and Werther,
1995); & U
g
= 3.0 m/s (Koenigsdor and Werther, 1995).
the core and annular regions, assuming plug ow and
neglecting axial dispersion, yields (Ambler et al., 1990)
o(G
c
f
c
)
oz
(I
ca
f
c
I
ac
f
a
) =
G
c
v
pc
of
c
ot
(57)
o(G
a
f
a
)
oz
(I
ca
f
c
I
ac
f
a
) =
G
a
v
pa
of
a
ot
(58)
for the core and annular regions, respectively, where
G
c
and G
a
are the upward and downward solids uxes
in the core and annular regions, respectively. I
ca
refers
to the mass of solids transferred from core to annulus
and I
ac
from annulus to the core, with (I
ca
-I
ac
) repre-
senting the net lateral ow from the core to the annu-
lus. f
c
and f
a
represent the mass fraction of tracer in the
core and annulus, respectively, while v
pc
and v
pa
are
average particle velocities in the core and annulus.
Values of I
ac
ranged from 0.1 to 10 kg/m
3
s for 90
and 106 mm sand particles (Ambler et al., 1990;
Smolders and Baeyens, 2000b), and increased as G
s
increased (Smolders and Baeyens, 2000b). Converted
into radial solids uxes, these interchange uxes (0.025
to 0.25 kg/m
2
s) fall into the same range of radial solids
uxes (0.1 to 1.5 kg/m
2
s) as those measured by sam-
pling methods (Zhou et al., 1995b), implying that lat-
eral solids mixing is mainly caused by particle cross
ow.
7 HEAT TRANSFER
7.1 Introduction
Favorable heat transfer is a major reason for the suc-
cess of CFB reactors in such applications as combus-
tion and calcination. CFB heat transfer has previously
been reviewed by Grace (1986b), Glicksman (1988),
Leckner (1990), Yu and Jin (1994), Basu and Nag
(1996), Glicksman (1997), and Molerus and Wirth
(1997). Temperature gradients tend to be small within
the riser due to vigorous internal mixing of particles,
while the gas and particle temperatures are locally
nearly equal, except at the bottom (Grace, 1986;
Watanabe et al., 1991). Because of wear caused by par-
ticle impacts, heat transfer surfaces are predominantly
vertical, and only this case is considered here. Many
CFB combustion systems are also equipped with exter-
nal uidized bed heat exchangers, cooling solids return-
ing to the bottom of the riser while they are being
recirculated in the external loop. However, these oper-
ate as bubbling beds exchangers, and the reader is
referred to Chapter 3. In this section we consider the
case where the bulk solids are hotter than the wall, as in
CFB combustion. It is straightforward to change to the
case where the transfer is in the opposite direction.
Suspension-to-wall heat transfer is controlled by the
ow of the solids and gas near the wall. Most studies
have employed small-scale columns at atmospheric
temperature and pressure. For simplicity, the experi-
mental risers have generally been idealized in that they
have been (Grace, 1996):
Smooth-walled, with roughness elements much
smaller than the diameter of the particles
Circular in cross section, i.e., without corners or
protruding surfaces
Vertical, i.e., with no tapered or oblique sections
Supplied with air only through a at multiorice
distributor at the bottom of the riser.
Most boilers operate at 800 to 900
C, but few experi-

mental investigations have been carried out at elevated
temperatures. Industrial CFB combustion systems
have risers that are usually rectangular in cross section,
manufactured from membrane water walls. Surfaces
may be rough, especially where refractory is present.
The lower section is commonly tapered or constricted.
In addition to primary gas from the base, secondary
gas may be injected through nozzles at higher levels.
Given these factors, data obtained in laboratory CFBs
are not always representative of industrial-scale CFB
reactors.
Heat transfer models are usually written in terms of
either clusters or dense wall layers, based on the hydro-
dynamics of fast uidization. For cluster models (Fig.
26), heat can be transferred between the suspension
and wall by (1) transient conduction to particle clusters
arriving at the wall from the bulk, supplemented by
radiation; (2) convection and radiation from the dis-
persed phase (gas containing a small fraction of solid
material). The various components are usually
assumed to be additive, ignoring interaction between
the convective and radiation components.
7.2 Convection
Any portion of the wall comes into intermittent
contact with clusters of particles, interspersed with per-
iods, usually brief, where that portion of the wall is in
contact with relatively dilute phase. Heat transfer from
the descending clusters is commonly described by
unsteady, one-dimensional conduction normal to the
wall. Because this transfer relies on renewal of fresh
particles from the bulk, this heat transfer mechanism
is often referred to as particle convection, even though
the transfer between the cluster and the wall is primar-
ily by conduction through the intervening gaseous
medium.
7.2.1 Inuence of Different Parameters on
Convective Heat Transfer
Suspension Density: The cross-sectional average
suspension bulk density, ,
susp
= ,
p
(1 c
av
) ,
g
c
av
, is
the most important determinant of heat transfer to
the wall. Experimental data from large-scale high-
temperature units are plotted in Fig. 27, with corre-
sponding operating conditions summarized in Table
7. The wall-to-suspension heat transfer coefcient in-
creases strongly with increasing ,
susp
, primarily be-
cause of the particle convection term.
Heat Transfer Surface: The walls of CFB com-
bustion chambers consist of membrane waterwalls,
i.e., parallel tubes connected longitudinally by ns
(Fig. 28). The protected channel between tubes is a
region of augmented particle downow and greater
particle concentration than the exposed tube surfaces
(Wu et al., 1991; Golriz, 1994; Lockhart et al., 1995;
Zhou et al., 1996b). Particles descend faster and tra-
vel further along the ns on average before being
stripped off and returning to the core (Wu et al.,
1991; Zhou et al., 1996b). These ndings help explain
the distribution of local heat transfer coefcients
around membrane surfaces (Andersson and Leckner,
1992; Golriz, 1994; Lockhart et al., 1995).
The conductivity and thickness of the membrane
water wall surface also play a role (Bowen et al.,
1991), because heat transferred to the n must be con-
ducted to the tube surface and then into the coolant
(usually boiling water). In practical systems, the heat
transfer coecient on the coolant side is almost always
much higher than on the suspension side, so that the
overall heat transfer coecient is barely aected by the
coolant ow rate. A model that accounts for the cool-
ant-side resistance, as well as the suspension-side and
Figure 26 Conceptual view of clusters of particles and gas
close to wall of CFB riser.
Figure 27 Heat transfer coecient vs. suspension density for large-scale CFB risers. For operating conditions see Table 7.
wall conduction resistances, has been developed by Xie
et al. (2003).
Descending Particles: To investigate the effect of
the descending wall layer of particles on the heat
transfer coefcient at the membrane water wall, Gol-
riz (1994) installed an angled deector 0.2 m into the
combustion chamber of a 12 MW
th
combustor 0.5,
1.0, 1.5, and 2.0 m above a heat ux meter located
4.0 m above the distributor. The heat transfer coef-
cient increased and the character of the signal chan-
ged with the obstacle diverting clusters and strands
away from the membrane wall, exposing the surface
to hotter particles. The measured heat transfer coef-
cient increased by 55% with the deector 0.5 m
above the heat ux meter. The enhancement of the
heat transfer decreased as the distance between the
meter and deector increased, disappearing at a se-
paration of 2 m. At 10 m above the distributor plate
where the suspension was more dilute, heat transfer
was unaffected by a deector 1.0 m above the heat
ux meter. Hyre and Glicksman (1995) found similar
augmentation by deecting descending particles from
a plane wall. Cao et al. (1994) increased the heat
transfer by 100% by placing a large horizontal ring
in a CFB.
Vertical Length of Heat Transfer Surface: Short
heat transfer surfaces lead to overestimates of heat
transfer to the wall (Wu et al., 1991; Nag and Moral,
1990) because clusters travel along the heat transfer
surface over considerable distances, causing the heat
transfer coefcient to decrease as temperature equili-
bration occurs between the cluster and the surface.
A number of researchers have measured descending
cluster velocities in the wall layer for a variety of ow
conditions and dierent wall geometries, with most
data between 0.5 and 2.0 m/s. The data have been
well summarized by Grith and Louge (1998) and
are correlated by equation (22).
The contact length for clusters, L
c
, can be 1.5 m or
more along the ns of membrane water walls (Wu et
al., 1989; Golriz and Leckner, 1992). L
c
is much
shorter for smooth walls (Wu et al., 1991; Burki et
al., 1993; Glicksman, 1997). The experimental results
suggest that the cluster residence time at the wall is
0.1 to 1 s for smooth walls and 0.5 to 3 s for
sheltered regions, including the ns of membrane
walls. However, for short heat transfer surfaces, e.g.,
0.01 m long, L
c
is only of order 0.01 s. Given the
inuence of length of heat transfer surface, consider-
able caution is required when applying data from short
surfaces to much longer surfaces (Grace, 1990, 1996).
Bulk Suspension Temperature: The heat transfer
coefcient increases with bulk temperature (e.g., see
Wu et al., 1989; Basu, 1990; Golriz and Sunden,
1994a,b), due to both the higher gas thermal conduc-
tivity and the increased radiation. At temperatures
> 500
C, especially for relatively dilute beds (suspen-

sion densities - 15 kg/m
3
), where radiation tends to
be the dominant mode of transfer, the increase in
heat transfer with temperature can be very signi-
cant. Figure 29 shows the local heat transfer coef-
cient along a 1.5 m long heat transfer surface for two
bulk temperatures (Wu et al., 1989). The 3040% in-
Table 7 Experimental Conditions for Large-Scale CFB Heat Transfer Studies
Unit, Capacity, Reference Size, m U
g
, m/s T
b
,

C d
p
, jm T
w
,

C
Chatham, 72 MW
th
, Couturier et al. (1993) 4 4 23 6.4 875 200 500
O
rebro, 165 MW
th
, Andersson et al. (1996) 4.6 12 33.5 35 700860 280 360
Emile Huchet, 125 MW
e
, Jestin et al. (1992) 8.6 11 33 NA 850 140 340
Nucla, 110 MW
e
, Nucla Report (1991) 6.9 7.4 34 2.65.1 774913 150 330
Chalmers, 12 MW
th
, Leckner (1991) 1.4 1.7 13.5 1.86.1 640880 220440 210
VW Wolfsburg, 145 MW
th
, Blumel et al. (1992) 7.6 5.2 31 6.2 850 NS 340
Flensburg, 109 MW
th
, Werdermann & Werther (1994) 5.1 5.1 28 6.3 855 220 340
Jianjiang, 50 MW
th
, Jin et al., (1999) 3.0 6.0 20 5.1 920 400 290
Figure 28 Plan view of section of membrane wall.
crease in heat transfer coefcient with increased tem-
perature is predominantly due to increased radiation
(see below).
Particle Size: The convective heat transfer coef-
cient decreases with increasing particle diameter
(Stro mberg, 1982; Fraley et al., 1983; Basu, 1990) for
small heat transfer sections where clusters have brief
residence times on the heat transfer surface. The in-
uence of d
p
diminishes with longer surfaces, since
the contact times on the surface are then much higher
than the thermal time constant for individual parti-
cles (Wu et al., 1989). Andersson (1996) found that
the vertical distribution of heat transfer coefcients
in a 12 MW
th
CFB boiler are independent of d
p
for
constant U
g
/u
t
.
Supercial Gas Velocity: Experimental evidence
(e.g., Wu et al., 1987; Furchi et al., 1988) indicates
little inuence of gas velocity for a constant suspen-
sion density. This occurs because gas convection is
much less important than particle convection for the
particle sizes and ow conditions of interest.
Pressure: For a given suspension density, the
heat transfer coefcient increases with pressure
(Xianglin et al., 1990; Mattmann, 1991). This effect is
smaller for smaller particles (Shen et al., 1991).
7.2.2 Convective Heat Transfer Models
Cluster Renewal Models: Most mechanistic
models for heat transfer in CFBs are extensions of
the model of Mickley and Fairbanks (1955). Des-
cending clusters and strands in the vicinity of the wall
surface are modeled as homogeneous semi-innite
media, separated from the wall by a thin gas layer.
Transient transfer of heat occurs from the moving
clusters to the wall, with the transfer rate depending
on the replacement frequency and thermal properties
of the clusters.
Emulsion Models: To simulate the coreannulus
structure, the cross section in emulsion models is di-
vided into an inner dilute core region where particles
are transported upwards, and a denser annular region
where particles descend along the wall, as in Fig. 26,
but without the clusters. The thickness of the solid
layer along the vertical heat transfer surfaces is often
approximated as uniform. However, for membrane
wall heat transfer surfaces, the annulus layer tends to
be thicker at the n than at the tube crest (Grace,
1990; Golriz 1992).
Descending particles are usually assumed to enter
the heat transfer zone at the temperature of the core.
Heat transfer from the particles in the wall layer results
in a thermal boundary layer. An example of the hor-
izontal temperature prole normal to the n and crest
of a membrane tube wall and normal to the refractory
wall 11.0 m above the primary air distributor plate in a
12 MW
th
CFB boiler (Golriz, 1992) appears in Fig. 30.
The thermal boundary layer thickness is smaller at the
summit of the tube than at the n.
The emulsion model is similar to the cluster renewal
model, but instead of clusters, the heat transfer surface
is assumed to be fully covered by a uniform emulsion
layer of voidage signicantly higher than c
mf
. The pre-
dictions from this model are in reasonable agreement
with experimental ndings (Luan et al., 1999).
Figure 29 Heat transfer coecient vs. distance from top of
heat transfer surface for two bulk temperatures. (Wu et al.,
1989a.)
Figure 30 Temperature prole at n, crest of tube of mem-
brane tube wall, and refractory wall from Golriz (1992). U
g
= 3.6 m/s, d
p
= 270 jm, T
b
= 860
C. Dashed line indicates

location of tube crest.
7.2.3 Empirical Correlations
Heat transfer coecients measured in small and large
CFB risers are approximately proportional to the
square root of suspension density, despite dierences
in operating conditions. Hence, in general form,
h
tot
= a,
n
susp
b (59)
Values of a, b, and n from the literature appear in
Table 8 with corresponding ranges of conditions.
7.3 Radiation
7.3.1 Some Experimental Investigations
Relatively few studies have been carried out at tem-
peratures high enough (> 600
C) that radiation
plays a signicant role. Basu and Konuche (1988)
and Wu et al. (1989) found that the radiative heat
transfer coecient increased with increasing tempera-
ture, while the contribution of radiation to the total
heat transfer increased with decreasing suspension den-
sity. In dierent studies, the radiative component has
been found to contribute more than 50% of the total
heat transfer at low particle loadings (e.g., Basu and
Konuche, 1988), whereas the contribution is 15 to
40% for higher suspension densities (e.g., Han and
Cho, 1999; Luan et al., 1999).
7.3.2 Radiative Heat Transfer Models
Packet Models: Models featuring packets con-
stitute the most common type of model. These usual-
ly assume that the radiative component is the sum of
radiation from clusters and from the dispersed phase.
The radiative heat transfer between clusters and the
parallel portion of the surface covered by them is
generally based on a gray body approximation. This
type of model is employed below.
Non uniform Emulsion Models: Luan et al.
(1999) treated the wall layer as a nongray, absorbing,
emitting, and anisotropically scattering medium. The
radiative heat ux at any distance, x, from the wall
in the emulsion layer is estimated by integrating the
local spectral radiation intensity. Convective and ra-
diative heat transfer are taken as additive; particles
are spherical and uniform in size. A stationary thin
surface gas layer is taken with its thickness estimated
from the correlation of Lints and Glicksman (1994).
The solid concentration is assumed to be uniform in
the core and non-uniform in the annulus region. The
emulsion layer thickness is estimated from Eq. (24),
while its voidage distribution is estimated from the
correlation of Zhang et al. (1994). For long surfaces
the temperature is assumed to vary only in the emul-
sion layer and to be given (Golriz, 1995) by
T T
w
T
b
T
w
= 1 0.023Re
p
0.094
T
b
T
w
_ _
0.294
z
H
_ _
_ _
exp 0.0054
x
d
p
_ _ _ _
(60)
For short surfaces, temperature variation in the emul-
sion layer is ignored. Temperature gradients in the core
are ignored in all cases. The gas and particle tempera-
tures are equal locally. The heat transfer surface is a
gray, diuse surface. The emulsion is taken as a non-
gray, absorbing, emitting, and scattering medium.
Radiation is neglected in the vertical direction.
Table 8 Summary of Empirical Correlations of Heat Transfer in CFB Risers
*
Investigators Correlation
Suspension density,
kg,m
3
Bulk temperature
C
Basu and Nag (1996) h
ov
= 40,
0.5
sus
5 - ,
sus
- 20 kg,m 750
C - T
b
- 850
Andersson & Leckner (1992) h
ov
= 30,
0.5
sus
580 750895
Golriz & Sunden (1994b) h
ov
= 88 9.45,
0.5
sus
770 800850
Divilio & Boyd (1994) h
c
= 23.2,
0.55
sus
5500 Cold condition
Andersson (1996) h
ov
= 70,
0.085
sus
> 2 kg,m
3
637883
h
ov
= 58,
0.36
sus
_ 2 kg,m
3
*
For more details see Golriz and Grace (2002).
The radiative heat ux predicted by the model for a
riser of 0.152 m square cross section indicates a rapid
decrease with increasing distance from the surface, sug-
gesting that only particles close to the wall contribute
signicantly to the radiative ux. The predictions of
the model are in reasonable agreement with experimen-
tal results for the conditions investigated.
Flamant et al. (1996) employed a similar method.
The predicted radiative heat ux for a 12 MW
th
13.5 m
high CFB boiler of 1.7 m1.4 m cross section indi-
cates that the radiative component contributes 25 to
50% of the total heat exchange for suspension tem-
peratures between 790 and 830
C.
7.4 Recommended Method for Estimating Total Heat
Transfer
Here we outline a practical engineering method (Golriz
and Grace, 2002) for estimating average suspension-to-
wall heat transfer in large CFB risers operating in the
fast uidization ow regime. Thin packets of emulsion,
well-mixed in the horizontal direction, descend along
the outside, separated from the outer wall by a thin
gap, o
g
. Six resistances shown in Fig. 31 are evaluated.
The suspension bulk temperature is T
b
, while the wall
temperature is T
w
. R
1
and R
2
in parallel correspond to
the portions of the wall that are uncovered at any
instant. The other four resistances apply to portions
covered by emulsion. R
3
and R
4
denote bulk-to-emul-
sion resistances, while R
5
and R
6
represent gas gap
resistances. R
2
, R
4
, and R
6
are radiation resistances
and need not be considered for T
b
and
T
w
- 500
C. The following steps are recommended:

1. Find the gas density (,
g
), gas thermal conduc-
tivity (k
g
), gas specic heat (c
pg
), particle density (,
p
),
particle thermal conductivity (k
p
), particle specic heat
(c
pp
), and particle emissivity (e
p
) at temperature
(T
b
T
w
),2 and estimate the wall emissivity (e
w
) at
the wall temperature, T
w
.
2. Estimate the cross-sectional average suspension
voidage, c
av
, and wall layer thickness, o(z), for the
height interval of interest, e.g., using the relationships
in Sec. 4.1.2 and Eq. (24) above.
3. Estimate the thickness of the gas lm at the wall
from the Lints and Glicksman (1993) correlation:
o
g
= 0.0282d
p
(1 c
av
)
0.59
(61)
4. Estimate the fractional wall coverage by clus-
ters. For this purpose we recommend an equation
(Golriz and Grace, 2002) tted to the data plotted by
Glicksman (1997) that includes the inuence of column
diameter, D, while also approaching the limiting value
of 1 in a rational manners as D increases:
f = 1 exp 25. 000(1 c
av
) 1
2
e
0.5D
e
0.5D
_ _ _ _
(62)
5. The gas convection coecient (h
cg
) is evaluated
using a standard steady state correlation for internal
Figure 31 Heat transfer model in network analogy form.
heat transfer for gas owing alone at a velocity equal
to its supercial velocity, U
g
, augmented by 15% to
account for some particles in the vicinity of the wall
during the times when clusters are absent (see
Glicksman, 1997) and also for the locally transient
nature of the exchange.
6. Estimate the emissivity of the suspension and
clusters using the equation of Brewster (1986):
e
susp
= e
c
=
1.5e
p
(1 e
p
)
1.5e
p
(1 e
p
)
2
_ _
_

1.5e
p
(1 e
p
)
(63)
A typical value of e
p
for silica sand is 0.6 (Flamant et
al., 1994).
7. Calculate the radiation heat transfer coecient
corresponding to the fraction of time (1 f ) when
there is no cluster at a given location on the heat trans-
fer surface. It is assumed that the riser is large enough
and the suspension dense enough that only suspension
is seen by the bare portions of the wall during these
intervals. We can then write
h
sr
=
o T
4
b
T
4
w
_ _
(T
b
T
w
) 1,e
susp
1,e
w
1
_ _ (64)
For very small risers or very dilute conditions where
exposed portions of the wall can see the opposite wall
or side walls, an electric network analogy can be
employed (Fang et al., 1995).
8. Estimate the lateral particle ux from a correla-
tion developed by Golriz and Grace (2002):
G
sh
= 0.0225 ln(,
susp
) 0.1093 (65)
where the cross-sectional average suspension density,
,
susp
, is in units of kg/m
3
. The remaining steps depend
on the size of the unit and the vertical length of the
heat transfer surface.
For large units (e.g., D > 1 m) and heat transfer
surfaces of height 1.5 m or more,
9a. Evaluate the emulsion temperature as
T
e
=
T
b
R
56
T
w
R
34
R
34
R
56
with R
34
=
R
3
R
4
R
3
R
4
and R
56
=
R
5
R
6
R
5
R
6
(66)
10a. Calculate the bulk-to-emulsion and emulsion-
to-wall heat transfer coecients as
h
rad
be
=
2o(T
4
b
T
4
e
)
(2,e
sus
1)(T
b
T
e
)
(67)
h
rad
ew
=
4o(T
4
e
T
4
w
)
(1,e
sus
1,e
w
1)(T
e
T
w
)
(68)
The coecient of 2 in equation (67) allows for the
radiation shielding eect of the diuse boundary
between the bulk suspension and the emulsion layer
facing the wall across the thin gas gap.
11a. We can now estimate the total heat transfer
coecient as
h
tot
= (h
gc
h
sr
)(1 f )
f
1
G
sh
C
pp
h
rad
be
1
(k
g
,o
g
) h
rad
ew
(69)
Figure 32 Predicted heat transfer vs. experimental data. Dashed lines indicate 25% deviations.
As shown in Fig. 32 (see also Golriz and Grace, 2002),
this approach gives good predictions for available heat
transfer data from large commercial units.
For cases where the surface is shorter than about 1.5
m or where there are appreciable changes in properties
such as r
susp
or d(z) over the heat exchange surface, both
T
e
and h
tot
vary with height. To nd local and average
heat transfer coecients, it is then necessary to integrate
over the height. For example, a heat balance over a slice
of height dz of emulsion (covered portion) for a riser of
circular cross section with ,
g
C
pg
_,
p
C
pp
yields
u
c
D
2
D
2
w
4
,
p
C
pp
(1 c)
dT
e
dz
= D
k
g
o
g
h
rad
ew
_ _
(T
e
T
w
)
D
w
G
sh
C
pp
h
rad
be
_ _
(T
b
T
e
)
(70)
where D
w
= D2o(z), with o(z) given by Eq. (24).
Beginning at the top of the heat exchange surface
where the boundary condition T
e
= T
b
is imposed,
one can integrate this equation downwards to the bot-
tom of the surface. Numerical integration is required
for small risers, not only because of the nonlinearity in
T
e
of the radiation terms in Eqs. (67) and (68), but also
because of variations of ,
susp
and o(z) with z. For the
special case where all radiation terms are unimportant
and where there is negligible variation of other terms
with height, Eq. (70) can be integrated to give
T
e
= T
eo
T
b
T
eo
( )e
z,z
th
(71)
where
T
eo
=
D(k
g
,o
g
)T
w
D
w
G
sh
C
pp
T
b
D(k
g
,o
g
) D
w
G
sh
C
pp
(72)
and
z
th
=
,
p
C
pp
(1 c
e
)v
pw
(D
2
D
2
w
)
4(Dk
g
,o
g
D
w
G
sh
C
pp
)
(73)
The voidage in the emulsion layer is estimated from an
equation proposed by Glicksman (1997):
c
e
= 1 1.23(1 c
av
)
0.54
(74)
while the particle velocity of descent in the wall layer,
v
pw
, is calculated from Eq. (23). Once T
e
is evaluated,
the local time mean heat ux to the wall at the corre-
sponding level is evaluated from
q
/
(z) = h
gc
h
sr
_ _
(1 f ) T
b
T
w
( )
k
g
,o
g
_ _
h
rad
ew
_ _
f (T
e
T
w
)
(75)
The local ux may then be integrated over the height to
give the total heat transfer coecient.
This approach is rational and reasonably accurate
except at the leading edge, but it makes a number of
assumptions. More comprehensive models are avail-
able (e.g., Xie et al., 2003) that adopt more rigorous
approaches, but these require considerably more com-
putational eort and are beyond the scope of this
chapter.
8 MASS TRANSFER
Mass transfer between gas and particles aects gas
solids contact eciencies in CFB risers. The mass
transfer from a single particle to the suspension in
CFB risers has been studied based on the sublimation
of naphthalene spheres (Halder and Basu, 1988; Li et
al., 1998), dehydration of 2-propanol (Masai et al.,
1985), adsorption of CCl
4
, naphthalene, H
2
S, and
NO (Kwauk et al., 1986; van der Ham et al., 1991,
1993; Vollert and Werther, 1994), and heat transfer
between a heat pulse and suspension (Dry et al.,
1987). For one-dimensional steady-state plug ow of
the gas, a mass balance of the adsorbed species in a
dierential volume element of the reactor (Kwauk et
al., 1986; Vollert and Werther, 1994) yields
U
g
dC
dz
a(1 c
av
)k
m
(C C
s
) = 0 (76)
where C
s
is the concentration of the tracer on particle
surface and a is the surface area per unit volume of the
particles (= 6,d
p
for spherical particles). If both c
av
and k
m
are assumed to be constant over a short dis-
tance dz, Eq. (77) can be integrated to obtain
k
m
=
d
p
U
g
6(1 c
av
)z
ln
C C
s
C
0
C
s
_ _
(77)
Typical axial proles of the lmmass transfer coef-
cient (Kwauk et al., 1986) show that k
m
initially
increases with height as voidage increases, suggesting
improved gas/solids contacting as the dense bed gives
way to a more dilute region. This k
m
is much lower
than for single particles (Venderbosch et al., 1998), as
estimated using the well-known RanzMarshall equa-
tion, and is generally of the same order or even below
that in gassolids bubbling uidized beds (Halder and
Basu, 1988; Vollert and Werther, 1994). A particle sur-
face eciency, dened as the eective particle surface-
to-volume ratio when the lmmass transfer coecient
is calculated based on the RanzMarshall correlation,
was only 0.01 to 0.13% for FCC particles and
decreased with increasing solids concentration (Dry et
al., 1987).
Relatively low gassolids contact eciency has also
been obtained in CFB risers based on the ozone
decomposition reaction (Jiang et al., 1991; Kagawa et
al., 1991; Sun and Grace, 1992; Ouyang et al., 1993).
The low contact eciency must be mainly attributed to
the formation of particle clusters and the coreannulus
ow structure in the riser.
To investigate mass transfer between clusters and
gas/dilute suspension, Li et al. (1998) constructed arti-
cial clusters by attaching small naphthalene balls to
metal wires and weaving them into dierent shapes.
The mass transfer coecient, k
m
, based on a single clus-
ter, was in good agreement with the RanzMarshall
equation for single particles, and k
m
increased with
decreasing cluster voidage; k
m
was found to be a func-
tion of cluster shape with higher values for paraboloid
and spherical clusters than for a cylindrical shape. A
nonuniform radial distribution of clusters in the riser
also led to a lower mass transfer coecient. For a
given number of particles in the riser, the gassolids
mass transfer rate increased as the fraction of dispersed
particles increased. Zethrus (1996) found that a simple
heterogeneous model that accounts for particle clusters
in the bottomdense region and coreannulus owstruc-
ture in the upper dilute region can reasonably predict
the gasparticle contact eciency in CFB risers.
9 CIRCULATING FLUIDIZED BEDS AS
CHEMICAL REACTORS
9.1 Introduction and Key Considerations
As noted in Sec. 2, most applications of CFB technol-
ogy involve chemical processes where the principal
reactions require solid particles, either as catalyst or
as reactant, and a gas. The major factors aecting
the performance of CFB reactors are related to the
ndings covered above:
There are pronounced radial (or lateral) gradients in
suspension density, with a greater concentration
of particles near the wall than near the axis of the
riser.
Because of this greater particle concentration near
the wall, concentrations of gaseous reactants
tend to be lower at the wall than in the interior
of the riser as shown in the examples of Fig. 33.
For the fast uidization ow regime, particles travel
downward near the wall, engendering axial dis-
persion of both solids and gas.
Figure 33 Nonuniform radial proles of concentration determined experimentally in CFB reactors: (a) concentration of ozone
undergoing decomposition in 0.25 mdia 10.5 m tall riser (Ouyang et al., 1995a); (b) concentration of NO
x
in
0.15 0.15 m7.3 m CFB pilot scale combustor (Brereton et al., 1995).
The riser is subject to signicant axial gradients in
solids holdup, with the extent of variation depen-
dent on such factors as supercial gas velocity
(U
g
), net solids circulation ux (G
s
), and riser
overall height/diameter (H,D) or height/width
(H,W) ratio.
The region at the bottom of the riser may operate as
a turbulent or bubbling bed, with a gradual
transition to fast uidization, dense suspension
upow, or dilute pneumatic conveying.
For some processes, e.g., CFB combustion, extra
gas is introduced at nozzles on the walls well
above the bottom primary distributor.
The exit geometry at the top of the riser inuences
solids reection and holdup in the upper part of
the reactor, aecting solids ow patterns, overall
solids holdup, and hence mixing.
Some relevant considerations with respect to the solids
recirculation system are
The eciency of cyclones (or other gassolids
separation devices where applicable) may greatly
aect the overall process performance.
Additional reaction (desired or undesirable) may
occur in the separators and in the return loop.
Table 9 Examples of Reactions Considered in CFB Reactor Studies and Dimensions of Columns Used in the
Experimental Studies
Reaction
M= model,
E = Exptl References Riser dimensions
A. Catalytic Reactions
Claus reaction M Puchyr et al., 1996, 1997 NA
FischerTropsch process E Shingles & McDonald, 1988 0.1 m id 13 m, then scaled up by
factors of 500 & 2.5
Fluid catalytic cracking M, E Fligner et al., 1994 0.3 m id 12 m
M Gao et al., 1999 NA
Methanol-to-olens M Schoenfelder et al., 1994 NA
M Gupta et al., 1999 NA
Maleic anhydride from n-butane M Pugsley et al., 1992 NA
E, M Golbig and Werther, 1996 21 mm i.d.
Oxidative coupling of CH
4
M Pugsley et al., 1996 NA
Ozone decomposition E Jiang et al., 1991 0.10 m id 6.3 m
M, E Ouyang & Potter, 1993, 1995a,b 0.25 m id 10.9 m
M, E Schoenfelder et al., 1996 0.4 m id 15.6 m
M, E Mao et al., 1999 0.25 m id 10.5 m
E Bolland & Nicolai, 1999 411 mm id 8.5 m
Partial oxidation of methane
to syngas
M Pugsley & Malcus, 1997 NA
Steam methane reforming M Matsukata et al., 1995 NA
B. Gas-Solid Reactions
Calcination E Legros et al., 1992 0.15 msq. 7.3 m
Combustion M, E Li et al., 1995 0.20 m square
M, E Talukdar & Basu, 1995a,b 0.20 msq. 6.5 m
0.51 msq. 14.3 m
3.96 msq. 21.4 m
M, E Mattisson & Lyngfelt, 1998 2.5 m
2
13.5 m
11 m
2
17 m
56 m
2
33.5 m
M, E Hartge et al., 1999 1.6 msq. 13.5 m
M, E Torii et al., 1999 0.4 msq. 20 m
Gasication M, E Jennen et al., 1999 0.3 m id 8 m
Roasting E Luckos & den Hoed, 2001 150 mm id 6.0 m
NA = not applicable.
Table 9 lists reactions/processes featured in CFB
reactor studies or in the development of reactor mod-
els, together with relevant references. The ozone
decomposition reaction is of no commercial interest,
but it is convenient for tests since it is essentially rst
order, irreversible, and able to proceed at room tem-
perature, and since the concentration of ozone can be
readily analyzed at low partial pressures. The other
reactions are all of commercial interest.
Understanding the hydrodynamics, mixing, and
transfer processes in CFB systems is essential to devis-
ing successful reactor models. In preparing models,
choices must be made. Overly simple models fail to
capture critical aspects of reactor behavior, whereas
fully comprehensive models that attempt to include
every facet that could aect the performance are likely
to be reaction-specic, time-consuming, and dicult to
apply. Models commonly contain a series of unproven
assumptions and/or empirical constants. Models of
intermediate complexity that are mechanistic, captur-
ing the major relevant features, are often the most use-
ful compromises for engineering design, optimization,
and control. In the longer term, more sophisticated
models, e.g., those based on multiphase computational
uid dynamic (CFD) codes, are likely to have increas-
ing impact, but (as noted above) they have not yet been
proven to give reliable predictions. Here we consider
models in order of increasing sophistication, abbreviat-
ing and extending the review of Grace and Lim (1997).
9.2 One-Dimensional Models
One-dimensional models constitute the simplest type
of CFB reactor model. Radial (or lateral) gradients
are completely ignored, with the gas and particle com-
position, as well as all hydrodynamic variables, treated
as invariant across the entire cross section at any level.
Some such models also assume that the hydrodynamics
are uniform in the axial direction, while others account
in some manner for the usual decrease in suspension
density with height and/or the presence of a relatively
dense zone, possibly subject to a dierent ow regime,
at the bottom. Key gas mixing and hydrodynamic
assumptions in one-dimensional models are summar-
ized in Table 10. Gas mixing assumptions cover a wide
spectrum from plug ow to relatively well mixed.
One-dimensional models are overly simplied
except to provide rough estimates. In practice, the pro-
nounced lateral gradients in CFB risers have important
consequences for reactions. For example, high particle
concentrations in the outer region can lead to elevated
local conversions of gas species toward the wall. When
there is downow of solids at the wall (i.e., fast uidi-
zation), concentration gradients can be pronounced,
and substantial backmixing occurs adjacent to the
wall. In CFB reactors with secondary gas addition
above the bottom of the reactor, gradients associated
with the wall jets can persist over considerable heights,
possibly even to the top of the reactor, with lateral
dispersion too slow to obliterate plumes originating
from the secondary jets. Lateral temperature gradients
can also be signicant.
Models that ignore axial gradients are further
removed from reality and are likely to require correc-
tion factors (e.g., Ouyang et al., 1993) to t experimen-
tal data. Models that account for axial variations in
hydrodynamic variables have found a variety of ways
of doing so, e.g.,
Table 10 One-Dimensional Steady-State CFB Reactor Models
Authors Axial dispersion
+
Axial hydrodynamic gradients
Hastaoglu et al., 1988 PF None
Gianetto et al., 1990 PF, ADPF None
Ouyang et al., 1993, 1995b PF, PM None
Marmo et al., 1996 PF, ADPF& (PMPF) Solids holdup decays exponentially
Lee & Hyppanen, 1989 PF c
av
= f (z)
Pagliolico et al., 1992 PF c
av
= f (z)
Weiss & Fett, 1986 TIS c
av
= f (z)
Muir et al., 1997 TIS c
av
= f (z) (each cell uniform)
Arena et al., 1995 TIS PF Two separate zones
Zhang et al., 1991 BB TIS Two separate zones
Jiang et al., 1991 TIS PF 5 compartments corresponding to regions
bounded by ring baes
+
ADPF = axially dispersed plug ow; BB = bubbling bed; PF = plug ow; PM= perfect mixing; TIS = well-
mixed tanks in series.
Making voidage a function of z, using empirical
evidence, as covered in Sec. 4
Introducing well-mixed compartments in series,
each compartment having a dierent solids con-
centration, in order to follow the concentration
prole in an approximate stepwise manner
Assuming ow regime transitions at one or more
levels in the riser, with dierent mixing represen-
tations above and below the transition level(s)
While accounting for axial gradients is helpful, neglect-
ing lateral or radial gradients seriously compromises
the ability of one-dimensional models to represent
CFB reactors.
9.3 CoreAnnulus Models
In coreannulus models, the relatively dense outer
region that is subject to solids downow on a time
mean basis, is treated as the annulus region, while
the more dilute inner dilute-upow region comprises
the core. Individually, the core and annulus are each
usually assumed to be one-dimensional, with radially
uniform voidages (e
a
and e
c
for the annulus and core,
respectively). This type of model was rst introduced
for gas mixing (Brereton et al., 1988) and has been
extended to coreannulus reactor models in a number
of studies, as summarized in Table 11.
Some of these models again ignore axial gradients.
For an nth-order solid-catalyzed gas phase reaction
with all of the gas assumed to pass through the core
and isothermal conditions, mole balances lead
(Kagawa et al., 1991; Marmo et al., 1996) to the gov-
erning equations
Annulus:
2K
ca
R
c
(C
a
C
c
)
R
2
R
2
c
k
n
(1 c
a
) C
n
a
= 0
(78)
Core:
U
g
R
2
R
2
c
dC
c
dz

2K
ca
(C
a
C
c
)
R
c
k
n
(1 c
c
) C
n
c
= 0 (79)
K
ca
is a core-to-annulus (interregion) mass transfer
coecient, with all other mass transfer resistances
(e.g., from the bulk to the particle surfaces within
each region) neglected. The nal term in each of
these equations accounts for reaction, with. k
n
being
the nth-order rate constant. The relevant boundary
condition is C
c
= C
0
at z = 0, where C
0
is the inlet
concentration of the reacting species. For a rst-
order reaction, it is straightforward to derive an ana-
lytical solution,
Conversion =
C
0
C
cH
C
0
= 1
exp k
+
1
(1 c
a
)(1
2
c
)K
/
k
+
1
(1 c
a
)(1
c
) K
/
(1 c
c
)
2
c
_ _ _ _
(80)
Table 11 Key Features of Some CoreAnnulus CFB Reactor Models
Authors
Axial dispersion
Axial gradients
Interregion
mass transf.
Core radius/
column radius Core Annulus
Kagawa et al., 1991 PF Stagnant None Fitted 0.85
Bi et al., 1992 PF Stagnant Allows separately for
ne & coarse particles
Rapid Varies with height
Marmo et al., 1996 PF Stagnant PF CA: c
av
= f (z) Fitted Presumably tted
Patience & Chaouki, 1993 PF Stagnant None Fitted Fitted
Werther et al., 1992 PF Intermittent
up and down
None ka = 0.23 s
1
by tting
0.85
Ouyang & Potter, 1994 PF PF rel. to solids None Fitted NS
Talukdar et al., 1994 PF PF TIS (turbt. & FF zones) Varied, then tted NS
Schoenfelder et al., 1994 PF rel. to
solids
PF rel. to solids 4 zones in series From gas mixing
expts
Fitted
Kunii & Levenspiel, 1998 PF Stagnant CA1-D decay zone To be tted To be tted
Puchyr et al., 1997 PF PF rel. to
descending
solids
Ignored Modied Higbie
penetration theory
Werther correlation
CA = core/annulus; FF = fast uidization; NS = not specied; PF = plug ow; PM= perfect mixing; TIS = tanks-in-series.
where
c
= R
c
,R (= dimensionless core radius), k
+
1
=
k
1
H,U
g
(= dimensionless rst-order rate constant or
Damkohler number), and K
/
= 2K
ca
HR
c
,(U
g
R
2
)
(= dimensionless coreannulus interregion mass trans-
fer coecient). Equations (78) to (80) are formally
similar to the corresponding equations for the two-
phase bubbling bed reactor model (Grace, 1986c),
with the dense phase of that model replaced by the
CFB annulus, and the bubbles by the core region. It
is therefore straightforward to apply analytical solu-
tions for cases provided for the earlier modelhalf-
order and consecutive rst-order reactionsto the
CFB two-region model. For more complex kinetics,
numerical solutions are required.
Conversions are plotted in Fig. 34 for four dimen-
sionless interregion mass transfer coecients
(K
/
= 0.01, 0.1, 1.0, and 10) and two average riser voi-
dages (c
av
= 0.90 and 0.96). In each case the corre-
sponding regional voidages are obtained from
approximate relationships given by Kagawa et al.
(1991), i.e., c
a
= 2c
av
1 and c
c
= 0.4 0.6c
av
; an over-
all balance on solids holdup then requires
c
= 0.714.
As expected, the conversion increases with increasing
kinetic rate constant and with increasing interregion
mass transfer. Increasing the column height, decreasing
the supercial gas velocity, or decreasing the voidage of
either region is also predicted to improve the conver-
sion. Kagawa et al. (1991) found that K
ca
= 0.001 m/s
gave the best t to experimental concentration proles.
Other workers (e.g., White et al., 1992; Zhao, 1992;
Patience and Chaouki, 1993; Schlichtaerle et al., 2001)
have assigned larger values of K
ca
, typical values being
of the order of 0.01 to 0.1 m/s. An analogy with gas
liquid annular ow (Senior and Brereton, 1992) gives
coecients of the order of 0.01 m/s. Use of a modied
Higbie penetration theory leads to similar values. The
interregion mass transfer coecient is important in
coreannulus models, but there are no reliable methods
for estimating the coecient. A value of 0.020.05 m/s
appears to be reasonable for most purposes.
As indicated in Table 11, models that use a core
annulus structure in combination with axial variation
in hydrodynamic properties dier widely in their
assumptions. Some (e.g., Marmo et al., 1996) impose
a separate region at the bottomto account for turbulent
or bubbling uidization in the bottom zone. Others
insert one or more additional zones, e.g., to account
for smoother transitions from the distributor region
to a coreannular structure, or to account for exit
eects. Some of the models fromthe literature featuring
multiple zones are shown schematically in Fig. 35.
There are several ways of allowing for axial voidage
variations. In some cases, e.g., Pugsley and Berruti,
1996; Gupta and Berruti, 1998; Gupta et al., 1999, a
Figure 34 Conversion vs. dimensionless rate constant from simple one-dimensional, two-region, axially uniform model for rst-
order chemical reaction with dierent overall voidages of 0.90 and 0.96 and dierent values of the dimensionless coreannulus
interregion mass transfer coecient.
series of equations are written that combine hydrody-
namic balances with empirical correlations to give
methods for predicting the evolution of voidages and
ows as a function of the axial coordinate, z. This
approach also incorporates correlations for a slip fac-
tor, discussed in Sec. 4.6.2 above.
9.4 More Sophisticated Models
The coreannulus models treated in the previous sec-
tion improve on the one-dimensional models covered
in the preceding section by making some allowance for
the dierence in behavior between the relatively dense
wall region and the dilute core of fast uidized beds.
However, the hydrodynamics are represented in a rela-
tively crude manner. As illustrated in Fig. 33, experi-
mental results show that reactant concentration varies
continuously across the entire cross section of the riser,
rather than there being a sharp discontinuity at a core
annulus boundary. Hence models are needed that pro-
vide for continuous variation across the riser, or at
least a greater number of intervals in the lateral direc-
tion. Such models include the following:
The model of Werther et al. (1992) still assumes a
coreannulus structure, but allows for radial dis-
persion in the central core, while disregarding
gradients in the annular wall zone.
Amos et al. (1993) included radial dispersion in the
core zone, but the core occupied the entire cross
section, with the solid particles recycling in a
separate region beyond the walls of the vessel.
A cluster/gas model (Fligner et al., 1994) with two
phasesa cluster phase containing all the parti-
cles as spherical clusters, each of voidage c
mf
, and
a gas phase devoid of particles. Mass transfer is
assumed to be controlled by the resistance at the
outer surface of the clusters.
Kruse et al. (1995) extended the model of
Shoenfelder et al. (1994) to include terms for
radial dispersion in both the core and the annu-
lus, with the radial dispersion coecient assumed
to be identical in both zones. These models were
further extended by Shoenfelder et al. (1996a) to
provide for continuous variations throughout the
entire cross section.
Ju (1995) developed a Monte Carlo model for a
CFB combustor with the riser divided into 40
cells, 20 in the core and 20 in the annulus.
Particles are introduced and tracked one by
one, with particles in the core only able to
move upwards or sideways subject to the laws
of chance, while those in the annulus can only
move downwards or sideways. Some particles
reaching the top are assumed to be reected
back down the riser. Particles devolatilize and
Figure 35 Schematic representation of several multizone models: (a) coreannulus freeboard approach of Kunii and
Levenspiel (1997); (b) four-zone approach of Neidel et al. (1995); (c) ve-zone approach of Schoenfelder et al. (1994).
then undergo combustion, allowing the heat
release pattern to be approximated by tracking
as few as 100 particles.
The probabilistic model of Abba et al. (2002)
extends the generalized bubbling turbulent
model of Thompson et al. (1999) to cover the
fast uidization ow regime, allowing for smooth
transition from bubbling through turbulent
uidization to fast uidization. A coreannulus
structure is assumed at the fast uidization
terminus, with hydrodynamic measurements of
regime transitions providing estimates of the
relative probabilities of each separate ow
regime.
In addition, there are several CFB reactor models that
are unrelated to coreannulus models. These are espe-
cially appropriate when the solids concentration and
gas velocity are high enough that the dense suspension
upow regime is reached, since, in the absence of
downow at the wall, there is no reason to be bound
by a coreannulus representation. Among these mod-
els are the following:
Cell-based models (e.g., Zhang et al., 1991;
Hyppanen et al., 1993; Muir et al., 1997) in
which mass and thermal balances allow changes
in concentration and energy to be tracked as a
function of height and time. Several of these
models also consider the recirculation loop in
which reactants and energy are recirculated and
reintroduced near the bottom of the riser.
Several attempts (e.g., Gao et al., 1999) have been
made to extend CFD (computational uid
dynamic) models (see Sec. 4.6.3 above) to include
chemical reaction terms.
Note that some of the cited models are intended to
provide dynamic predictions that can be used for con-
trol purposes or to simulate startup, shutdown, or the
response to upsets of CFB reactors.
Virtually all of the above models assume radial
symmetry in risers of circular cross section. However,
there are various sources of asymmetry in practice:
The solids return system almost invariably returns
solids to one side at the base of the reactor.
Similarly the solids feed system is generally
asymmetric, feeding solids to one side of the
reactor, or, for large units, to a limited number
of discrete feed positions on the periphery of the
reactor.
The draw-o system at the exit of the reactor is
again usually located at one side.
The cross section of many CFB reactors, e.g., of
atmospheric pressure combustors and calciners,
are rectangular. Corners are regions of additional
solids downow, as discussed in Sec. 4.5.1.
Secondary gas injection at some distance above the
bottom of the reactor, commonly found in CFB
combustors, causes nonsymmetric jets and
plumes that require some distance to dissipate.
9.5 Recommendations
While multiphase CFD models will no doubt make
valuable future contributions to predicting the perfor-
mance of CFB reactors, they are currently unable to
make reliable predictions. Simpler reactor models are
currently as reliable, while also being much easier to
use. For steady-state modeling of reactors operating in
the fast uidization ow regime, coreannulus models
coupled with mechanistic hydrodynamic relationships,
as in the approach of Gupta et al. (1999), provide
reasonable starting points. When the gas velocity is
low enough that the ow regime has not fully achieved
fast uidization, then the probabilistic approach pio-
neered by Abba et al. (2003) provides a useful tool.
When radial dispersion and gradients are important,
the approach developed by Shoenfelder et al. (1996a) is
recommended. This model also appears to provide a
rational method for dealing with cases where the ow
regime is dense suspension upow rather than fast ui-
dization.
NOMENCLATURE
A
c
= cross-sectional area of column, m
2
Ar = Archimedes number, ,
g
(,
p
,
g
)d
3
p
g,j
2
g
a = coreannulus interfacial area per unit volume,
m
2
/m
3
a. b = constants in Eq. (59), W,m
2-3n
K kg
n
and
W,m
2
K, respectively
C = tracer concentration, mol/m
3
for gas and
kg,m
3
for solids
C
0
= tracer concentration at injection point, mol/m
3
for gas and kg/m
3
for solids
C = cross-sectional average tracer concentration,
mol/m
3
C
c
= tracer gas or reactant concentration in annulus
region, mol/m
3
C
a
= tracer gas or reactant concentration in core
region, mol/m
3
C
i
= initial tracer concentration, mol/m
3
C
pc
= specic heat of cluster, J/kg K
C
pg
= specic heat of gas, J/kg K
C
pp
= specic heat of particles, J/kg K
C
s
= tracer concentration on surface of tracer
particles, mol/m
3
D = column diameter, equivalent diameter, or
width of riser, m
D
gb
= axial gas backmixing coecient, m
2
/s
D
gd
= axial gas dispersion coecient, m
2
/s
D
ge
= eective gas dispersion coecient, m
2
/s
D
gr
= radial gas dispersion coecient, m
2
/s
D
pr
= radial solids dispersion coecient, m
2
/s
D
pz
= axial solids dispersion coecient, m
2
/s
d
p
= mean particle diameter, m
d
+
p
= dimensionless particle diameter, (= Ar
1,3
)
E( = dimensionless residence time distribution
function
e
c
= emissivity of cluster
e
p
= emissivity of particles
e
susp
= emissivity of bulk suspension
e
w
= emissivity of wall surface
Fo = Fourier number = t
c
,R
2
w
c
f = fraction of total time during which wall
surface is covered by clusters
f
a
= mass fraction of tracer in the annulus
f
c
= mass fraction of tracer in the core
g = acceleration due to gravity, m/s
2
G
a
= downward solids ux in annular region,
kg,m
2
s
G
c
= upward solids ux in core region, kg/m
2
s
G
s
= solids net circulation ux or solids
entrainment ux, kg/m
2
s
G
s.max
= maximum solids circulation ux, kg/m
2
s
G
s.CA
= saturation carrying capacity, kg/m
2
s
H = total height of riser, m
h
cc
= heat transfer coecient due to cluster
convection, W/m
2
K
h
cg
= heat transfer coecient due to gas convection
when local surface is not covered by a cluster,
W/m
2
K
h
rc
= heat transfer coecient due to cluster
radiation, W/m
2
K
h
rs
= heat transfer coecient due to radiation to
dispersed suspension, W/m
2
K
h
tot
= total heat transfer coecient, W/m
2
K
I
ac
= mass transfer ux from annulus to core per
unit length, kg/m
3
s
I
ca
= mass transfer ux from core to annulus per
unit length, kg/m
3
s
K
/
= dimensionless interregion mass transfer
coefficient = 2K
ca
HR
c
,(UR
2
)
K
ca
= core-to-annulus interregion mass transfer
coecient, m/s
k = thermal conductivity, W/m K
k
c
= eective thermal conductivity of cluster, W/
m K
k
g
= thermal conductivity of gas, W/m K
k
1
= rst-order order rate constant, s
1
k
+
1
= dimensionless rst-order rate constant =
k
1
H,U
k
m
= mass transfer coecient, m/s
k
p
= thermal conductivity of particles, W/m K
L = vertical distance between injection and
sampling points, m
L
c
= average length of travel of cluster along wall, m
M = radial gas velocity nonuniformity index
m = mass of particles, kg
n = order of reaction or exponent in Eq. (59)
P = pressure, Pa
Pe
ge
= axial gas Peclet number, (= U
g
L,D
ge
)
Pe
gr
= radial gas Peclet number, (= U
g
D,D
gr
)
Pe
pr
= radial particle Peclet number (= U
g
D,D
pr
)
Pe
pz
= axial particle Peclet number (= U
g
L,D
pz
)
R = radius or hydraulic radius of column, m
r = radial coordinate, m
R
c
= radius of dilute core region, m
Re
p
= particle Reynolds number, (= ,
g
U
g
d
p
,j
g
)
R
w
= thermal gas lm conduction
resistance = o
g
,k
g
. K/W
T = emptying time, s
T
b
= bulk suspension temperature, K
T
c
= cluster temperature, K
T
w
= wall temperature, K
t = time, s
t
c
= residence time of cluster at wall, s
U
CA
= type A (accumulative) choking velocity, m/s
U
CC
= type C (classical) choking velocity, m/s
U
DSU
= transition from fast uidization to dense
suspension upow, m/s
U
g
= supercial gas velocity, m/s
U
mf
U
mp
= supercial gas velocity corresponding to
minimum pressure drop, m/s
U
se
= onset velocity of fast uidization at which
signicant solids entrainment occurs, m/s
U
slip
= slip velocity between gas and particles, m/s
U
tr
= transport velocity, m/s
U
+
= dimensionless supercial velocity,
(= Re
g
,Ar
1,3
)
u
c
= cluster velocity, m/s
u
g
= interstitial gas velocity, m/s
u
gc
= average gas velocity in core region, m/s
u
t
= terminal settling velocity of individual
particles, m/s
v
p
= particle axial velocity, m/s
v
pa
= average particle velocity in the annulus, m/s
v
pc
= average particle velocity in the core, m/s
v
pw
= downwards particle velocity at wall, m/s
W
0
= total solids inventory in a uidized bed, kg
x = distance from wall, m
z = vertical coordinate measured from bottom of
riser, m
z
i
= axial location of tracer injection, m
Greek letters
c
= ,
c
C
pc
k
c
. W
2
s,m
2
K
4
o
g
= thickness of gas lm between cluster and wall
surface, m
c
= dimensionless core radius = R,R
c
c = voidage
[c] = overall average voidage for the riser as a
whole
cc = local time average voidage
c
a
. c
c
= voidage of annulus, core
c
av
= cross-sectional average voidage at a given
height
c
cl
= cluster voidage
c
CA
= voidage at U
CA
c
c
= emulsion layer voidage
c
mf
= bed voidage at minimum uidization
c
s
= solids volume fraction
c
sav
= cross-sectional average solids fraction
c
sd
= solids volume fraction is dense bottom region
c
+
s
= solids volume fraction at top exit
j
g
= gas viscosity, kg/m s
,
b
= average bulk bed density, kg/m
3
,
c
= density of cluster, kg/m
3
,
g
= gas density, kg/m
3
,
p
= particle density, kg/m
3
,
susp
= suspension density = ,
p
(1 c) ,
g
c. kg,m
3
= dimensionless time, (= t,t or t,t
m
)
t = mean residence time, s
o = StefanBoltzmann constant,
5.67 10
8
Wm
2
K
4
o
t
= variance of tracer concentration, s
2
REFERENCES
Abba IA, Grace JR, Bi HT, Thompson ML. An integrated
approach to uidized bed reactor modeling, AIChE J, in
press, 2003.
Adams CK. Gas mixing in fast uidized beds. In: Basu P,
Large JF, eds. Circulating Fluidized Bed Technology II.
Ambler PA, Milne BJ, Berruti F, Scott DS. Residence time
distribution of solids in a circulating uidized bed: experi-
mental and modeling studies. Chem Eng Sci
45:21792186, 1990.
Amos G, Rhodes MJ, Mineo H. Gas mixing in gassolids
risers. Chem Eng Sci 48:943949, 1993.
Andersson B-A
. Eect of bed particle size on heat transfer in

circulating uidized bed boilers. Powder Technol
87:239248, 1996.
Andersson B-A
, Leckner B. Experimental methods of esti-

mating heat transfer in circulating uidized bed boilers.
Int J Heat and Mass Transfer 35:33533362, 1992.
Andersson B-A
, Brink K-E, Gustafsson L. Eect of furnace

size on CFB wall heat transfer. In: Kwauk M, Li J, eds.
Circulating Fluidized Bed Technology V. Beijing: Science
Press. 1996, pp 539544.
Aquillon J, Shakourzadeh K, Guigon P. A new method for
local solid concentration measurement in circulating ui-
Arastoopour H, Yang Y. Experimental studies on dilute gas
and cohesive particles ow behavior using a laser-Doppler
anemometer. In: Potter OE, Nicklin DJ, eds. Fluidization
VII. New York: Engineering Foundation, 1992,
pp 723730.
Arena U, Cammarota A, Pistone L. High velocity uidiza-
tion behavior of solids in a laboratory scale circulating
bed. In: Basu P, ed. Circulating Fluidized Bed
Technology. Toronto: Pergamon Press, 1986, pp 119125.
Arena U, Cammarota A, Marzocchella A, Massimilla L.
Solids ow structures in a two-dimensional riser of a
circulating uidized bed. J Chem Eng Japan 22:236241,
1989.
Arena U, Marzocchella A, Massimilla L, Malandrino A.
Hydrodynamics of circulating uidized beds with risers
of dierent shape and size. Powder Technol 70:237247,
1992.
Arena U, Chirone R, DAmore M, Miccio M, Salatino P.
Some issues in modelling bubbling and circulating
uidized bed coal combustors. Powder Technol
82:301316, 1995.
Avidan AA. Fluid catalytic cracking. In: Grace JR, Avidan
AA, Knowlton TM, eds. Circulating Fluidized Beds.
London: Chapman and Hall, 1997a, pp 466-487.
Avidan AA, Yerushalmi J. Solids mixing in an expanded top
uid bed. AIChE J 31:835841, 1985.
Azzi M, Turlier P, Large JF, Bernard JR. Use of a momen-
tum probe and gamma-densitometry to study local
properties of fast uidized beds. In: Basu P, Horio M,
Hasatani M, eds. Circulating Fluidized Bed Technology
III. Oxford: Pergamon Press, 1991, pp 189194.
Bader R, Findlay J, Knowlton TM. Gas solid ow patterns
in a 30.5-cm-diameter circulating uidized bed. In: Basu
P, Large JF, eds. Circulating Fluidized Bed Technology
II. Oxford: Pergamon Press, 1988, pp. 123137.
Bai D, Kato K. Quantitative estimation of solids holdups at
dense and dilute regions of circulating uidized beds.
Bai D, Jin Y, Yu ZQ, Yao WH. A study on the performance
characteristics of the circulating Fluidized Bed. Chem
React Eng and Technol (in Chinese). 3:2432 1987.
Bai D, Jin Y, Yu ZQ, Zhu JX. The axial distribution of the
cross-sectionally averaged voidage in fast uidized beds.
Powder Technol 71:5158, 1992a.
Bai D, Yi J, Jin Y, Yu ZQ. Residence time distributions of
gas and solids in a circulating uidized bed. In: Potter OE,
Nicklin DJ, eds. Fluidization VII, New York: Engineering
Foundation, 1992b, pp 195202.
Bai D, Nakagawa N, Shibuya E, Kinoshita H, Kato K. Axial
distribution of solid holdups in binary solids circulating
uidized beds. J Chem Eng Jap 27:271275, 1994.
Bai D, Issangya AS, Grace JR. Characteristics of gas-ui-
dized beds in dierent ow regimes. Ind Eng Chem
38:803811, 1999.
Balasubramanian N, Srinivasakannan C. Hydrodynamic
aspects of a circulating uidized bed with internals. Ind
Eng Chem Res 37:25482552, 1998.
Basu P. Heat transfer in high temperature fast uidized beds.
Chem Eng Sci 45: 10, 31233136, 1990.
Basu P. Combustion of coal in circulating uidized bed boi-
lers: a review. Chem Eng Sci 54:55475557, 1999.
Basu P, Konuche F. Radiative heat transfer from a fast ui-
dized bed. In: Basu P, Large JF, eds. Circulating Fluidized
Bed Technology II. Oxford: Pergamon Press, 1988, pp
245254.
Basu P, Nag PK. Heat transfer to walls of a circulating ui-
dized bed furnace. Chem Eng Sci 51:126, 1996.
Beaud F, Louge M. Similarity of radial proles of solid
volume fraction in a circulating uidized bed. In: Large
JF, Lague rie C, eds. Fluidization VIII. New York:
Berruti F, Chaouki J, Godfroy L, Pugsley TS, Patience GS.
Hydrodynamics of circulating uidized bed risers: a
review. Can J Chem Eng 73:579602, 1995.
Bi HT. A discussion on the transport velocity. Chem Eng
Comm, 189:942958, 2002.
Bi HT, Fan LS. Regime transitions in gassolid circulating
uidized beds. Paper #101e, AIChE Annual Meeting, Los
Angeles, Nov. 1722, 1991.
Bi HT, Fan LS. On the existence of turbulent regime in gas
solid uidization. AIChE J 38:297301, 1992.
Bi HT, Grace JR. Flow regime maps for gassolids uidiza-
tion and upward transport. Int J Multiphase Flow
21:12291236, 1995.
Bi HT, Grace JR. Flow patterns in high-velocity uidized
beds and pneumatic conveying. Can J Chem Eng
77:223230, 1999.
Bi HT, Zhu JX. Static instability analysis of circulating ui-
dized beds and the concept of high-density risers. AIChE
J 39:12721280, 1993.
Bi HT, Jiang P, Jean RH, Fan LS. Coarse-particle eects in a
multisolid circulating uidized bed reactor for catalytic
reactions, Chem Eng Sci 47: 31133124, 1992.
Bi HT, Grace JR, Zhu JX. On types of choking in pneumatic
systems. Int J Multiphase Flow 19:10771092, 1993.
Bi HT, Grace JR, Zhu JX. Regime transitions aecting gas
solids suspensions and uidized beds. Chem Eng Res Des
73:154161, 1995.
Bi HT, Zhou J, Qin SZ, Grace JR. Annular wall layer thick-
ness in circulating uidized bed risers. Can J Chem Eng
74:811814, 1996.
Bi HT, Ellis N, Abba IA, Grace JR. A state-of-the-art review
of gassolids turbulent uidization. Chem Eng Sci
55:47894825, 2000.
Bierl TW, Gajdos LJ, McIver AE, McGovern JJ. Studies in
support of recirculating uidized bed reactors for the
processing of coal. DOE Rept. Ex-C-78-01-2449, 1980.
Blumel WP, Kaferstein P, Rummel A, Morl P.
Wirbelschichtsysteme. VGB Conference, 1992.
Bolland O, Nicolai R. Describing mass transfer in circulating
uidized beds by ozone decomposition, AIChE Symp Ser
95(321):5260, 1999.
Bolton LW, Davidson JF. Recirculation of particles in fast
uidized beds. In: Basu P, Large JF, eds. Circulating
Fluidized Bed Technology II. Oxford: Pergamon Press,
1988, pp 139146.
Bowen BD, Fournier M, Grace JR. Heat transfer in mem-
brane waterwalls. Int J Heat Mass Transf 34:10431057,
1991.
Brereton CMH. Fluid mechanics of high velocity uidized
beds. PhD diss., University of British Columbia,
Vancouver, 1987.
Brereton C. Combustion Performance. In: Grace JR, Avidan
Chapman and Hall, London. 1997, pp 369416.
Brereton CMH, Grace JR. Microstructural aspects of the
behavior of circulating uidized beds. Chem Eng Sci
48:25652572, 1993.
Brereton CMH, Grace JR. End eects in circulating uidized
bed hydrodynamics. In: Avidan AA, ed. Circulating
Fluidized Bed Technology IV. New York: AIChE, 1994,
pp 137144.
Brereton CHM, Grace JR, Yu J. Axial gas mixing in a
circulating uidized bed. In: Basu P, Large JF, eds.
Circulating Fluidized Bed Technology II. Oxford:
Brereton CMH, Lim CJ, Grace JR, Luckos A, Zhu J. Pitch
and coke combustion in a circulating uidized bed. Fuel
74:14151423, 1995.
Brewster MQ. Eective absorptivity and emissivity of parti-
culate media with application to uidized bed. Trans
ASME 108:710713, 1986.
Brobecker V, Salvaterra A, Ocone R, Geldart D. Solid ux
measurements in a circulating uidized bed riser. In:
Kwauk M, Li J, Yang WC, eds. Fluidization X. New
Bu J, Zhu J. Inuence of ring-type internals on axial pressure
distribution in circulating uidized bed. Can J Chem Eng
77:2634, 1999.
Burkell JJ, Grace JR, Zhao J, Lim CJ. Measurement of solids
circulation rates in circulating uidized beds. In: Basu P,
Large JF, eds. Circulating Fluidized Bed Technology II.
Burki V, Hirchberg B, Tuzla K, Chen JC. Thermal develop-
ment for heat transfer in circulating uidized beds. AIChE
Annual Meeting, St. Louis, 1993.
Caloz Y, Reh L, Cahen C, Evrard R, Piedfer O. Local solids
velocities and their uctuations in CFB units of dierent
sizes. In: Werther J, ed. Circulating Fluidized Bed
TechnologyVI. Frankfurt: DECHEMA, 1999, pp849854.
Cao C, Bai D, Jin Y, Yu Z. Mechanism of heat transfer
between an immersed vertical surface and suspension in
a circulating uidized bed. In: Kwauk M. Fluidization 94
Science and Technology. Beijing: Science Press, 1994,
pp 180187.
Chen J, Cao H, Liu T. Catalyst regeneration in uid catalytic
cracking. In: Kwauk M, ed. Fast Fluidization. Adv Chem
Eng 20:389419, 1994.
Cheng Y, Wei F, Yang G, Jin Y. Inlet and outlet eects on
ow patterns in gassolid risers. Powder Technol
98:151156, 1998.
Chesonis DC, Klinzing GE, Shah TT, Dassori CG.
Hydrodynamics and mixing of solids in a circulating ui-
dized bed. Ind Eng Chem Res 29:17851792, 1990.
Cho YJ, Namkung W, KimSD, Park S. Eect of secondary air
injection on axial solid holdup distribution in a circulating
uidized bed. J Chem Eng Japan 27:158164, 1996.
Contractor RM. Duponts CFB technology for maleic anhy-
dride. Chem Eng Sci 54: 56275632, 1999.
Coronella CJ, Deng J. A novel method for isokinetic mea-
surement of particle ux within the riser of a circulating
Couturier M, Doucette B, Stevens D, Poolpol S, Razbin V.
Temperature, gas concentration and solid mass ux pro-
les within a large circulating uidized bed combustor. In:
Anthony EJ, ed. Proc 11
th
Int Conf Fluidized Bed
Combustion. New York: ASME, 1991, pp 107114.
Couturier MF, Steward FR, Poolpol S. Experimental deter-
mination of heat transfer coecients in a 72 MW
th
circu-
lating uidized bed boiler. In: Rubow LN, ed. Proc 12th
Int Conf on Fluidized Bed Combustion. New York:
ASME, 1993, pp 12151222.
Cui H, Mostou N, Chaouki J. Comparison of measurement
techniques of local particle concentration for gassolid
uidization. In: Kwauk M, Li J, Yang WC, eds.
Fluidization X. New York: Engineering Foundation,
2001, pp 779786.
Davidson JF. Circulating uidised bed hydrodynamics.
de Diego LF, Gaya n P, Ada nez J. Modelling of the ow
structure in circulating uidized beds. Powder Technol
85:1927, 1995.
Dean R, Mauleon JL, Letzsch W. TOTAL introduces new
FCC process. Oil Gas J 168176, Oct 11, 1982.
Diguet S. PhD diss., LInstitut National Polytechnique de
Toulouse, France, 1996. Data provided in Smolders and
Baeyens (2000).
Divilio RJ, Boyd TJ. Practical implication of the eect of
solids suspension density on heat transfer in large-scale
CFB boilers. In: Avidan AA, ed. Circulating Fluidized
Bed Technology IV. NewYork: AIChE, 1994, pp 334339.
Drahos J, Cermak J, Guardani R, Schugerl K.
Characterization of ow regime transition in a circulating
Dry RJ, Beeby CJ. Applications of CFB technology to gas
solid reactions. In: Grace JR, Avidan AA, Knowlton TM,
eds. Circulating Fluidized Beds. London: Chapman and
Hall, 1997, pp 441465.
Dry RJ, White CC. Gas residence-time characteristics in a
high-velocity circulating uidized bed of FCC catalyst.
Dry RJ, Christensen IN, White CC. Gas-solids contact e-
ciency in a high-velocity uidized bed. Powder Technol
52:243250, 1987.
Du B, Wei F, Wang F. Eect of particle size on solids mixing
in a FCC riser. In: Werther J, ed. Circulating Fluidized
Bed Technology VI. Frankfurt: DECHEMA, 1999, pp
405410.
Fabre A, Molodtsof Y, Koniuta A. Flow structure charac-
terization in a 1 sq m CFB. In: Kwauk M, Li J, eds.
Circulating Fluidized Bed Technology V. Beijing:
Science Press, 1997, pp 4853.
Fang Y, Huang J, Zhang J, Wang Y. The study of hydro-
dynamics of circulating uidized bed under ambient and
high temperature condition. In: Werther J, ed. Circulating
Fluidized Bed Technology VI. Frankfurt: DECHEMA,
1999, pp 3338.
Fang ZH, Grace JR, Lim CJ. Radiative heat transfer in
circulating uidized beds. J Heat Transfer 117: 963968,
1995.
Flamant G, Lu JD, Variot B. Radiation heat transfer in
uidized beds: a comparison of exact and simplied
approaches. J Heat Transfer, 116: 652659, 1994.
Flamant G, Variot B, Golriz MR, Lu JD. Radiative heat
transfer in a pilot scale circulating uidized bed boiler,
In: Kwauk M, Li J, eds. Circulating Fluidized Beds
Technology V. Beijing: Science Press, 1996, pp 563568.
Fligner M, Schipper PH, Sapre AV, Krambeck FJ. Two
phase cluster model in riser reactors: impact of radial
density distribution on yields. Chem Eng Sci
49:58135818, 1994.
Fraley LD, Lin YY, Hsiao KH, Slobakken A. Heat transfer
coecient in a circulating uidized reactor. ASME paper
83-HT-92, 1983.
Furchi JCL, Goldstein L, Lombardi G, Mohseni M.
Experimental local heat transfer in circulating uidized
bed. In: Basu P, Large JF, eds. Circulating Fluidized
Bed Technology II. Oxford: Pergamon Press, 1988, pp
1329.
Fusey I, Lim CJ, Grace JR. Fast uidization in a concentric
circulating bed. In: Basu P, ed. Circulating Fluidized Bed
Technology. Oxford: Pergamon Press, 1986, pp 409416.
Galtier PA, Pontier RJ, Patureaux TE. Near full-scale cold
ow model for the R2R catalytic cracking process. In:
1989, pp 1724.
Gan N, Jiang DZ, Bai DR, Jin Y, Yu ZQ. Concentration
proles in fast uidized bed with blu-body (in Chinese).
J Chem Eng Chinese Univ 3:273277, 1990.
Gao J, Xu C, Lin S, Yang G, Guo Y. Advanced model for
turbulent gassolid ow and reaction in FCC riser
reactors. AIChE J 45:10951113, 1999.
Gartside RJ. QCa new reaction system. In: Grace JR,
Shemilt LW, Bergougnou MA, eds. Fluidization VI.
Gayan P, de Diego LF, Adanez J. Radial gas mixing in a fast
Gianetto A, Pagliolico S, Rovero G, Ruggeri B. Theoretical
and practical aspects of circulating uidized bed reactors
for complex chemical systems. Chem Eng Sci
45:22192225, 1990.
Gidaspow D. Multiphase Flow and Fluidization: Continuum
and Kinetic Theory Descriptions. London: Academic
Press, 1994.
Glicksman LR. Circulating uidized bed heat transfer.
In: Basu P, Large JF, eds. Circulating Fluidized
Bed Technology II. Oxford: Pergamon Press, 1988,
pp 1329.
Fluidized Beds. London: Chapman and Hall, 1997,
pp 261311.
Glicksman LR, Westphalen D, Brereton CMH, Grace JR.
Verication of the scaling laws for circulating uidized
beds. In: Basu P, Horio M, Hasatani M, eds.
Circulating Fluidized Bed Technology III. Oxford:
Glicksman LR, Hyre M, Woloshun K. Simplied scaling
relationships for uidized beds. Powder Technol
77:177199, 1993.
Godfroy L, Larachi F, Chaouki J. Position and velocity of a
large particle in gas/solid riser using the radioactive par-
ticle tracking technique. Can J Chem Eng 77:253261,
1999a.
Godfroy L, Patience GS, Chaouki J. Radial hydrodynamics
in risers. Ind Eng Chem 38:8189, 1999b.
Golbig KC, Werther J. Selective synthesis of maleic anhy-
dride in a riser-regenerator system. In: Kwauk M, Li J,
eds. Circulating Fluidized Bed Technology V. Beijing:
Golriz MR. Thermal and uid-dynamic characteristics of
circulating uidized bed boilers, Lic Eng thesis,
Chalmers University of Technology, Go teborg, Sweden,
1992.
Golriz MR. Inuence of wall geometry on temperature dis-
tribution and heat transfer in circulating uidized bed
boilers, In: Avidan AA, ed. Circulating Fluidized Bed
Technology IV. New York: AIChE, 1994, pp 693700.
Golriz MR. An experimental correlation for temperature
distribution at the membrane wall of CFB boiler. In:
Heinschel KJ, ed. Proc 13th Int Conf Fluidized Bed
Combustion. New York: ASME, 1995, pp 499507.
Golriz MR, Grace JR. Predicting heat transfer in large-scale
CFB boilers, In: Grace JR, Zhu J, de Lasa HI, eds.
Circulating Fluidized Bed Technology VII. Ottawa:
CSChE, 2002, pp 121128.
Golriz MR, Leckner B. Experimental studies of heat transfer
in a circulating uidized bed boiler. Proc Int Conf
Engineering Application Mechanics. Sharif University of
Technology, Vol. 3, 1992, pp 167174.
Golriz MR, Sunde n B. A method for temperature measure-
ments in circulating uidized bed combustors. Exp
Thermal Fluid Science 9:274282, 1994a.
Golriz MR, Sunde n B. An experimental investigation of
thermal characteristics in a 12 MWth CFB boiler. Exp
Heat Transfer J 7:217233, 1994b.
gassolid and other two-phase suspensions. Can J Chem
Eng 64:353363, 1986a.
Grace JR. Heat transfer in circulating uidized beds. In:
Basu P, ed. Circulating Fluidized Bed Technology.
Toronto: Pergamon Press, 1986b, pp 6380.
Grace JR. Modelling and simulation of two-phase uidized
bed reactors. In: de Lasa HI, ed. Chemical Reactor
Design and Technology. Dordrecht: Martinus Nijho,
1986c, pp 245289.
Grace JR. Heat transfer in high velocity uidized beds. In:
Hetsroni G, ed. Proc 9th International Heat Transfer
Conference, Jerusalem, 1990, pp 329339.
Grace JR. Inuence of riser geometry on particle and uid
dynamics in circulating uidized beds risers. In: Kwauk
M, Li J, eds. Circulating Fluidized Bed Technology.
Beijing: Science Press, 1996, pp 1628.
Grace JR, Lim KS. Reactor modelling for high-velocity ui-
dized beds. In: Grace JR, Avidan AA, Knowlton TM,
eds. Circulating Fluidized Beds. London: Chapman and
Hall, 1997, pp 504524.
Grace JR, Avidan AA, Knowlton TM. eds. Circulating
Fluidized Beds. London: Chapman and Hall, 1997.
Grace JR, Issangya AS, Bai DR, Bi HT, Zhu JX. Situating
the high-density circulating uidized beds. AIChE J
45:21082116, 1999.
Graf R. Circulating uidized beds in the ue gas scrubbing
developments, applications and operating experiences in
the years 1970 to 2000. In: Werther J, ed. Circulating
1999, pp 601607.
Graham RG, Freed BA, Bergougnou MA. Scale-up and
commercialization of rapid biomass pyrolysis for fuel
and chemical production. In: Klass DL, ed. Energy
from Biomass and Wastes XIV. Chicago: Inst Gas
Technol, 1991, pp 10911104.
Grassler T, Wirth KE. Radial and axial proles of solids
concentration in a high-loaded riser reactor. In: Werther
J, ed. Circulating Fluidized Bed Technology VI.
Frankfurt: DECHEMA, 1999, pp 6570.
Grith AE, Louge MY. The scaling of cluster velocity at the
wall of circulating uidized bed risers. Chem Eng Sci
53:24752477, 1998.
Gupta SK, Berruti F. Modeling considerations for large scale
high density risers. In: Fan LS, Knowlton TM, eds.
1998, pp 205212.
Gupta SK, Berruti F. Evaluation of the gassolid suspension
density in CFB risers with exit eects. Powder Technol
108:2131, 2000.
Gupta SK, Pugsley T, Berruti F. A process simulator for
circulating uidized bed chemical reactors. In: Werther
Hage B, Werther J. The guarded capacitance probea tool
for the measurement of solids ow patterns in laboratory
and industrial uidized bed combustors. Powd Tech
93:235245, 1997.
Halder PK, Basu P. Mass transfer from a coarse particle to a
fast bed of ne solids. AIChE Symp Ser 84(262):5867,
1988.
Han GY, Cho. Radiative heat transfer in a circulating ui-
dized bed coal combustor. Powder Technol 102:266273,
1999.
Harris AT, Thorpe RB, Davidson JF. The measurement of
particle residence time distributions in circulating ui-
dized beds. In: Grace JR, Zhu J, de Lasa HI, eds.
CSChE, 2002, pp. 145152.
Harris BJ, Davidson, JF. Velocity proles, gas and solids, in
fast uidized beds. In: Potter OE, Nicklin DJ, eds.
1992, pp 219226.
Harris BJ, Davidson JF. A core/annulus deposition model.
In: Avidan AA, ed. Circulating Fluidized Bed Technology
Harris BJ, Davidson JF, Xue Y. Axial and radial variation of
ow in circulating uidized bed risers. In: Avidan AA, ed.
AIChE, 1994, pp 103110.
Hartge EU, Werther J. Gas distributors for circulating ui-
dized bed combustors. In: Fan LS, Knowlton TM, eds.
1998, pp 213220.
Hartge EU, Li Y, Werther J. Analysis of the local structure
of the two phase ow in a fast uidized bed. In: Basu P,
ed. Circulating Fluidized Bed Technology. Toronto:
Pergamon, 1986a, pp 153160.
Hartge EU, Li Y, Werther J. Flow structures in fast ui-
dized beds. In: Ostergaard K, Sorensen K, eds.
1986b, pp 345352.
Hartge EU, Luecke K, Werther J. The role of mixing in the
performance of CFB reactorsCFB combustion as an
example. In: Werther J, ed. Circulating Fluidized Bed
Technology VI. Frankfurt: DECHEMA, 1999,
pp 411416.
Hartge EU, Rensner D, Werther J. Solids concentration and
velocity in circulating uidized beds. In: Basu P, Large JF,
eds. Circulating Fluidized Bed Technology II. Oxford:
Hastaoglu MA, Berruti F, Hassam MS. A generalized gas
solid reaction model for circulating uidized bedsan
application to wood pyrolysis. In: Basu P, Large JF,
Helmrich H, Schurgerl K, Janssen K. Decomposition of
NaHCO
3
in laboratory and bench scale circulating ui-
dized bed reactors. In: Basu P, ed. Circulating Fluidized
Bed Technology. Oxford: Pergamon Press, 1986, pp
161166.
Herb B, Dou S, Tuzla K, Chen JC. Solid mass uxes in
circulating uidized beds. Powder Technol 70:197205,
1992.
Herbert PM, Gauthier TA, Briens CL, Bergougnou MA.
Application of ber optic reection probes to the mea-
surement of local particle velocity and concentration in
gassolid ow. Powder Technol 80:243252, 1994.
Hiltunen M, Myo ha nen K. Extremely rapid cooling of hot
gas containing sticky components by the means of
Fluxow technology. In: Potter OE, Nicklin DJ, eds.
1992, pp 841848.
Hirschfelder H, Vierrath H. Electricity and syngas from bio-
mass and wastes applying CFB gasication. In: Werther J,
ed. Circulating Fluidized Bed Technology VI. Frankfurt:
DECHEMA, 1999, pp 459467.
Horio M, Kuroki H. Three-dimensional ow visualization of
dilutely dispersed solids in bubbling and circulating ui-
Horio M, Morishita K, Tachibana O, Murata N. Solid dis-
Technology II. Oxford: Pergamon Press, 1988, pp
147154.
Horio M, Ishii H, Kobukai Y, Yanmanishi N. A scaling law
for circulating uidized beds. J Chem Eng Japan
22:587592, 1989.
Horio M, Mori K, Takei Y, Ishii H. Simultaneous gas and
solid velocity measurements in turbulent and fast ui-
dized beds. In: Potter OE, Nicklin DJ, eds.
1992, pp 757762.
Huang W, Zhu J. An experimental investigation on solid
acceleration length in the riser of a long circulating ui-
dized bed. Chin J Chem Eng 9:7076, 2001.
Husain P, Weber P, Orth A, Eichberger H. Development and
experience with uidized bed based direct reduction pro-
cesses. In: Werther J, ed. Circulating Fluidized Bed
Technology VI. Frankfurt: DECHEMA, 1999, pp
593598.
Hyppanen T, Lee YY, Kettunen A, Riiali J. Dynamic simu-
lation of a CFB based utility power plant. Proc 12
th
Intern
Fluidized Bed Combustion Conf. New York: ASME,
1993, pp 11211127.
Hyre M, Glicksman LR. Experimental investigation of heat
transfer enhancements in circulating uidized beds. Proc
4th ASME/JSME Thermal Eng. Conf, Maui, Hawaii,
1995.
Issangya AS, Bai D, Bi HT, Zhu J, Grace JR. Suspension
densities in a high-density circulating uidized bed riser.
Chem Eng Sci 54:54515460, 1999.
Issangya AS, Bai DR, Grace JR, Zhu JX. Solids ux proles
in a high density circulating uidized bed riser. In: Fan
LS, Knowlton TM, eds. Fluidization IX. New York:
Issangya AS, Grace JR, Bai D, Zhu J. Further measurements
of ow dynamics in a high-density circulating uidized
bed riser. Powder Technol 111:104113, 2000.
Issangya AS, Grace JR, Bai D, Zhu J. Radial voidage varia-
tion in CFB risers. Can J Chem Eng 79:279286, 2001.
Jennen T, Hiller R, Ko neke D, Weinspach PM. Modeling of
gasication of wood in a circulating uidized bed. In:
Werther J, ed. Circulating Fluidized Bed Technology
VI. Frankfurt: DECHEMA, 1999, pp.431436.
Jestin L, Meyer P, Schmitt G, Morin JX. Transfert de cha-
leur dans le foyer du lit uidise circulant de 125 MW
e
de
Carlin. Saint-Denis: Elect. de France, 1992.
Jiang P, Fan LS. On the turbulent radial transfer of particles
in a CFB riser. In: Werther J, ed. Circulating Fluidized
Bed Technology VI. Frankfurt: DECHEMA, 1999,
pp 8388.
Jiang P, Inokuchi K, Jean R-H, Bi HT, Fan LS. Ozone
decomposition in a catalytic circulating uidized bed reac-
tor. In: Basu P, Horio M, Hasatani M, eds. Circulating
1991a, pp 557562.
Jiang P, Bi HT, Jean RH, Fan LS. Bae eects on perfor-
mance of catalytic circulating uidized bed reactors.
AIChE J 37:13921400, 1991b.
Jiang P, Bi HT, Liang SC, Fan LS. Hydrodynamic behavior
of circulating uidized bed with polymeric particles.
AIChE J 40:193206, 1994.
Jin X, Lu J, Liu Q, Li Y, Xing X, Yue G. Investigation on the
heat transfer in a CFB boiler. In: Werther J, ed.
Circulating Fluidized Bed Technology VI. Frankfurt:
DECHEMA, 1999, pp 355360.
Jin Y, Yu Z, Qi C, Bai D. The inuence of exit structures on
the axial distribution of voidage in fast uidized bed. In:
Kwauk M, Kunii D, eds. Fluidization 88 Science and
Technology. Beijing: Science Press, 1988, pp 165173.
Jin Y, Zheng Y, Wei F. State-of-the-art review of downer
reactors, In: Grace JR, Zhu J, de Lasa HI, eds.
CSChE, 2002, pp 4060.
Johnsson F, Svensson A, Leckner B. Fluidization regimes in
circulating uidized bed boilers. In: Potter OE, Nicklin
Johnsson F, Zhang W, Johnsson H, Leckner B. Optical and
momentum probe measurements in a CFB furnace. In:
Kwauk M, Li J, eds. Circulating Fluidized Bed
Johnsson F, Sterne us J, He Q, Leckner B. Fluidization
regimes in a CFB riserinuence of in-bed tubes. In:
Kwauk M, Li J, Yang WC, eds. Fluidization X. New
Johnsson F, Johansson A, Sterne us J, Leckner B, Hartge
EU, Budinger S, Fehr M, Werther J, Sekret R, Bis Z,
Nowak W, Noskievic P, Ochodek T, Gadowski J,
Jablonski J, Walkowiak R, Kallner P, Engel E,
Stro mberg L, Hyppa nen T, Kettunen A. An investigation
of in-furnace processes in a 235 MW
e
CFB boiler. In:
Grace JR, Zhu J, de Lasa HI, eds. Circulating Fluidized
Bed Technology VII. Ottawa: CSChE, 2002, pp 607614.
Johnsson H, Johnsson F. Measurements of local solids
volume-fraction in uidized bed boilers. Powder Technol
115:1325, 2001.
Ju DWC. Modelling of steady-state heat release, oxygen pro-
le and temperature prole in circulating uidized bed
combustors. MASc diss., University of British
Columbia, Vancouver, Canada, 1995.
Kagawa H, Mineo H, Yamazaki R, Yoshida K. A gassolid
contacting model for fast-uidized bed. In: Basu P, Horio
M, Hasatani M, eds. Circulating Fluidized Bed
Technology III. Oxford: Pergamon Press, 1991, pp
551556.
Karri SBR, Knowlton TM. The eect of pressure on CFB
riser hydrodynamics. In: Kwauk M, Li J, eds. Circulating
Fluidized Bed Technology V. Beijing: Science Press, 1997,
pp 103109.
Karri SBR, Knowlton TM. Flow direction and size segrega-
tion of annulus solids in a riser. In: Fan LS, Knowlton
TM, eds. Fluidization IX. New York: Engineering
Karri SBR, Knowlton TM. A comparison of annulus solids
ow direction and radial solids mass ux proles at low
and high mass uxes in a riser. In: Werther J, ed.
DECHEMA, 1999, pp 7176.
Kashiwa BA, Yang WC. Computational uid dynamics for
the CFBR: Challenges that lie ahead. In: Grace JR, Zhu J,
deLasa H, eds. Circulating Fluidized Bed Technology VII.
Ottawa: CSChE, 2002, p. 2739.
Kato K, Takarada T, Tamura T, Nishino K. Particle hold-up
distribution in a circulating uidized bed. In: Basu P, Horio
M, Hasitani M, eds. Circulating Fluidized Bed Technology
Kehlenbeck R, Yates J, diFelice R, Hofbauer H, Rauch R.
Novel scaling parameter for circulating uidized beds.
AIChE J 47:582589, 2001.
Kim J. Hydrodynamic behavior of solid transport for a
closed loop circulating uidized bed with secondary air
injection. Korean J Chem Eng 16:840842, 1999.
Koenigsdor R, Werther J. Gassolids mixing and ow
structure modeling of the upper dilute zone of a circulat-
ing uidized bed. Powder Technol 82:317329, 1995.
Kojima T, Ishihara KI, Yang GL, Furusawa T.
Measurement of solids behavior in a fast uidized bed. J
Chem Eng Japan 22:341346, 1989.
Konrad K. Dense-phase pneumatic conveying: a review.
Kostazos AE, Weinstein H, Gra R. The eect of the loca-
tion of gas injection on the distribution of gas and catalyst
in a riser. In: Fan LS, Knowlton TM, eds. Fluidization
IX. New York: Engineering Foundation, 1998, pp
221228.
Krol S, de Lasa H. CREC-GS-optical probe for particle
cluster characterization. In: Fan LS, Knowlton TM, eds.
1998, pp 565572.
Kruse M, Schoenfelder H, Werther J. A two-dimensional
model for gas mixing in the upper dilute zone of a circu-
lating uidized bed. Can J Chem Eng 73:620634, 1995.
Kunii D, Levenspiel O. Circulating uidized-bed reactors.
Chem Eng Sci 52:24712482, 1997.
Kunii D, Levenspiel O. Conversion expressions for FF
reactors. In: Fan LS, Knowlton TM, eds. Fluidization
IX. New York: Engineering Foundation, 1998,
pp 677684.
Kuroki H, Horio M. The ow structure of a three-dimen-
sional circulating uidized bed observed by the laser sheet
technique. In: Avidan AA, ed. Circulating Fluidized Bed
Technology IV. Oxford: Pergamon Press, 1994, pp 7784.
Kwauk M. ed. Fast Fluidization. Vol. 20, Advances in
Chemical Engineering Series. San Diego: Academic
Press, 1994.
Kwauk M, Wang ND, Li YC, Chen BY, Shen ZY. Fast
uidization at ICM. In: Basu P, ed. Circulating
Fluidized Bed Technology. Oxford: Pergamon Press,
1986, pp 3362.
Lafaneche` re L, Jestin L. Study of a circulating uidized bed
furnace behavior in order to scale it up to 600 MW
e
. In:
Heinschel KJ, ed., Proc 13
th
Combustion. New York: ASME 2:971980, 1995.
Leckner B. Heat transfer in circulating uidized bed boilers.
Fluidized Bed Technology III. Toronto: Pergamon
Press. 1990, pp 2737.
Leckner B, Golriz MR, Zhang W, Andersson BA, Johnsson
F. Boundary layersrst measurements in the 12 MW
CFB research plant at Chalmers University. Anthony
EJ, ed. Proc 11
th
Fluidized Bed Comb Conf. New York:
ASME, 1991 pp 771778.
Lee YY, Hyppanen T. Coal combustion model for circulat-
ing uidized bed boilers. Proc 10
th
Intl Conf Fluidized
Bed Combustion. New York: ASME, 1989, pp 753764.
Lee YY. Design considerations for CFB boilers. In: Grace
JR, Avidan AA, Knowlton TM, eds. Circulating
417440.
Legros R, Lim CJ, Brereton CMH, Grace JR. Calcination of
lime mud in a circulating uidized bed. J Pulp Paper Sci
18:7173, 1992.
Leung LS. Vertical pneumatic conveying: a ow regime
diagram and a review of choking versus non-choking
systems. Powder Technol 25:185190, 1980.
Levenspiel, O. Chemical Reaction Engineering 2d ed. New
Levenspiel O, Turner JCR. The interpretation of residence-
time experiments. Chem Eng Sci 25:16051609, 1970.
Li H, Xia Y, Tung Y, Kwauk M. Micro visualization of two-
phase structure in a fast uidized bed. In: Basu P, Horio
177182.
Li J, Weinstein H. An experimental comparison of gas back-
mixing in uidized beds across the regime spectrum.
Chem Eng Sci 44:16971705, 1989.
Li J, Zhang X, Zhu J, Li J. Eects of cluster behavior on gas
solid mass transfer in circulating uidized beds. In: Fan
LS, Knowlton TM, eds. Fluidization IV. New York:
Li X, Luo Z, Ni M, Cen K. Modeling sulfur retention in
circulating uidized bed combustors. Chem Eng Sci 50:
22352242, 1995.
Li Y, Kwauk M. The dynamics of fast uidization. In: Grace
JR, Matsen JM, eds. Fluidization. New York: Plenum
Press, 1980, pp 537544.
Li Y, Wu P. Study on axial gas mixing in a fast uidized bed.
1991, pp 581586.
Li Y, Zhang X. Circulating uidized bed combustion. In:
Kwauk M, ed. Fast Fluidization. Adv Chem Eng
20:333388, 1994.
Lim KS, Zhou J, Finley C, Grace JR, Lim CJ, Brereton
CMH. Cluster descending velocity at the wall of circulat-
ing uidized bed risers. In: Kwauk M, Li J, eds.
Circulating Fluidized Bed Technology V. Beijing:
Lin W, Weinell CF, Hansen PFB, Dam-Johansen K.
Hydrodynamics of a commercial scale CFB boiler
study with radioactive particles. Chem Eng Sci
54:54955506, 1999.
Lints T, Glicksman LR. The structure of particle clusters
near the wall of a circulating uidized bed. AIChE
Symp Ser 89(296):3552, 1993.
Lischer DJ, Louge M. Optical ber measurements of particle
concentration in dense suspensions: calibration and simu-
lation. Applied Optics 31:51065113, 1992.
Liu JZ. Particle and gas dynamics of high density circulating
uidized beds. PhD diss., University of British Columbia,
Vancouver, 2001.
Liu JZ, Grace JR, Bi HT, Morikawa H, Zhu JX. Gas dis-
persion in fast uidization and dense suspension upow.
Chem Eng Sci 54:54515460, 1999.
Lockhart C, Zhu J, Brereton CMH, Lim CJ, Grace JR. Local
heat transfer, solid concentration and erosion around
membrane tubes in a cold model circulating uidized
bed. Int J Heat Mass Transfer, 38:24032410, 1995.
Louge M, Chang H. Pressure and voidage gradients in ver-
tical gassolid risers. Powder Technol. 60:197201, 1990.
Louge M, Lischer DJ, Chang H. Measurements of voidage
near the wall of a circulating uidized bed riser. Powder
Technol 62:269276, 1990.
Louge MY, Briscout V, Martin-Letellier S. On the dynamics
of pressurized and atmospheric circulating uidized bed
risers. Chem Eng Sci 54:18111824, 1999.
Luan W, Lim CJ, Brereton BD, Bowen BD, Grace JR.
Experimental and theoretical study of total and radiative
heat transfer in circulating uidized beds. Chem Eng Sci
54:37493764, 1999.
Luckos A, den Hoed P. Fluidized-bed roasting of ilmenite.
In: Kwauk M, Li J, Yang WC, eds. Fluidization X. New
Luo G, Yang QL. Axial gas dispersion in fast uidized beds.
Proc 5
th
National Fluidization Conference, Beijing, 1990,
pp 155158.
Malcus S, Pugsley TS. Investigation of the axial variation of
the CFB riser hydrodynamics by means of electrical capa-
citance tomography (ETC). In: Kwauk M, Li J, Yang
Mao QM, Li XG, Andrews JRG, Potter OE. Catalytic che-
mical reaction in CFB riser. In: Werther J, ed. Circulating
1999, pp 521526.
Marmo L, Manna L, Rovero G. Comparsion among several
predictive models for circulating uidized bed reactors.
In: Lague rie C, Large JF, eds. Fluidization VIII. New
Martin MP, Turlier P, Bernard JR. Gas and solids behavior
in cracking circulating uidized beds. Powder Technol
70:249258, 1992.
Masai M, Tanaka S, Tomomasa Y, Nishiyama S, Tsuruya S.
Reactant-catalyst contact in riser-tube reactor. Chem Eng
Comm 34:153159, 1985.
Mastellone ML, Arena U. The eect of particle size and
density on solids distribution along the riser of a cir-
culating uidized bed. Chem Eng Sci 54:53835391,
1999a.
Mastellone ML, Arena U. The inuence of particle size and
density on the radial gas mixing in the dilute region of the
circulating uidized bed. Can J Chem Eng 77:231237,
1999b.
Mathiesen V, Solberg T, Hjertager BH. Experimental and
computational study of multiphase ow behavior in a
circulating uidized bed. Int J Multiph Flow
26:387419, 2000.
Matisson T, Lyngfelt A. A sulphur capture model for circu-
lating uidized-bed boilers. Chem Eng Sci 53:11631173,
1998.
Matsen JM. Design and scale-up of CFB catalytic reactors.
In: Grace JR, Avidan AA, Knowlton TM, eds.
Circulating Fluidized Beds. London: Chapman and
Hall, 1997, pp 489503.
Matsen TM. Mechanisms of choking and entrainment.
Powder Technol 32:2133 1982.
Matsukata M, Matsushita T, Ueyama K. A circulating ui-
dized bed CH
4
reformer: performance of supported Ni
catalysts. Energy and Fuels 9:822828, 1995.
Matsumoto S, Marakawa M. Statistical analysis of the tran-
sition of the ow pattern in vertical pneumatic conveying.
Int J Multiphase Flow 13:123129, 1987.
Matsuno Y, Yamaguchi H, Oka T, Kage H, Higashitani K.
The use of optic probes for the measurement of dilute par-
ticle concentration: calibration and application to gas-ui-
dized bed carryover. Powder Technol 36:215221, 1983.
Mattmann W. Konvektiver Wa rmeu bergang in vertikalen
gas-feststo-stro mungen. Dr Ing diss., Universita t
Erlagen-Nu rnberg, 1991.
Mayer-Schwinning G, Herden H. Use of CFB technology for
cleaning o-gases from waste-to-energy plants. In:
VI. Frankfurt: DECHEMA, 1999, pp 633637.
Mickley HS, Fairbanks, DF. Mechanism of heat transfer in
uidized beds, AIChE J 1:374384, 1955.
Miller AL, Gidaspow D. Dense vertical gassolid ow in a
pipe. AIChE J 38:18011815, 1992.
Molerus O, Wirth KE. Heat Transfer in Fluidized Beds.
Monceaux L, Azzi M, Molodtsof Y, Large JF. Overall and
local characterization of ow regimes in a circulating ui-
dized bed. In: Basu P, ed. Circulating Fluidized Bed
Technology. Toronto: Pergamon Press, 1986, pp 185191.
Mori S, Kato K, Kobayashi E, Liu D, Hasatani M, Matsuda
H, Hattori M, Hirama T, Takeuchi H. Eect of apparatus
design on hydrodynamics of circulating uidized-bed.
AIChE Symp Ser 88(289):1725, 1992.
Muir JR, Brereton CMH, Grace JR, Lim CJ. Dynamic mod-
eling for simulation and control of a circulating uidized
bed combustor. AIChE J 43:11411152, 1997.
Nag PK, Moral MNA. Eect of probe size on heat transfer
at the wall in circulating uidized beds. J Inst Energy Res
14:965974, 1990.
Namkung W, Kim SD. Gas backmixing in a circulating ui-
Namkung W, Kim SD. Radial gas mixing in a circulating
Naruse I, Kumita M, Hattori M, Hasatani M.
Characteristics of ow behavior in riser of circulating ui-
dized bed with internal nozzle. In: Basu P, Horio M,
Neidel W, Gohla M, Borghardt R, Reimer H. Theoretical and
experimental investigation of mix-combustion coal/biofuel
in circulating uidized beds. In: Lague rie C, Large JF, eds.
Fluidization VIII Preprints, 1995, pp 573583.
Nicolai R, Reh L. Experimental study on cold-model CFB
uid dynamics. In: Kwauk M, Li J, eds. Circulating
Fluidized Bed Technology V. Beijing: Science Press,
1997, pp 5459.
Nieuwland JJ, Meijer R, Kuipers JAM, van Swaaij WPM.
Measurements of solids concentration and axial solids
velocity in gassolid two-phase ows. Powder Technol
87:127139, 1996.
Nucla Circulating Atmospheric Fluidized Bed
Demonstration ProjectFinal Report, DOE/MC/
251373046, 1991.
Ohki K, Shirai T. Particle velocity in uidized bed. In:
Keairns, DL ed. Fluidization Technology, Vol. 1,
Washington DC: Hemisphere, 1976, pp 95110.
Ouyang S, Lin J, Potter OE. Ozone decomposition in a 0.254
m diameter circulating uidized bed reactor. Powder
Technology 74:7378, 1993.
Ouyang S, Lin J, Potter OE. Circulating uidized bed as a
catalytic reactor: experimental study. AIChE J
41:15341542, 1995a.
Ouyang S, Li XG, Potter OE. Investigation of ozone decom-
position in a circulating uidized bed on the basis of core
annulus model. In: Lague rie C, Large JF, eds.
Fluidization VIII Preprints, 1995b, pp 457465.
Pagliolico S, Tiprigan M, Rovero G, Gianetto A. Pseudo-
homogeneous approach to CFB reactor design. Chem
Eng Sci 47:22692274, 1992.
Pa rssinen JH, Zhu J, Yan AJ, Bu JJ. Flow development in a
high-ux and long riser. In: Kwauk M, Li J, Yang WC,
eds. Fluidization X. New York: Engineering Foundation,
2001, pp 261268.
Patience GS, Chaouki J. Solids circulation rate determined
by pressure drop measurements. In: Basu P, Horio M,
Patience GS, Chaouki J. Gas phase hydrodynamics in the
riser of a circulating uidized bed. Chem Eng Sci
48:31953205, 1993.
Patience GS, Chaouki J. Solids hydrodynamics in the fully
developed region of CFB risers. In: Lague rie C, Large JF,
eds. Fluidization VIII Preprints, 1995, pp 3340.
Patience GS, Chaouki J, Kennedy G. Solids residence time
distribution in CFB reactors. In: Basu P, Horio M,
Patience GS, Chaouki J, Berruti F, Wong R. Scaling consid-
erations for circulating uidized bed risers. Powder
Technol. 72:3137, 1992.
Pienemann B, Stockhausen W, McKenzie L. Experience with
the circulating uid bed for gold roasting and alumina
calcination. In: Potter OE, Nicklin DJ, eds. Fluidization
VII. New York: Engineering Foundation, 1992,
pp 921928.
Plass L. Future R&D directions required in uidization and
uidparticle systems for sustainable development in che-
mical industries. In: Kwauk M, Li J, Yang WC, eds.
2001, pp 114.
Puchyr DMJ, Mehrotra AK, Behie LA, Kalogerakis N.
Hydrodynamic and kinetic modelling of circulating ui-
dized bed reactors applied to a modied Claus plant.
Chem Eng Sci 51:52515262, 1996.
Puchyr DMJ, Mehrotra AK, Behie LA, Kalogerakis N.
Modelling a circulating uidized bed riser reactor with
gassolids downow at the wall. Can J Chem Eng,
75:317326, 1997.
Pugsley TS, Berruti F. A predictive hydrodynamic model for
circulating uidized bed risers. Powder Technol 89:57-69,
1996.
Pugsley T, Lapointe D, Hirschberg B, Werther J. Exit eects
in circulating uidized bed risers. Can J Chem Eng
75:10011010, 1997.
Pugsley TS, Malcus S. Partial oxidation of methane in a
circulating uidized-bed catalytic reactor. Ind Eng Chem
Res 36:45674571, 1997.
Pugsley TS, Patience GS, Berruti F, Chaouki J. Modeling the
catalytic oxidation of n-butane to maleic anhydride in a
CFB reactor. Ind Eng Chem Res 31:26522660, 1992.
Qi CM, Fregeau JR, Farag IH. Hydrodynamics of the com-
bustion loop of N-circulating uidized bed combustion.
AIChE Symp Ser 88(289):2632, 1992.
Qin S, Liu G. Application of optical bers to measurement
and display of uidized systems. In: Kwauk M, Kunii D,
eds. Fluidization: Science and Technology. Beijing:
Raghunathan K, Ghosh-Dastidar A, Fan LS. High tempera-
ture reactor system for study ultrafast gassolid reactions.
Rev Sci Instrum 64(7):19891993, 1993.
Ran X, Wang Z, Wei F, Jin Y. Particle velocity and solids
fraction proles of swirling ow in a riser. In: Werther J,
ed. Circulating Fluidized Bed Technology VI. Frankfurt:
DECHEMA, 1999, pp 131136.
Reh L. Fluidized bed processing. Chem Eng Prog
67(2):5863, 1971.
Reh L. The circulating uid bed reactora key to ecient
gas/solid processing. In: Basu P, ed. Circulating Fluidized
Bed Technology. Toronto: Pergamon Press, 1986, pp
105118.
Reh L, Li J. Measurement of voidage in uidized beds by
optical probes. In: Basu P, Horio M, Hasatani M, eds.
Rhodes MJ. Modeling the ow structure of upward-owing
gassolid suspensions. Powder Technol 60:2738, 1990.
Rhodes MJ, Geldart D. The hydrodynamics of re-circulating
uidized beds. In: Basu P, ed. Circulating Fluidized Bed
Technology. Oxford: Pergamon Press, 1986, pp 193200.
Rhodes MJ, Laussmann P. Characterising non-uniformities
in gasparticle ow in the riser of a circulating uidized
Rhodes MJ, Hirama T, Cerutti G, Geldart D. Non-unifor-
mities of solids ow in the risers of circulating uidized
beds. In: Grace JR, Bergougnou MA, Shemilt LW, eds.
1989, pp 7380.
of the 305 mm diameter riser of a cold model circulating
uidized bed. In: Basu P, Horio M, Hasatani M, eds.
Pergamon Press, 1991a, pp 171176.
Rhodes MJ, Zhou S, Hirama T, Cheng H. Eects of operat-
ing conditions on longitudinal solids mixing in a cir-
culating uidized bed riser. AIChE J 37:14501458,
1991b.
Rhodes MJ, Wang XS, Cheng H, Hirama T, Gibbs BM.
Similar proles of solids ux in circulating uidized-bed
risers. Chem Eng Sci 47:16351643, 1992a.
of a circulating uidized bed. Powder Technol
70:207214, 1992b.
Rhodes MJ, Sollaart M, Wang XS. Flow structures in a fast
uid bed. Powder Technol 99:194200, 1998.
Saberi B, Shakourzadeh K, Guigon P. Local solid concentra-
tion measurement by ber optics: application to circulat-
ing uidized beds. Trans I Chem Eng 76:748752, 1998.
Schlichthaerle P, Werther J. Axial pressure proles and solids
concentration distributions in the CFB bottom zone.
Chem Eng Sci 54:54855493, 1999.
Schlichthaerle P, Hartge EU, Werther J. Interphase mass
transfer and gas mixing in the bottom zone of a circulat-
ing uidized bed. In: Kwauk M, Li J, Yang WC, eds.
2001, pp 549556.
Schmidt HW. Advanced circulating uid bed technology for
alumina calcination. In: Werther J, ed. Circulating
1999, pp 587592.
Schoenfelder H, Werther J, Hinderer J, Keil F. A multi-stage
model for the circulating uidized bed reactor. AIChE
Symp Ser 90(301):92104, 1994.
Schoenfelder H, Kruse M, Werther J. Two-dimensional
model for circulating uidized bed reactors. AIChE J
42:18751888, 1996.
Schouten JC, Zijerveld RC, van den Bleek CM. Scale-up of
bottom-bed dynamics and axial solids-distribution in cir-
culating uidized beds of Geldart-B particles. Chem Eng
Sci 54:21032112, 1999.
Schu gerl K. Experimental comparison of mixing processes in
two- and three-phase uidized beds. Drinkenburg AAH,
ed. Proceedings Intern Symp on Fluidization.
Amsterdam: Netherlands University Press, 1967, pp
782794.
Schuurmans HJA. Measurements in a commercial catalytic
cracking unit. Ind Eng Chem Proc Des Dev 19:267271,
1980.
Senior RC, Brereton CHM. Modelling of circulating ui-
dized bed solids ow and distribution. Chem Eng Sci
47:281296, 1992.
Senior RC, Grace JR. Integrated particle collision and tur-
bulent diusion model for dilute gassolid suspensions.
Shen X, Zhou N, Xu Y. Experimental study of heat transfer
in a pressurized circulating bed. In: Basu P, Horio M,
Synthol CFB reactors. In: Basu P, Large JF, eds.
Simons SJR, Seville JPK, Clift R, Gilboy WB, Hosseini-
Ashra ME. Application of gamma-ray tomography to
gas uidized and spouted beds. In: Beck MS,
Campogrande E, Morris M, Williams M, Waterfall RC,
eds. Tomographic Techniques for Process Design and
Operation. Southampton: Computational Mechanics
Publications, 1993, pp. 227238.
Sinclair JL. Hydrodynamic modeling. In: Grace JR, Avidan
London: Chapman and Hall, 1997, pp 149180.
Sinclair JL, Jackson R. Gas-particle ow in a vertical pipe
with particle-particle interactions. AIChE J 35:1473-1486,
1989.
Smith TN. Limiting volume fractions in vertical pneumatic
transport. Chem Eng Sci 33:745749, 1978.
Smolders K, Baeyens J. Overall solids movement and solids
residence time distribution in a CFB-riser. Chem Eng Sci
55:41014116, 2000.
Sobocinski DJ, Young N, de Lasa HI. New ber-optic
method for measuring velocites of strands and solids
hold-up in gassolids downow reactors. Powder
Technol 83:111, 1995.
Squires AM. Origins of the fast uidized bed. In: Kwauk M,
ed. Fast Fluidization. Adv Chem Eng 20:137, 1994.
Srivastava A, Agrawal K, Sundaresan S, Reddy Karri SB,
Knowlton TM. Dynamics of gas-particle ow in
1998.
Sterne us J, Johnsson F, Leckner B, Palchonok GI. Gas and
solids ow in circulating uidized bedsdiscussion on
turbulence. Chem Eng Sci 54:53775382, 1999.
Steynberg AP, Shingles T, Dry ME, Jager B, Yukawa Y.
Sasol commercial scale experience with Synthol FFB
and CFB catalytic FischerTropsch reactors. In: Basu P,
Horio M, Hasatani M, eds. Circulating Fluidized Bed
527532.
Stro mberg L. Fast uidized bed combustion of coal. Proc 7
th
Int Conf Fluidized Bed Combustion. Philadelphia. 1982,
pp 11521163.
Sun B, Gidaspow D. Computation of circulating uidized-
bed riser ow for Fluidization VIII benchmark test. Ind
Eng Chen Res 38:787792, 1999.
Sun GL, Grace JR. Eect of particle size distribution in
dierent uidization regimes. AIChE J 38:716722, 1992.
Sun G, Chao Z, Fan Y, Shi M. Hydrodynamic behavior in
the bottom region of a cold FCC riser. In: Werther J, ed.
DECHEMA, 1999, pp 179184.
Takeuchi H, Hirama T. Flow visualization in the riser of a
circulating uidized bed. In: Basu P, Horio M, Hasatani
M, eds. Circulating Fluidized Bed Technology III.
Takeuchi H, Hirama L, Chiba T, Biswas J, Leung LS. A
quantitative regime diagram for fast uidization.
Tallon S, Davies CE. Attenuation of sound in beds of parti-
culate materials and application to measurement of ow
rate. In: Fan LS, Knowlton TM, eds. Fluidization IX.
Talukdar J, Basu P. Modelling of nitric oxide emission from
a circulating uidized bed furnace. In: Lague rie C, Large
JF, eds. Fluidization VIII Preprints, Vol. 2, 1995a, pp
829837.
Talukdar J, Basu P. A simplied model of nitric oxide emis-
sion from a circulating uidized bed combustor. Can J
Chem Eng 73:635643, 1995b.
Talukdar J, Basu P, Joos E. Senstitivity analysis of a perfor-
mance predictive model for circulating uidized bed boiler
furnaces. In: Avidan AA, ed. Circulating Fluidized Bed
Tanner H, Li J, Reh L. Radial proles of slip velocity
between gas and solids in circulating uidized beds.
AIChE Symp Ser 90(301):105113, 1994.
Thompson ML, Bi HT, Grace JR. A generalized bubbling/
turbulent uidized bed reactor model. Chem Eng Sci
54:21752185, 1999.
Torii I, Tagashira K, Arakawa Y, Arima K. One-dimen-
sional steady-state simulation model of CFBC. In:
VI. Frankfurt: DECHEMA, 1999, pp 437442.
Tung Y, Li Y, Kwauk M. Radial voidage proles in a fast
uidized bed. In: Kwauk M, Kunii D, eds. Fluidization
88: Science and Technology. Beijing: Science Press, 1988,
pp 139145.
van Bruegel JW, Stein JJM, deVries RJ. Isokinetic sampling
in a dense gassolids stream. Proc Instn Mech Engrs
184:1823, 1969.
van der Ham A, Prins G, van Swaaij WPM. Hydrodynamics
and mass transfer in a regularly packed circulating ui-
dized bed. In: Basu P, Horio M, Hasatani M, eds.
Pergamon, 1991, pp 605611.
van der Ham A, Prins GJ, van Swaaij WPM. Hydrodynamics
of a pilot-plant scale regularly packed circulating uidized
bed. AIChE Symp Ser 89(296):5372, 1993.
van der Meer EH, Thorpe RB, Davidson JF. The inuence of
exit geometry for a circulating uidised bed with a square
cross-sectional riser. In: Kwauk M, Li J, eds. Circulating
Fluidized Bed Technology V. Beijing: Science Press, 1997,
pp 575580.
van der Meer EH, Thorpe RB, Davidson JF. Dimensionless
groups for practicable similarity of circulating uidised
beds. Chem Eng Sci 54:53695376, 1999.
van Zoonen D. Measurements of diusional phenomena and
velocity proles in a vertical riser. Proc Symp Interaction
betw. Fluids and Particles. London: Instn Chem Engrs,
1962, pp 6471.
Venderbosch RH, Prins W, van Swaaij WPM. Inuence of
the local catalyst activity on the conversion in a riser reac-
tor. In: Fan LS, Knowlton TM, eds. Fluidization IX. New
Vollert J, Werther J. Mass transfer and reaction behavior of
a circulating uidized bed reactor. Chem Eng Technol
17:201209, 1994.
Wang Q, Zhou J, Tu J, Luo Z, Li X. Residence time in
circulating uidized bed. Proc Fluidized Bed
Combustion 1:16, 1992.
Wang T, Lin ZJ, Zhu CM, Liu DC, Saxena SC. Particle
velocity measurements in a circulating uidized bed.
AIChE J 39:14061410, 1993.
Wang XS, Gibbs, BM. Hydrodynamics of a circulating ui-
dized bed with secondary air injection. In: Basu P, Horio
Technology III. Oxford: Pergamon Press, 1991,
pp 225230.
Wang XS, Rhodes MJ, Gibbs BM. Solids ux distribution in
a CFB riser operating at elevated temperatures. In: Large
JF, Lague rie C, eds. Fluidization VIII. New York:
Watanabe T, Chen Y, Hasatani M, Xie Y, Naruse I. Gas
solid heat transfer in riser. In: Basu P, Horio M, Hasatani
Weeks ST, Rohrer JW. Commercial demonstration of bio-
mass gasication. Tappi J 80(5):147152, 1997.
Wei F, Yang GQ, Jin Y, Yu ZQ. The characteristic of cluster
in a high density circulating uidized bed. Can J Chem
Eng 73:650655, 1995a.
Wei F, Jin Y, Yu ZQ, Chen W, Mori S. Lateral and axial
mixing of the dispersed particles in CFB. J Chem Eng
Japan 28:506510, 1995b.
Wei F, Lu FB, Jin Y, Yu ZQ. Mass ux proles in a high
density circulating uidized bed. Powder Technol
91:189195, 1997.
Wei F, Lin H, Yi C, Wang Z, Jin Y. Proles of particle
velocity and solids fraction in a high-density riser.
Weimer AW, Gyure DC, Clough DE. Application of a
gamma-radiation density gauge for determining hydrody-
namic properties of uidized beds. Powder Technol
44:179194, 1985.
Weinstein H, Li J. An evaluation of the actual density in the
acceleration section of vertical risers. Powder Technol
57:7779, 1989.
Weinstein H, Feindt HJ, Chen L, Gra RA, Pell M,
Contractor M, Jordan SP. Riser gas feed nozzle cong-
uration eects on the acceleration and distribution of
solids. In: Large JF, Lague rie C, eds. Fluidization VIII.
Weinstein H, Shao M, Schnitzlein M. Radial variation in
solid density in high velocity uidization. In: Basu P, ed.
Circulating Fluidized Bed Technology. Toronto:
Weiss V, Fett FN. Modeling the decomposition of sodium
bicarbonate in a circulating uidized bed reactor. In: Basu
P, ed. Circulating Fluidized Bed Technology. Toronto:
Werdermann CC, Werther J. Heat transfer in large-scale cir-
culating uidized bed combustors of dierent sizes. In:
Werther J. Fluid mechanics of large-scale CFB units. In:
Werther J, Hirschberg B. Solids motion and mixing. In:
119148.
Werther J, Hartge EU, Kruse M. Gas mixing and interphase
mass transfer in the circulating uidized bed. In: Potter
OE, Nicklin DJ, eds. Fluidization VII. New York:
Werther J, Hartge EU, Rensner D. Measurement techniques
for gassolid uidized bed reactors. Int Chem Eng
33:1826, 1993.
Westphalen D, Glicksman L. Lateral solids mixing measure-
ments in circulating uidized beds. Powder Technol
82:153168, 1995.
White CC, Dry RJ, Potter OE. Modelling gas-mixing in a
9 cm diameter circulating uidized bed. In: Potter OE,
Nicklin DJ, eds. Fluidization VII. New York:
Win KK, Nowak W, Matsuda H, Hasatani M, Kruse M,
Werther J. Radial gas mixing in the bottom part of a
multi-solid uidized bed. J Chem Eng Japan 27:696698,
1994.
Winter F, Mickal V, Hofbauer H, Brunner C, Aichernig C,
Liegl J. Scale-up of CFB-uid dynamics for Group C par-
ticles from the laboratory-scale to the industrial-scale. In:
Werther J, ed. Circulating Fluidized Bed Technology VI.
Wirth KE, Gruber U. Fluid mechanics of circulating ui-
dized beds with small density ratio of solids to uid. In:
Kwauk M, Li J, eds. Circulating Fluidized Bed
Wirth KE, Seiter M, Molerus O. Concentration and velocity
of solids in areas close to the walls in circulating uidized
systems. VGB Kraftwerkstechnik 10:824828, 1991.
Wu RL, C.J. Lim J, Chaouki, JR Grace. Heat transfer from
a circulating uidized bed to membrane water-wall sur-
faces. AIChE J 33:18881893, 1987.
Wu RL, Grace JR, Lim CJ, Brereton CMH. Suspension-to-
surface heat transfer in a circulating uidized bed com-
bustor AIChE J 35:16851691, 1989.
Wu RL, Lim CJ, Grace JR, Brereton CMH. Instantaneous
local heat transfer and hydrodynamics in a circulating
uidized bed. Int J Heat Mass Transfer 34:20192027,
1991.
Xie D, Bowen BD, Grace JR, Lim CJ. Two-dimensional
model of heat transfer in circulating uidized beds.
Intern J Heat Mass Transf. In press, 2003.
Xu G, Sun G, Nomura K, Li J, Kato K. Two distinctive
variational regions of radial particle concentration pro-
les in circulating uidized bed risers. Powder Technol
101:91100, 1999.
Yang GL, Huang Z, Zhao L. Radial gas dispersion in a fast
uidized bed. In: Kunii D, Toei R, eds. Fluidization.
Cambridge: Cambridge University Press, 1983, pp
145152.
Yang H, Gautam M, Mei J. Gas velocity distribution in a
circular circulating uidized bed riser. Powder Technol
78:221229, 1994.
Yang WC. A mathematical denition of choking phenom-
enon and a mathematical model for predicting choking
velocity and choking voidage. AIChE J 21:10131021,
1975.
Yang WC. From pneumatic transport to circulating uidized
beds: the importance of operating maps. In: Kwauk M, Li
J, Yang WC, eds. Fluidization X. New York: Engineering
Yang YL, Jin Y, Yu ZQ, Wang ZW. Investigation on slip
velocity distributions in the riser of dilute circulating ui-
Yerushalmi J, Cankurt NT. Further studies of the regimes of
Yerushalmi J, Cankurt NT, Geldart D, Liss B. Flow regimes
in vertical gassolid contact systems. AIChE Symp Ser
174(176):112, 1978.
Yoshida K, Mineo H. High velocity uidization. In:
Doraiswamy LK, Mujumdar AS, eds. Transport in
Fluidized Particle Systems. Amsterdam: Elsevier, 1989,
pp 241285.
Younis HF, Glicksman LR, Hyre MR. Measurement
and prediction of the hydrodynamics of binary mix-
tures in a cold model pressurized circulating uidized
bed. In: Werther J, ed. Circulating Fluidized Bed
Technology VI. Frankfurt: DECHEMA, 1999,
pp 3945.
Yous Y, Gau G. Aerodynamique de le coulement vertical
de suspensions concentre es gaz-solideI. Re gimes
de coulement et stabilite aerodynamique. Chem Eng Sci
29:19391946, 1974.
Yu Z, Jin Y. Heat and mass transfer. Adv Chem Eng 20:
203237, 1994.
Zenz FA. Two-phase uidized-solid ow. Ind Eng Chem
41:28012806, 1949.
Zethrus B. A theoretical model for gasparticle contact
eciency in circulating uid bed risers. Powder Technol
88:133142, 1996.
Zhang DZ, VanderHeyden WB. High-resolution three-
dimensional numerical simulation of a circulating ui-
Zhang H, Johnston PM, Zhu JX, de Lasa HI, Bergougnou
MA. A novel calibration procedure for a ber optic solids
concentration probe. Powder Technol 100:260272, 1998.
Zhang L, Li TD, Zheng, QY, Lu CD. A general dynamic
model for circulating uidized bed combustion with wide
particle size distributions. Proc 11
th
International
Fluidized Bed Combustion Conf. New York: ASME,
1991, pp 12891294.
46:30453052, 1991.
Zhang W, Johnsson F, Leckner B. Characteristics of the
lateral particle distribution in circulating uidized bed
boilers. In: Avidan AA, ed. Circulating Fluidized Bed
Zhang W, Johnsson F, Leckner B. Fluid-dynamic boundary
layer in CFB boilers. Chem Eng Sci 50:201210, 1995.
Zhang W, Johnsson F, Leckner B. Momentum probe and
sampling probe for measurement of particle ow proper-
ties in CFB boilers. Chem Eng Sci 52:497509, 1997.
Zhang YF, Arastoopour H, Wegerer DA, Lomas DA,
Hemler CL. Experimental and theoretical analysis of gas
and particle dispersion in large scale CFB. In: Avidan
AA, ed. Circulating Fluidized Bed Technology IV. New
York: AIChE, 1993, pp 473478.
Zhao J. Nitric oxide emissions from circulating uidized bed
combustion, PhD diss., University of British Columbia,
Vancouver, Canada, 1992.
Zheng CG, Tung YK, Li HZ, Kwauk M. Characteristics of
fast uidized beds with internals. In: Potter OE, Nicklin
Zheng QY, Zhu DH, Feng JK, Fan JH. Some aspects of
hydrodynamics in N-shaped circulating uidized bed
combustor. Manaker AM, ed. Proc 10
th
Intern Conf
Fluidized Bed Combustion. New York: ASME, 1989,
Vol. 2, pp 157161.
Zheng QY, Xing W, Lou F. Experimental study on radial
gas dispersion and its enhancement in circulating ui-
dized beds. In: Potter OE, Nicklin DJ, eds.
1992, pp 285293.
Zheng Z, Zhu J, Grace JR, Lim CJ, Brereton CMH. Particle
motion in circulating and revolving uidized beds via
microcomputer-controlled colour-stroboscopic photo-
graphy. In: Potter OE, Nicklin DJ, eds. Fluidization
VII. New York: Engineering Foundation, 1992, pp
781789.
Zhou J. Circulating uidized bed hydrodynamics in a riser of
square cross-section. PhD diss., University of British
Columbia, Vancouver, 1995.
Zhou J, Grace JR, Qin S, Brereton CHM, Lim CJ, Zhu J.
Voidage proles in a circulating uidized bed of square
cross-section. Chem Eng Sci 49:32173226, 1994.
Zhou J, Grace JR, Lim CJ, Brereton CMH. Particle velocity
proles in a circulating uidized bed of square cross-sec-
tion. Chem Eng Sci 50:237244 1995a.
Zhou J, Grace JR, Lim CJ, Brereton CMH, Qin S, Lim KS.
Particle cross-ow, lateral momentum ux and lateral
velocity in a circulating uidized bed. Can J Chem Eng
73:612619, 1995b.
Zhou J, Grace JR, Brereton CMH, Lim CJ. Inuence of wall
roughness on the hydrodynamics in a circulating uidized
bed. AIChE J 42:11531156, 1996a.
Zhou J, Grace JR, Brereton CMH, Lim CJ. Inuence of
membrane walls on particle dynamics in a circulating ui-
dized bed. AIChE J 42:35503553, 1996b.
Zhu JX, Yu JQ, Jin Y, Grace JR, Issangya AS. Cocurrent
downow circulating uidized bed (downer) reactors. Can
J Chem Eng 73:612619, 1995.
20
Other Nonconventional Fluidized Beds
Wen-Ching Yang
1 INTRODUCTION
In a conventional uidized bed, uid is passed through
a bed of solids via a distributor plate. At a uid velo-
city beyond the minimum uidization or minimum
bubbling velocity, visible bubbles appear. The uid
thus passes through the bed in two phases, the bubble
and the emulsion phases. The bubble-induced solids
mixing and circulation provide the liquidlike behavior
of a bed of otherwise immobile solids. The liquidlike
behavior of a uidized bed allows continuous feeding
and withdrawal of bed material. The vigorous mixing
of solids in the bed gives rise to a uniform bed tem-
perature even for a highly exothermic or endothermic
reaction. This leads to easier control and operation.
The advantages of a uidized bed, compared to other
modes of contacting such as a packed bed, are numer-
ous, and they are described in detail in Chapter 3,
Bubbling Fluidized Beds, and Chapter 26,
LiquidSolids Fluidization. Because of the inherent
advantages of uidized beds, they are widely employed
in various industries for both physical and chemical
operations.
The conventional uidized beds also possess some
serious deciencies, however. The bubbles that are
responsible for many benets of a uidized bed repre-
sent the uid bypassing and reduction of uidsolids
contacting. The rapid mixing of solids in the bed leads
to nonuniform solids residence time distribution in the
bed. The rigorous solids mixing in the bed also leads to
attrition of bed material and increases the bed material
loss from elutriation and entrainment. Thus for many
industrial applications, the conventional uidized beds
have been modied to overcome those disadvantages.
Those modications, in many ways, alter substantially
the operational characteristics of the uidized beds and
also change the design and engineering of the beds. It is
the intent of this chapter to document four of the non-
conventional uidized beds in detail: the spouted bed,
the recirculating uidized bed with a draft tube, the
jetting uidized bed, and the centrifugal uidized bed.
2 SPOUTED BED
The words spouted bed and spouting were rst coined
by Mathur and Gishler (1955) at the National
Research Council of Canada during the development
of a technique for drying wheat. The rst extensive
assimilation of the literature came from the publication
of Spouted Beds by Mathur and Epstein (1974). A
more recent review can be found in Epstein and
Grace (1997).
A classical and conventional spouted bed is shown
in Fig. 1. The uid is supplied only through a centrally
located jet. If the uid velocity is high and the bed is
low enough, the uid stream will punch through the
bed as a spout as shown in Fig. 1. The spout uid will
entrain solid particles at the spoutannulus interface
and form a fountain above the bed. The spout uid
will also leak through the spoutannulus interface into
the annulus to provide aeration for the particles in the
annulus. The spouted bed is usually constructed as a
cylindrical vessel with a conical bottom as shown in
Fig. 1 to eliminate the stagnant region. Spouting in a
conical vessel has also been employed. Solid particles
can be continuously fed into a spouted bed through the
concentric jet or into the annulus region and continu-
ously withdrawn from the annulus region, just as in a
uidized bed.
Not all beds are spoutable. For beds with the ratio
of nozzle diameter to column diameter D
i
,D, above a
certain critical value, there is no spouting regime. The
bed transfers from the xed bed directly into the ui-
dized state with increasing gas velocity. To achieve a
stable nonpulsating spouted bed, the nozzle-to-parti-
cle-diameter ratio, D
i
,d
p
, should be less than 25 or
30. There are also the so-called minimum spouting
velocity and the maximum spoutable bed height
requirements.
2.1 Minimum Spouting Velocity
According to Epstein and Grace (1997), the Mathur
Gishler equation shown below remains the simplest
equation to estimate the minimum spouting velocity,
good to 15%for cylindrical vessels up to about 0.5 m.
For D _ 0.5 m
(U
ms
)
0.5
=
d
p
D
_ _
D
i
D
_ _
1,3
2gH(,
p
,
f
)
,
f
_
(1)
For nonspherical particles, the diameter, d
p
, to be used
in Eq. (1) should be the diameter of a sphere with equal
volume. For closely sized near-spherical particles, the
volumesurface mean diameter should be employed.
For prolate spheroids, the smaller of the two principal
dimensions is best used as the particle diameter in
Eq (1).
For D > 0.5 m
U
ms
= 2.0D(U
ms
)
0.5
D in meters (2)
The maximum value of the minimum spouting velo-
city, U
m
, occurs when the bed height is at the maxi-
mum spoutable bed depth or
U
m
=
d
p
D
_ _
D
i
D
_ _
1,3
2gH
m
(,
p
,
f
)
,
f
_
(3)
The U
m
is closely related to the minimum uidization
velocity, U
mf
, of the particles by
U
m
U
mf
= b = 0.9 to 1.5 (4)
A comprehensive review of correlations proposed
for the minimum spouting velocity can be found in
Mathur and Epstein (1974). A general correlation for
the minimum spouting velocity was also suggested by
Littman and Morgan (1983).
The minimum spouting velocity has been shown by
King and Harrison (1980) to decrease markedly with
increasing pressure up to 20 bar. A modied version of
the Mathus and Gishler correlation, shown in Eq. (1),
was proposed by King and Harrison (1980) as
(U
ms
)
0.5
=
,
f
,
air(p=1)
_ _
0.2
d
p
D
_ _
D
i
D
_ _
1,3
2gH(,
p
,
f
)
,
f
_
(5)
where ,
air(p=1)
is the density of air at one atmosphere
and room temperature.
Figure 1 A classical and conventional spouted bed.
For single particle size in a rectangular spouted bed,
Anabtawi et al. (1992) showed that the minimum
spouting velocity had little dierence compared to
that in cylindrical beds predictable within the varia-
tions from dierent correlations available for cylindri-
cal beds. A dimensionless correlation for predicting the
minimum spouting velocity in a rectangular spouted
bed with a mixture of binary particle sizes was recently
proposed by Anabtawi (1998).
For a given bed height, the minimum spouting velo-
city decreases with increasing pressure following the
Mathur and Gishler equation shown in Eq. (1).
Experimental data obtained by He et al. (1988b) indi-
cated that Mathur and Gishler equation under-pre-
dicted the U
ms
by about 50% for the heavy steel
balls and about 39.5% for the large glass beads. The
deviation is smaller, about 26%, for the small glass
beads. This corresponds to a U
ms
dependence of gas
density of ,
0.36
f
for the steel balls and large glass beads
and of ,
0.22
f
for the small glass beads. Thus for correct
prediction of the pressure eect on U
ms
, the exponent
on ,
f
in the Mathur and Gishler equation should have
dierent values depending on the particle Reynolds
number.
2.2 Maximum Spoutable Bed Height
The maximum spoutable bed height, H
m
, can be solved
by combining Eqs. (3) and (4) with the Wen and Yu
(1966) equation
(Re)
mf
=
d
p
U
mf
,
f
j
= 33.7
1 35.9 10
6
Ar
_
1
_ _
(6)
to give (Epstein and Grace, 1997)
H
m
=
D
2
d
p
D
D
i
_ _
2,3
568b
2
Ar
1 35.9 10
6
Ar
_
1
_ _
2
(7)
McNab and Bridgwater (1977) found that Eq. (7) with
b = 1.11 tted the experimental data best. For high-
temperature applications, take b = 0.9 to be conserva-
tive (Ye et al., 1992; Li, 1992). More correlations were
reviewed in Mathur and Epstein (1974) and Littman et
al. (1979).
By dierentiating Eq. (7) with respect to Ar,
dH
m
,dAr, we can nd that there is critical particle
size at Ar = 223,000, calculatable as
(d
p
)
crit
= 60.6
j
2
g(,
p
,
f
),
f
_ _
1,3
(8)
For particles larger than the critical particle shown in
Eq. (8), the maximum spoutable bed height, H
m
,
decreases with increases in particle size, and with par-
ticles smaller than the critical particle, H
m
increases
with increasing d
p
. Equation (8) is only good for
gassolids spouting and is not applicable to liquid
solids spouting. For liquidsolids spouting, the maxi-
mum spoutable bed height decreases with increases in
particle size for all cases. For gassolids spouting, the
critical Archimedes number Ar = 223,000 corresponds
to a critical Reynolds number of (Re)
mf
= 67, obtain-
able from the Wen and Yu equation, Eq. (6). Thus for
gassolids spouting, the critical particle diameter is
usually in the range of 1.0 to 1.5 mm.
The maximum spoutable bed height, H
m
, increases
with increasing pressure based on Eq. (7). Thus the
region of spoutability is greater at higher pressure.
Recent experimental data obtained by He et al.
(1998b) provided the evidence for the trend.
However, it was found that the McNab and
Bridgwater (1977) modication of Eq. (7) over-pre-
dicted the H
m
up to 36.5% for the steel balls and the
smaller glass beads and under-predicted the H
m
for the
large glass beads by 10.3%.
2.3 Spout Diameter
The average spout diameter has been correlated
empirically by McNab (1972) with the equation
D
s
=
2.00G
0.49
D
0.68
,
0.41
b
(9)
Equation (9) is a dimensional equation where SI units
should be used. Equation (9) is good to 5.6% at
room temperature. He et al. (1998a) employed a
ber-optic probe to measure the spout diameter in a
semicylindrical and a full cylindrical spouted bed and
found that the presence of the at front plate in the
semicylindrical bed considerably distorted the spout
shape. They found that the McNab equation under-
estimated the spout diameter in a full cylindrical bed
by an average of 35.5%. Under pressure, the McNab
equation can also introduce error up to 65.5% at a
pressure of 343 kPa (He et al., 1998b). Spout diameter
tends to increase with increasing pressure.
For elevated temperatures, Wu et al. (1987) sug-
gested the dimensionally consistent semiempirical
equation
D
s
= 5.61
G
0.433
D
0.583
j
0.133
(,
f
,
b
g)
0.283
_ _
(10)
2.4 Voidage Distribution
The voidage at the annulus at minimum spouting, c
a
,
can be expected to be close to the voidage at minimum
uidization, c
mf
. For narrowly sized spherical particles,
this voidage is usually around 0.42. For nonspherical
particles, the voidage will be slightly higher. He et al.
(1994) found that the voidage in the annulus was some-
what higher than the loose-packed voidage of a packed
bed and increased with increasing spouting gas ow.
The voidage in the spout decreases from 1 at the
spout inlet almost linearly with height until it reaches
around 0.7 at the top of the spout. Models are avail-
able for prediction of the voidage distribution for both
the annulus and spout (Lim and Mathur, 1978).
2.5 Fluid Flow Distribution
By applying Darcys law in the annulus of the spouted
bed, Mamuro and Hattori (1968) proposed the follow-
ing equation for the calculation of the supercial uid
velocity, U
a
, in the annulus of the spouted bed at
height z.
U
a
U
mf
= 1 1
z
H
m
_ _
3
(11)
Equation (11) is expected to apply even for H _ H
m
(Grbravcic et al., 1976) and for annulus Reynolds
number one or two orders of magnitude larger than
the upper limit of Darcys law (Epstein et al., 1978).
By continuity at any level of the bed, the gas ow
balance can be written as
U
sz
D
2
s
U
a
(D
2
D
2
s
) = UD
2
(12)
Thus at any level, the fraction of the total uid ow
passing through the annulus region can be calculated
from Eq. (12) if the spout diameter is known. The
supercial velocity at minimum spouting, U = U
ms
,
can be calculated from Eq. (1). In practice, the operat-
ing velocity of a spouted bed is typically 10 to 50%
higher than U
ms
. Under those conditions, the excess
gas above that required for minimum spouting can
be assumed to pass through the spout as a rst approx-
imation.
Typical gas streamlines in the annulus are deter-
mined by Epstein and Grace (1997) to be like that
shown in Fig. 2. A gas recirculation zone was also
observed immediately adjacent to the gas inlet due to
the venturi eect above the jet nozzle. The gas ow in
the spout is essentially in plug ow and in the annulus,
in dispersed plug ow similar to that in a packed bed.
Because of the extensive communication of gas
between the spout and the annulus regions, the gas
residence time distribution deviates substantially from
both plug ow and perfect mixing.
2.6 Particle Movement and Fountain Height
Solids continuity dictates that, at any bed level, the
particle owing up in the spout has to be balanced
by the particles moving down in the annulus, or
W = ,
p
A
s
(1 c
s
)v
s
= ,
p
A
a
(1 c
a
)v
w
(13)
Equation (13) neglects the radial variation of particle
velocities in the spout and in the annulus. The particle
Figure 2 Typical gas streamlines in the annulus of a spouted
bed.
velocity in the annulus is further assumed to equal to
the particle velocity observed at the wall, v
w
.
He et al. (1994) applied a ber-optic probe system to
measure the vertical particle velocities in the spout, the
fountain, and the annulus of a full-column spouted
bed. They found that radial proles of vertical particle
velocities in the spout were of near Gaussian distribu-
tion rather than parabolic as reported by earlier
researchers. In the annulus, vertical particle velocities
decreased with decreasing height because of cross-ow
of particles from the annulus to the spout. On the
contrary, vertical particle velocities increased with
decreasing height owing to the reduction of annular
cross-sectional area. In the fountain core region, the
particles decelerated, attained zero velocity at top of
the fountain, and then rained down around the sur-
rounding region. In the radial direction, the particle
velocities decreased with increasing radial distance
from the axis. The semitheoretical model proposed
by Grace and Mathur (1978) as shown in Eq. (14)
was found to predict the fountain height quite well.
H
F
= c
1.46
bs
v
2
0 max
2g
,
p
,
p
,
f
(14)
2.7 Pressure Gradient in the Annulus
Experimental data obtained by He et al. (1998b)
showed that the longitudinal pressure prole in the
annulus was independent of pressure with steel balls
and large and small glass beads as bed materials.
2.8 Conical Spouted Beds
Olazar and his associates (1992, 1993ac, 1995, 1998)
studied the design, operation, and performance of a
conical spouted bed and found that the conical
spouted bed is especially useful for hard-to-handle
solids that are irregular in texture or sticky. A conical
spouted bed is depicted in Fig. 3. The conical spouted
bed exhibits pronounced axial and radial voidage pro-
les that are quite dierent from the cylindrical
spouted beds.
2.8.1 Minimum Spouting Velocity
(Re)
D
o
.ms
= 0.126Ar
0.5
D
b
D
o
_ _
1.68
tan

2
_ _ _ _
0.57
(15)
where D
b
=upper diameter of the stagnant bed
D
i
=diameter of the bed bottom
D
o
=diameter of the inlet
(Re)
Do.ms
=Reynolds number at minimum
spouting base of D
o
=included angle of the cone
2.8.2 Bed Voidage Along the Spout Axis
The bed voidage along the spout axis at r = 0 was
found to be parabolic and dependent on the system
variables employing an optical ber probe (San Jose
et al., 1998). It can be calculated from
c(0) = 1 E
z
H
_ _
2
(16)
where E is an empirical parameter varying between 0.3
and 0.6 and is empirically correlated as follows
E = 1.20
D
b
D
o
_ _
0.12
H
o
D
i
_ _
0.97
U
U
ms
_ _
0.7
0.25
(17)
Figure 3 Conical spouted bed.
where H
o
is the height of the stagnant bed
and D
b
= D
i
2H
o
tan

2
_ _
(18)
2.8.3 Bed Voidage at the Wall
The bed voidage was found by San Jose et al. (1998) to
decrease radially toward a minimum at the wall. The
bed voidage at the wall can be calculated from the
equation
c(w) = c
o
1
H z
H
_ _
0.5
(19)
where c
o
=loose bed voidage
c(w) =bed voidage at the wall
H =height of the developed bed
2.8.4 Bed Voidage Correlation
The general bed voidage correlation can be expressed
as
c =
c(0) c(w)
1 exp (r r
s
),27.81r
2.41
s
_ _ c(w) (20)
where r
s
is the radius of the spout.
3 RECIRCULATING FLUIDIZED BEDS WITH
A DRAFT TUBE
The recirculating uidized bed with a draft tube con-
cept is illustrated in Fig. 4. This concept was rst called
a recirculating uidized bed by Yang and Keairns
(1974). Several other names have also been used to
describe the same concept: the uid-lift solids recircu-
lator (Buchanan and Wilson, 1965), the spouted uid
bed with a draft tube (Yang and Keairns, 1983;
Hadzismajlovic et al., 1992), the internally circulating
uidized bed (Milne et al., 1992; Lee and Kim, 1992);
or simply a circulating uidized bed (LaNauze, 1976).
The addition of a tubular insert, a draft tube, in a
spouted uid bed changes the operational and design
characteristics of an ordinary spouted bed. Notably,
there is no limitation on the so-called maximum spou-
table bed height. Theoretically, a recirculating ui-
dized bed with a draft tube can have any bed height
desirable. The so-called minimum spouting velocity
will also be less for a recirculating uidized bed with a
draft tube because the gas in the draft tube is conned
and does not leak out along the spout height as in an
ordinary spouted bed.
There is considerably more operation and design
exibility for a recirculating uidized bed with a
draft tube. The downcomer region can be separately
aerated. The gas distribution between the draft tube
and the downcomer can be adjusted by changing the
design parameters at the draft tube inlet. Because the
draft tube velocity and the downcomer aeration can be
individually adjusted, the solid circulation rate and
particle residence time in the bed can be easily con-
trolled. Stable operation over a wide range of operat-
ing conditions, solids circulation rates up to 100 metric
tons per hour, and a solids loading of 50 (weight of
solids/weight of air) in the draft tube have recently
been reported by Hadzismajlovic et al. (1992). They
used a 95.3 cm diameter bed with a 25 cm diameter
draft tube using 3.6 mm polyethylene particles. A
detailed discussion of the recirculating uidized bed
with a draft tube is recently published by Yang (1999).
Operating conditions for a recirculating uidized
bed can be exible as well. The bed height can be
lower than the draft tube top or just cover the draft
Figure 4 Recirculating uidized bed with draft tube oper-
ated as a pneumatic transport tube.
tube top so that a spout can penetrate the bed as in a
spouted bed. The bed height can also be substantially
higher than the draft tube top, so that a separate ui-
dized bed exists above the draft tube. Rather than
operating the draft tube as a dilute-phase pneumatic
transport tube, one can uidize the solids inside the
draft tube at lower velocities to induce the necessary
recirculation of the solids. Several studies were con-
ducted in this fashion (Ishida and Shirai, 1975;
LaNauze, 1976; LaNauze and Davidson, 1976). The
draft tube wall can also be solid or porous, although
most of the studies in the literature employ a solid-wall
draft tube. Clain and Fane (1983) reported that a
porous draft tube was suitable for applications in ther-
mal disinfestation of wheat where control of particle
movement and good gas/solid contacting could be
accomplished at a modest pressure drop. The concept
can also be employed as a liquidsolids and liquid
gassolids contacting device (Oguchi and Kubo,
1973). The design and operation of a recirculating ui-
dized bed with a draft tube are discussed below.
3.1 Draft Tube Operated as a Fluidized Bed
The schematic for this system, where the draft tube is
operated as a uidized bed rather than a dilute phase
pneumatic transport tube, is shown in Fig. 5. A math-
ematical model for the system was developed by
LaNauze (1976). The driving force for solids circula-
tion in this case was found to be the density dierence
between draft tube and downcomer. The solids circu-
lation rate was also found to be aected only by the
distance between the distributor and the draft tube and
not by the draft tube length or height of bed above it.
Because of the lower velocity in the draft tube, the
draft tube diameter tends to be larger compared to
when the draft tube is operated in a dilute phase pneu-
matic transport mode. One disadvantage of operating
the draft tube as a uidized bed is that if the draft tube
diameter is too small or the draft tube is too high, the
draft tube tends to operate in a slugging bed mode. In
fact, the mathematical model developed by LaNauze
(1976) described below assumes that the draft tube is a
slugging uidized bed.
The pressure balance for the dense phase in the
downcomer in the circulating uidized system shown
in Fig. 5 can be expressed as
P
1-4
= ,
b
gH
mf
(1 c
bd
)
t
d
S
d
A
d
(21)
A similar expression can be written for the pressure
balance in the draft tube as
P
2-3
= ,
b
gH
mf
1 c
bf
( )
t
r
S
r
A
r
(22)
Combining Eqs. (21) and (22), we have
,
b
gH
mf
c
br
c
bd
( ) =
t
d
S
d
A
d
t
r
S
r
A
r
(23)
Experimental evidence indicates that the voidage of the
solids ow down the downcomer is close to that of
minimum uidization, thus c
bd
= 0.
The bubble voidage in the draft tube, c
br
, was cal-
culated on the basis of the velocity of a rising gas slug
in a slugging bed relative to its surrounding solids. The
total gas supercial velocity in the draft tube, U
fr
, can
be derived as
U
fr
= U
slug
c
br
U
mf

V
sr
c
mf
1 c
mf
(24)
The slug velocity U
slug
, is dened as the rising velocity
of the slug relative to the particle velocity at its nose
and can be expressed as
U
slug
= v
p
0.35
gD
_
(25)
and
v
p
= (U
fr
U
mf
) V
sr
(26)
Substituting Eq. (25) into (24), we have
c
br
=
U
fr
U
mf
( ) V
sr
c
mf
, 1 c
mf
( )
U
fr
U
mf
( ) V
sr
0.35

gD
_ (27)
The ow rate of particles in the downcomer and the
draft tube are related by a mass balance as
V
sr
,
p
A
r
= V
sd
,
p
A
d
= W
sd
A
d
= W
sr
A
r
(28)
Figure 5 Recirculating uidized bed with draft tube oper-
ated as a uidized bed.
By solving Eqs. (24) and (27) simultaneously, the mass
ux can be calculated, provided the wall shear stress is
known as a function of particle supercial volume ow
rate. Botterill and Bessant (1973) have proposed sev-
eral relationships for shear stress, but these are not
general. LaNauze (1976) also proposed a method of
measuring this shear stress experimentally.
A similar application of the concept as a slugging
lifter of solids was studied by Singh (1978) based on the
two-phase theory of uidization and the properties of
slugs.
3.2 Draft Tube Operated as a Pneumatic Transport
Tube
Most of the applications for the recirculating uidized
bed with a draft tube operate the draft tube as a dilute
phase pneumatic transport tube. Typical experimental
pressure drops across the downcomer, P
1-4
, and the
draft tube, P
2-3
, show that they are essentially simi-
lar. Thus successful design of a recirculating uidized
bed with a draft tube requires development of mathe-
matical models for both downcomer and draft tube.
An applicable model is described below for general
application.
3.2.1 Downcomer and Draft Tube Pressure Drop
Downcomer Pressure Drop When the downcomer
is uidized, the downcomer pressure drop can be cal-
culated as in an ordinary uidized bed as
P
1-4
= L(1 c
d
),
p
(29)
When the downcomer is less than minimally uidized,
the pressure drop can be estimated with a modied
Ergun equation substituting gassolid slip velocities
for gas velocities (Yoon and Kunii, 1970), as shown
in Eq. (30).
P
1-4
=
L
g
150
j(U
gd
U
pd
)(1 c
d
)
2
d
2
p
2
s
c
2
d
_
1.75
,
f
(U
gd
U
pd
)
2
(1 c
d
)
d
p
s
c
d
_
(30)
The voidage in the downcomer, c, can be assumed to
be the same as the voidage at minimum uidization,
c
mf
, which can be determined in a separate uidized
bed. The agreement between the calculated and the
experimental values is usually better than 10%
(Yang and Keairns, 1978a). When the downcomer is
not minimally uidized, the bed voidage depends on
the amount of aeration and solid velocity. Use of the
voidage at minimum uidization is only a rst approx-
imation.
Draft Tube Pressure Drop The pressure drop
across the draft tube, P
2-3
, is usually similar to that
across the downcomer, P
1-4
, in magnitude. Thus
for a practical design basis, the total pressure drop
across the draft tube and across the downcomer can
be assumed to be equal. In most operating condi-
tions, the pressure drop at the bottom section of the
draft tube has a steep pressure gradient due primarily
to the acceleration of solid particles from essentially
zero vertical velocity. The acceleration term is espe-
cially signicant when the solid circulation rate is
high or when the draft tube is short.
The model suggested by Yang (1977) for calculating
the pressure drop in vertical pneumatic conveying lines
can be applied here to estimate the acceleration pres-
sure drop. The acceleration length can be calculated
from numerical integration of the equation
L =
_
U
pr2
U
pr1
U
pr
dU
pr
3
4
C
DS
c
4.7
r
,
f
(U
gr
U
pr
)
2
(,
p
,
f
)d
p
g
f
p
U
2
pr
2D
_ _
(31)
The solid friction factor, f
p
, can be evaluated with the
equation proposed by Yang (1978):
f
p
= 0.0126
(1 c
r
)
c
3
r
(1 c
r
)
(Re)
t
(Re)
p
_ _
0.979
(32)
The lower limit of integration, U
pr1
, is derived from
W
sr
= U
pr
,
p
(1 c
r
) (33)
with c
r
= 0.5, and the upper limit, U
pr2
, by the equa-
tion
U
pr
= U
gr
U
t
1
f
p
U
2
pr
2gD
_ _
c
4.7
r
_
(34)
The total pressure drop in the acceleration region can
then be calculated as
P
2-3
=
_
L
0
,
p
(1 c
r
) dL
_
L
0
2f
g
,
f
U
2
gr
gD
dL
_
L
0
f
p
,
p
(1 c
r
)U
2
pr
2gD
dL
,
p
(1 c
r
)U
2
pr
g
_ _
atL
(35)
If the draft tube height is less than the acceleration
length, the integration of Eq. (35) is carried out
through the whole length of the draft tube. If the
draft tube height is larger than the acceleration length,
the integration of Eq. (35) is carried out for the total
acceleration length, and the extra pressure drop for the
rest of the draft tube can then be included to give the
total pressure drop in the draft tube. The suggested
equations have been applied to actual experimental
data satisfactorily (Yang and Keairns, 1978a).
3.2.2 Gas Bypassing Phenomenon
Because of dierent design and operating parameters,
the distribution of the total ow between the draft tube
side and the downcomer side can be very dierent.
According to Yang (1999), the important design para-
meters are the area ratio between the downcomer and
the draft tube, the diameter ratio between the draft
tube and the draft tube gas supply or the diameter of
the solid feeding tube, the distance between the distri-
butor plate and the draft tube inlet, the area ratio of
the draft tube gas supply and the concentric solids
feeder, and the design of the downcomer gas supply
nozzle. In addition to the design parameter, the oper-
ating parameters will also aect gas bypassing. The
relative strength of the concentric jets of the draft
tube gas supply and the solids feeder determines the
half-angle of the combined jet, and the jet velocity
determines the jet penetration. The jet velocity of the
downcomer gas supply nozzles is also important if the
jets are horizontal and directed toward the draft tube.
The gas bypassing phenomenon was studied by
Stocker et al. (1989) by measuring the dierential pres-
sure drops between the draft tube and the downcomer.
A more rigorous investigation was conducted by Yang
and Keairns (1978a) by injecting gas tracer, carbon
dioxide or helium, continuously at dierent locations
and taking gas samples from both the draft tube side
and the downcomer side. The actual amounts of gas
passing through the draft tube and the downcomer
were then obtained by solving mass conservation equa-
tions for tracer gas.
Except for the conical distributor plate, no simple
gas bypassing relationship exists. No rigorous theore-
tical model has thus far been proposed. The quanti-
tative gas bypassing information is usually determined
experimentally. Qualitatively, the gas bypasses from
the draft tube side into the downcomer side for
small draft-tube-to-downcomer area ratios and vice
versa. Gas bypasses exclusively from the downcomer
side to the draft tube side when the distance between
the distributor plate and the draft tube inlet is small.
For a conical distributor plate, the angle of the con-
ical plate (= 45
and 60
) does not seem to aect the

gas bypassing characteristics. A more detailed dis-
cussion of gas bypassing phenomena is presented in
Yang (1999).
3.2.3 Solids Circulation Mechanisms and the Solids
Circulation Rate
Both solids circulation mechanisms and the solids cir-
culation rate are important in designing and operating
a recirculating uidized bed with a draft tube. For
commercial applications in the area of coating and
encapsulation of solid particles, such as in coating of
pharmaceutical tablets and in coating seeds for delayed
germination and controlling the release rate of fertili-
zers, the particle residence time and cycle time are
important considerations. The performance based on
cycle time distribution analysis for coating and granu-
lation was studied by Mann and Crosby (1973, 1975),
Mann (1983), and Turton et al. (1999).
Two mechanisms for solids circulation have been
observed experimentally (Yang and Keairns, 1978a).
High-speed movies (1000 to 1500 frames per second)
taken at the inlet and the midsection of the draft tube
with a sand bed revealed that solids transport inside
the draft tube was not a conventional pneumatic trans-
port, where uniform solid suspension prevailed, but a
slugging type transport. The high-speed movies taken
at the outlet of a 23 m/s air jet showed that the air jet
issuing from the jet nozzle supplying air to the draft
tube was composed of bubbles rather than a steady jet.
The bubble grew from the mouth of the nozzle until its
roof reached the draft tube; then the sudden suction
from the draft tube punctured the roof. A continuous
stream of dilute solids suspension passed through the
roof into the draft tube. Simultaneously, another bub-
ble was initiated. As this bubble grew, it pushed a slug
of solids into the draft tube. The high-speed movies
taken at the midsection of the draft tube exhibited
alternate sections of dilute suspension and solids slug
occupying the total cross section of the draft tube.
A steady jet without bubbling can be maintained in
a sand bed between the jet nozzle and the draft tube
inlet with high jet velocities of the order of 60 m/s and
without downcomer aeration. Once the downcomer is
aerated, the solids circulation rate increases dramati-
cally and the steady jet becomes a bubbling jet.
Apparently, the inward-owing solids have enough
momentum to shear the gas jet periodically into
bubbles.
When the polyethylene beads (density = 907 kg,m
3
,
average size = 2800 jm) and the hollow epoxy spheres
(density = 210 kg,m
3
, average size = 2800 jm) were
used as the bed material, a steady jet between the jet
nozzle and the draft tube was always observed for all
experiments conducted.
Solids circulation rate was found to be strongly
aected by the design conguration at the bottom of
the draft tube and the downcomer owing to changes in
gas bypassing characteristics (Yang and Keairns,
1983). This coupling eect indicates that an under-
standing of the gas bypassing characteristics is essen-
tial. Except for simple cases, the dependency of the gas
bypassing characteristics on design and operating
parameters are still not amenable to theoretical treat-
ment as discussed earlier. A recent study by Alappat
and Rane (1995) on the eects of various design and
operational parameters on solids circulation rate essen-
tially arms the above conclusions.
When only the draft tube gas ow is employed with-
out downcomer aeration, the solid circulation rate
depends primarily on the entrainment rate of the jets.
Aeration of the downcomer can be provided with
greatly increased solid circulation rate. At lower down-
comer aeration, the solid circulation rate is essentially
similar to that without downcomer aeration. At higher
downcomer aeration, however, a substantial increase
in solid circulation rate can be realized. Apparently, a
minimum aeration in the downcomer is required in
order to increase substantially the solid circulation
rate.
As expected, the closer the distance between the
distributor plate and the draft tube inlet, the lower
the solids circulation rate. This is not only because of
the physical constriction created by locating the distri-
butor plate too close to the draft tube inlet but also
because of the dierent gas bypassing characteristics
observed at dierent distributor plate locations as dis-
cussed earlier. When the distance between the distribu-
tor plate and the draft tube inlet becomes large, it can
create start-up problems discussed in Yang et al.
(1978).
3.3 Design Procedure for a Recirculating Fluidized
Bed with a Draft Tube
From the experimental evidence, the design of a recir-
culating uidized bed with a draft tube involves the
specication of a number of design parameters and
an understanding of the coupling eects between the
design and the operating variables. A procedure is pre-
sented here for the design of a bed to give a specied
solids circulation rate. This design procedure assumes
that the solids and gas characteristics, feed rates, and
operating temperature and pressure are given. The
design parameters to be specied include the vessel
diameter, draft tube diameter, draft tube height, gas
distributor, and distributor position. These parameters
can be specied using the solids circulation rate model,
experimental data on gas bypassing, and process
requirements (e.g., selection of gas velocity in the bed
above the draft tube).
Experience indicates that a simple theoretical model
to predict gas bypassing that takes into account all the
design and operating variables cannot be developed.
Empirical correlation, however, can be obtained by
conducting experiments with tracer gas injection for
a given distributor plate design at dierent operating
conditions and at dierent distances from the draft
tube inlet.
It is also assumed that the total gas ow into the
bed is known. When the operating uidizing velocity
is selected for the uidized bed above the draft tube,
the diameter of the vessel is determined. The nal
design decisions include selection of the draft tube
diameter, the distributor plate design, the separation
between the draft tube and the distributor plate, and
the draft tube height. Selection of the draft tube
height may be determined by other considerations,
such as solids residence time, though it also aects
the solids circulation rate. A gas distributor is selected
to be compatible with the process and to maintain the
gas velocity in the downcomer near U
mf
. The draft
tube diameter is then selected by using the solids cir-
culation rate model to obtain the desired circulation
rate.
The design procedures are thus:
1. Assume a solid circulation rate per unit draft
tube area, W
sr
, and calculate the particle velocity in
the downcomer, U
pd
, from the equation
W
sr
= U
pd
1 c
mf
( ),
p
A
d
A
r
(36)
2. If the two-phase theory applies, the slip velocity
between the gas and the particles in the downcomer
must equal the interstitial minimum uidizing velocity
as
U
fd
U
pd
=
U
mf
c
mf
(37)
where U
fd
and U
mf
are positive in the upward direction
and U
pd
is positive in the downward direction.
Calculate U
fd
from Eq. (37) and the pressure drop
across the downcomer, P
1-4
, from Eq. (30) assuming
c
d
= c
mf
.
3. Use trial and error with Eqs. (32), (33), and (34)
to evaluate U
pr
, c
r
, and f
p
.
4. Numerically integrate Eq. (31) to obtain the
particle acceleration length and Eq. (35) to obtain the
pressure drop across the draft tube, P
2-3
.
5. Compare the pressure drop across the down-
comer, P
1-4
, and that across the draft tube, P
2-3
.
If they are not equal, repeat procedures 1 through 5
until P
1-4
P
2-3
.
3.4 Startup and Shutdown Considerations
Both cold ow experiments and actual pilot-plant
experience show that, if operating conditions and
design parameters are not selected carefully, start-up
(initiating solids circulation) might be a problem (Yang
et al., 1978). The primary design parameters that will
aect the start-up are the distance between the grid and
the draft tube inlet (L
d
) and the diameter ratio between
the draft tube and the draft tube gas supply nozzle
(D,d
D
). The maximum allowable distance, L
d
, can be
determined by applying the jet penetration equation
suggested by Yang and Keairns (1978b):
L
j
d
D
= 6.5
,
f
,
p
,
f
U
2
j
gd
D
_ _
1,2
(38)
where U
j
is the gas velocity issuing from the draft tube
gas supply. For high-temperature and high-pressure
operations, Eq. (41), to be discussed later, should be
used for calculating L
j
. Another consideration is that
the jet boundary at the end of jet penetration should
correspond to the physical boundary of the draft tube
inlet. Merrys expression (1975) for jet half-angle can
be used for this purpose:
cot() = 10.4
,
f
d
D
,
p
d
p
_ _
0.3
(39)
or
L
d
=
(Dd
D
)
2 tan()
(40)
The distance between the distributor plate and the
draft tube inlet, L
d
, selected for the design should be
the smaller one of those estimated from Eqs. (38) and
(40).
A start-up technique described by Hadzismajlovic et
al. (1992) is also worthy of consideration if the draft
tube gas supply is retractable. The draft tube gas sup-
ply nozzle can be inserted into the draft tube during
start-up and shutdown. This will reduce the diculty
described here during start-up. After start-up, the sup-
ply nozzle can be lower to below the draft tube inlet at
a predetermined height to provide the normal opera-
tion conguration. This will prevent solids from drain-
ing into the gas supply nozzle during shutdown. Of
course, if the draft tube gas supply nozzle is not mova-
ble due to hostile operating conditions, the technique
cannot be used. Then the design precautions discussed
above during start-up should be followed.
3.5 Multiple Draft Tubes
Studies in the past always concentrate on beds with a
single draft tube. A literature survey failed to uncover
any reference on the operation of multiple draft tubes.
Even in the area of conventional spouted beds, the
references on multiple spouted beds are rare. Foong
et al. (1975) reported that the multiple spouted bed
was inherently unstable owing to pulsation and regres-
sion of the spouts. Similar instability was also observed
by Peterson (1966), who found that vertical baes
covering at least one-half of the bed height were neces-
sary to stabilize the operation. In an industrial envir-
onment where solids are processed in large vessels,
multiple draft tubes may be both necessary and bene-
cial. Exploratory tests in a 2D bed with three draft
tubes were reported by Yang and Keairns (1989). The
design methodology proposed earlier for beds with a
single draft tube is still applicable here for beds with
multiple draft tubes.
3.6 Industrial Applications
The application of the recirculating uidized bed with
a draft tube was probably rst described by Taskaev
and Kozhina (1956) for a coal devolatilizer. Dry coal is
introduced into the devolatilizer below the bottom of
the draft tube through a coal feeding tube concentric
with the draft tube gas supply. The coal feed and
recycled char at up to 100 times the coal feed rate
are mixed inside the draft tube and carried upward
pneumatically in dilute phase at velocities greater
than 4.6 m/s. The solids disengage in a uidized bed
above the top of the draft tube and then descend in an
annular downcomer surrounding the draft tube as a
packed bed at close to minimum uidization velocity.
Gas is introduced at the base of the downcomer at a
rate permitting the downward ow of the solids. The
recirculating solids eectively prevent agglomeration
of the caking coal as it devolatilizes and passes through
the plastic stage.
A seeded coal process was later developed by
Curran et al. (1973) using the same concept to smear
the liquid raw coal undergoing the plastic transition
onto the seed char and the recirculating char during
low-temperature pyrolysis. Westinghouse successfully
demonstrated a rst stage coal devolatilizer with cak-
ing coals in a pilot-scale Process Development Unit
employing a similar concept where the downcomer
was uidized and the jet issuing from the draft tube
was immersed in a uidized bed above the draft tube
(Westinghouse, 1977). The same concept was also pro-
posed for extending uidized bed combustion technol-
ogy for steam and power generation (Keairns, et al.,
1978). The British Gas Council has also developed the
concept for oil and coal gasication (Horsler and
Thompson, 1968; Horsler et al., 1969). The develop-
ment eventually resulted in a large-scale recirculating
uidized bed hydrogenator gasifying heavy hydrocar-
bon oils (Ohoka and Conway, 1973). McMahon (1972)
also described a reactor design for oil gasication using
a multiplicity of draft tubes. The Dynacracking process
developed by Hydrocarbon Research Inc. in the 1950s
(Rakow and Calderon, 1981) for processing heavy
crude oil also utilized an internal draft tube. More
recently, coal gasication in a recirculating uidized
bed with a draft tube was described by Judd et al.
(1984) and a coalwater mixture combustion by Lee
and Kim (1992).
Other industrial applications of the concept include
that for coating tablets in the pharmaceutical industry
(Wurster et al., 1965), for drying of dilute solutions
containing solids (Hadzismajlovic, 1989), and for mix-
ing and blending (Decamps et al., 1971/1972;
Matweecha, 1973; Solt, 1972; Krambrock, 1976).
Both Conair Waeschle Systems and Fuller Company
supply commercial blenders based on the concept. The
concept was also proposed as a controllable solids fee-
der to a pneumatic transport tube (Decamps et al.,
1971/1972; Silva et al., 1996).
Although most of the experimental data reported
here were obtained with large particles, Geldart class
B and D powders, it is believed that the concept can
equally be applicable for any ne aeratable and free-
owing solids, Geldarts class A powders.
A similar concept has also been used for liquid
solids and liquidgassolids contacting devices
(Oguchi and Kubo, 1973; Fan et al., 1984) and bio-
reactors (Chisti, 1989). Bubble columns tted with
draft tubes have also been employed in the chemical
process industries as airlift reactors for gasliquid con-
tacting operations. Examples are the low-waste conver-
sion of ethylene and chlorine to dichloroethane,
biological treatment of high-strength municipal and
industrial euent, and bioreactors. Critical aspects of
the design and operation of bubble columns with draft
tubes have recently been reviewed by Chisti and Moo-
Young (1993). Freedman and Davidson (1964) also
carried out a fundamental analysis for gas holdup
and liquid circulation in a bubble column with a
draft tube. Extensive experimentation in a bubble col-
umn with a draft tube was conducted by Miyahara et
al. (1986), and an in-depth analysis by Siegel et al.
(1986). The eects of geometrical design on perfor-
mance for concentric-tube airlift reactors were studied
by Merchuk et al. (1994).
4 JETTING FLUIDIZED BEDS
In a gas uidized bed, the introduction of gas is usually
accomplished through distributors of various designs.
Any time the gas is distributed through orices or noz-
zles, a jetting region appears immediately above the
grid. A large uctuation of bed density occurs in this
zone, indicating extensive mixing and contacting of
solids and gas. If the chemical reactions between gas
and solids are fast, much of the conversion may occur
in this jetting region.
Another type of uidized bed, where the jetting phe-
nomenon is an important consideration, is the spouted
uid bed, where a large portion of gas goes through a
fairly large nozzle. Because of the dominant eect of
this jetting action provided by the large nozzle, this
type of uidized bed can be more appropriately called
a jetting uidized bed, especially when the jet does
not penetrate through the bed like that in a spouted
bed. A schematic of a typical jetting uidized bed is
shown in Fig. 6. Jetting, bubbling dynamics, and solid
circulation are important hydrodynamic phenomena
governing the performance and operation of large-
scale jetting uidized beds. They are the focus of our
attention in this section. A more extensive discussion
can be found in Yang (1999).
4.1 Jet Penetration and Bubble Dynamics
Gas jets in uidized beds were critically reviewed in
Sec. 11 of Chapter 3, Bubbling Fluidized Beds.
Most of the data available now are from jets smaller
than 25 mm. The discussion here will emphasize pri-
marily the large jets, up to 0.4 m in diameter, and
operation at high temperatures and high pressures.
The gas jets can also carry solids and are referred to as
gassolid two-phase jets in this discussion.
4.1.1 Momentum Dissipation of a Gas-Solid
Two-Phase Jet
Gas velocity proles in a gassolid two-phase jet inside
a uidized bed were determined using a pitot tube by
Yang and Keairns (1980). The velocity proles were
integrated graphically, and gas entrainment into a jet
was found to occur primarily at the base of the jet. A
reasonably consistent universal velocity prole can be
obtained by plotting (U
jr
U
jb
),(U
jm
U
jb
) vs. r,r
1,2
,
comparable with the Tollmien solution for a circular
homogeneous jet in an innite medium (Abramovich,
1963; Rajaratnam, 1976).
4.1.2 Jet Penetration and Jet Half Angle
Jet Penetration Jetting phenomena were studied
by Yang and Keairns (1978b) in a semicircular col-
umn 30 cm in diameter using hollow epoxy spheres
(r
p
= 210 kg/m
3
) as the bed material and air as the
uidizing medium to simulate the particle/gas density
ratio in actual operating conditions at 1520 kPa and
1280
K. A two-phase Froude number, dened as

,
f
U
2
j
,(,
p
,
f
)gd
0
and derived from both the momen-
tum balance and the dimensionless analysis, was
found to correlate jet penetration data well. The cor-
relation was extended to cover the high-pressure jet
penetration data of Knowlton and Hirsan (1980) at
pressures up to 5300 kPa for uidized beds of sand
(,
p
= 2629 kg/m
3
), FMC char (,
p
= 1158 kg/m
3
),
and siderite (,
p
= 3988 kg/m
3
). A subsequent analysis
by Yang (1981) indicated that the two-phase Froude
number originally suggested could be modied
slightly to account for the pressure effect.
L
j
d
o
= 7.65
1
R
cf
,
f
(,
p
,
f
)
U
2
j
gd
o
_ _
0.472
(41)
where
R
cf
= (U
cf
)
p
,(U
cf
)
atm
(42)
In the absence of (U
cf
)
p
and (U
cf
)
atm
, (U
mf
)
p
and
(U
mf
)
atm
can be employed.
The limiting form of Eq. (41) at atmospheric pres-
sure (101 kPa), where the correction factor R
cf
= 1,
approaches the correlation originally proposed for
atmospheric conditions shown in Eq (38):
Jet Half-Angle The jet half-angle can be calcu-
lated from the experimentally measured bubble size
and jet penetration depth as follows:
= tan
1
D
B
d
0
2L
j
_ _
(43)
Experimentally observed jet half-angles range from 8
to 12
for the experimental data mentioned above.

These compare to 10
suggested by Anagbo (1980)

for a bubbling jet in liquid.
4.1.3 Bubble Dynamics
To describe the jet adequately, the bubble size gener-
ated by the jet needs to be studied. A substantial
amount of gas leaks from the bubble to the emulsion
phase during the bubble formation stage, particularly
when the bed is less than minimally uidized. A model
developed on the basis of this mechanism predicted the
Figure 6 Schematic of a jetting uidized bed.
experimental bubble diameter well when the experi-
mental bubble frequency was used as an input. The
experimentally observed bubble frequency is smaller
by a factor of 3 to 5 than that calculated from the
Davidson and Harrison model (1963), which assumed
no net gas interchange between the bubble and the
emulsion phase. This discrepancy is due primarily to
the extensive bubble coalescence above the jet nozzle
and the assumption that no gas leaks from the bubble
phase.
High-speed movies were used to document the phe-
nomena above a 0.4 m diameter jet in a 3 meter dia-
meter transparent semicircular jetting uidized bed
(Yang et al., 1984b). The movies were then analyzed
frame by frame to extract information on bubble fre-
quency, bubble diameter, and jet penetration depth.
The process of bubble formation is very similar to
that described in Kececioglu et al. (1984), but it was
much more irregular in the large 3 m bed. A model was
developed to describe this phenomenon by assuming
that the gas leaks out through the bubble boundary at
a supercial velocity equivalent to the supercial mini-
mum uidization velocity. For a hemispherical bubble
in a semicircular bed, the rate of change of bubble
volume can be expressed as
dV
B
dt
= G
j
U
mf
D
2
B
2
(44)
where V
B
= D
3
B
,12 for a hemispherical bubble.
Equation (44) can be reduced to show the changes
of bubble diameter with respect to time in Eq. (45):
D
2
B
dD
B
(4G
j
2U
mf
D
2
B
)
=
dt
(45)
Integrating Eq. (45) with the boundary condition that
D
B
= 0 at t = 0 gives
t =
1
2U
mf
G
j
2G
j
U
mf
_
_
ln
2G
j
D
B
2G
j
U
mf
_
2G
j
D
B
2G
j
U
mf
_
_ _
D
B
_
(46)
The maximum bubble size, where the total gas leakage
through the bubble boundary equals the total jet ow,
can be obtained from either Eq. (44) or Eq. (45):
(D
B
)
max
=
2G
j
U
mf
_
(47)
The total amount of gas leakage from the bubble at a
bubble size D
B
is then
F =
G
j
2U
mf
G
j
2G
j
U
mf
_
_
ln
2G
j
D
B
2G
j
U
mf
_
2G
j
D
B
2G
j
U
mf
_ D
B
_ _
D
3
B
12
(48)
Equation (46) and the experimental bubble frequency,
n = 1,t, were used to predict the expected bubble dia-
meter. The predicted bubble diameters are very close to
those actually observed. Theoretically, Eqs. (46) and
(48) can be solved to obtain both the bubble frequency
and the bubble diameter if the total gas leakage at the
moment of bubble detachment, F, is known. The bub-
ble growth equations can be similarly derived for a
circular jet in a three-dimensional bed. The same
experimental observation and conclusions described
above for a semicircular bed are expected to hold as
well. Bubble frequency from the jet was also studied
using a force probe in the same bed. The results were
published in Ettehadieh et al. (1988).
The validity of extrapolating the data obtained in a
semicircular model to a circular one is also of concern.
Not much research has been carried out in this area.
Preliminary research results by Whiting and Geldart
(1980) indicated that, for coarse, spoutable solids
(Geldarts group D powders), semicircular columns
could provide information very similar to that from
the circular ones.
4.2 Gas Mixing Around the Jetting Region
4.2.1 Gas Mixing Around Single Jets
Gas exchange between the jet and the outside emulsion
phase was studied by tracer gas injection and by inte-
gration of gas velocity proles in the jet at various
heights above the jet nozzle in a 28.6 cm diameter
bed with a 3.5 cm jet using polyethylene beads as bed
material (Yang et al., 1984a). The concentration pro-
les obtained at dierent elevations were found to be
approximately similar if the local tracer concentration
is normalized with the maximum tracer concentration
at the axis, C,C
m
, and plotted against a normalized
radial distance, r,(r
1,2
)
c
, where (r
1,2
)
c
is the radial posi-
tion where the tracer concentration is just half the
maximum tracer concentration at the axis. Thus in a
permanent amelike jet in a uidized bed, not only the
velocity proles in the jet but also the gas concentra-
tion proles are similar.
The gas mixing between the jetting region and the
emulsion phase and the gas ow pattern around the jet
were determined by solving the tracer gas conservation
equation numerically along with the axial velocity pro-
les in the jet obtained with a pitot tube. It is con-
cluded that the gas mixing in a jetting uidized bed
with a permanent amelike jet is due primarily to con-
vection and that diusion plays a negligible role.
The resulting velocity proles and the ow pattern
inside and around the jet indicated that the gas ow
direction is predominantly from the emulsion phase
into the jet at distances close to the jet nozzle. This
ow can be from the aeration ow in the emulsion
phase, as in the cases of high jet velocity or high aera-
tion ow, or from the ow recirculated from the upper
part of the jet. The entrainment of gas into the jet
occurs immediately above the jet nozzle. The extent
of this region depends on both the aeration ow out-
side the jet and the jet velocity. Increases in aeration
ow and jet velocity tend to increase the height of this
region. Beyond this gas entrainment region, the gas in
the jet is then expelled from the jet along the jet height.
The gas expelled at the lower part of the jet is recircu-
lated back into the jet, setting up a gas recirculation
pattern at the lower part of the jet. The extent of this
recirculation pattern increases with increases in jet
velocity and with decreases in aeration ow outside
the jet.
The axial velocity proles, calculated on the basis of
Tollmien similarity and experimental measurement in
Yang and Keairns (1980), were integrated across the
jet cross section at dierent elevations to obtain the
total jet ow across the respective jet cross sections.
The total jet ows at dierent jet cross sections are
compared with the original jet nozzle ow, as shown
in Fig. 7. Up to about 50% of the original jet ow can
be entrained from the emulsion phase at the lower part
of the jet close to the jet nozzle. This distance can
extend up to about 4 times the nozzle diameter. The
gas is then expelled from the jet along the jet height.
4.2.2 Gas Mixing Around Concentric Jets
Gas mixing phenomena around a concentric jet were
investigated by Yang et al. (1988) in a large semicircu-
lar cold ow model, 3 meters in diameter and 10 meters
high, with a triple concentric jet nozzle assembly of 25
cm in diameter (see also Yang, 1998). A dividing gas
streamline was observed experimentally that prevents
the gas mixing between the jetting region and the emul-
sion phase until at higher bed heights. This dividing
gas streamline corresponds roughly to the boundary
of down-owing solids close to the walls, to be dis-
cussed later. Several observations were made based
on this study. Regardless of the incoming jet ow
rate, the gases that are injected through the concentric
jets essentially remain in the core of the reactor and do
not fully mix with the gas in the dense solid down-ow
region of the bed. Similarly, the gas injected through
the conical grid sections is not entrained by the incom-
ing jets. Partial entrainment and mixing of these gases
occur at locations where bubble formation and bubble
coalescence take place. On the contrary, the mixing
among the concentric jets occurs quite fast and is
usually completed within the jet penetration length.
4.3 Solids Circulation in Jetting Fluidized Beds
The solids circulation pattern and solids circulation
rate are important hydrodynamic characteristics of
an operating jetting uidized bed. They dictate directly
the solids mixing and the heat and mass transfer
between dierent regions of the bed.
In many applications the performance of uidized
beds is frequently controlled by the hydrodynamics
phenomena occurring in the beds. Applications such
as the uidized bed combustion and gasication of
fossil fuels are the cases in point. In those applications,
the rates of fuel devolatilization and nes combustion
are of the same order of magnitude as the mixing phe-
nomena in a uidized bed. The mixing and contacting
of the gases and solids very often are the controlling
factors in reactor performance. This is especially true
in large commercial uidized beds where only a limited
number of discrete feed points for reactants is allowed
because of economic considerations. Unfortunately,
solids mixing in a uidized bed has not been studied
extensively, especially in large commercial uidized
beds, because of experimental diculties.
4.3.1 Solids Circulation Pattern
Yang et al. (1986) have shown that, based on the tra-
versing force probe responses, three separate axial
solids ow patterns can be identied. In the central
core of the bed, the solid ow direction is all upward,
induced primarily by the action of the jets and the
rising bubbles. In the outer regions, close to the vessel
walls, the solid ow is all downward. A transition zone,
in which the solids move alternately upward and
downward, depending on the approach and departure
of the large bubbles, was detected in between these two
regions.
The solids circulation patterns were investigated
with a force probe. Since the force probe is directional,
the upward solids movement will produce a positive
response from the probe and vice versa, the magnitude
of the response being an indication of the magnitude of
the solids circulation rate. The number of major peaks
per unit time is closely related to the actual bubble
frequency in the bed. The force probe data allow the
identication of three major solids ow regions in the
3 m model as shown in Fig. 8. At the central portion of
the bed, the solids ow is induced upward primarily by
jetting action at the lower bed height and by large
bubbles at the higher bed height. At the outer region
next to the vessel wall, the solids ow is all downward.
The region has a thickness of approximately 0.25 m.
Between these two regions the solids ow is alterna-
tively upward and downward, depending on the
approach and departure of large bubbles. No stagnant
region was evident anywhere in the bed.
In addition to the three solids circulation regions
readily identiable, the approximate jet penetration
depth and bubble size can also be obtained from Fig.
8. The jetting region can be taken to be the maximum
average value of jet penetration depth. From the jet
Figure 7 Gas entrainment into a 3.5 cm semicircular jet. (Adapted from Yang, 1998.)
boundary at the end of the jetting region, an initial
bubble diameter can be estimated. This value can be
taken to be the minimum value of the initial bubble
diameter. The diameter of a fully developed bubble can
be obtained from the bubble boundary in the devel-
oped-bubble region, as shown in Fig. 8. The central
region is thus divided further into three separate
regions axially: the jetting region, the bubble-develop-
ing region, and the developed-bubble region. Bubbles
were observed to coalesce in the bubble-developing
region during analysis of the motion pictures taken
through the transparent front plate.
4.3.2 Solids Circulation Rate
The results of the force probe measurement indicate
that the solid circulation rates increase with increasing
jet ow rates. A simple mechanistic model was devel-
oped to correlate the solids circulation data. The model
assumes that solids circulation inside the bed is
induced primarily by bubble motion. The solids circu-
lation pattern inside the bed can be divided into two
major regions radially. In the center of the bed, the
particle movement is predominantly upward and is
induced by bubbles disengaged from the central jet.
This region has a radius similar to the radius of the
average bubble size. In the outer region, the particles
move primarily downward. In the meantime, the par-
ticles in both regions exchange with each other across
the neighboring boundary at a constant rate of
W
z
g,cm
2
-s. This mechanistic model is shown schema-
tically in Fig. 9. Material balance in a dierential ele-
ment dz as shown in Fig. 9 gives,
in the bubble street region,
K
oX
/
J
oz
R
2
i
1 c
mf
( ),
p
oX
/
J
ot
2R
i
W
z
(X
/
J
X
J
) = 0
(49)
and in the annular region,
K
oX
J
oz
R
2
0
R
2
i
_ _
1 c
mf
( ),
p
oX
J
ot
2R
i
W
z
X
J
X
/
J
_ _
= 0
(50)
where
K = nV
B
f
w
(1 c
w
),
p
(51)
The data do not show any clear dependence on the
axial position, z. The axial dependence is thus assumed
to be negligible. Equations (49) and (50) are reduced
from partial dierential equations to ordinary dier-
ential equations. If we consider only the annular
region, Eq. (50) reduces to
dX
J
dt

2R
i
W
z
(R
2
o
R
2
i
)(1 c
mf
),
p
X
J
X
/
J
) = 0
_
(52)
Since both X
J
and X
/
J
are independent of z, the rela-
tionship between X
J
and X
/
J
can be approximated by
the material balance of the coarse particles injected
into the bed to serve as the tracer.
Figure 8 Solids circulating pattern in a jetting uidized bed.
Solving for X
/
J
we have
X
/
J
=
W
t
R
2
i
H(1 c
mf
),
p
R
o
R
i
_ _
2
1
_ _
1 c
mf
1 c
i
_ _
X
J
(53)
Substituting Eq. (53) into Eq. (52), after some mathe-
matical manipulation we get
dX
J
dt
PX
J
Q = 0 (54)
where
P =
2R
i
W
z
(R
2
o
R
2
i
)(1 c
mf
),
p
2W
z
R
i
(1 c
i
),
p
(55)
Q =
2W
z
W
i
R
i
(R
2
o
R
2
i
)H(1 c
mf
)(1 c
i
),
2
p
(56)
Equation (54) can be integrated with the boundary
condition that X
J
= 0 at t = 0 to give
X
J
X
o
J
=
1
t
w
P
Pt 1 exp(Pt) [ ]
_ _
(57)
The equilibrium tracer concentration in the bed after
complete mixing can be expressed as
X
0
J
=
W
t
H,
p
R
2
i
(1 c
i
) (R
2
o
R
2
i
)(1 c
mf
)
_ _ (58)
The voidage inside the bubble street, c
i
, can be calcu-
lated as
c
i
= c
mf
f
B
(1 c
mf
) = c
mf

nV
B
R
2
i
U
A
(1 c
mf
)
(59)
where f
B
is the volumetric fraction of bubbles occupy-
ing the bubble street region at any instant; it can be
evaluated from
f
B
=
nV
B
R
2
i
U
A
(60)
If the bubble frequency, bubble diameter, and bubble
velocity are known, the solids mixing rate can be cal-
culated.
In correlating the data, the solid exchange rate
between the two regions, W
z
, was assumed to be con-
stant. Comparison of the calculated and the experi-
mentally observed tracer concentration proles was
good.
The solids mixing study by injection of tracer par-
ticles indicated that the axial mixing of solids in the
bubble street is apparently very fast. Radial mixing
ux depends primarily on the bubble size, bubble velo-
city, and bubble frequency, which in turn depend on
the size of the jet nozzle employed and the operating jet
velocity.
4.4 Solid Entrainment Rate into Gas and GasSolid
Two-Phase Jets
A mathematical model for solid entrainment into a
permanent amelike jet in a uidized bed was pro-
posed by Yang and Keairns (1982). The model was
supplemented by particle velocity data obtained by
following movies frame by frame in a motion analyzer.
The particle entrainment velocity into the jet was
found to increase with increases in distance from the
jet nozzle, to increase with increases in jet velocity, and
to decrease with increases in solid loading in the gas
solid two-phase jet.
High-speed movies indicated that the entrained par-
ticles tended to bounce back to the jet boundary more
readily under high solid loading conditions. This may
Figure 9 Schematic of a mathematical model for solids cir-
culation in a jetting uidized bed.
explain why the entrainment rate decreases with
increases in solid loading in a two-phase jet. A ready
analogy is the relative diculty in merging into a rush-
hour trac as compared to merging into a light trac.
A simple model for solid entrainment into a perma-
nent amelike jet was described in Sec. 11.5.1 of
Chapter 3 Bubbling Fluidized Beds.
4.5 Scale-up Considerations
The development of commercial uidized bed proces-
sors generally requires intermediate stages of testing on
physical models simulating commercial equipment.
Simulation (or scale-up) criteria derived from uidized
bed momentum-conservation relations may be applied
to determine the design and operating conditions for
physical models discussed in Chapter 13 Fluidized
Bed Scale-up. Such criteria, while not totally estab-
lished at this time, have been applied to simulate a
pressurized uidized bed gasier having a large vertical
central jet for fuel feeding, combustion, and gasica-
tion (Yang et al., 1995).
Physical models of commercial uidized bed equip-
ment provide an important source of design informa-
tion for process development. A physical model of a
commercial uidized bed processor provides a small-
scale simulation of the uid dynamics of a commercial
process. While commercial processes will typically oper-
ate at conditions making direct observation of bed uid
dynamics dicult (high temperature, high pressure,
corrosive environment), a physical model is designed
to allow easy observation (room temperature and pres-
sure, nonreactive atmosphere, transparent vessel).
Cold ow studies have several advantages.
Operation at ambient temperature allows construction
of the experimental units with transparent plastic
material that provides full visibility of the unit during
operation. In addition, the experimental unit is much
easier to instrument because of operating conditions
less severe than those of a hot model. The cold
model can also be constructed at a lower cost in a
shorter time and requires less manpower to operate.
Larger experimental units, closer to commercial size,
can thus be constructed at a reasonable cost and within
an aordable time frame. If the simulation criteria are
known, the results of cold ow model studies can then
be combined with the kinetic models and the intrinsic
rate equations generated from the bench-scale hot
models to construct a realistic mathematical model
for scale-up.
The need for physical modeling of uidized bed
processors is dictated by the state of the art of uidized
bed scale-up technology. In general, no rational proce-
dure exists for scaling up a new uidized bed processor
concept that precludes the need for physical modeling.
Many empirically developed rules of thumb for ui-
dized bed scale-up exist in specic areas of uidized
bed application that are not generally applicable.
Existing mathematical modeling approaches are them-
selves based heavily on empirical descriptions of ui-
dized bed uid dynamics. These bubbling bed models
can be applied only where condence exists for the
empirical bubble ow description built into the
model. Fluidized bed processors operate over such a
broad range of uid dynamic regimes that this con-
dence rarely exists for new concepts.
Yang et al. (1995) described the application of this
scale-up approach. Comprehensive testing programs
were performed on two relatively large-scale simula-
tion units for a period of several years: both a 30 cm
diameter (semicircular) Plexiglas cold model and a 3 m
diameter (semicircular) Plexiglas cold model operated
at atmospheric pressure.
The results are highly signicant to the development
of uidized bed technology because they represent a
case study of a rational uidized bed development
approach. The extensive data generated are unique in
their equipment dimensions, pushing existing models
and correlations to new extremes and oering new
insights into large-scale equipment behavior. The
understanding of the hydrodynamic phenomena devel-
oped from the cold ow model studies, and the analy-
tical modeling reported was integrated with parallel
studies that investigated coal gasication kinetics, ash
agglomeration, charash separation, and nes recycle
to develop an integrated process design procedure.
4.6 Applications
The primary applications for large-scale jetting ui-
dized beds are in the area of coal gasication as
described by Yang et al. (1995), Kojima et al. (1995),
and Tsuji and Uemaki (1994). Smaller scale applica-
tions are for uidized bed coating and granulation dis-
cussed in Chapter 17, Applications for Coating and
Granulation.
5 ROTATING FLUIDIZED BEDS
Rotating uidized beds make use of the centrifugal
force, which can reach many times the gravitational
force, to increase the minimum uidizing velocity of
the particles and minimize the formation of bubbles
even at a very high gas ow rate. This can make the
rotating uidized beds very compact compared to con-
ventional uidized beds so that they can be utilized in
special applications. Pfeer et al. (1986) described an
application using a rotating uidized bed as a high-
eciency dust lter. Tsutsumi et al. (1994) investigated
the reduction of nitrogen oxides with soot emitted
from diesel engines using a centrifugal uidized bed.
The fundamental governing equations have been stu-
died by Kao et al. (1987), Chen (1987), Fan et al.
(1985), and Chevray et al. (1980). A schematic of a
rotating uidized bed is presented in Fig. 10.
5.1 Minimum Fluidization Velocity
Because both the centrifugal force and the drag force
on the particles in the centrifugal uidized bed vary
with radial position, there are three minimum uidiza-
tion velocities that can be dened (Qian et al., 1998).
The surface minimum uidization velocity is dened as
the point when the inner surface of the bed is uidized.
The pressure drop across the bed can be calculated on
the basis of a xed bed because the bed is essentially a
xed bed except at the surface. The point where the
whole bed is uidized (at the distributor plate) is called
the critical minimum uidization velocity. In this case,
the pressure drop can be determined as in a uidized
bed. Then there is the average minimum uidization
velocity calculated from these two extremes.
When we make use of the Wen and Yu modication
of the Ergun equation (1966), the surface minimum
uidization velocity, U
mfi
, can be derived as
U
mfi
,
f
d
p
j
r
o
r
i
= (33.7)
2
0.0408
,
f
(,
p
,
f
)d
3
p
o
2
r
i
j
2
_ _
0.5
33.7
(61)
Similarly, the critical minimum uidization velocity,
U
mfc
, can be expressed as
U
mfc
,
f
d
p
j
= (33.7)
2
0.0408
,
f
(,
p
,
f
)d
3
p
o
2
r
o
j
2
_ _
0.5
33.7
(62)
The average minimum uidization velocity can then be
calculated from
U
mf
,
f
d
p
j
= 33.7
C
2
C
1
_ _ _
2
0.0408
,
f
(,
p
,
f
)d
3
p
o
2
j
2
C
3
C
1
_
0.5
33.7
C
2
C
1
(63)
Figure 10 Schematic of horizontal and vertical rotating uidized beds.
where
C
1
= r
2
o
1
r
i
1
r
o
_ _
C
2
= r
o
ln
r
o
r
1
_ _
C
3
=
(r
2
o
r
2
i
)
2
(64)
5.2 Pressure Drop
For U
g
_ U
mfi
, the bed is a packed bed and the
pressure drop can be calculated from the packed bed
equation
dP
dr
= o
U
g
r
o
r
_ _
[
U
g
r
o
r
_ _
2
(65)
or
P = oU
g
r
o
ln
r
o
r
i
_ _
[U
2
g
r
2
o
1
r
i
1
r
o
_ _
(66)
where
o =
1650(1 c)j
d
2
p
[ =
24.5(1 c),
f
d
p
(67)
When U
g
_ U
mfc
, the bed is completely uidized and
the pressure drop can be calculated from the uidized
bed equation.
dP
dr
= (,
p
,
f
)(1 c)ro
2
(68)
or
P = (,
p
,
f
)(1 c)o
2
(r
2
o
r
2
i
)
2
(69)
With U
mfi
- U
g
- U
mfc
, the bed is partially uidized,
and the pressure drop can be calculated by summing
the pressure drop across the uidized bed and the
packed bed:
P = (,
p
,
f
)(1 c)o
2
(r
2
pf
r
2
i
)
2
oU
g
r
o
ln
r
o
r
pf
_ _
[U
2
g
r
2
o
1
r
pf
1
r
o
_ _
(70)
Qian et al. (1998) found that when a sintered metal
distributor plate was used for gas distribution, the
experimental pressure drop could be predicted fairly
well with the theoretical equation mentioned above.
However, when a distributor with slotted openings
was used, the experimental pressure drop is just
about 70% of that calculated from the theoretical
equation. Apparently, with a slotted distributor plate,
the bed was not completely uidized. The bed above
the webs was not uidized. Chen (1987) also reported
two dierent types of pressure drop in the literature.
One shows a pressure drop curve similar to that of a
conventional uidized bed, while the other one exhibits
a maximum pressure drop in the uidized region.
NOTATION
A
a
= cross-sectional area of annulus at any level
A
d
= cross-sectional area of the downcomer
A
r
= cross-sectional area of the draft tube
Ar = Archimedes number, ,
f
(,
p
,
f
)d
3
p
g,j
2
A
s
= cross-sectional area of spout at any level
C = tracer gas concentration
C
DS
= drag coecient of a single particle
C
m
= maximum tracer gas concentration at the
jet axis
D = draft tube diameter
= uidized bed or spouted bed diameter
D
b
= upper diameter of the stagnant bed in
conical spouted bed
D
B
= bubble diameter
(D
B
)
max
= maximum bubble diameter
d
D
= diameter of draft tube gas supply
D
i
= diameter of spout nozzle in spouted beds
= diameter of the bed bottom in conical
spouted bed
d
o
= diameter of jet nozzle
D
o
= diameter of the inlet in conical spouted bed
d
p
= mean solid particle diameter
(d
p
)
crit
= critical particle size in spouted bed
D
s
= spout diameter
D
s
= average spout diameter
F = total amount of gas leakage during bubble
formation from a jet
f
B
= volumetric fraction of bubble in bubble
street
f
p
= solid friction factor
f
w
= wake fraction of the bubble
G = supercial mass ux of spouting uid
G
j
= total gas ow rate through the jet
G
r
= total gas ow rate in the draft tube
H = bed height of a uidized bed or a spouted
bed
= height of the developed bed in the conical
spouted bed
H
F
= fountain height, m
H
m
= maximum spoutable bed height
H
mf
= bed height at minimum uidization
H
o
= height of stagnant bed in conical spouted
bed
L = height of draft tube or height of
downcomer
L
d
= distance between the distributor plate and
the draft tube inlet
L
j
= jet penetration length
L = distance required for acceleration of
particles
n = bubble frequency
P = pressure drop
P
1-2
= pressure drop between 1 and 2 (see Fig. 5)
P
1-4
P
2-3
P
3-4
r = radial distance from the jet axis or from
the spout axis
r
i
= radius of inner surface of granule bed
r
o
= radius of rotating uidized bed
r
pf
= radius of interface of uidized and packed
beds
r
s
= radius of the spout
r
1,2
= radial distance where gas velocity is one-
half the maximum gas velocity at the jet
axis
(r
1,2
)
c
= radial distance where tracer gas
concentration is one-half the maximum at
the jet axis
R
i
= radius of bubble street, R
i
= D
B
,2
R
o
= bed radius
(Re)
Do.ms
= Reynolds number at minimum spouting
based on D
o
(Re)
mf
= Reynolds number at minimum uidization,
= d
p
U
mf
,
f
,j
(Re)
p
= Reynolds number based on the slip
velocity, = d
p
(U
gr
U
pr
),
f
,j
(Re)
t
= Reynolds number based on the terminal
velocity, = d
p
U
t
,
f
,j
S
d
= total wall area in the downcomer
S
r
= total wall area in the draft tube
t = time
t
w
= total time required to inject all tracer
particles
U = supercial uid velocity
U
a
= supercial uid velocity in annulus of a
spouted bed at any level
U
A
= absolute bubble velocity
(U
cf
)
atm
= complete uidization velocity at
(U
cf
)
p
= complete uidization velocity at pressure P
U
fr
= supercial uid velocity in the draft tube
U
g
= supercial gas velocity
U
gd
= supercial gas velocity in the downcomer
U
gr
= supercial gas velocity in the draft tube
U
j
= supercial jet nozzle velocity
U
jb
= gas velocity at jet boundary
U
jm
= maximum gas velocity at the jet axis
U
jr
= gas velocity at radial distance r from the
jet axis
U
m
= maximum value of the minimum spouting
velocity at the maximum spoutable bed
height
U
mf
= supercial minimum uidization velocity
U
mfc
= critical minimum uidization velocity
U
mfi
= surface minimum uidization velocity
(U
mf
)
atm
= supercial minimum uidization velocity at
U
mf
)
p
= supercial minimum uidization velocity at
pressure P
U
ms
= minimum spouting velocity
(U
ms
)
0.5
= minimum spouting velocity for vessel
diameter less than 0.5 meters
U
pd
= solid particle downward velocity in the
downcomer
U
pt
= solid particle velocity in the draft tube
U
slug
= rising velocity of the gas slug relative to
the particle velocity at its nose
U
sz
= supercial upward uid velocity in spout
or fountain core at any level
U
t
= terminal velocity of a single solid particle
V
B
= volume of a gas bubble
V
0 max
= particle velocity along axis at bed surface
v
s
= local upward particle velocity in spout or
fountain core at any level
V
sd
= net upward supercial volumetric ow rate
of particles in the downcomer
V
sr
= net upward supercial volumetric ow rate
of particles in the draft tube
v
w
= downward particle velocity at column wall
W = mass downow rate of solids in annulus at
any level = mass upow rate of solids in
spout at same level
W
sd
= mass ux of particles in the downcomer,
W
sd
= V
sd
,
p
W
sr
= mass ux of particles in the draft tube,
W
sr
= V
sr
,
p
W
t
= cumulative weight of tracer particles
injected after time t
W
z
= radial solids mixing ux
X
o
j
= tracer particle weight fraction in the bed
after complete mixing
X
j
. X
/
j
= tracer particle weight fractions in annulus
and in bubble street, respectively
z = axial coordinate
Greek Letters
c = bed voidage
c(0) = bed voidage at r = 0
c(w) = bed voidage at the wall in conical spouted
bed
c
a
= annulus voidage at minimum spouting
c
bd
= bubble voidage in the downcomer
c
br
= bubble voidage in the draft tube
c
bs
= spout voidage at bed surface
c
d
= voidage in the downcomer
c
i
= voidage in bubble street
c
mf
c
r
= voidage in the draft tube
c
o
= loose bed voidage in conical spouted bed
c
s
= voidage in spout or fountain core at any
level
c
w
= voidage in bubble wake
j = viscosity of the uid
,
air(p=1)
= density of air at one atmosphere and room
temperature
,
b
= bed density or bulk density
,
f
= density of the uid
,
p
= solid particle density
s
= sphericity of the solid particle
o = rotating speed
= jet half angle
= included angle of the cone in conical
spouted bed
t
d
= particlewall shear stress in the downcomer
t
r
= particlewall shear stress in the draft tube
REFERENCES
Abramovich GN. The Theory of Turbulent Jets. Cambridge,
MA: M.I.T. Press, 1963.
Alappat BJ, Rane VC. Studies on the eects of various
design and operational parameters on solid circulation
rate in a recirculating uidized bed. Can J Chem Eng
73:248252, 1995.
Anabtawi MZ. Minimum spouting velocity for binary
mixture of particles in rectangular spouted beds. Can J
Chem Eng 76:132136, 1998.
Anabtawi MZ, Uysal BZ, Jumah RY. Flow characteristics in
a rectangular spout-uid bed. Powder Technol
69:205211, 1992.
Anagbo PE. Derivation of jet cone angle from bubble theory.
Chem Eng Sci 35:14941495, 1980.
Botterill JSM, Bessant DJ. The ow properties of uidized
solids. Powder Technol 8:213222, 1973.
Buchanan RH, Wilson B. The uid-lift solids recirculator.
Mech Chem Eng Trans (Australia) 117124, May 1965.
Chen YM. Fundamentals of a centrifugal uidized bed.
AIChE J 33:722728, 1987.
Chevray R, Chan YNI, Hill FB. Dynamics of bubbles and
entrained particles in the rotating uidized beds. AIChE
26:390398, 1980.
Chisti MY. Airlift Bioreactors. New York: Elsevier, 1989.
Chisti MY, Moo-Young M. Improve the performance of air-
lift reactors. Chem Eng Progr 3845, June 1993.
Clain JK, Fane AG. Spouting with a porous draft-tube.
Can J Chem Eng 61:356363, 1983.
Curran GP, Pasek B, Pell M, Gorin E. Pretreatment of
bituminous coals for pressure gasication. Proceedings
of Fluidized Bed Combustion Symposium, American
Chemical Society Meeting, Chicago, August 1973.
Decamps F, Dumont G, Goossens W. Vertical pneumatic
conveyor with a uidized bed as mixing zone. Power
Technol 5:299306, 1971/1972.
Epstein N, Grace JR. Spouting of particulate solids. In:
Fayed ME, Otten L, eds. Handbook of Powder Science
and Technology. New York: Chapman and Hall, 1997,
pp 532567.
Epstein N, Lim CJ, Mathur KB. Data and models for ow
distribution and pressure drop in spouted beds. Can J
Chem Eng 56:436447, 1978.
Ettehadieh B, Yang WC, Haldipur GB. Motion of solids,
jetting and bubbling dynamics in a large jetting uidized
Fan LS, Hwang SJ, Matsuura A. Some remarks on hydro-
dynamic behavior of a draft tube gasliquidsolid ui-
dized bed. AIChE Symp Ser 80(234):91, 1984.
Fan LT, Chang CC, Yu YS, Takahash T, Tanaka Z.
Incipient uidization condition for a centrifugal uidiza-
tion bed. AIChE J 31:9991009, 1985.
Foong SK, Barton RK, Ratclie JS. Characteristics of multi-
ple spouted beds. Mech Chem Eng Trans (Australia)
MCII(1,2):712, 1975.
Freedman W, Davidson JF. Hold-up and liquid circulation
in bubble columns. Trans Inst Chem Eng 47:T251T262,
1969.
Grace JR, Mathur KB. Height and structure of the fountain
region above spouted beds. Can J Chem Eng 56:533537,
1978.
Grbavcic ZB, Vukovic DV, Zdanski FK, Littman H.
Fluid ow pattern, minimum spouting velocity and
pressure drop in spouted beds. Can J Chem Eng
54:3342, 1976.
Hadzismajlovic DzE, Povrenovic DS, Grbavcic ZB, Vukovic
DV. A spout-uid bed dryer for dilute solutions contain-
ing solids. In: Grace JR, Shemilt LW, Bergougnou MA,
eds. Fluidization VI. New York: Engineering Foundation,
1989, pp 277284.
Hadzismajlovic DzE, Grbavcic ZB, Povrenovic DS, Vukovic
DV, Garic RV. The hydrodynamic behavior of a 0.95 m
diameter spout-uid bed with a draft tube. In: Potter OE,
Nicklin DJ, eds. Fluidization VII. New York: Engineering
He YL, Lim CJ, Grace JR, Zhu JX, Qin SZ. Measurements
of voidage proles in spouted beds. Can J Chem Eng
72:229234, 1994a.
He YL, Qin SZ, Lim CJ, Grace JR. Particle velocity proles
and solid ow patterns in spouted beds. Can J Chem Eng
72:561568, 1994b.
He YL, Lim CJ, Qin SZ, Grace JR. Spout diameters in full
and half spouted beds. Can J Chem Eng 76:702706,
1998a.
He YL, Lim CH, Grace JR. Hydrodynamics of pressurized
spouted beds. Can J Chem Eng 76:696701, 1998b.
Horsler AG, Thompson BH. Fluidization in the development
of gas making process. Proceedings of Tripartite Chem
Engr Conf. Montreal, 1968, p 51.
Horsler AG, Lacey JA, Thompson BH. High pressure
uidized beds. Chem Eng Progr 65:5964, October 1969.
Ishida M, Shirai T. Circulation of solid particles within the
uidized bed with a draft tube. J Chem Eng (Japan)
8:477481, 1975.
Judd MR, Masson H, Meihack W. Solid circulation and
gasication experiments in a uidized bed with a draft
tube. In: Kunii D, Toei R, eds. Fluidization. New York:
Kao J, Pfeer R, Tardos GI. On partial uidization in rotat-
ing uidized beds. AIChE J 33:858861, 1987.
Keairns DL, Yang WC, Newby RA, Hamm JR, Archer DH.
Circulating bed boiler concepts for steam and power
generation. Proceedings of 13th Intersociety Energy
Conversion Eng Conf. Warrendal: Society of
Automotive Engineers, 1978, pp 540547.
Kececioglu I, Yang WC, Keairns DL. Fate of solids fed
pneumatically through a jet into a uidized bed. AIChE
J 30:99110, 1984.
King DF, Harrison D. The minimum spouting velocity of a
spouted bed at elevated pressure. Powder Technol
26:103107, 1980.
tration in semi-cylindrical gas-uidized beds. In: Grace
JR, Matsen JM, eds. Fluidization. New York: Plenum
Press, 1980, pp 315324.
Kojima T, Yoshitake H, Kimura T, Matsukata M, Uemiya
S. Contribution of local reactions in the grid zone to the
performance of a jetting uidized bed gasier of coal char.
Energy Fuels 9:379383, 1995.
Krambrock W. Mixing and homogenizing of granular bulk
material in a pneumatic mixer unit. Powder Technol
15:199206, 1976.
LaNauze RD. A circulating uidized bed. Powder Technol
15:117127, 1976.
LaNauze RD, Davidson JF. The ow of uidized solids. In:
Keairns DL, ed. Fluidization Technology. Vol. 2.
Washington: Hemisphere, 1976, pp 113124.
Lee WJ, Kim SD. Hydrodynamics and CWM combustion
characteristics in an internally circulating uidized bed
combustor. In: Potter OE, Nicklin DJ, eds. Fluidization
479486.
Li Y. Spouted bed hydrodynamics at temperatures up to
580
o
C, M.A.Sc. thesis, University of British Columbia,
Vancouver, Canada, 1992.
Lim CJ, Mathur KB. Modeling of particle movement in
spouted beds. In: Davidson JF, Keairns DL, eds.
Fluidization. Cambridge: Cambridge University Press,
1978, pp 104109.
Littman H, Morgan III MH. A general correlation for the
minimum spouting velocity. Can J Chem Eng 61:369373,
1983.
Littman H, Morgan III MH, Vukovic DV, Zdanski FK,
Grbavcic ZB. Prediction of the maximum spoutable
height and the average spout to inlet tube diameter ratio
in spouted beds of spherical particles. Can J Chem Eng
57: 684687, 1979.
Mamuro T, Hattori H. Flow pattern of uid in spouted beds.
J Chem Eng Japan 1:15, 1968.
Mann U. Analysis of spouted-bed coating and granulation.
1. Batch operation. Ind Eng Chem Process Des Dev
22:288292, 1983.
Mann U, Crosby EJ. Cycle time distribution in continuous
systems. Chem Eng Sci 28:623627, 1973.
Mann U, Crosby EJ. Cycle time distribution measurements
in spouted beds. Can J Chem Eng 53:579581, 1975.
Mathur KB, Epstein N. Spouted Beds. New York: Academic
Press, 1974.
Mathur KB, Gishler PE. A technique for contacting gases
with coarse solid particles. AIChE J 1: 157164, 1955.
Matweecha DM. Blending apparatus. U.S. Patents 3,648,985
(1972); 3,729,175 (1973).
McMahon JF. Fluidized bed reactor. U.S. Patent 3,825,477
(1972).
McNab GS. Prediction of spout diameter. Brit Chem Eng
Proc Tech 17:532, 1972.
Merchuk JC, Ladwa N, Cameron A, Bulmer M, Pickett
A. Concentric-tube airlift reactors: eects of geometri-
cal design on performance. AIChE J 40:11051117,
1994.
AIChE J 21:507510, 1975.
Merry JMD. Fluid and particle entrainment into vertical jets
in uidized beds. AIChE J 22:315323, 1976.
Milne BJ, Berruti F, Behie LA. Solids circulation in an
internally circulating uidized bed (ICFB) reactor. In:
Potter OE, Nicklin DJ, eds. Fluidization VII. New
Miyahara T, Hamaguchi M, Sukeda Y, Takahashi T. Size of
bubbles and liquid circulation in a bubble column with a
draft tube and sieve plate. Can J Chem Eng 64:718725,
1986.
Oguchi U, Kubo J. Liquidsolid particles or liquidgas
solid particle contacting method. U.S. Patent 3,754,993
(1973).
Ohoka I, Conway HL. Progress in the gasication of heavy
hydrocarbon oils in a recirculating uidized bed hydroge-
nator. Proceedings of the 5th Synthetic Pipeline Gas
Symposium, Chicago, 1973.
Olazar M, San Jose MJ, Aguayo AT, Arandes JM, Bilbao J.
Stable operation conditions for gassolid contact regimes
in conical spouted beds. Ind Eng Chem Res 31:17841792,
1992.
Pressure drop in conical spouted beds. Chem Eng J
51:5360, 1993a.
Design factors of conical spouted beds and jet spouted
beds. Ind Eng Chem Res 32:12451250, 1993b.
Olazar M, San Jose MJ, Penas FJ, Aguayo AT, Bilbao J. The
stability and hydrodynamics of conical spouted beds with
binary mixtures. Ind Eng Chem Res 32:28262834, 1993c.
Olazar M, San Jose MJ Llamosas R, Alvarez S, Bilbao J.
Study of local properties in conical spouted beds using an
optical ber probe. Ind Eng Chem Res 34:40334039,
1995.
Peterson WS. Multiple spouted beds. Can. Patent No.
739,660 (1966).
Pfeer R, Tardos GI, Gal E. The use of a rotating uidized
bed as a high eciency dust lter. In: Ostergaard K,
Sorengsen A, eds. Fluidization V. New York:
Qian GH, Bagyi I, Pfeer R, Shaw H, Stevens JG. A para-
metric study of a horizontal rotating uidized bed using
slotted and sintered metal cylindrical gas distributors.
Rajaratnam N. Turbulent Jets. New York: Elsevier, 1976.
Rakow MS, Calderon M. The dynacracking processan
update. Chem Eng Prog 77:3136, February 1981.
San Jose MJ, Olazar M, Alvarez S, Bilbao J. Local bed
voidage in conical spouted beds. Ind Eng Chem Res
37:25532558, 1998.
Siegel MH, Merchuk JC, Schugerl K. Air-lift reactor analy-
sis: interrelationships between riser, downcomer, and gas
liquid separator behavior, including gas recirculation
eects. AIChE J 32:15851596, 1986.
Silva EMV, Ferreira MC, Freire JT. Mean voidage measure-
ments and uid dynamics analysis in a circulating ui-
dized bed with a spouted bed type solids feeding system.
Proceedings of 2nd European Thermal-Sciences and 14th
UIT National Heat Transfer Conference, Rome, 1996.
Singh B. Theory of slugging lifters. Powder Technol
21:8189, 1978.
Solt PE. Airlift blending apparatus, U.S. Patent 3,647,188
(1972).
Stocker RK, Eng JH, Svrcek WY, Behie LA. Gas residence
time distribution studies in a spouted bed with a draft
tube. In: Grace JR, Shemilt LW, Bergougnou MA, eds.
1989, pp 269276.
Taskaev ND, Kozhina ME. Semicoking of Kok-Yangak coal
in a circulating bed. Trudy Akad Nauk Kirgiz S.S.R.
7:109, 1956.
Tsuji T, Uemaki O. Coal gasication in a jet-spouted bed.
Can J Chem Eng 72:504510, 1994.
Tsutsumi A, Demura M, Yoshida K. Reduction of nitrogen
oxides with soot emitted from diesel engines by a centri-
fugal uidized bed. AIChE Symposium Serier
90(301):152, 1994.
Turton R, Tardos GI, Ennis BJ. Fluidized bed coating and
granulation. In: Yang WC, ed. Fluidization, Solids
Handling, and Processing. Westwood: Noyes, 1999,
pp 331434.
1966.
Westinghouse Electric Corporation. Advanced coal gasica-
tion system for electric power generation. Quarterly
Progress Report, NTIS No. FE-1514-61, January 1977.
Whiting KJ, Geldart D. A comparison of cylindrical and
semi-cylindrical spouted beds of coarse particles. Chem
Eng Sci 35:14991501, 1980.
Wu SWM, Lim CJ, Epstein N. Hydrodynamics of spouted
beds at elevated temperatures. Chem Eng Commun
62:251268, 1987.
Wurster DE, Lindlof JA. Particle coating apparatus. U.S.
Patent 3,241,520 (1966).
Wurster DE, Lindlof JA, Battista JR. Process for preparing
agglomerates. U.S. Patent 3,207,824 (1965).
Yang WC. A unied theory on dilute phase pneumatic trans-
port. J Powder Bulk Solids Tech 1:8995, 1977.
Yang WC. A correlation for solid friction factor in vertical
pneumatic conveying lines. AIChE J 24:548552, 1978.
Yang WC. Jet penetration in a pressurized uidized bed.
I&EC Fundamentals 20:297300, 1981.
Yang WC. Comparison of jetting phenomena in 30-cm and
3-m diameter semicircular uidized beds. Powder Technol
100:147160, 1998.
Yang WC. Engineering and applications of recirculating and
jetting uidized bed. In: Yang WC, ed. Fluidization,
Solids Handling, and Processing. Westwood: Noyes,
1999, pp 236330.
Yang WC, Keairns DL. Recirculating uidized bed reactor
data utilizing a two-dimensional cold model. AIChE
Symposium Series 70(141):2740, 1974.
Yang WC, Keairns DL. Estimating the acceleration pressure
drop and the particle acceleration length in vertical and
horizontal pneumatic transport lines. Proceedings of the
Pneumotransport 3, Bedford (England): BHRA Fluid
Engineering, 1977, pp D7-89D7-98.
Yang WC, Keairns DL. Design of recirculating uidized
beds for commercial applications. AIChE Symposium
Series 74(176):218228, 1978a.
Yang WC, Keairns DL. Design and operating parameters for
a uidized bed agglomerating combustor/gasier. In:
Davidson JF, Keairns DL, eds. Fluidization.
Cambridge: Cambridge University Press, 1978b,
pp 208214.
Yang WC, Keairns DL. Momentum dissipation of and gas
entrainment into a gassolid two-phase jet in a uidized
bed. In: Grace JR, Matsen JM, eds. Fluidization. New
York: Plenum Press, 1980, pp 305314.
Yang WC, Keairns DL. Solid entrainment rate into gas and
gassolid, two-phase jets in a uidized bed. Powder
Technol 33:8994, 1982.
Yang WC, Keairns DL. Studies on the solid circulation rate
and gas bypassing in spouted uid-bed with a draft tube.
Can J Chem Eng 61:349355, 1983.
Yang WC, Keairns DL. Operational characteristics of a 2-D
bed with three draft tubes. In: Grace JR, Shemilt LW,
Bergougnou MA, eds. Fluidization. New York:
Yang WC, Margaritis PJ, Keairns DL. Simulation and mod-
eling of startup and shutdown in a pilot-scale recirculating
uidized bed coal devolatilizer. AIChE Symposium Series
74(176):87100, 1978.
Yang WC, Keairns DL, McLain DK. Gas mixing in a jetting
uidized bed. AIChE Symposium Series 80(234):3241,
1984a.
Yang WC, Revay D, Anderson RG, Chelen EJ, Keairns DL,
Cicero DC. Fluidization phenomena in a large-scale cold-
ow model. In: Kunii D, Toei R, eds. Fluidization. New
York: Engineering Foundation, 1984b, pp 7784.
Yang WC, Ettehadieh B, Haldipur GB. Solids
circulation pattern and particles mixing in a
large jetting uidized bed. AIChE J 32:1994
2001, 1986.
Yang WC, Ettehadieh B, Anestis TC, Kettering RE,
Haldipur GB, Holmgren JD. Gas mixing in a large jetting
uidized bed. In: Kwauk M, Kunii D, eds.
Fluidization88: Science and Technology, Beijing:
Yang WC, Newby RA, Keairns DL. Large scale uidized
Technol 82:331346, 1995.
Ye B, Lim CJ, Grace JR. Hydrodynamics of spouted and
spout-uidized beds at high temperature. Can J Chem
Eng 70: 840847, 1992.
Yoon SM, Kunii D. Gas ow and pressure drop through
moving beds. I&EC Process Des Develop 9:559565,
1970.
21
Standpipes and Nonmechanical Valves
T. M. Knowlton
1 INTRODUCTION
Plants that process solids are known to have more
problems achieving design capacity than plants hand-
ling only liquids and gases. Merrow (1985) listed the
major sources of problems in operating solids proces-
sing plants in his study on the reasons for their poor
performance. He found that one of the primary trouble
spots in most solids processing plants was in getting
the solids to ow smoothly and consistently. Thus in
many instances the key to successful process operation
is how well the solids transport systems have been
designed. Preventing problems from occurring by
good design of the elements of a transportation system
can speed plant start-up and minimize downtime.
This chapter discusses two important elements of a
solids transport system: standpipes and nonmechanical
valves. Although they are very simple in conguration,
trying to design and operate these devices without a
basic understanding of their principles of operating can
lead to much frustration and wasted time. Describing
how standpipes and nonmechanical valves operate is
the purpose of this chapter.
2 STANDPIPES
A standpipe is essentially a length of pipe through
which solids ow. Solids can ow through a standpipe
in either dilute or dense-phase ow. Standpipes can be
vertical, angled, or a mixture of angled and vertical
pipes called a hybrid standpipe. The standpipe was
invented by a research group working at the Jersey
Standard Company in the 1940s (Campbell et. al.,
1948) trying to develop an FCC unit to produce
high-octane aviation gasoline in World War II.
The purpose of a standpipe is to transfer solids from
a region of lower pressure to a region of higher pres-
sure. This is schematically shown in Fig. 1, where
solids are being transferred from a uidized bed at
pressure P
1
to another uidized bed operating at P
2
,
which is higher than P
1
.
Solids can be transferred by gravity from a low
pressure to a higher pressure if gas ows upward rela-
tive to the downward owing solids. This relative gas
solids ow will then generate the sealing pressure drop
required for the system. The direction of the actual gas
ow in the standpipe relative to the standpipe wall can
be either up or down and still have the relative gas
solids velocity, v
r
, directed upwards. This is sometimes
dicult to understand, but it can be explained with the
aid of Fig. 2, and the denition of the relative velocity.
The relative gassolids velocity, v
r
, is dened as
v
r
v
s
v
g
1
where v
s
is the solids velocity, and v
g
is the interstitial
gas velocity (v
g
U=").
It is generally easier to visualize what is occurring in
a solids transfer system by mentally traveling along
with the solids. Therefore the positive reference direc-
tion for determining v
r
in this chapter will be the direc-
tion in which the solids are owing. For standpipe
ow, this direction is downward.
In Fig. 2 solids are being transferred downward in a
standpipe from pressure P
1
to a higher pressure P
2
.
Solids velocities in Fig. 2 are denoted by the length
of the bold arrows, gas velocities by the length of the
dashed arrows, and the relative gassolids velocity by
the length of the thin-lined arrows.
For Case 1 in Fig. 2, solids are owing downward,
and gas is owing upward relative to the standpipe
wall. The relative velocity is directed upward and is
equal to the sum of the solids velocity and the gas
velocity, i.e.,
v
r
v
s
v
g

v
s
v
g
2
For Case 2, solids are owing down the standpipe
relative to the standpipe wall. Gas is also owing
down the standpipe relative to the standpipe wall,
but at a velocity less than that of the solids. For this
case, the relative velocity is also directed upward and is
equal to the dierence between the solids velocity and
the gas velocity, i.e.,
v
r
v
s
v
g
3
In both cases, if one were riding down the standpipe
with the solids, the gas would appear to be moving
upward relative to your reference point.
The gas owing upward relative to the solids gen-
erates a frictional pressure drop. The relationship
between the pressure drop per unit length (P=Lg)
and the relative velocity for a particular material is
determined by the uidization curve for that material.
Normally, this uidization curve is generated in a ui-
dization column where the solids are not owing.
However, the relationship also applies for solids ow-
ing in a standpipe.
Nearly all standpipe transfer systems use either
Geldart group A or Geldart group B solids. The ui-
dization curve for Geldart group B solids diers from
that for group A solids. For both types of solids, as the
relative gas velocity through the bed increases from
zero, the P=Lg through the bed increases linearly
with v
r
. This region is called the packed bed region. At
some v
r
, the P generated by the gas owing through
the solids is equal to the weight of the solids per unit
area, and the solids become uidized. The relative velo-
city at this point is termed the interstitial minimum
Figure 1 Overow standpipe.
Figure 2 Concept of relative gassolids velocity.
uidization velocity, v
mf
, or U
mf
="
mf
. The P=Lg at
v
mf
is designated as P=Lg
mf
and is often referred to
as the uidized bed density at minimum uidization,
because P=Lg has the units of density. For more
detailed discussion on this subject, see Sec. 1.3 in
Chapter 3 Bubbling Fluidized Beds.
Increases in v
r
above v
mf
do not lead to further
increases in P=Lg. For Geldart group B materials,
nearly all of any gas ow in excess of that required
at v
mf
goes into the formation of bubbles. Therefore,
as v
r
increases beyond v
mf
, P=Lg remains almost
constant and then begins to decrease as the bubble
volume in the bed increases.
For Geldart group A materials, as v
r
is increased
above v
mf
the solids expand without bubble generation
over a certain velocity range. Because of this bubbleless
expansion, P=L decreases over this velocity range.
The velocity where bubbles begin to form in group A
materials is called the minimum bubbling velocity, v
mb
.
Typical uidization curves for Geldart group A and
Geldart group B materials are shown in Fig. 3.
Standpipes generally operate in three basic ow
regimes: packed bed ow, uidized bed ow, and a
dilute-phase ow called streaming ow.
1. Packed Bed Flow. In packed bed ow v
r
is less
than v
mf
, and the voidage in the standpipe is more or
less constant. As v
r
is increased, P=L
g
increases more
or less linearly in packed bed ow. When a standpipe is
operating in the moving packed bed ow regime, a
ow condition is sometimes reached that causes the
solids to stop momentarily and then start again. This
often causes the standpipe to vibrate, and a loud chat-
tering noise can often be heard. This type of ow is
called stickslip ow. It should be avoided, but no
method presently exists to predict when it will occur.
2. Fluidized Bed Flow. In uidized bed ow, v
r
is
equal to or greater than v
mf
. The voidage in the stand-
pipe can (and generally does) change along the length
of the standpipe, and P=Lg does not change with
increasing v
r
. There are two kinds of uidized-bed
ow, bubbling and nonbubbling. When a group B
solid is uidized, it always operates in the bubbling
uidized bed mode because bubbles are formed at all
relative velocities above v
mf
. However, for group A
solids, there is an operating window corresponding to
a relative velocity between v
mf
and v
mb
, where the
solids are uidized but no bubbles are formed in the
standpipe. A standpipe operating with group A solids
and with a relative velocity above v
mb
operates in the
bubbling uidized bed mode.
Bubbles, especially large bubbles, are undesirable in
a standpipe. If a standpipe is operating in the bubbling
uidized bed mode so that the solids velocity, v
s
, is less
than the bubble rise velocity, u
b
, then bubbles will rise
and grow by coalescence. The bubbles rising against
the downowing solids hinder and limit the solids ow
rate (Knowlton and Hirsan, 1978; Eleftheriades and
Judd, 1978) because the solids ow area is reduced
by the presence of bubbles. The larger the bubbles
are, the greater the hindrance to solids ow.
When the solids velocity in the standpipe is greater
than the bubble rise velocity, the solids will carry the
bubbles down the standpipe relative to the standpipe
wall. It is also possible for bubbles to coalesce and
hinder solids ow when they are traveling downward.
In this case, the small bubbles are carried downward
faster than the larger bubbles. When they catch up to a
larger bubble, they coalesce, which results in even lar-
ger bubbles and thus hindrance to the solids ow.
Bubbles also reduce the P=Lg or density of the
solids in the standpipe. Thus a standpipe operating in
the bubbling regime will require a longer length to seal Figure 3 Fluidization curvesgroup A and group B solids.
the same dierential pressure than a standpipe in
which the same solids are slightly above minimum ui-
dizing conditions.
Therefore for optimum uidized-bed standpipe
operation, for group B solids, v
r
should be maintained
just slightly above v
mf
; for group A solids, v
r
should be
maintained in a range just slightly below v
mb
to just
slightly above v
mb
.
The relative velocity range where it is best to operate
standpipes with both group A and group B solids is
shown schematically in Fig. 4. In both cases, operation
in these areas will either reduce the formation of bub-
bles or allow standpipe operation with only small bub-
bles.
Because it is dicult to prevent small bubbles from
occurring in standpipes, small-diameter standpipes are
more dicult to operate than large-diameter stand-
pipes. In small-diameter standpipes, even small bub-
bles have a diameter that is a large fraction of the
diameter of the standpipe, and these bubbles can hin-
der solids ow in small standpipes. In large standpipes,
this same small bubble diameter will not signicantly
aect solids ow (Fig. 5)
3. Streaming Flow. Underow standpipes (espe-
cially cyclone diplegs) sometimes operate in a dilute-
phase streaming ow characterized by high voidages.
A substantial amount of gas can be carried down the
standpipe when operating in this mode (Geldart and
Broodryk, 1991). Excessive gas ow down a dipleg is
usually undesirable. It can be minimized by reducing
the mass ux in the dipleg and/or increasing the pres-
sure drop that the dipleg has to seal. This usually
means increasing the immersion of the dipleg into the
uidized bed.
Figure 4 Optimum standpipe operating regions for group A and group B solids.
There are two basic types of standpipe congurations,
the overow standpipe (Fig. 6A) and the underow
standpipe (Fig. 6B). The overow standpipe is so
named because the solids overow from the top of
the uidized bed into the standpipe, and there is no
bed of solids above the standpipe. In the underow
standpipe, the solids are introduced into the standpipe
from the underside, or bottom, of the bed or hopper,
and a bed of solids is present above the standpipe. With
this denition, a cyclone dipleg is classied as an over-
ow standpipe because there is no bed of solids above
the entrance to the dipleg.
With two types of standpipe congurations and the
two typical standpipe ow regimes (uidized and
packed bed), there are four dierent types of stand-
pipes:
1. An underow packed bed standpipe
2. An underow uidized bed standpipe
3. An overow uidized bed standpipe
4. An overow packed bed standpipe
All of these standpipe are used extensively in industry
except for the overow packed bed standpipe. It is
possible for this type of standpipe to operate, but it
is much harder to operate and control than the others.
Therefore, it is not used.
Each type of standpipe can be constrained (solids
ow limited) at either the top or the bottom. However,
top-constrained standpipes are relatively rare, and
nearly all standpipes used in uidized bed standpipe
systems are constrained at the bottom. The standpipes
discussed in this chapter will all be bottom-constrained
standpipes.
In any gassolids ow system, a pressure drop loop
can be dened so that the sum of the pressure drop
Figure 5 Small and large bubbles in standpipes.
Figure 6 Overow and underow standpipes.
components around the loop is zero. In many (but not
all) pressure drop loops, the standpipe is the dependent
part of the loop. This means that the pressure drop
across the standpipe will automatically adjust to bal-
ance the pressure drop produced by the other indepen-
dent components (which do not automatically adjust
their pressure drops) in the loop. How the standpipe
pressure drop adjustment is made is dierent for over-
ow and underow standpipes.
Consider the overow standpipe system shown in
Fig. 7A. Solids are being transferred at a low rate
from the upper uidized bed to the lower uidized
bed against the pressure dierential P
2
-P
1
. This pres-
sure drop consists of the pressure drop in the lower
uidized bed from the standpipe exit to the top of
the bed (P
lb
), the pressure drop across the distributor
supporting the upper uidized bed (P
d
), and the pres-
sure drop across the upper uidized bed (P
ub
), i.e.,
P
2
-P
1
P
lb
P
d
P
ub
4
This pressure drop must be balanced by the pressure
drop generated in the overow standpipe. If the stand-
pipe is operating in the uidized bed mode at a P=Lg
equal to (P=Lg
mf
, the solids height in the standpipe,
H
sp
, will adjust so that the pressure buildup generated
in the standpipe, P
sp
, will equal the product of
(P=Lg
mf
and H
sp
, i.e.,
P
sp
P
2
-P
1
P
lb
P
d
P
ub
P
Lg

mf
H
sp
5
If the gas ow rate through the two beds is increased,
P
d
increases, while the pressure drops across the two
uidized beds essentially remain constant. Therefore
P
2
-P
1
increases to P
0
2
-P
1
. The pressure drop across
the overow standpipe will also increase to P
0
2
-P
1
.
The standpipe pressure drop increases because the
height of uidized solids in the standpipe increases
from H
sp
to H
0
sp
, i.e.,
P
sp
P
0
2
-P
1

P
Lg

mf
H
0
sp
6
This is shown schematically in Fig. 7B, and the change
is also reected in the pressure diagram. If the increase
in the pressure drop across the distributor is so much
that H
sp
must increase to a value greater than the
standpipe height available to seal the pressure dier-
ential, the standpipe will not operate.
As indicated above, the most common perception of
an overow standpipe is that it consists of a uidized
dense phase at the bottom (the height of which is pro-
portional to the pressure drop across the dipleg), and a
dilute phase in the upper part (which does not contri-
bute signicantly to the sealing pressure drop in the
standpipe). However, recent testing has shown that
this picture is not necessarily correct for overow
standpipes operating at appreciable solids mass uxes.
Geldart and Broodryk (1991) reported that at high
mass uxes in an overow standpipe, there is little or
no dierence between the densities in the top and bot-
tom sections of the standpipe. This is illustrated in Fig.
8 which shows data obtained in a dipleg at PSRI
(1995). As seen in this gure, the typical dense phase
at the bottom of the overow standpipe (cyclone dip-
leg) only occurs at very low solid mass uxes. At med-
ium to high solid mass uxes, the standpipe density
was found to be relatively evenly distributed through-
out the entire length of the standpipe.
When a cyclone dipleg is operating with a dense
phase at the bottom and a dilute phase at the top,
the pressure drop through the dilute-phase section at
the top is generally assumed to be zero when calculat-
ing the required dipleg length. This is not the case, Figure 7 Overow standpipe operation.
however, and at high mass uxes through the dipleg,
this dilute-phase pressure drop can be large enough
that with a suciently long dipleg and a low pressure
drop across the dipleg, the dilute phase alone can gen-
erate the required sealing pressure drop. This situation
is shown schematically in Fig. 9.
When a dipleg operates in streaming ow, substan-
tial amounts of gas can be dragged down the dipleg by
the owing solids. The amount of gas being transferred
may be very large (as much as 1/3 of the gas entering
the cyclone can be transferred down the dipleg under
certain conditions; PSRI, 1995). In many cases, it is
not desirable to have so much gas being transferred
down the standpipe with the solids. By increasing the
pressure drop across the standpipe (increasing the
P=Lg across the standpipe) or reducing the solids
mass ux through the standpipe, the amount of gas
being transferred down the standpipe can be substan-
tially reduced. Practically, this means that the gas ow
down the standpipe or dipleg can be decreased by
immersing the standpipe further in the uidized bed
(increasing P=Lg), increasing the diameter of the
standpipe (reducing the mass ux), or reducing the
length of the standpipe, if possible (increasing P=Lg).
Wirth (1995) conducted a study of a standpipe oper-
ating between a cyclone and a loop seal at the bottom
of a standpipe. He also found that if the solids mass
ux in the standpipe increased above a certain value,
the standpipe lost its dense-phase seal at the bottom.
This increased mass ux was accompanied by increas-
ing amounts of gas owing through the standpipe. If
the solids mass ux was increased further beyond a
certain threshold value, an even greater ow of gas
was observed in the standpipe, with the result that
even the upward-owing part of the loop seal became
dilute. Wirth found that this situation could be con-
trolled by decreasing the solids ux through the stand-
pipe. This could be accomplished in two ways, by
decreasing the rate of solids owing around the unit,
or more practically, by increasing the diameter of the
standpipe. Increasing the diameter of the standpipe
decreases the solids mass ux in the downcomer and
therefore the dilute-phase density in the standpipe.
When the dilute-phase density is decreased below a
Figure 8 Cyclone dipleg pressure drop proles.
critical value, a dense phase appears at the bottom of
the standpipe.
Streaming ow is also more likely to occur with long
standpipes. Therefore to prevent this type of situation
from occurring and causing poor loop seal operation in
combustors, it is recommended that a large-diameter
downcomer be used.
In Fig. 10A, solids are being transferred through an
underow standpipe (operating in the packed bed
mode) from the upper uidized bed to the freeboard
of the lower uidized bed against the dierential pres-
sure P
2
-P
1
. The dierential pressure P
2
-P
1
consists of
the pressure drop across the gas distributor of the
upper uidized bed P
d
. However, there is also a pres-
sure drop across the solids ow control valve P
v
.
Therefore the standpipe pressure drop P
sp
must
equal the sum of P
d
and P
v
, i.e.,
P
sp

P
Lg

H
sp
P
d
P
v
P
2
-P
1
P
v
7
Thus for this packed bed underow standpipe case, the
standpipe must generate a pressure drop greater than
P
2
-P
1
. This is shown as Case I in the pressure diagram
of Fig. 10B.
If the gas ow rate through the column is
increased, P
d
will increase. If P
v
remains constant,
then P
sp
must also increase to balance the pressure
drop loop. This is shown as Case II in the pressure
diagram of Fig. 10B. Unlike the overow standpipe
case, the solids level in the standpipe cannot rise to
increase the pressure drop in the standpipe. However,
the P=Lg in the standpipe must increase in order to
balance the pressure drop around the loop. This
occurs in a packed bed standpipe because of an
increase in v
r
in the standpipe. This can be visualized
with the aid of Fig. 10C.
For Case I, the pressure drop in the bed was satis-
ed by having the standpipe operate at point I on the
P=Lg versus v
r
curve, as shown in Fig. 10C. When
the pressure drop across the distributor increased,
the v
r
in the standpipe adjusted to generate a higher
Figure 9 How diplegs operate in streaming ow.
Figure 10 Underow standpipe operation.
P=Lg, (P=Lg
II
, to balance the higher pressure
drop.
If the pressure drop across the distributor is
increased so that the product of (P=Lg
mf
(the max-
imum P=Lg possible in the standpipe) and the stand-
pipe length, H
sp
, is less than the sum of P
v
and P
d
,
then the underow standpipe will not seal.
As stated above, the standpipe is not always the
component that adjusts to balance the change in pres-
sure drop in a loop (i.e., it is not always the dependent
part of the pressure drop loop). In uidized catalytic
crackers, uidized underow standpipes are used that
do not adjust for changes in pressure drop. This type of
standpipe is one of the most widely used standpipes in
industry. How does this type of uidized underow
standpipe operate? Unlike the underow standpipe in
the previous example, the standpipe pressure drop will
not adjust to balance pressure drop changes, because
the standpipe is operating in the uidized bed mode. In
the uidized mode, changes in the relative velocity will
not cause changes in the pressure drop in the standpipe
as in an underow nonuidized standpipe. Also,
because it is an underow standpipe and operating
full of solids, changes in the bed height to balance the
pressure drop are not possible. Therefore this type of
standpipe is designed so that its length is long enough to
generate more pressure, or head, than required. The
excess pressure generated by the standpipe is then
burned up across the slide valve in order to balance
pressure drop changes in the other loop components.
With the uidized underow standpipe, aeration gas
is added to the standpipe to maintain the solids in a
uidized state as they ow down the standpipe. As the
solids ow down the uidized underow standpipe
from a low pressure to a higher pressure, the gas in
the standpipe is compressed, which causes the solids to
move closer together. When the standpipe is operating
at low pressures, the percentage change in gas density
from the top of the standpipe to the bottom can be
signicant. If aeration is not added to the standpipe to
prevent this, the solids can deuidize near the bottom
of the standpipe (Fig. 11A). Deuidization of solids in
the standpipe results in less pressure buildup in the
standpipe and a reduction in the solids ow rate
around the loop.
To maintain the solids in a uidized underow
standpipe in a uidized state, aeration gas is added
to the standpipe. Adding the correct amount of gas
uniformly (every 2 to 3 meters) in a commercial ui-
dized underow standpipe will prevent deuidization
Figure 11 Adding aeration prevents deuidization.
at the bottom of the standpipe (Fig. 11B). If the mate-
rial owing in the standpipe is a Geldart group A mate-
rial, it is required that the aeration be added uniformly
along the standpipe. If the aeration is added at only one
location (i.e., at the bottom of the standpipe), a large
bubble will form in the standpipe at the aeration point
(Fig. 12A). If the bubble is large enough, it can restrict
the ow of solids down the standpipe. The large bubble
forms because it is dicult for the aeration gas to
permeate through the very ne solids moving through
the standpipe. Therefore it requires a signicant area
for the gas to dissipate through the very ne particles at
the same rate that it is being added through the aeration
tap. If the aeration gas is added at several locations,
then the bubble size is signicantly reduced, and stand-
pipe operation is signicantly improved (Fig. 12B).
Note in Figs. 12A and 12B that the bubbles are
shown extending downward from the aeration point
in the direction of ow of the solids. This occurs
because the momentum of the solids is much greater
than the buoyancy force of the bubble and elongates
the aeration bubble in the direction of ow.
For Geldart group B solids, it is often unnecessary
to add aeration at several locations along the standpipe
to maintain the standpipe in uidized ow. Adding
aeration at the bottom of the standpipe operating
with group B solids is sometimes sucient. This is
because the gas can permeate through the larger
group B particles much easier than through the
group A particles (group A particles have a signi-
cantly larger surface area and produce more drag for
the same gas ow conditions).
The amount of aeration required to maintain solids
in a uidized state throughout the standpipe was pre-
sented by Karri and Knowlton (1993) as
Q 1000
P
b
P
t
1
mf
sk

sk

8
In a commercial uidized underow standpipe, the
amount of aeration theoretically required is added in
equal increments via aeration taps located approxi-
mately 2 to 3 meters apart. Care should be taken not
to overaerate the standpipe. If this occurs, large bub-
bles are generated in the standpipe that hinder solids
owing down the standpipe. Thus standpipes can be
overaerated as well as underaerated.
As indicated above, it is detrimental to have bubbles
in standpipes. For uidized bed underow standpipes
with the standpipe entrance in the uidized bed, bub-
bles can be sucked down the standpipe at its entrance if
nothing is done to prevent this from occurring. This is
especially true when the bed consists of Geldart group
Figure 12 Single and multiple aeration points (group A solids).
A solids. When solids ow from a uidized bed into the
top of an underow uidized bed standpipe, the solids
are accelerated from a low velocity near 0 m/s in the
bed to as much as 2 m/s in the standpipe. This sudden
increase in solids velocity can carry bubbles with the
solids down into the standpipe and degrade standpipe
operation. To prevent this, a cone is often added to the
top of the standpipe to minimize the solids velocity at
the standpipe entrance and minimize bubble carry-
under. Experience has shown that the diameter of
the standpipe inlet cone should be at least four times
the area of the standpipe (King, 1992). Standpipe inlet
cones typically have an included angle of from 25 to 35
degrees.
In many uidized beds, a sparger type of gas distri-
butor is used to uidize the bed. The sparger consists
of a pipe with nozzles in it inserted into the bottom of
the uidized bed. Solids ow down through the distri-
butor and into the standpipe. Another technique to
prevent bubbles from entering the standpipe can be
used with sparger grids. Instead of having the stand-
pipe entrance in the bed, the standpipe entrance is
located below the sparger grid. As the solids ow
between the sparger grid and the standpipe entrance,
the bubbles dissipate and do not enter the standpipe.
Generally, an aeration ring is added around the stand-
pipe to ensure that the solids are uidized as they enter
the standpipe.
If Geldart group A solids are deuidized when they
enter the standpipe, they are extremely dicult to
reuidize (Ross, 1997). Therefore it is necessary to pre-
vent this deuidization. It is somewhat puzzling why it
is so dicult to reuidize group A solids. In a labora-
tory uidized bed, these solids are relatively easy to
uidize. The reason for the dierence is that in the
standpipe the solids are owing at a relatively high
solids velocity and have a high momentum. Adding
gas to owing solids streams gives a dierent result
from in a laboratory bed in which the net solids velo-
city is zero. It is more dicult for gas to permeate
through a moving solids stream possessing signicant
momentum. The gas tends to segregate and cannot
permeate the moving solids mass easily. This leads to
problems with gas distribution in owing uidized
solids systems (Rall et al., 1999).
2.1 Geldart Group C Solids Flow in Standpipes
Geldart group C particles typically have an average
particle size of less than about 30 microns. These par-
ticles are also called cohesive particles because inter-
particle forces are greater than body forces for these
materials, which causes the particles to clump together.
Flour is a typical Geldart group C material. The rela-
tive strength of the interparticle forces in these small
materials causes the particles to bridge and to stop
owing in standpipes.
However, there is technique that allows group C
particles to ow through standpipes. If gas is pulsed
into the bottom of a standpipe operating with group C
particles, the particles can be made to ow. The theory
behind this technique is that the pressure aeration
pulse is transmitted through out the standpipe and
breaks the solids bridge in the standpipe. Before the
bridge can form again, another pulse of aeration is
added. This technique has been used with good results
in standpipes operating with 10 micron oil shale
(Knowlton, 1991). Typically, the pulsing sequence
will be a pulse duration of from 0.1 to 0.5 seconds
followed by a no-pulse period of from 3 to 10 seconds.
This sequence is repeated (by using a timer and a sole-
noid valve) as long as necessary. For very long stand-
pipes, it may be necessary to add a second pulsing
point.
In a 30 cm diameter underow standpipe with a
slide valve at the bottom to control the solids ow
rate, it was found that this type of pulsing was neces-
sary to start group C solids owing in the standpipe
and to keep them owing at low solids velocities. After
the solids had reached a relatively high velocity in the
standpipe, it was found that the pulsing could be shut
o and the solids would ow well.
2.2 Standpipes in Recirculating Solid Systems
Many chemical processes recirculate solids. Catalytic
systems recirculate catalyst in a reaction/regeneration
cycle. First the catalyst is used to supply heat or a
reactant to the process; it is then transferred to a sepa-
rate vessel to regenerate the catalyst, and then it is
returned to the reactor. Circulating uidized bed com-
bustors recirculate fuel and ash around a loop to burn
the fuel completely. A system with a cyclone collecting
entrained solids above a uidized bed and returning
the solids to the bed via the cyclone dipleg is also a
recirculating solid system. All of these recirculation
systems employ standpipes.
There are two basic types of solids recirculation
systems, automatic and controlled. In the automatic
system, the solids are recirculated around the loop at
their natural recirculation rate without being con-
trolled. In the controlled recirculation system, a valve
is used to control the solids ow rate. In the automatic
system, overow standpipes are almost always
employed. In the controlled recirculation systems, the
standpipe is an underow standpipe. It can be either a
uidized or a nonuidized underow standpipe.
2.3 Automatic Solid Recirculation Systems
One of the simplest types of automatic solids recircula-
tion systems is the cyclone/dipleg system utilized to
collect entrained solids and return them to the bed.
This type of system is shown in Fig. 13 for a uidized
bed containing both primary and secondary internal
cyclones. External cyclone return systems are also
automatic recycle systems. All solids entering the pri-
mary and secondary cyclones are returned to the bed
automatically via the primary and secondary diplegs.
The pressure balance around the system for return-
ing solids to the bed via the secondary cyclone dipleg is
P
fluid bed
P
cy 1
P
cy 2
P
dip 2

dip 2
H
dip 2
g
9
For solids to be returned to the bed in the secondary
cyclone dipleg, the secondary dipleg length, H
dip2
,
should be at least
H
dip 2

fluid bed
P
cy 1
P
cy 2
dip 2
g
10
As shown above, if the primary or secondary cyclone
pressure drop increases, the immersion of the second-
ary dipleg in the uidized bed increases, or the solids
density in the secondary cyclone dipleg decreases, the
solids level in the overow standpipe (secondary
cyclone dipleg) increases. Conversely, if the primary
or secondary cyclone pressure drop decreases, the
immersion of the secondary cyclone dipleg in the ui-
dized bed decreases, or the solids density in the second-
ary cyclone dipleg increases, the solids level in the
secondary cyclone dipleg decreases to adjust.
2.4 Controlled Solid Recirculation Systems
In a controlled solids recirculation system, a valve is
used to control the solids ow rate in the system. In
recirculation systems that are controlled, the standpipe
is almost always an underow (either uidized or non-
uidized) standpipe. A typical type of controlled recir-
culation system is that used by FCC units in petroleum
reneries. This type of recirculation loop (or something
very similar) is used by other catalytic and noncatalytic
processes as well.
The primary standpipes in FCC units are under-
ow uidized bed standpipes, not overow uidized
bed standpipes. Standpipes in FCC units can be either
completely vertical, completely angled, or a combina-
tion of vertical and angled sections. Nearly all FCC
units incorporate two standpipes in their loop sys-
tems. The solids ow rate around the system is con-
trolled by a slide valve or a cone valve in each
standpipe.
A typical FCC riser/recirculation system developed
by UOP (called a side-by-side unit) is shown in Fig. 14.
A pressure drop balance around this unit gives
P
regen
P
sp 1
P
sv 1
P
riser
P
stripper
P
sp 2
P
sv 2
0
11
In this unit, hot catalyst is introduced into a dilute-
phase riser where it is contacted with crude oil. The
hot solids vaporize the oil and catalytically crack it
into lower molecular weight hydrocarbons in the
riser. The catalyst reactivity is signicantly reduced
by carbon deposition during this step. Therefore the Figure 13 Fluid bed cyclone recirculation system.
catalyst is transferred to a regenerator where the car-
bon is burned o of the catalyst. This heats up the
catalyst and also restores its reactivity. However,
before the catalyst is returned to the regenerator via
an angled standpipe/slide valve combination, hydro-
carbon vapors are removed from the catalyst in a
steam stripper. From the regenerator, the catalyst is
transferred down another angled standpipe/slide
valve combination and is injected into the bottom of
the riser.
In most FCC units, the solids ow rate around the
system is controlled by a slide valve. FCC units
designed by the Kellogg Brown and Root Company
use another type of valve called a cone, or plug
valve, to control the solids ow rate (Wrench et al.,
1985).
Solids ow in standpipes in FCC units such as those
shown in Fig. 14 are generally controlled in the follow-
ing manner. In standpipe 1, slide valve 1 controls the
temperature at the outlet of the riser by varying the
ow of regenerated solids to the riser. The slide valve
in standpipe 2 controls the uidized solids level in the
stripper (King, 1992). Generally, slide valves in FCC
standpipes are operated with a pressure drop between
about 15 kPa (the minimum required for good control
of the catalyst) and 100 kPa. Higher pressure drops
result in excessive valve wear. Good valve design will
result in the valve operating between 25 and 75% open,
with the valve port area about 25 to 50% of the stand-
pipe open area. A typical design is to for the valve to
operate in the middle of the ranges shown above (i.e.,
at about 50 kPa pressure drop and 50% open) as
reported by King (1992).
Underow uidized standpipes in FCC units are
operated in a vertical conguration, a completely
angled conguration, or a hybrid conguration in
Figure 14 Schematic drawing of side-by-side UOP catalytic cracking unit.
which both vertical and angled sections are present
(Fig. 15). Angling a standpipe is a convenient way to
transfer solids between two points that are separated
horizontally as well as vertically. However, it has been
found (Karri and Knowlton, 1993; Yaslik, 1993) that
long angled underow uidized standpipes do not per-
form as well as vertical standpipes.
Sauer et. al. (1984) and Karri and Knowlton
(1993) studied hybrid angled standpipe operation
using transparent standpipes to allow visual observa-
tion of the ow. Both found the gas and solids
separated in the standpipe, with the gas bubbles ow-
ing up along the upper portion of the standpipe while
the solids owed down along the bottom portion of
the standpipe (Fig. 16). The pressure buildup in the
hybrid standpipe was lower than that in the vertical
standpipe, and Karri and Knowlton (1993) reported
that the maximum solids mass ux possible in a
hybrid angled underow uidized standpipe was less
than that attainable in a vertical underow uidized
standpipe (Fig. 17). The principal reason for this is
that the rising bubbles in the angled section of the
standpipe become relatively large at a low solids
ow rate (and low aeration rate). At a certain solids
Figure 15 Vertical, angled, and hybrid uidized underow standpipes.
Figure 16 Gas and solids ow in an angled standpipe.
mass ux, the bubbles become large enough to bridge
across the vertical section at the top of the standpipe,
hindering the solids ow. When this occurs, the max-
imum solids ow rate in the hybrid angled standpipe
has been achieved.
Karri et. al. (1995) showed that the solids ow rate
through a hybrid angled standpipe can be increased if a
bypass line (Fig. 18) is added between the top of the
angled section of the standpipe and the freeboard of
the bed above it. The eect of adding the bypass line at
the top of the standpipe can be seen in Fig. 17. The
bypass line allows the bubble gas from the angled sec-
tion to bypass the vertical section of the pipe so that
large bubbles are not formed there. Thus the solids
ow rate can be increased. Karri et. al. (1995) reported
that if the bypass was used, the solids ow rate could
be increased to such a value that the solids velocity in
the hybrid standpipe became greater than the bubble
rise velocity, and the bubbles were carried down the
standpipe with the solids. When the bubbles were being
carried down the standpipe by the solids, the bypass
line could then be closed and the standpipe would
operate without slugging in the vertical section.
Even though vertical standpipes can transfer solids
more eciently than hybrid angled standpipes, true
angled standpipes (those containing no vertical sec-
tion) are commonly operated satisfactorily in large
FCC units with Geldart group A catalyst. However,
these standpipes are relatively short and are designed
so that the mass ux through them is not too high, so
that they can be operated satisfactorily. Yaslik (1993)
found that a long angled standpipe had a limited solids
circulation rate relative to vertical standpipes. Thus
when operating a hybrid angled standpipe or a true
angled standpipe it is essential (1) to keep the solids
mass ux through the standpipe below a value that will
lead to slugging, and (2) to keep the line as short as
possible so that the large gas slugs will not have as
great a length in which to form.
2.5 Nonmechanical Solids Flow Devices
A nonmechanical solids ow device is one that uses
only aeration gas in conjunction with its geometrical
shape to cause particulate solids to ow through it.
Nonmechanical solids ow control devices have sev-
eral advantages over mechanical solids ow devices:
1. They have no moving mechanical parts, which
are subject to wear and/or seizure. This feature is espe-
cially benecial when operating at elevated tempera-
tures and pressures.
2. They are inexpensive because they are con-
structed of ordinary pipe and ttings.
Figure 17 Comparison of ow through a vertical and hybrid standpipe.
3. They can be fabricated in-house, which avoids
the long delivery times often associated with the pur-
chase or replacement of large mechanical valves.
Nonmechanical devices can be operated in two dier-
ent modes:
1. In the valve mode to control the ow rate of
particulate solids
2. In an automatic solids ow-through mode
There is often confusion as to how these modes dier,
and what kind of nonmechanical devices should be
used for a particular application. Each mode of opera-
tion is discussed below.
2.6 Nonmechanical Valve Mode Operation
In the valve mode of operation, the solids ow rate
through the nonmechanical device is controlled by
the amount of aeration gas added to it. The most com-
mon types of nonmechanical valves are the L-valve
and the J-valve. These devices are shown schematically
in Fig. 19. The primary dierence between these
devices are their shapes and the directions in which
they discharge solids. Both devices operate on the
same principle. It is harder to fabricate a smooth
180-degree bend for a typical J-valve. Therefore the
J-valve can be approximated and congured more sim-
ply by the geometry shown in Fig. 19C.
The most common nonmechanical valve is the
L-valve, because it is easiest to construct, and also
because it is slightly more ecient than the J-valve
(Knowlton et al., 1981). Because the principle of opera-
tion of nonmechanical valves is the same, nonmechani-
cal valve operation is presented here primarily through
a discussion of the characteristics of the L-valve.
Solids ow through a nonmechanical valve because
of drag forces on the particles produced by the aera-
tion gas. When aeration gas is added to a nonmecha-
nical valve, gas ows downward through the particles
and around the constricting bend. This relative gas
solids ow produces a frictional drag force on the par-
ticles in the direction of ow. When this drag force
exceeds the force required to overcome the resistance
Figure 18 Bypass to increase ow in a hybrid standpipe.
Figure 19 The most common nonmechanical valves.
to solids ow around the bend, the solids ow through
the valve.
The actual gas ow that causes the solids to ow
around the L-valve is not just the amount of aeration
gas added to the valve. If gas is traveling down the
moving packed bed standpipe (which occurs in most
cases) with the solids, the amount of gas that ows
around the L-valve bend, Q
T
, is the sum of the stand-
pipe gas ow, Q
sp
, and the aeration gas ow, Q
A
, as
shown in Fig. 20A. If the gas is owing up the stand-
pipe (which would be the case when the standpipe is
operating with low solids ow rates and/or large
solids), then the amount of gas owing around the
bed, Q
T
, is the dierence between the aeration gas
ow and the gas owing up the standpipe, as shown
in Fig. 20B.
When aeration is added to a nonmechanical valve,
solids do not begin to ow immediately. The initial
aeration gas added is not enough to produce the fric-
tional force required to start solids ow. Above the
threshold amount of gas required to initiate solids
ow, additional aeration gas added to the valve causes
the solids ow rate to increase, and reducing the
amount of aeration to the valve causes the solids
ow rate to decrease. In general, there is little hyster-
esis in the aeration-vs.-solids ow rate curve for a non-
mechanical valve.
Nonmechanical valves work best with materials
having average particle sizes between 100 and 5000
microns. These materials are in Geldart groups B
and D. Materials with average particle sizes greater
than about 2000 microns require substantial amounts
of gas to generate the drag forces required to make the
solids ow around the constricting bend. This is
because larger solids have less surface area available
for the generation of the drag forces required to pro-
duce ow through a nonmechanical valve. These larger
materials work best in nonmechanical valves if there
are smaller particles mixed in with the larger ones. The
smaller particles ll the void spaces between the larger
particles and decrease the voidage of the solids mix-
ture, thereby increasing the drag on the entire mass of
solids when aeration is added, and causing the solids to
move through the constricting bend.
In general, Geldart group A materials (with average
particle sizes from approximately 30 to 100 microns)
do not work in L-valves. Group A materials retain air
in their interstices and remain uidized for a substan-
tial period of time when they are added to the stand-
pipe attached to the nonmechanical valve. Because
they remain uidized, they ow through the constrict-
ing bend like water, and the L-valve cannot control the
solids. Although most group A materials cannot be
used with nonmechanical devices operating in the con-
trol mode, materials at the upper end of the group A
classication that contain few nes (particles smaller
than 44 microns) can and have been controlled in non-
mechanical valves.
Geldart group C materials (with average particle
sizes less than about 30 microns) have interparticle
forces that are large relative to body forces and are
very cohesive (our is a typical example). These mate-
rials do not ow well in any type of pipe, and do not
ow well in L-valves.
Nonmechanical valves are used extensively in
CFBC systems where Geldart group B solids are
used. They are not used in FCC circulating systems
where Geldart group A solids are used.
Figure 20 Gas ow around L-valve bend.
A nonmechanical device operating in the valve
mode is always located at the bottom of an underow
standpipe operating in moving packed bed ow. The
standpipe is usually fed by a hopper, which can either
be uidized or nonuidized. Knowlton and Hirsan
(1978) and Knowlton et al. (1978) have shown that
the operation of a nonmechanical valve is dependent
upon the pressure balance and the geometry of the
system.
Consider the circulating uidized bed/L-valve
return system shown in Fig. 21. The high-pressure
point in such a recycle loop is at the L-valve aeration
point. The low-pressure common point is at the bot-
tom of the cyclone. The pressure drop balance around
the recycle loop is such that
P
L-valve
P
CFB
P
cy
P
sp
P
surge hopper
12
The surge hopper pressure drop is generally negligible,
so that the above equation may be written
P
sp
P
L-valve
P
CFB
P
cy
13
The moving packed bed standpipe is the dependent
part of the pressure drop loop in that its pressure
drop adjusts to exactly balance the pressure drop pro-
duced by the sum of the pressure drops on the inde-
pendent side of the loop. However, there is a maximum
pressure drop per unit length (P=Lg) that the moving
packed bed standpipe can develop. This maximum
value is the uidized bed pressure drop per unit length,
P=Lg
mf
, for the material.
The independent pressure drop can be increased by
increasing any or all of the pressure drops across the
CFBC, the cyclone, or the L-valve. For a constant gas
velocity in the riser, as the solids ow rate into the bed
is increased, the independent part of the pressure drop
loop increases. The moving packed bed standpipe pres-
sure drop then increases to balance this increase. It
does this by automatically increasing the relative gas
solids velocity in the standpipe. Further increases in
the solids ow rate can occur until the P=Lg in the
moving packed bed standpipe reaches the limiting
value of P=Lg
mf
. Because of its reduced capacity
to absorb pressure drop, a short standpipe reaches its
maximum P=Lg at a lower solids ow rate than a
longer standpipe. Thus the maximum solids ow rate
through a nonmechanical valve depends upon the
length of standpipe above it.
As indicated above during the discussion of how an
underow packed bed standpipe operates, the pressure
drop in such a standpipe is generated by the relative
velocity, v
r
, between the gas and solids. When v
r
reaches the value necessary for minimum uidization
of the solids, a transition from packed bed to uidized
bed ow occurs. Any further increase in v
r
results in
the formation of bubbles in the standpipe. These bub-
bles hinder the ow of solids through the standpipe
and cause a decrease in the solids ow rate.
To determine the minimum standpipe length
required for a particular solids ow rate, it is necessary
to estimate the pressure drop on the independent side
of the pressure drop loop at the solid ow rate
required. The minimum length of standpipe necessary,
L
min
, is
L
min

P
independent
P=Lg
mf
14
The actual length of standpipe selected for an L-valve
design should be greater than L
min
to allow for the
possibility of future increases in the solids ow require-
ments and to act as a safety factor. Standpipe lengths
are typically designed to be 1.2 to 2 times L
min
, depend-
ing on the length of the standpipe.
To determine the diameter of an L-valve, it is neces-
sary to select the linear solids velocity desired in the
standpipe. Nearly all L-valves operate over a linear Figure 21 CFB/L-valve loop conguration.
particle velocity range of between 0 to 0.3 m/s in the
standpipe. Velocities greater than about 0.3 m/s can
result in stickslip ow in the standpipe. Although an
L-valve may theoretically be designed for any linear
solids velocity in the standpipe above it, a value near
0.15 m/s is usually selected to allow for substantial
increases or decreases in the solids ow rate.
It is desirable to add aeration to an L-valve as low in
the standpipe as possible. This results in maximizing
standpipe length and the minimizing L-valve pressure
drop, both of which increase the maximum solids ow
rate through the L-valve. However, if aeration is added
too low in the standpipe of the L-valve, gas bypassing
results and solids ow control can be insensitive and/or
ineective.
Knowlton and Hirsan (1978) found that aeration
was most eective if it was added at a length-to-dia-
meter (L=D) ratio of 1.5 or more above the centerline
of the horizontal section of the L-valve. Aeration
added at the centerline of the horizontal section or at
the bottom of the centerline of the standpipe was
found to bypass directly to the top of the horizontal
section. Thus it was not being eciently utilized to
drag the solids through the constricting bend. The
aeration tap locations and the solids ow rate versus
aeration rate curves for each aeration tap location
investigated in their study are shown in Fig. 22. In
summary, L-valve aeration should be added above
the constricting bend. To assure good operation (i.e.,
to prevent bypassing), it should be added at an L=D of
greater than 1.5 above the centerline of the horizontal
section of the L-valve.
In order for an L-valve to operate properly, the L-
valve horizontal section length must be kept between a
minimum and maximum length. The minimum length
must be greater than the horizontal length, H
min
, to
which the solids ow owing to their angle of repose.
For design purposes, the minimum horizontal length
should be between 1.5 to 2 times H
min
. In general, the
shorter the L-valve, the better the operation of the
device.
If the L-valve is operating properly, solids ow
through it in small pulses at a relatively high fre-
quency, and for all practical purposes, the solids ow
is steady. If the L-valve horizontal section is too long,
solids ow can become intermittent (slugging, stop-
ping, and then surging again). In L-valves with long
horizontal sections, a slug of solids builds up to such a
size that it blocks the pipe, and solids ow stops
momentarily. Gas pressure then builds up behind the
slug until it becomes so great that the slug collapses
and the solids surge momentarily through the L-valve.
Another slug then builds up after the gas pressure is
released, and the pattern repeats. This cycle does not
generally occur if the L-valve horizontal length is less
than an L=D ratio of about 8 to 10. Adding additional
Figure 22 Eect of aeration location on L-valve operation.
gas to the horizontal section prevents slug formation in
long L-valves and results in smoother solids ow. The
extra gas does not aect control of the valve.
The extra gas for the horizontal section can be
added in several ways. Adding aeration taps to the
bottom of the L-valve works well except that the taps
tend to become plugged if the gas to the taps is shut
o. Two other methods have been adopted by various
L-valve operators. In one, gas is added to the hori-
zontal section via taps installed along the top of the
L-valve section. This prevents plugging of the taps
when gas to them is shut o. However, this gas is
not as eective as gas added to the bottom, because
some of it bypasses along the top of the L-valve.
Another method is to insert a tube with holes drilled
along its bottom into the lower part of the horizontal
section. This tube can be inserted and withdrawn as
required. For hot L-valves, this tube may warp and
move to the top of the pipe. Therefore, it should be
prevented from moving to the top of the section by
means of a loose clamp that can allow for longitudi-
nal expansion.
Knowlton and Hirsan (1978) have determined the
eect of varying geometrical and particle parameters
on the operation of the L-valve. The eects of these
parameters are summarized here. L-valve aeration
requirements increase with
1. Increasing vertical section diameter
2. Increasing particle size
3. Increasing particle density
L-valve pressure drop increases with
1. Increasing solids ow rate
2. Increasing particle density
3. Decreasing horizontal section diameter
The aeration rate, solids ow rate, and pressure
drop relationships in an L-valve have been esti-
mated by Geldart and Jones (1991) and Yang and
Knowlton (1993). The Yang and Knowlton method
is somewhat involved. It relates L-valve ow to a
variable solids area (A
0
). This technique rst
requires estimating or measuring the bulk density
at minimum uidization, the terminal velocity of
the average particle size of solids owing through
the L-valve, and the L-valve pressure drop. From
these values and the L-valve diameter and horizon-
tal length, the relationship between Q
T
, Q
sp
, and
A
0
, and W
s
(the solids ow rate through the L-
valve) is determined by solving the following four
equations by trial and error:
1. The Jones and Davidson (1965) equation relat-
ing valve pressure drop to solids ow rate and valve
open area, i.e.,
P
Lvalve

1
2
p
1 "
mf
W
s
C
D
A
0

2
15
2. The relationship between the amount of aera-
tion added to the L-valve (Q
A
), the amount of gas
owing down the standpipe (Q
sp
), and the amount
actually owing around the L-valve bend (Q
T
):
Q
T
Q
a
Q
sp
16
3. An equation estimating the amount of gas ow-
ing down the standpipe (this equation assumes no slip
between the solids and gas in the standpipe):
Q
T

W
s
"
mf
p
1 "
mf
17
4. The empirical equation relating Q
T
and the
variable L-valve area (A
0
):
Q
T
=4D
2
v
L
v
1:9 7:64
U
t
A
0
=4D
2
v
L
v
18
This technique relates L-valve aeration and solids ow
rate to within about 40%. The procedure was devel-
oped from data on L-valves ranging in size from 50 to
150 mm in diameter.
If possible, it is recommended that the basic pres-
sure drop as a function of aeration relationship be
obtained from a small L-valve test unit.
Alternatively, these parameters may be extrapolated
from the data presented by Knowlton and Hirsan
(1978).
People are often surprised that L-valve aeration gas
requirements at high temperatures are small when
compared to aeration requirements at low tempera-
ture. The reason for this is that the viscosity of a gas
increases at high temperatures. Therefore, the drag on
the particles for the same external aeration rate
increases. This results in an increase in the solids ow
rate through the L-valve. Because the viscosity of a gas
can increase by a factor of 2 to 2.5 as the temperature
is increased, the amount of gas required to ow a cer-
tain solids ow rate at high temperature can be as little
as half of that at low temperatures for the same aera-
tion rate.
In addition, if particles owing through the L-valve
are Geldart group B particles that lie near the AB
boundary, increasing system temperature can cause
the particles to cross to the A side of the boundary
(Grace, 1986). If this occurs, the L-valve may then
experience ow problems at high temperatures not
observed at low temperature.
2.7 Automatic Solids Flow Devices
Nonmechanical devices can also be used to pass solids
through them automatically. In the automatic mode,
they serve as a simple owthrough device without con-
trolling the solids. If the solids ow rate to the device is
changed, the device automatically adjusts to accommo-
date the changed ow rate. These devices are primarily
used to assist in sealing pressure (in conjunction with a
standpipe) and to reroute the solids where desired.
There are several types of these nonmechanical devices.
The most popular are the straight cyclone dipleg, a loop
seal (also often called a J-valve, siphon seal, or uo-
seal), and an L-valve (in this case valve is a misnomer
since in this mode it is not controlling the solids ow).
In circulating uidized bed systems, the most fre-
quent application of automatic nonmechanical devices
is to recycle collected cyclone nes back to the CFB.
The cyclone discharge is at a lower pressure than the
desired return point in the CFB, so the solids must be
transferred against a pressure gradient. In the past,
lockhoppers or rotary valves were sometimes used to
transfer these solidsgenerally with less than satisfac-
tory results. The best and simplest way to return the
nes is to use a nonmechanical device. The simplest
nonmechanical device used to recycle cyclone nes
back to a reactor is a straight cyclone dipleg.
However, the dipleg must be immersed in a uidized
dense-phase bed (or tted with a trickle valve or
another control device at its discharge end if it is not
immersed in the bed) in order to perform properly.
Since cyclones in CFBs are generally external to the
bed and located some distance from the CFB, a verti-
cal cyclone dipleg cannot be used. Angled diplegs tend
to slug and perform poorly for essentially the same
reason as angled standpipes (see above). Therefore,
loop seals, seal pots, and L-valves are employed to
return the solids to the bed. A V-valve, another type
of solids return device, can also be used in this manner.
A discussion of the various types of automatic non-
mechanical devices used is given below.
2.7.1 Seal Pot
A seal pot is essentially an external uidized bed into
which the cyclone nes discharge via a straight dipleg
(Fig. 23A). The solids and gas from the cyclone and the
uidizing gas for the seal pot are discharged via a
Figure 23 Schematic drawing of seal pot and loop seal.
downwardly angled overow line to the desired return
point. Seal pots have been used for many years and are
reliable devices. However, one must be sure that the
udizing gas velocity is high enough that the lateral
solids transfer rate in the uidized bed is designed to
be greater than the solids ow rate through the dipleg
for good operation. This is generally accomplished by
increasing the uidizing gas velocity in the seal pot
until satisfactory operation is achieved.
With a seal pot, the solids in the dipleg rise to a
height necessary to seal the pressure drop around the
recycle solids transfer loop. In general, seal pot transfer
problems are associated more with the standpipe than
the seal pot itself. These problems depend to a large
extent on the particle size of the solids. Geldart groups
A and B particles can be transferred with relative ease.
Cohesive group C particles are dicult to transfer
unless auxiliary means (vibration, pulsing) are used
to assist the nonmechanical device. For low solids
ow rates, it is recommended that a straight horizontal
opening be used at the dipleg exit. Experience has
shown that diplegs with mitered bends require a rela-
tively high solids mass ux through them to operate
properly.
Aerating standpipes is sometimes necessary in order
to achieve optimum solids ow through them.
However, too much aeration is just as detrimental as
too little aeration. Too much aeration leads to bubble
formation and slugging. Often seal pots operate satis-
factorily with no additional aeration added to the
standpipe, especially with group B solids. However, it
is usually wise to add aeration taps into the standpipe
discharging into the seal pot. They will be there if
needed and do not have to be used if they are not.
An aeration point at the bottom of the standpipe
and aeration points approximately every 2 to 3 m of
solid seal height should be sucient. It is also best to
use a separate aeration control rotameter for each
aeration point, especially in research units. In a com-
mercial unit, restriction orices are generally sucient
to ensure equal ows to each aeration location.
2.7.2 Loop Seal
The loop seal (Fig. 23B) is essentially a variation of the
seal pot. Like the seal pot, it is composed of a stand-
pipe and a uidized bed section. However, the solids
from the standpipe enter the uidized bed from the
side. This allows the uidized bed portion of the device
to be smaller in diameter, resulting in a smaller transfer
device, lower uidization gas requirements, and a more
ecient operation.
The height of the vertical ow portion of the loop
seal can be increased to insulate the operation of the
loop seal from the pressure uctuations in the bed to
which it is discharging. Increasing the height of the
vertical portion of the loop seal also helps prevent
blowing the seal leg of the loop seal if a pressure
upset occurs. However, a large pressure upset will
still blow the seal and cause other operational pro-
blems as well.
It is essential to uidize the upow section of the
loop seal in order for Geldart group B solids to ow
smoothly through the loop seal. For Geldart group A
solids, little or no uidization gas may be required to
be added to the upow section, because they may not
deuidize when owing around the bed. With both
types of solids, the minimum amount of aeration gas
necessary to produce smooth, steady ow is what
should be used. Too much gas results in slugging and
unsteady ow. It is also recommended that aeration at
the base of the loop seal be separated so that the
amount of aeration added below the bottom of the
dipleg and the amount added for the upow section
can be varied independently of each other.
As noted above, Wirth (1995) found that using a
cyclone dipleg too small in diameter caused an exces-
sive amount of gas to be carried down the dipleg. At
high mass uxes, the amount of gas ow was so great
that it caused the upward-owing part of the loop seal
to become dilute. This problem was alleviated by
increasing the dipleg diameter to decrease the mass
ux in the dipleg. Because the amount of gas carried
down the dipleg is proportional to the solids mass ux
in the downcomer, increasing the dipleg diameter
decreased the solids mass ux for the same solids
ow rate, and also decreased the amount of gas carried
down the dipleg.
2.7.3 N-Valve
A novel type of nonmechanical device called the N-
valve was used by Hirama et. al. (1986). However,
this type of device is not recommended because gas
channeling back from the angled section can cause
large bubbles in the standpipe just as was found with
the hybrid standpipes discussed earlier.
2.7.4 V-Valve
The V-valve (Fig. 24) is a close relative of the loop seal
and also acts as an automatic ow device at the base of
standpipes operating in the overow mode (Li et. al.,
1982). The V-valve consists of an angled diverging sec-
tion connected to a standpipe A circular aperture in
the standpipe allows the solids to ow from the stand-
pipe into the diverging section. The included angle of
the diverging section is usually smallbetween 5 and
10 degrees. Greater angles cause uneven distribution of
the gas and solids.
As with the loop seal, the V-valve will not operate if
the solids in the upow (diverging) section are not
uidized. When the solids in the diverging section are
uidized, solids ow from the standpipe, through the
aperture and diverging section, into the uidized bed.
As with the loop seal, the upow section of the V-valve
prevents pressure surges from causing the dipleg to
blow.
2.7.5 L-valve
The L-valve can also operate automatically at the
bottom of an overow standpipe. Its operation is
somewhat dierent than that of the other nonmecha-
nical devices. Chan et al. (1988) showed how the
L-valve operates in the automatic mode. This opera-
tion is described below. Consider the L-valve shown
in Fig. 25A, in which solids are owing through the
L-valve at rate W
1
while aeration is being added to
the L-valve at a constant rate Q
A
. The solids above
the aeration point are owing in the uidized bed
mode and are at an equilibrium height, H
1
. For
Geldart group B solids, the particles below the aera-
tion point are owing in the moving packed bed
mode. The fraction of aeration gas required to
produce solids mass ow rate W
1
ow around the
elbow with the solids, while the remaining portion
of the aerating gas ows up the standpipe.
If the solids ow rate into the standpipe above
the L-valve increases from W
1
to W
2
(Fig. 25B), the
solids level in the standpipe initially rises because
solids are being fed to the L-valve faster than they
are being discharged. The increased height of solids in
the standpipe causes the pressure at the aeration point
to increase relative to the bed. If a constant aeration
ow to the aeration point is maintained by a critical
orice or a control valve, a greater fraction of the
aeration gas ows around the L-valve elbow, causing
an increase in solids ow through the L-valve. If
enough aeration gas is being added to the L-valve,
the system reaches equilibrium at a point where a
larger fraction of the aeration gas is owing around
the L-valve bend to cause solids to ow at rate W
2
.
The height of solids in the standpipe above the
L-valve reaches equilibrium at increased height, H
2
,
and the system is again in balance. If the solids ow
Figure 24 Schematic drawing of V-valve.
Figure 25 Depiction of automatic L-valve operation.
rate decreases from W
1
to W
3
(Fig. 25C), the solids
level in the standpipe above the L-valve falls, decreas-
ing the pressure at the aeration point relative to the
bed. This results in less gas owing around the
L-valve bend. The reduced ow of gas through
the L-valve will result in a lower solids ow rate.
Balance is reached when solids are owing through
the L-valve at rate W
3
, and the solids in the standpipe
are at equilibrium height, H
3
.
In the description above, gas in the standpipe was
assumed to be traveling upward relative to the stand-
pipe wall. It is also possible for gas in the standpipe to
travel downward relative to the standpipe wall. In fact,
gas travels downward with the solids for most stand-
pipe operation.
The L-valve may not work in the automatic mode if
it is used to discharge Geldart group A solids into a
dilute-phase environment. This is because group A
solids may not deaerate and will ush through the L-
valve and not form a pressure seal. However, Geldart
group A solids will work in an automatic L-valve if it is
discharging into a dense bed. An automatic L-valve
works well with Geldart group B solids when dischar-
ging into both dense-phase and dilute-phase media.
The L-valve will not operate automatically over an
innite range of solid ow rates. At some increase in
the solids ow rate in the example given above, not
enough gas is available to uidize the solids in the
standpipe, and they will deuidize. This causes the
standpipe to ll with deuidized solids to a level
which balances the pressure drop loop. The automatic
L-valve can also work if the solids in the standpipe are
in the packed bed mode. However, because the P=Lg
in a packed bed can vary over a wide range, the height
of solids required for pressure sealing in the standpipe
can also vary. This makes it dicult to control the
height of solids in the standpipe, and they can easily
back up into the cyclone, resulting in signicant loss of
material. To prevent this from happening, more aera-
tion must be added to the L-valve.
Although the automatic L-valve is geometrically
simpler and less expensive to install than the loop
seal, for the reason given above, the automatic L-
valve can be more dicult to operate when using
group B materials. Loop seal operation requires
fewer adjustments, and is often preferred over the L-
valve for automatic operation with group B materials.
For group A materials, either automatic device can be
used.
As noted above, it is generally best to design a
nonmechanical device by obtaining the required
data in a cold-ow test unit using the actual solids
to be transferred. If this is not possible, solids of
similar size and density can be used. Nonmechanical
valve or automatic nonmechanical device design has
not yet reached a stage where it can be done analyti-
cally with great condence. Cold modeling in a unit
with 76 to 100 mm diameter transfer lines is relatively
inexpensive and minimizes operational problems
when incorporating the device into a pilot plant or
commercial system.
2.7.6 Cyclone Diplegs and Trickle Valves
As noted above, cyclone diplegs are really overow
standpipes. It is important that they be designed cor-
rectly. Poor operation of cyclone diplegs usually results
in poor collection eciency for the cyclone to which it
is attached.
Diplegs attached to rst-stage cyclones generally
give few operational problems because the solids ux
rate down them is high. Typical design uxes for a
rst-stage cyclone dipleg in FCC units are 350 to 750
kg/s-m
2
. Second stage cyclone diplegs are the ones that
tend to have ow problems because the solids in them
are ner (sometimes approaching cohesive group C
size), and because the solids mass ux through the dip-
leg is low. Both factors result in more sluggish solids
ow, which can sometimes lead to blockage. The mini-
mum dipleg diameter recommended for all commercial
cyclone diplegs is about 100 mm. This diameter is large
enough to prevent most bridging. Obviously, smaller
research units will have smaller diameter diplegs, with
the associated plugging problems.
Cyclone diplegs are often designed with a mitered
bend at its end (Fig. 26A). This type of bend generally
oers no problem for a rst-stage dipleg because of the
high mass uxes through it, but it can cause problems
in a second-stage dipleg. When a mitered bend is used,
the bend causes a restriction in the ow path. If the
solids mass ux is not high enough, the dipleg can
plug. It is much better to use a straight end dipleg
(Fig. 26B) instead of a mitered bend if the mass ux
through the dipleg is low (less than 100 kg/s-m
2
).
During start-up, uidizing gas preferentially ows
up the dipleg of a cyclone until a bed is established
to seal it. This occurs because gas does not have to
undergo the cyclone entrance and cyclone barrel pres-
sure loss when passing up the dipleg. Many processes
operate with high gas ows in the unit during start-up.
First stage diplegs generally have enough solids ow
through them to allow a seal to be established in spite
of the upward gas ow in the dipleg. However, the
solids ow through a second stage dipleg is generally
too small (and the gas ow even greater than through
the rst-stage dipleg) to establish a seal unless a device
to prevent this gas ow is attached to the end of the
dipleg. A device called a trickle valve (Fig. 26C) is
generally used to prevent this bypass gas ow. A trickle
valve is essentially a loosely hung plate that hangs ver-
tically in front of the dipleg discharge opening. Trickle
valves allow the establishment of a solids seal upon
start-up but can cause problems during operation.
Typical problems are plugging (if the trickle valve is
not designed correctly) or erratic pulsating ow. It is
preferable to avoid trickle valves and to modify the
start-up procedure to allow a lower gas ow rate
through the unit until the second stage cyclone dipleg
is sealed. However, this is not always possible, and
trickle valves are often a necessary evil.
In order for the trickle valve to function properly
and not bind during operation, it is attached to the
dipleg by loose hanger rings. The dipleg opening is
generally inclined about 3 to 5 degrees from the verti-
cal so that the apper plate exerts a positive closing
force equal to the moment of its weight about the pivot
point of the valve.
Geldart and Kerdoncu (1992) and Bristow and
Shingles (1989) studied trickle valve operation in cold
ow models with 100 mm diameter diplegs. Four types
of solids ow through the trickle valve were found to
occur: (1) constant trickling, (2) dumping, (3) trick-
lingdumping, and (4) ooding. Constant trickling is
the most desirable type of ow because it is smoother
and because dipleg blockage is least likely to occur.
Constant trickling occurs when solids discharge conti-
nually through the valve maintaining a constant height
of solids in the dipleg. Flow in this regime was found to
occur at high solids velocities in the dipleg (between
about 2 and 25 cm/s), high aeration rates equivalent to
about 3 cm/s in the dipleg and at the location shown in
Fig. 26C (other aeration locations were not nearly as
eective), high cyclone dierential pressures, and high
trickle valve opening torques. All of these parameters
tend to increase the amount of gas aeration or leakage
rate into the dipleg. Bristow and Shingles (1989) also
found that gas leakage through the trickle valve and/or
aeration was absolutely necessary for satisfactory
trickle valve operation.
The dumping mode of trickle valve operation occurs
at low solids velocities, low aeration and/or leakage
rates, low cyclone pressure drops, and low valve tor-
ques. These are the opposite of what causes the more
desirable constant trickling ow. This mode of ow
causes more frequent blockage of solids and may
cause excessive wear in the upper part of the dipleg
discharge.
Tricklingdumping is the regime intermediate
between constant trickling and dumping. Material con-
tinuously trickles out of the valve while the solids per-
iodically dump when the level in the dipleg reaches the
critical height to cause the valve to open.
Flooding occurs when the solids ow into the dipleg
is greater than the solids discharge through the valve.
This causes solids to back up into the cyclone.
Flooding is generally caused by extremely high solids
ow rates or excessive leakage or aeration.
Trickle valve diplegs in commercial units often
exhibit erosion in the upper part of the dipleg and/
or in the upper part of the apper. This is because gas
leakage is preferentially through the upper part of the
trickle valve (Fig. 27A). The relatively high leakage
Figure 26 Dipleg end congurations.
gas velocities (up to 60 cm/s in the dipleg, but much
higher in the space between the apper and the dip-
leg) carry solids with them and eventually erode the
dipleg and apper (a horseshoe pattern occurs on
the apper) in this area. This can be prevented by
adding a plate across the upper part of the dipleg
(Fig. 27B). This reduces the leakage rate and prevents
erosion. Bristow and Shingles (1989) found that a
plate covering at least one-third of the dipleg area
and up to one-half of the area was eective in redu-
cing the gas leakage rate into the dipleg. However,
covering less of the dipleg opening than this amount
was ineective, while covering too much of the open-
ing resulted in blockage of the solids. Although cover-
ing the upper part of the dipleg opening can prevent
erosion, erosion probably occurs because the trickle
valve is operating in the dumping mode. It is much
better to modify the dipleg/trickle valve operation to
cause constant trickling ow to occur than to install
the plate restrictions.
NOMENCLATURE
A
0
= L-valve variable ow area, m
2
C
D
= orice coecient
D = pipe diameter, m
D
v
= L-valve diameter, m
g = gravitational constant, 9.81 m/s
2
H
dip2
= secondary cyclone dipleg length, m
H
dip2
= secondary cyclone dipleg length, m
H
min
= minimum length of L-valve, m
H
sp
= standpipe height, m
H
0
sp
= changed standpipe height, m
L = length, m
L
min
= minimum standpipe length required to
seal, m
L
sp
= standpipe length, m
P = pressure, Pa
P
0
= changed pressure, Pa
P
b
= pressure at bottom of standpipe, Pa
P
t
= pressure at top of standpipe, Pa
P = pressure drop, Pa
P
H
= higher packed bed pressure drop, Pa
P
CFB
= CFB pressure drop, Pa
P
cy
= cyclone pressure drop, Pa
P
cy 1
= primary cyclone pressure drop, Pa
P
cy2
= secondary cyclone pressure drop, Pa
P
d
= distributor pressure drop, Pa
P
dip2
= pressure drop across secondary cyclone
dipleg, Pa/m
P
fluid bed
= uidized bed pressure drop, Pa
P
independent
= independent part of pressure drop loop,
Pa
P
lb
= lower uidized bed pressure drop, Pa
P
ls
= loop seal pressure drop, Pa
P
L-valve
= L-valve pressure drop, Pa
P
regen
= regenerator pressure drop, Pa
P
riser
= riser pressure drop, Pa
P
sp
= standpipe pressure drop, Pa
P
sp1
= rst standpipe pressure drop, Pa
P
sp2
= second standpipe pressure drop, Pa
P
stripper
= stripper pressure drop, Pa
P
surge
= surge vessel pressure drop, Pa
P
surge hopper
= pressure drop across surge hopper, Pa/m
P
sv1
= rst slide valve pressure drop, Pa
P
sv2
= second slide valve pressure drop, Pa
P
ub
= upper uidized bed pressure drop, Pa
P
v
= valve pressure drop, Pa
P=Lg = pressure drop per unit length, Pa/m
P=Lg
mf
= pressure drop per unit length at
minimum uidization, Pa/m
Q = aeration required to prevent
deuidization, m
3
/s
Q = volumetric ow rate of gas, m
3
/s
Q
A
= aeration added to L-valve, m
3
/s
Q
sp
= amount of gas owing down standpipe,
m
3
/s
Figure 27 Trickle valve leakage and erosion prevention.
Q
T
= total L-valve aeration, m
3
/s
U
mf
= supercial gas velocity at minimum
uidization, m/s
v
g
= interstitial gas velocity, Ug=", m/s
v
mb
= minimum bubbling velocity, m/s
v
mf
= interstitial minimum uidization
velocity, U
mf
="
v
r
= relative gas/solid velocity, m/s
v
s
= solids velocity, m/s
W
s
= solids ow rate through L-valve, kg/s
" = voidage
"
mf
dip2
= secondary cyclone dipleg density, kg/m
3
mf
= uidized-bed density at minimum
uidization, kg/m
3
p
= particle density, kg/m
3
sk
= skeletal density, kg/m
3
t
= uidized bed density at top of
standpipe, kg/m
3
REFERENCES
Bristow TC, Shingles T. Cyclone dipleg and trickle valve
operation. In: Grace JR, Shemilt LW, Bergougnou MA,
eds. Fluidization VI, 1989, pp 161168.
Campbell DL, Martin HZ, Tyson CW. U.S. Patent
2,451,803, 1948.
Chan I, Findlay J, Knowlton TM. Operation of a nonmecha-
nical L-valve in the automatic mode. Paper presented at
the Fine Particle Society Meeting, Santa Clara, CA, 1988.
Eleftheriades CM, Judd MR. The design of downcomers
joining gas-uidized beds in multistage systems. Powder
Technology 21:217225. 1978.
Geldart D, Broodryk N. Studies on the behaviour of cyclone
diplegs. Presented at the Annual Meeting of the AIChE,
Los Angeles, CA, November 1821, 1991.
Geldart D, Jones P. The behaviour of L-valves with granular
powders. Powder Technology 67:163172, 1991.
Geldart D, Kerdoncu A. The behaviour of secondary and
tertiary cyclone diplegs. Presented at the AIChE Annual
Meeting, Miami Beach, 1992.
Grace JR. Contacting modes and behaviour classication of
gassolid and other two-phase suspensions. Can J Chem
Eng 64:353364, 1986.
Hirama T, Takeuchi H, Horio M. Nitric oxide emission from
circulating uidized bed coal combustion. Proceedings of
the 9th International Fluidized Bed Combustion
Conference, Boston, pp 898910, 1986.
Jones DRM, Davidson JF. Rheol Acta, 4:180191, 1965.
Karri SBR, Knowlton TM. Comparison of group A solids
ow in hybrid angled and vertical standpipes. In: Avidan
Ar, ed. Circulating Fluidized Bed Technology IV, 1993,
pp 253259.
Karri SBR, Knowlton TM, Litcheld J. Increasing solids ow
rates through a hybrid angles standpipe using a bypass
line. Proceedings of the Eighth International Conference
on Fluidization, Tours, France, 1995.
King D. Fluidized catalytic crackers, an engineering review.
In: Potter, OE, Nicklin, DJ, eds. Fluidization VII, 1991,
pp 1526.
Knowlton TM, Aquino MRY. A comparison of several
lift-line feeders. Paper presented at the Second
Congress of Chemical Engineering, Montreal, Canada,
1981.
Knowlton TM, Findlay JG. Determination of ow-regime
boundaries for cohesive particles. Final Report for the
US Department of Engergy, Contract No.: DE-AC22-
88PC88951, October, 1992.
Knowlton TM, Hirsan I. L-valves characterized for solids
ow. Hydrocarbon Processing 57:149161, 1978.
Knowlton TM, Leung LS, Hirsan I. The eect of aeration
tap location on the performance of a J-valve. In:
Davidson, JF Keairns, DL, eds. Fluidization II, 1978, p
128136.
Li X, Liu D, Kwauk M. Pneumatically controlled multistage
uidized bedsII. Proceedings of the Joint Meeting of
Chemical Engineers, SIESC and AIChE. Beijing: Chem
Industry Press, 1982, pp 382394.
Merrow EW. Linking R & D to Problems Experienced in
Solids Processing. Chemical Engineering Progress, pp
1422, May 1985.
PSRI Technical Report RR-181. An investigation of the
eect of solids mass ux on the axial density prole and
the amount of gas entrained in a submerged cyclone dip-
leg. Chicago, IL, 1995.
Rall R, Pell M. Use of gasliquid contacting internals for
uidized bed stripping applications. In: Werther J, ed.
Proceedings of the Sixth Internation Conference on
Circulating Fluidized Beds. Wu rzburg, Germany,
August 2227, 1999.
Ross J. Personal communication, 1997.
Sauer RA., Chan IH, Knowlton TM. The eects of system
and geometrical parameters on the ow of Class B solids
in overow standpipes. AIChE Symposium Series 234, 80,
pp 110, 1984
Wirth KE. Fluid mechanics of the downcomer in circulating
uidized beds. Proceedings of the Eighth International
Conference on Fluidization, Tours, France, 1995.
Wrench RE, Wilson JE, Guglietta G. Design features for
improved cat cracker operations. Presented at the rst
South American Ketjen Catalyst Seminar, Rio de
Janeiro, Brazil, 1985.
Yang W-C, Knowlton TM. L-valve equations. Powder
Yaslik AD. Circulation diculties in long angled standpipes.
In: Avidan, A ed. Circulating Fluidized Bed Technology
IV, 1993, pp 484492.
22
Cyclone Separators
T. M. Knowlton
1 INTRODUCTION
There are three primary reasons for separating gases
from solids in solids processing plants: (1) to minimize
emissions for environmental purposes, (2) to protect
other processing equipment (turbines, etc.) from part-
icle-laden streams, and (3) to stop unwanted gassolids
reactions from occurring. The primary separator in
most uidized bed processes is a cyclone. The eciency
of separation achieved by the cyclone depends upon
the nature of the process. If the process stream is dry
and the particles are relatively large and free owing,
collection is relatively easy. If it is not, or there is spe-
cial chemistry or extreme temperature involved, parti-
culate collection can be challenging. It is the purpose of
this chapter to describe the basic operation of cyclone
separators and how dierent conditions aect their
performance.
2 CYCLONES
A cyclone is a device that separates particulate solids
from a uid stream by a radial centrifugal force exerted
on the particles. This force separates the solids from
the gas by driving the solids to the cyclone wall, where
they slide to the bottom solids outlet and are collected.
Cyclones are used with almost all uidized beds to
remove solids from exit gas streams.
Particles entering a cyclone are subjected to a sub-
stantial centrifugal separating force. the force exerted
in the particles relative to the gravitational force is
proportional to (U
i
2
gr
o
, where g is the gravitational
acceleration. For a cyclone with an inlet width that is
20% of the diameter of the cyclone (40% of the
radius), the number of gs that the particles experience
(assuming the particles are all at the center line of the
cyclone inlet) is
U
2
i
g0:8r
o
U
2
i
9:810:4D
b
1
For a cyclone with a barrel diameter of 1 meter and
operating with an inlet gas velocity of 20 m/s, over 100
gs will be exerted on the entering particles.
Cyclones can be very small or very large. The
smallest cyclones range from approximately 1 to 2
cm in diameter and the largest up to about 10 m in
diameter. The number of cyclones used for a single
uidized bed can vary from one to up to 22 sets of
rst-stage and second-stage cyclones (44 cyclones
total).
The advantages of cyclones are that they (1) have no
moving parts, (2) are relatively inexpensive to con-
struct, (3) have relatively low pressure drops, and (4)
have maintenance costs that are low. The limitation of
typical cyclones is that they have reduced collection
eciencies for particles approximately 10 microns in
diameter and smaller. Special cyclones can be operated
and designed that can collect particles down to 3
microns in size with eciencies approaching 90%.
This type of collection system usually consists of
many very small cyclones in parallel and is often called
a multiclone.
Cyclones have been used to remove particulates
from gas streams since the middle of the 19th century
(Rietema and Verver, 1961). Early researchers (Rosin
et al., 1932; Alexander, 1949; Stairmand, 1951; Ter
Linden, 1949; Iinoya, 1953; Lapple, 1951; van
Tongren, 1936) conducted the rst experiments
designed to understand the operation of this mechani-
cally simple but operationally complex device. Many
of these experiments (Ter Linden, 1949; Lapple, 1951;
Stairmand, 1951, etc.) resulted in cyclone designs based
on relative cyclone dimensions. More recently, Zenz
(1975) developed an empirical cyclone design proce-
dure that has achieved popular acceptance in the
United States.
In Germany, Barth (1956) developed a theoretical
model of cyclone operation, which was expanded by
Dietz (1981). Muschelknautz (1970a, 1970b) continued
the work of Barth and developed a cyclone design
technique that is used more in Europe than in the
remainder of the world.
2.1 Cyclone Types
A cyclone designed to separate solids from a gas must
have a gassolids inlet that will cause an axial rotation
giving rise to a centrifugal force. It must also have an
axial gas discharge port and a solids discharge port for
the collected solids.
There are two primary types of cyclones: the
reverse-ow cyclone and the uniow cyclone. The
reverse-ow (Fig. 1) is by far the most common. It is
called a reverse-ow cyclone because the gassolids
mixture enters the cyclone tangentially at its periphery,
spirals around the barrel, and then the gas reverses
ow and exits through a gas outlet tube (also called
the vortex nder or the vortex tube) at the top of the
cyclone. The solids spiral down around the barrel of
the cyclone at an angle of approximately 15 degrees
and enter the cyclone cone attached to the bottom of
the barrel. The solids exit the reverse-ow cyclone at
the bottom of the cyclone cone.
In the uniow cyclone (Fig. 2) the gassolids mix-
ture enters the cyclone and spirals around the barrel,
and then the gas exits the cyclone at the bottom in the
center of the uniow cyclone. The solids also exit the
uniow cyclone at the bottom, but along the wall of
the cyclone. A small amount of gas (generally 1 to 3%
of the inlet gas ow) is also withdrawn with the solids
to improve collection eciency. Gauthier and co-
workers (Gauthier et al., 1990a, 1990b, 1992) exten-
sively studied uniow cyclones, and reported on
various aspects of their operation. One application of
this type of cyclone is as a short-contact time reactor.
Solids spend less time in a uniow cyclone than in a
reverse-ow cyclone.
Over the decades that cyclones have been used,
many dierent reverse-ow cyclone geometries have
been tried to improve eciency, prevent particle attri-
tion, prevent erosion of the cyclone wall, or prevent
particle buildup on the cyclone surfaces. However,
there are a few basic types that have emerged as the
most popular over the years. Some of these cyclone
types are shown in Fig. 3. The cyclones shown in this
gure are the tangential inlet cyclone, the volute inlet
cyclone, and the axial inlet cyclone. This last type of
cyclone uses axial swirl vanes to impel the gassolids
mixture into rotary centrifugal motion.
In some cyclones, the roof of the cyclone is angled
downward in a helical fashion. In some cyclones with
standard roofs, the solids that enter the cyclone do not
move downward far enough by the time they traverse
one revolution and then collide with the incoming
solids. This leads to particle bouncing and reduced
cyclone eciency as well as slightly higher pressure
drops. To circumvent this unwanted particle interac-
tion at the inlet, cyclone roofs are sometimes angled
downward to force the solids downward and below the
incoming solids. In some applications, helical roofs are
used to prevent the buildup of sticky material of the
cyclone roof by forcing their incoming solids to rub
against the top of the cyclone.
Figure 1 Typical reverse ow cyclone.
2.2 Flow Patterns in Cyclones
In a tangential-inlet reverse-ow cyclone, the cyclone
inlet translates the linear inlet gas ow into a rotating
vortex ow. As shown in Fig. 1, the gassolids mixture
enters an annulus region between the outer wall of the
cyclone and the outer wall of the gas outlet tube. As
the gassolids mixture spirals downwards, it sets up a
vortex with an axial direction downward toward the
solids outlet.
The rotating vortex in the main body of the cyclone
below the gas outlet tube sets up a tangential velocity
that varies with the radius of the cyclone. As the
radius, r
w
, decreases from the wall toward the center
of the cyclone, the tangential velocity increases to a
maximum that occurs at a radius slightly less than
the outer radius of the gas outlet tube, r
o
, as shown
in Fig. 4. At radii much smaller than that of the gas
outlet tube, the tangential gas velocity decreases to a
much lower value at the center.
The inner vortex (often called the core of the vortex)
rotates at a much higher velocity than the outer vortex.
In the absence of solids, the radius of this inner vortex
has been measured to be 0.4 to 0:8r
o
. With axial-inlet
cyclones, the inner core vortex is aligned with the axis
of the gas outlet tube. With tangential or volute
cyclone inlets, however, the vortex is not exactly
aligned with the axis. The nonsymmetric entry of the
tangential or volute inlet causes the axis of the vortex
to be slightly eccentric from the axis of the cyclone.
This means that the bottom of the vortex is displaced
some distance away from the axis, and can pluck o
and reentrain dust from the solids sliding down the
cyclone cone if the vortex gets too close to the wall
of the cyclone cone.
At the bottom of the vortex, there is substantial
turbulence as the gas ow reverses and ows up the
middle of the cyclone into the gas outlet tube. As indi-
cated above, if this region is too close to the wall of the
Figure 2 Schematic drawing of uniow cyclone and multiclones.
Figure 3 Schematic drawing of tangential, volute, and axial
inlet cyclones.
cone, substantial reentrainment of the separated solids
can occur, so it is very important that cyclone design
take this into account.
Hugi and Reh (1998) have recently reported that (at
high solids loadings) enhanced cyclone eciency
occurs when the solids form a coherent, stable strand
at the entrance to a cyclone. The formation of such a
strand is dependent upon several factors. They
reported that they obtained a higher cyclone eciency
for smaller (d
p;50
40 microns) solids than for larger
solids (d
p;50
125 microns). This is not what theory
would predict. However, they also found that the smal-
ler particles formed coherent, stable strands more read-
ily than the larger particles, which explained the reason
for the apparent discrepancy.
Muschelknautz et al. (1996) also proposed a
mechanistic model of cyclone operation. In this
model, the gas can carry only a maximum amount of
solids (called the critical loading). At any solids load-
ing in excess of this critical loading, the solids are
immediately separated from the gas at the inlet to the
cyclone, as indicated in Fig. 5. The solids remaining in
the gas are then separated in the cyclone barrel and in
the inner vortex below the gas outlet tube as if the
cyclone were operating at a lower solids loading.
3 SERIES AND PARALLEL CYCLONE
ARRANGEMENTS
3.1 Series
A single cyclone can sometimes give sucient gas
solids separation for a particular process or applica-
tion. However, solids collection eciency can usually
be enhanced by placing cyclones in series. Cyclones in
series are typically necessary for most processes to
minimize particulate emissions or to minimize the
loss of expensive solid reactant or catalyst. Two
cyclones in series are most common, but sometimes
three cyclones in series are used. Series cyclones can
be very ecient. In uidized catalytic cracking (FCC)
regenerators, two stages of cyclones can give ecien-
cies of up to and even greater than 99.999%.
Typically, rst-stage cyclones will have an inlet gas
velocity less than that of second-stage cyclones. The
lower inlet velocity of rst-stage cyclones results in
lower particle attrition rates and lower wall erosion
rates. After most of the solids are collected in the
rst stage, a higher velocity is generally used in
second-stage cyclones to increase the centrifugal force
on the solids and increase collection eciency. Erosion
rates are generally low in the second stage because of
the vastly reduced ux of solids into the second stage
cyclone.
Sometimes it is better to use a single high-eciency
cyclone than operate with two cyclones in series. This
situation can occur when the rst-stage cyclone is very
ecient and the solids owing to the second stage
cyclone are very small and cohesive (Geldart group C
solids). Cohesive solids are dicult to transfer in
cyclone diplegs and may cause the solids to bridge
momentarily or not ow at all. Second-stage diplegs
operating with trickle or apper valves and cohesive
solids are particularly sensitive to bridging. The inabil-
ity to ow solids down the second-stage diplegs causes
the solids to pass straight through from the inlet to the
outlet, thus destroying cyclone eciency.
In addition, ow of gas up the dipleg of second-
stage cyclone diplegs can also be a major problem. If
this gas ow is excessive during normal operation, the
diplegs will not operate well, and erosion of the
cyclone cone and dipleg trickle valve apper plate
may result. This problem occurs because the solids
ux in a second-stage dipleg is very low (of the
order of 10 to 20 kg/s-m
2
). For a dipleg density of
approximately 800 kg/m
3
, the velocity of the solids in
the dipleg (calculated as the mass ux through the
dipleg divided by the suspension density in the dipleg)
Figure 4 Typical radial tangential velocity prole in a
cyclone.
is only 0.025 m/s for a ux in the dipleg of 20 kg/s-
m
2
. Gas can easily ow up the dipleg against this very
low velocity. Conversely, in rst stage cyclone diplegs
where the solids ux may be as high as 750 kg/s-m
2
,
the solids velocity is so high (approximately 0.94 m/s
at a dipleg density of 800 kg/m
3
) that gas is carried
down the dipleg with the solids.
Because of these recurring problems with second-
stage cyclone diplegs, companies sometimes choose to
use a single high-eciency cyclone in place of two
cyclone stages. Even though the overall eciency
using a single cyclone is slightly less than using two
cyclone stages, they nd that this cyclone arrangement
optimizes the operation of the process by minimizing
upsets and reducing downtime using a single high-
eciency cyclone. It is also a much simpler system as
well and saves on initial capital costs.
3.2 Parallel
Several small cyclones are placed in parallel when it is
not possible to t a single large cyclone into the
available height, or when extremely high centrifugal
forces are required. It is dicult to ensure equal dis-
tribution of gas and solids into parallel cyclones. This
can lead to cyclone ineciencies and increased wear on
the cyclones taking the bulk of the solids ow.
Increasing the pressure drop across the cyclones will
improve the solids distribution but is not always
sucient to ensure equal solids ow into each cyclone.
Typically, when the number of parallel cyclones is
small (less than 6 to 8), each cyclone will have an indi-
vidual inlet and outlet duct. However, when the num-
ber of cyclones exceeds this number, the cyclones
generally have common inlet and outlet plenums and
a common collection bin. Parallel arrangements of this
type frequently result in the operating problems of
equalizing the gas and solid ow rates into each
cyclone, and preventing backow from the common
chamber into one or more cyclones and reentrainment
of solids from the common collection hopper back into
one or more cyclone outlets. To prevent these two
problems, a device is often placed at the outlet of
each parallel cyclone. This device (generally proprie-
tary) reduces the area available for ow and acts as a
check valve to prevent backow and dust reentrain-
ment into the cyclone.
Placing a large number of cyclones in parallel in a
common bin can result in distribution problems
because it will be easier for the gas and solids to ow
through the closest cyclone than one located some dis-
tance away from the inlet. Multiple inlets to the com-
mon vessel reduce this problem, but result in increased
complexity and cost.
Figure 5 Muschelknautz critical loading plot.
Because of problems in the operation of small par-
allel cyclones, it is generally found that they result in a
collection eciency less than or (if designed well) equal
to that of a large, single cyclone processing the same
amount of gas and solids.
In the FCC and similar processes there are many
units operating with two cyclone stages in series, but
with several (2 to 18) cyclones in each stage in parallel.
These cyclones can have very high eciencies even
though they operate in parallel. However, each rst-
stage cyclone generally has a dipleg that is immersed in
the bed (there is no common collection chamber). Each
second stage cyclone has a dipleg that usually has a
trickle valve on its discharge. The trickle valve eec-
tively is a check valve that prevents backow of most
of the gas through the dipleg on startup. In addition,
the pressure drop through the cyclone stages is gener-
ally sucient to prevent maldistribution.
There has been very little work in the literature on
the operation of parallel cyclones. However, what
work does exist indicates that too low a pressure
drop results in poor gassolids distribution, and that
too many parallel cyclones in a common hopper are
not recommended.
Broodryk and Shingles (1995) studied parallel
cyclone operation with cyclones in the freeboard
above a turbulent uidized bed. They found that too
low an overall pressure drop (caused by inlet gas velo-
cities of less than about 15 m/s) resulted in preferential
ow of gas through two of their three cyclones. They
also found that poor trickle valve design and/or opera-
tion could induce an imbalance in the gas ow into the
cyclones. When preferential gas ow occurred, it
resulted in signicant solids carryover and reduced
cyclone eciency.
Koman (1953) studied the eciency of several dif-
ferent small-diameter (38 and 51 mm diameter) parallel
cyclone congurations and reported the relative e-
ciencies shown in Table 1. Komans data show that
placing cyclones in a common hopper reduced the col-
lection eciency, and that adding more cyclones to a
common hopper also caused a reduction in eciency.
Smellie (1942), using three 200 mm diameter cyclones,
and Whiton (1941), using 9 to 20 cyclones 610 mm in
diameter, also reported poorer cyclone performance
with parallel cyclone operation. However, OMara
(1950) reported very little dierence between single
cyclone and parallel cyclone eciencies These results
indicate that parallel cyclones can be designed and
operated with the same eciencies as a single cyclone,
but that it takes good and careful design of the
cyclones to accomplish this.
4 INTERNAL AND EXTERNAL CYCLONES
Cyclones can either be placed in the freeboard above
the uidized bed or be located outside of the uidized-
bed vessel. There are advantages and disadvantages to
each type of placement, and the optimum type of
placement depends on what is best for a particular
process.
Internal cyclones have the advantages that they
require no inlet piping (their inlets can be open to
the freeboard), require no high-pressure shell, and
have straight cyclone diplegs. Internal cyclones are
generally smaller in diameter than external cyclones
because their size is limited by the headspace available
in the freeboard above the uidized bed. These size
limitations result in using several smaller cyclones in
parallel instead of one large cyclone. In addition, it is
dicult to aerate second-stage cyclone diplegs (gener-
ally an advantageous technique) when internal
cyclones are used. Aerating secondary cyclone diplegs
can improve the operation of the dipleg signicantly.
The advantages of external cyclones are that they
can be much larger than internal cyclones, that they
are more accessible than internal cyclones, and that
their diplegs can be aerated more easily. The disadvan-
tages of external cyclones are that they require a pres-
sure shell and that external cyclone diplegs generally
require a section with an angled or a horizontal pipe to
return the solids to the bed. The angled or horizontal
dipleg sections can result in poor dipleg operation if
not designed correctly.
5 CYCLONE INLET DESIGN
The design of the cyclone inlet can greatly aect
cyclone performance. It is generally desired to have
the width of the inlet (L
w
) be as narrow as possible
so that the entering solids will be as close as possible
to the cyclone wall where they can be collected.
Table 1 Efciencies of Parallel Cyclones
Conguration Eciency, %
Individual cyclone 96.0
Seven cyclones in parallel
(individual hoppers)
95.3
Seven cyclones in parallel
(common hopper)
94.1
Fourteen cyclones in parallel
(common hopper)
92.2
However, narrow inlet widths require that the height of
the inlet (H) be very long in order to give an inlet area
required for the desired inlet gas velocities. Therefore a
balance between narrow inlet widths and the length of
the inlet height has to be made. Typically, low-loading
cyclones (cyclones with inlet loadings less than
approximately 2 to 5 kg/m
3
) have height-to-width
ratios (H=L
w
) of between 2.5 and 3.5. For high-loading
cyclones, this inlet aspect ratio can be increased to as
high as 7 or so with the correct design. Such high inlet
aspect ratios require that the cyclone barrel length
increase substantially.
A common cyclone inlet is a rectangular tangential
inlet with a constant area along its length. This type of
inlet is satisfactory for many cyclones, especially those
operating at low solids loading. However, a better type
of inlet is one in which the inner wall of the inlet is
angled toward the outer cyclone wall at the cyclone
inlet (Fig. 6). This induces solids momentum toward
the outer wall of the cyclone. The bottom wall of the
inlet is angled downward so that the area decrease
along the inlet ow path is not too rapid and accelera-
tion is controlled. In addition, the entire inlet can be
angled slightly downward to give enhanced eciencies.
This type of inlet is superior to the constant area tan-
gential inlet, especially for higher solids loadings
(greater than 2 to 5 kg/m
3
).
Hugi and Reh (1998) report that continuous accel-
eration of the solids throughout the inlet is desired for
improved eciency, and that the angled inlet described
above achieves this. If the momentum of the solids is
sucient and the solids are continuously accelerating
along the length of the inlet, they will form the stable,
coherent strand important for high collection ecien-
cies.
Another superior inlet for high solids loadings is the
volute cyclone inlet. At high inlet loadings in a tangen-
tial cyclone inlet, the gassolids stream expands rapidly
from its minimum width at the point of contact. This
rapid expansion disturbs the laminar gas ow around
the gas outlet tube and causes ow separation around
the tube. This results in lowered cyclone eciency. But
when a volute inlet is used, the expanding solids stream
is further from the gas outlet tube and enters at an
angle so that the solids do no induce as much ow
separation or asymmetrical ow around the gas outlet
tube. This cyclone eciency is not aected to as great a
degree.
The nature of the gassolids ow in the inlet ducting
to the cyclone can aect cyclone eciency signicantly.
If the solids in the inlet salt out on the bottom and
result in dune formation and the resulting unsteady
or pulsing ow, cyclone eciency is adversely aected.
To minimize the possibility of this occurring, it is
recommended that the inlet line to the cyclone operate
above the saltation velocity (Gauthier et al. 1990b),
which will prevent the solids from operating in the
dune or pulsing ow regime. If this is not possible,
then the inlet line can be angled downward (approxi-
mately 15 to 20 degrees) to let gravity assist in the ow
of the solids. Keeping the inlet line as short as possible
can also minimize any pulsing of the solids ow.
6 EFFECT OF SOLIDS LOADING
Cyclones can collect solids over a wide range of load-
ings. Traditionally, solids loadings have been reported
as either kg of solids per m
3
of gas (kg/m
3
), or as kg of
solids per kg of gas (kg
s
=kg
g
). However, loading based
on mass is probably not the best way to report solid
loadings for cyclones. This is because the volume of
solids processed by a cyclone at the same mass loading
can vary greatly, depending on the density of the
solids. For example, many polymers have a bulk den-
sity of approximately 400 kg/m
3
and iron ore has a
bulk density of approximately 2400 kg/m
3
. This is a
ratio of 6. Therefore, a cyclone operating with polymer
would have to process six times the volume of solids Figure 6 Angled cyclone inlet.
that a cyclone operating with iron ore would process at
the same mass loading. If the cyclone operating with
the polymer were to be designed to operate at high
loadings on a mass basis, it would probably plug. In
addition, the diplegs below the cyclone operating with
the polymer may experience operational problems
because of the high volumetric loading. However,
because almost all solids have been and are still
reported on a mass basis, they will continue to be
reported on this basis in this chapter.
At ambient conditions, cyclones have been operated
at solid loadings as low as 0:02 kg=m
3
(0.0125 kg
s
=kg
g
)
and as high as 64 kg=m
3
(50 kg
s
=kg
g
) or more. This is a
factor of 3200. In general, cyclone eciency increases
with increasing solids loading. This is because at higher
loadings, very ne particles are trapped in the inter-
stices of the larger particles and this entrapment
increases the collection eciency of the small particles.
Even though collection eciencies are increased with
increased loading, cyclone loss rates are also increased
as loading is increased. This is because the cyclone
eciency increase is almost always less than the
increase in the solids loading.
When solids are added to a cyclone the overall
cyclone pressure drop decreases (Yuu et al., 1978).
This eect is signicant, as it often results in a pressure
drop reduction of greater than 50% when solids are
added. The reason for this decrease is that when solids
are present at the wall of the cyclone, the increased
drag on the gas due to the solids at the wall causes
the cyclone tangential velocity to decrease. This results
in a decrease in cyclone pressure drop. This condition
occurs for low and medium solids loading. At higher
solids loading the cyclone pressure drop begins to
increase with loading as the pressure drop due to solids
acceleration becomes greater than the pressure drop
due to the reduction in the tangential velocity. The
decrease in cyclone pressure drop at lower loadings
and the subsequent increase in cyclone pressure drop
at higher loadings are illustrated in Fig. 7.
7 GAS OUTLET TUBE
To prevent solids from the cyclone inlet from bypass-
ing directly into the outlet of the cyclone, a tube the
same diameter as the gas outlet is extended into the
cyclone to a level equal to or below the bottom of
the solids inlet. This prevents solids from bypassing
directly into the outlet of the cyclone. This tube is
called a gas outlet tube (it is also called a vortex nder
or a vortex tube). The gas outlet tube does increase the
eciency of a low-loading cyclone relative to a cyclone
that does not have a gas outlet tube. However, many
Figure 7 Cyclone pressure drop versus loading.
cyclone designers use a longer gas outlet tube than is
required. For low loadings, the length of the gas outlet
tube does not have to be greater than the height of the
solids inlet. For high loadings, the optimum gas outlet
tube length has been shown to be less than the height
of the solids inlet. The optimum gas outlet tube length
depends upon the gas velocity and the solids loading
into the cyclone. At very high loadings, the optimum
gas outlet tube length has been reported by Lewnard et
al. (1993) to be approximately half that of the cyclone
inlet height, or 0:5H, as shown in Fig. 8.
Decreasing the diameter of the gas outlet tube gen-
erally increases the eciency of a cyclone. This is due
to increasing the length and rotating velocity of the
vortex below the gas outlet tube. However, if the
cyclone is not long enough to contain the increased
vortex length and it extends too close to the solids
owing on the wall of the cone, then cyclone eciency
will decrease as the vortex plucks o solids from the
wall of the cone, so it is imperative that the cyclone
have a sucient length before reducing the diameter of
the gas outlet tube.
Hugi and Reh (1998) operated cyclones at loadings
ranging from 10 to 50 kg
s
=kg
g
(13 to 64 kg=m
3
). They
report that if the roof of the cyclone is extended, no gas
outlet tube is required, and that this type of cyclone is
superior to a cyclone in which the roof is not extended
and a gas outlet tube is used. Muschelknautz et al.
(1999) agreed with this assessment, but found that
even higher eciencies were possible if the cyclone
roof was raised and a vortex tube was used in the
cyclone.
Muschelknautz et al. (1999) have also reported that
if the gas outlet tube for volute-entry cyclones is moved
o-center, a cyclones eciency can be improved sig-
nicantly The reason for this improvement appears to
be that the solids entry for a tangential or volute entry
cyclone is at the side and is not symmetrical. This
causes the vortex to be slightly o-center. If the gas
outlet tube is moved slightly, it eliminates eddy forma-
tion and improves cyclone eciency.
8 INLET GAS VELOCITY
The inlet gas velocity for cyclones can range from a
low of approximately 10 m/s to a high of over 40 m/s.
The gas velocity used depends on the application of the
cyclone. Often cyclones are used to separate a fragile,
expensive material such as a catalyst from a gas
stream. High inlet gas velocities would result in exces-
sive attrition of the catalyst, so for this type of cyclone
operation the inlet velocities are reduced to minimize
the attrition of the expensive solids. In this case, the
gas velocity in the outlet tube is designed to be much
higher than the velocity in the inlet. Most of the solids
are collected in the barrel of the cyclone, even at the
relatively low inlet gas velocity, and most of the
remaining solids are separated in the high-velocity vor-
tex below the gas outlet tube.
In second-stage cyclones, the inlet gas velocity is
generally higher than the inlet gas velocity of the rst
stage. In the second-stage cyclone, the loading is low,
and an increased gas velocity leading to a higher cen-
trifugal force is required for high separation ecien-
cies. Also the loading in the second-stage cyclone is so
low that erosion and attrition will be relatively small,
and so high velocities can be utilized.
High cyclone inlet gas velocities also result in very
high cyclone pressure drops. Because the pressure drop
across the cyclone is proportional to the square of the
inlet gas velocity (i.e., P1
g
gU
2
, lowering the inlet
gas velocity can signicantly reduce the pressure drop
across the cyclone. This will lead to reduced energy
Figure 8 Eect of outlet tube penetration on cyclone e-
ciency. (From Lewnard et al., 1994.)
expenditures and will also reduce the length of cyclone
diplegs required for sealing the cyclones.
9 CYCLONE LENGTH
As described above, the cyclone length should be great
enough to contain the vortex below the gas outlet tube.
It is generally advisable to have the cyclone somewhat
longer than required so that modications to the gas
outlet tube can be made if required. Either the barrel
or the cone can be increased in length to contain the
vortex. However, cyclone barrels can be made too
long. If the barrel is too long, the rotating spiral of
solids along the wall can lose its momentum. When
this happens, the solids along the wall can be reen-
trained into the rotating gas in the barrel and cyclone
eciency will be reduced.
Homan et al. (2001) studied the eect of cyclone
length on cyclone eciency and showed that the e-
ciency of a cyclone increases with length. However,
they also found that after a certain length, cyclone
eciency decreased. They reported that cyclone e-
ciency suddenly decreased after a certain cyclone
length, which in their cyclone was at a length-to-dia-
meter ratio of 5.65 (although many researchers employ
this length-to-diameter ratio as a correlating parameter
to make the length parameter dimensionless, it is likely
that it is the actual length of the cyclone that is impor-
tant). Homan et al. (2001) stated that the probable
reason for the sudden decrease in cyclone eciency
was the central vortex touching and turning on the
cyclone cone. When this occurred, the eciency col-
lapsed and caused increased solids reentrainment.
Homan et al. (2001) also reported that cyclone
pressure drop decreased with increasing cyclone
length. This probably occurs for the same reason that
cyclone pressure drop decreases with increasing
cyclone loading. For long cyclones, the increased
length of the cyclone wall results in a longer path for
the gas to travel. This causes more resistance to the
ow of the gas in the cyclone (much as a longer pipe
produces more resistance to gas ow than a shorter
pipe) that results in reducing the tangential velocity
in the cyclone and therefore the cyclone pressure
drop. Homan et al. (2001) were able to predict this
eect with various models.
10 CYCLONE WALL ROUGHNESS
Rough cyclone walls reduce cyclone eciency and
decrease the pressure drop in a cyclone. Iinoya (1953)
added coarse sandpaper to the wall of a cyclone and
found that cyclone pressure drop decreased. The coar-
ser the sandpaper, the lower the pressure drop. Yuu et
al. (1978) also found that sandpaper glued on the wall
reduced cyclone pressure drop and attributed this to a
reduction in the tangential velocity due to the higher
drag on the rough walls.
Ionoya (1953) added a 1.2 mm diamond mesh liner
to a cyclone and found that the liner signicantly
reduced cyclone eciency. When the mesh liner cov-
ered 87% of the cyclone inner wall, cyclone eciency
was decreased by 9.5% relative to the cyclones e-
ciency without the diamond mesh liner.
Large weld beads, etc., can also reduce cyclone e-
ciency. If the solids ow along the wall of a cyclone
encounter a large protuberance such as a weld bead,
the weld bead acts as a type of ski jump and causes
the solids to be deected further into the center of the
cyclone where they can be thrown into the vortex and
carried out of the cyclone. In small pilot or research
cyclones this is especially common, because the dis-
tance between the wall of the cyclone and the vortex
tube is very small. Because of their detrimental eect
on cyclone eciency, weld beads should be ground o
to make the cyclone inner wall smooth.
In high-temperature processes, cyclones are often
lined with refractory both to minimize heat loss and
to protect the metal surfaces from abrasion. These
refractor surfaces are not as smooth as metal, but
after a few days of operation, the refractory becomes
smoother because of the abrasive action of the solids.
With very small laboratory or pilot cyclones, some
solids (large polymer beads, spherical particles, etc.)
can sometimes bounce o of the cyclone wall immedi-
ately across from the cyclone inlet and be deected into
the vortex. Very large particles can be found in the gas
outlet stream of the cyclone with these very small
cyclones and with particles that bounce. To increase
cyclone eciency with these types of solids, the cyclone
barrel diameter can be increased. This increases the
distance between the cyclone vortex and the wall and
prevents most of the solids from bouncing back into
the vortex.
11 CYCLONE DIAMETER
In theory, a smaller diameter cyclone should be able to
collect smaller particles because it can develop a higher
centrifugal force. However, using smaller cyclones gen-
erally means that many have to be used in parallel to
accommodate large gas ows. The problem with par-
allel cyclones (as indicated above) is that it is dicult
to get even distribution of solids into all of the
cyclones. If maldistribution occurs, this can cause inef-
ciencies that can negate the natural advantage of the
smaller cyclones.
Cyclone diameters can be very large. Perhaps the
largest cyclones are those used in circulating uidized
bed combustors, where cyclone diameters approach 10
meters. Large diameter cyclones also result in very long
cyclones, and so these large-diameter long-length
cyclones are really not feasible as internal cyclones in
uidized beds (they make the vessel too tall).
When using small cyclones in pilot or demonstra-
tion plants, particle bouncing or rebounding from the
wall can be a problem. Because the cyclones are small,
particles can bounce o the cyclone wall into the vor-
tex and be carried out the cyclone exit. This is espe-
cially true for plastic particles that can bounce large
distances after striking a wall. Therefore, for very small
cyclones, it is recommended that the cyclone diameters
be increased over the typical diameter dimensions to
counter this eect. As an example, increasing the dia-
meter of a 150 mm diameter laboratory cyclone to 300
mm in diameter should solve the most egregious boun-
cing problems. For particles that do not bounce as
much as plastic, smaller diameter increases can be
made.
12 CYCLONE DIMENSIONS AND DESIGN
Using a force balance (obtained by equating the time
that it takes for particles of a particular size to traverse
a width equal to the width of the cyclone inlet to the
time that the gas resides in the cyclone), Rosin et al.
(1932) developed a relation between the fractional e-
ciency (the eciency of collection of a particular par-
ticle size of a cyclone to system geometrical and system
parameters. The relation developed,
d
p;th

9L
w
N
s
U
i
s
2
relates the theoretical size that can be collected by a
cyclone to the width of the cyclone inlet, L
w
, the vis-
cosity of the gas, , eective number of spiral paths
taken by the gas, N
s
, the inlet gas velocity U
i
, the
particle density,
p
, and the gas density,
g
. Ecient
cyclones collect particles of smaller particle size, so
cyclone eciency, E
o
is proportional to 1=d
p;th
, or
E
o
/
1
d
p;th
3
Therefore, cyclone eciency is increased by increasing
the eective number of spiral paths taken by the gas
(this increases the residence time of the gas in the
cyclone), the inlet gas velocity (this increases the cen-
trifugal force on the particles), and the particle density
(which also increases the centrifugal force on a part-
icle). Cyclone eciency is decreased by larger inlet
widths (it takes longer for the solids to get to the
wall for greater inlet widths) and increasing gas visc-
osity (it takes longer for the particles to get to the wall
through a more viscous gas). Note that the equation
predicts that gas density will not have a large eect on
cyclone eciency. This is because of the term (
p
-
g
).
Because
p
is so much larger than
g
, this relation
predicts that gas density should have no major eect
on cyclone eciency.
One equation that has been used extensively in
cyclone design procedures is that of Lapple (1940).
This equation replaces the smallest theoretically col-
lected particle size with a particle size, D
p;th
, collected
by the cyclone with a theoretical eciency of 50%,
thus
D
p;th

9L
w
N
s
U
p
s
4
According to Eq. (4) (which applies at relative low
solid loadings in the cyclone), the higher the inlet
gas velocity, the greater the cyclone eciency. In
practice, however, this is not the case. In a real
cyclone, eciency increases up to a limiting value of
gas velocity and then begins to decrease. The reason
for the decrease in cyclone eciency at high gas velo-
cities is that the inner vortex becomes longer as gas
velocity increases and causes the pressure drop of the
gas entering the gas outlet tube to increase. The
increased pressure drop across the gas outlet tube
causes the inner vortex length to increase. At some
vortex length, the vortex approaches the side of the
cyclone cone. When this happens, the vortex will
begin to pluck o the solids owing on the side of
the cone, and cyclone eciency will begin to decrease.
If the vortex length is too long, so that it impinges
against the side of the cone, severe erosion of the
cyclone will result.
The cyclone design procedure developed by Zenz
(1975) utilizes Eq. (4) as the basis of the design proce-
dure. The procedure consists of several steps:
1. Calculate D
p;th
from Eq. (4).
2. Estimate the eective number of eective spiral
gas paths in the cyclone, N
s
.
3. Partition the particle size entering the cyclone
into at least 10 dierent cuts, to give particle
sizes of the cuts (d
pi
) from d
pl
to d
pn
.
4. Calculate the ratio of d
pi
=D
p;th
for each cut and
determine the low-loading cyclone collection
eciency for each cut size.
5. Determine the overall cyclone collection e-
ciency (E
o
) for very low loading cyclones (low
loading is dened to be 1 grain/ft
3
(0.0023 kg/
m
3
) by the relation
E
o

X
n
i1
x
i
E
oi
5
where:
x
i
the weight fraction of each individual
cut size
E
oi
the collection efficiency of the cyclone
for each individual cut size, %:
6. Modify the low-loading cyclone collection e-
ciency for the eect of loading.
7. Determine cyclone pressure drop.
A detailed description of this approach is described
below.
12.1.1 Estimation of N
s
The number of eective spiral paths taken by the gas in
the outer vortex of the cyclone can be estimated by the
curve in Fig. 9. The number of eective cyclone spirals
cannot be determined in most cyclones because they
are made of metal. However, N
s
has been experi-
mentally determined by back-calculation from
observed cyclone eciencies The gas velocity used to
determine N
s
in Eq. (3) should be the maximum of the
inlet gas velocity, U
i
, or the outlet gas velocity, U
o
.
The curve to predict N
s
is valid only for very dilute
(low loading) cyclones where individual particles can
be assumed to be present in the cyclone (Zenz, 2001).
The assumption is generally valid for cyclone loadings
of no greater than approximately 1 grain/ft
3
(0.002 kg/
m
3
).
12.1.2 Partition Cuts
When dividing the overall particle size distribution
entering the cyclone into several cuts, it is advisable
to use a minimum of at least 10 cuts. It is also better
not to divide the distribution into equal size cuts.
Instead, it is better to bias the cuts so that the small
end of the particle size distribution has more cuts than
the large end. This is because small particles are more
dicult to collect than large particles, and it is the
small particles that will determine the collection e-
ciency of the cyclone. For each d
pi
in the partition,
there will also be a corresponding weight fraction x
i
,
for each cut.
12.1.3 Calculate d
pi
=D
p;th
and Determine E
oi
After D
p;th
has been calculated and the particle size
distribution has been partitioned, the ratio of
Figure 9 The eect of gas velocity on the number of solid spirals in a cyclone. For maximum velocity, use greater of inlet or
outlet velocity.
d
pi=
D
p;th
is calculated for each of the cuts in the biased,
partitioned particle size distribution. From the values
of D
pi
=D
p;th
, the low-loading collection eciency (E
oi
)
for each of the cuts can then be determined from
Fig. 10. The overall low-loading cyclone collection e-
ciency is then calculated from the individual cut collec-
tion eciencies by using Eq. (5), i.e.,
E
o

X
n
i1
x
i
E
oi
12.1.4 Determine the Effect of Loading on Cyclone
Efciency
The eect of solids loading can be calculated by using
the curves shown in Fig. 11. These curves are empirical
curves obtained from a considerable amount of data.
The curves are plotted on a graph with solids loading
on the abscissa and cyclone eciency on the ordinate.
To use the curves, rst nd the curve corresponding to
the low-loading cyclone eciency calculated above.
This eciency value will be at the lowest loading on
the plot. Follow this curve to the actual loading of he
cyclone, and read o the actual eciency of the
cyclone when the low-loading curve intersects with
the actual cyclone loading.
12.1.5 Calculate Cyclone Pressure Drop
Cyclone pressure drop is calculated by summing
several individual types of pressure-drop terms. The
individual pressure drops are listed and described
below:
Figure 10 Cyclone eciency versus d
pi
=D
p
;
th
.
Figure 11 Cyclone eciency versus loading.
1. Contraction Pressure Drop. The contraction
pressure drop applies to the pressure drop produced
by gas contracting from a large area to a small area.
For internal cyclones, the most common application of
the contraction pressure drop is for the contraction
from the large area of the freeboard of a uidized
bed to the smaller area of the cyclone inlet. For exter-
nal cyclones, the contraction pressure drop does not
usually apply from the freeboard of the bed to the
cyclone inlet. For external cyclones, the contraction
pressure drop will usually have to be applied twice,
from the area of the freeboard of the bed to the area
of the exit gas line, and from the area of the exit gas
line to the area of the cyclone inlet. The contraction
pressure drop is given by
P
fig
0:5
g
U
2
i
U
2
f
K
fi
U
2
i

6
In Eq. (6) K
fi
is the contraction coecient due to gas
owing from a larger area to a smaller area. It is a
function of the ratio of the smaller area (A
s
) to the
larger area (A
1
). The contraction coecient in the pres-
sure-drop equation for an internal cyclone is a function
of the ratio of the inlet area, A
i
, to the area in the
freeboard, A
f
. Values of this coecient for various
area ratios are shown in Table 2.
For external cyclones, people often measure the
pressure drop from a point immediately before the
cyclone inlet to a point immediately above the cyclone
in the gas outlet tube. This measurement does not
include the contraction pressure drop as gas ows
from the freeboard of the bed into the exit gas line
from the uidized bed. For dipleg seal height calcula-
tions, this pressure drop should be taken into account.
It is better to measure the total cyclone pressure drop
as the dierence between the pressure in the freeboard
and the pressure in the gas outlet tube.
2. Acceleration of Solids Pressure Drop. The
solids entering the gas with the cyclone must be accel-
erated from a very low gas velocity in the freeboard of
the uidized bed to the inlet gas velocity of the cyclone.
This term is generally the largest of the individual
cyclone pressure drop terms listed here, and is given by
P
fip
LU
pi
U
pi
U
pf

7
For small particles having a low terminal velocity, U
t
,
U
pi
is usually taken as U
i
, and U
pf
as U
f
. In general,
this approximation is used for particle sizes less than
about 100 microns. Hugi and Reh (1998) recently mea-
sured the solid and gas velocities in the inlet of a
cyclone and reported that there is generally signicant
slip between them, even for small particle sizes (40
microns). This probably occurs because of strand or
agglomerate formation in the cyclone inlet. However,
if no information is available regarding this slip, using
the general guidelines given above is probably the best
way to proceed.
3. Barrel Friction Pressure Drop. The pressure
drop due to solids owing along the wall of the cyclone
barrel is
P
bf

2f
g
U
2
i
D
b
N
s
d
hi
8
where f is the Fanning friction factor (generally
ranging between 0.003 and 0.008) and d
hi
is the
hydraulic diameter of the cyclone inlet given by the
relation
d
hi

4Inlet area)
Inlet perimeter
9
The Reynolds number for the barrel pressure drop
friction factor should be based on this diameter. This
Reynolds number is given by
Re
hi

d
hi
U
i
10
4. Gas Reversal Pressure Drop. The gas reversal
pressure drop is due to the gas reversing its direction
and its subsequent acceleration in the gas vortex in the
cyclone. It is given by
P
r

g
U
2
i
2
11
5. Outlet Exit Contraction Pressure Drop. The
outlet contraction pressure drop occurs because of
the gas contracting from the area of the cyclone barrel
to the area of the gas outlet tube. It is estimated by
P
o
0:5
g
U
2
o
U
2
b
K
o
U
2
o

12
Table 2 Values of the
Contraction Coefcient for
Various Small and Large Area
Ratios
Ratio of the smaller
to the larger area K
fi
or K
o
0.0 0.50
0.1 0.47
0.2 0.43
0.3 0.395
0.4 0.35
K
o
is obtained from Table 2 for vales of the ratio of the
outlet area to the inlet area calculated from (D
o
=D
b
2
.
When calculating dipleg seal heights, it is not neces-
sary to take into account the exit contraction pressure
drop in the calculations. This is because the pressure at
the top of the dipleg is actually the pressure drop inside
the cyclone, not the pressure in the gas outlet tube.
This is not really an issue in most calculations because
the exit contraction pressure drop is usually very low.
Also, it is conservative to include this pressure drop in
the calculation because it will result in slightly longer
dipleg lengths.
The sum of these ve pressure drop terms estimates
the total cyclone pressure drop. The cyclone pressure
drop estimated by these ve terms assumes that the
cyclone pressure drop is at a high enough loading so
that the cyclone pressure-drop reduction due to loading
is at its minimum, as shown in Fig. 7. If the cyclone
loading is lower than this (on the left-hand side of the
curve in Fig. 7), then a correction or an adjustment for
this higher pressure drop at low loadings must be made.
To determine if the solids loading is suciently low
so that the pressure drop must be adjusted, one must
use Fig. 12. To use Fig. 12, rst calculate the solids
mass ux (in kg/s-m
2
) at the inlet of the cyclone. If the
value of the mass ux at the cyclone inlet is less than
approximately 100 kg/s-m
2
, then the calculated pres-
sure drop must be adjusted. The amount of the adjust-
ment is determined from the value on the ordinate of
the gure. The ordinate is a ratio of the true cyclone
pressure drop to the calculated pressure drop. If the
value of the calculated inlet mass ux is so low that the
ratio is signicantly greater than one, then the calcu-
lated cyclone pressure drop must be multiplied by the
value on the ordinate. As an example, assume that the
value of the inlet solids ux is 10 kg/s-m
2
. The value of
the ordinate for this inlet mass ux is approximately
1.2. Thus the calculated pressure drop from the proce-
dure outlined above should be multiplied by 1.2 to
obtain the real pressure drop.
The cyclone design procedure that has just been
described is a tool that can be used to estimate cyclone
eciency. As with most calculation procedures, there is
always a dierence between what is measured and
what is calculated. As Zenz (2001) notes, the primary
dierence between the measured eciency and the cal-
culated eciency using this method is due to the cor-
rect estimation of the actual particle size entering the
cyclone. Agglomeration can occur in ne dusts to such
an extent that the actual cyclone eciency can be much
greater than calculated.
13 VORTEX LENGTH
The spinning gas from the outer annulus ows into the
inner rotating core (the vortex) over a certain length of
the cyclone. The length of this vortex, L
v
, is taken to be
the length within the cyclone where the vortex gas can
exert a positive driving force (Bryant et al., 1983). This
Figure 12 Cyclone pressure drop correction curve.
length is called the vortex length of the cyclone. If the
cyclone is so short that the bottom of the vortex length
is close to and/or touches the cone, solids will be
plucked from the wall of the cone and entrained into
the exiting gas in the vortex, so it is of great interest to
be able to estimate the length of the spinning vortex.
Alexander (1949) developed an empirical equation
to predict L
v
:
L
v
2:3D
o
D
2
B
A
i
!
1=3
13
As Bryant et al. (1983) note, this equation seems incor-
rect, because as D
o
gets smaller, it would be expected
that L
v
would get larger, not smaller as Eq. (13) pre-
dicts. Bryant et al. (1983) then developed an equation
for the vortex length that seems more directionally
correct:
L
v
D
B
2
A
o
=A
i
1=2
14
This equation is plotted in Fig. 13. For most cyclones,
the minimum practical value of A
o
=A
i
is approximately
0.4. At values much lower than this, the practical out-
let diameter becomes too small for most cyclones.
In the same study, Bryant et al. (1983) conducted
tests with a cylindrical cyclone with no cone that was
tted with a sealed piston lled with solids. When the
piston was raised in the cylinder the solids were sud-
denly entrained from the piston at a critical height
below the gas outlet tube of approximately 1:6D
b
.
When the piston was lowered, the vortex remained
attached to the bed surface until a second critical
distance was reached. This distance corresponded to
a value of approximately 2D
b
. Systematic experiments
have not yet been conducted to determine whether
the vortex length is actually a function of the dia-
meter of the cyclone. To investigate this, experiments
on dierent diameter cyclones would have to be con-
ducted.
Bryant et al. (1983) also reported that in tests with
sticky particles that adhered to the wall of the cyclone,
particles adhered to the cyclone cone at a distance
greater than 1:6D
b
below the gas outlet tube; but
above this length the cyclone was clean. This result
indicates that it would be better to have short cyclones
when operating with sticky cyclones.
Homan et al. (1995) studied the cyclone vortex
length and found that separation eciency was less
for cyclones with their natural vortex ending in the
cone instead of the barrel. They also reported that
the vortex length (swirl intensity) decreased with
increasing solids loading and was a strong function
of the cyclone length. Akiyama and Marui (1989), in
Figure 13 Vortex length vs. A
o
=A
i
.
their investigation of uniow cyclones, also reported
that the vortex swirl intensity decreased with increasing
cyclone length.
13 VORTEX STABILIZERS
As described above, the vortex of a cyclone will precess
(or wobble) about the center axis of the cyclone. This
motion can bring the vortex into close proximity to the
wall of the cone of the cyclone and pluck of and reen-
train the collected solids owing down along the wall
of the cone. Sometimes an inverted cone or a similar
device is added to the bottom of the cyclone in the
vicinity of the cone and dipleg to stabilize and x
the vortex. If placed correctly, the vortex will attach to
the cone and the vortex movement will be stabilized,
thus minimizing the eciency loss due to plucking the
solids o the wall of the cyclone.
14 EFFECT OF PRESSURE AND
TEMPERATURE
Most processes operate at high temperatures and/or
high pressures, so it is important to know how cyclones
operate at these conditions. Ecient cyclones are able
to collect very small particle sizes, so that cyclone e-
ciency is proportional to 1=D
p;th
, i.e.,
E
o
/
1
D
p;th
N
s
U
p
9L
w
s
15
The eects of temperature and pressure manifest them-
selves in the way that they aect gas density and gas
viscosity. From Eq. (15), it can be seen that cyclone
eciency is theoretically related to gas density and gas
viscosity as
E
o
/
r
16
As pressure is increased, gas density will increase.
However, the term
p
g
does not change with
increases in gas density because particle density is so
much greater than the gas density (typically about 2000
kg/m
3
versus approximately 20 kg/m
3
at high pressure)
that it dominates this term. Therefore, it would be
expected that gas density would have little or no eect
on cyclone eciency. Conversely, cyclone eciency
would be expected to decrease with system temperature
because gas viscosity increases with increasing tem-
perature.
Knowlton and Bachovchin (1978) studied the eect
of pressure on cyclone performance and found little
change in overall cyclone eciency with pressure over
a pressure range from 0 to 55 barg. However, fractional
eciency curves for the same study (Fig. 14) showed
that cyclone eciency decreased with pressure for par-
ticle sizes less than about 20 to 25 microns. For particle
Figure 14 Cyclone fractional eciency as a function of system pressure.
sizes greater than about 25 microns, there was no eect
of pressure on cyclone eciency. Karri (1998) also
showed that cyclone eciency decreased with pressure,
but only for particle sizes less than about 15 microns.
The eect of temperature on cyclone eciency was
studied by both Parker et al. (1981) and Patterson and
Munz (1989). Both studies showed that cyclone e-
ciency decreased with increasing gas viscosity. As with
the studies at high pressure, Patterson and Munz (1989)
reported that only the collection eciency of particles
less than about 10 microns was reduced because of
operation at high temperature (Fig. 15).
NOMENCLATURE
A
f
= Freeboard area above uidized beds, m
2
A
i
= Cyclone inlet area, m
2
A
1
= Larger of the ow areas in gas contraction
ow, m
2
A
o
= Area of cyclone gas outlet, tube, m
2
A
s
= Smaller of the ow areas in gas contraction
ow, m
2
D
b
= Cyclone barrel diameter, m
D
p;th
= Theoretical particle size collected by a
cyclone at 50% eciency, m
D
o
= Gas outlet tube diameter, m
d
hi
= Hydraulic diameter of cyclone inlet, m
d
p;50
= Median particle size, m
d
p;th
= Theoretical particle size capable of being
collected by a cyclone, m
E
o
= Cyclone eciency, %
E
oi
= Low-loading cyclone collection eciency, %
f = Fanning friction factor
H = Cyclone inlet height,
K
fi
= Contraction coecient for ow from the
freeboard to the cyclone inlet
K
o
= Contraction coecient for ow from the
cyclone barrel to the cyclone outlet
L = Solids loading, kg of solids/m
3
of gas
L
v
= Vortex length below gas outlet tube, m
L
w
= Cyclone inlet width, m
N
s
= Eective number of solid spirals in a cyclone
Re
hi
= Reynolds number at the cyclone inlet based
on d
hi
r
o
= Radius of cyclone, m
U = Supercial gas velocity, m/s
U
f
= Gas velocity in freeboard of uidized bed, m/s
U
i
= Cyclone inlet gas velocity, m/s
U
o
= Gas velocity in cyclone outlet tube, m/s
Up
i
= Particle velocity at solids inlet, m/s
U
pf
= Particle velocity in freeboard of uidized bed,
m/s
U
t
= Particle terminal velocity, m/s
x
i
= Weight fraction of cut i in particle size
distribution
P = Pressure drop, cm H
2
O
Figure 15 Cyclone fractional eciency as a function of system temperature.
P
fig
= Gas contraction pressure drop from
freeboard to cyclone inlet, cm H
2
O
P
fig
= Solids acceleration pressure drop, cm H
2
O
P
bf
= Cyclone barrel friction pressure drop, cm H
2
O
P
o
= Cyclone exit pressure drop, cm H
2
O
= Gas viscosity, kg/(m-s)
p
= Particle density, kg/m
3
g
= Gas density, kg/m
3
REFERENCES
Akiyama T, Marui T. Dust collection eciency of a straight-
through cyclone: eects of duct length, guide vanes and
nozzle angle for secondary rotational ow. Powder
Technol 58:181, 1989.
Alexander RM. Fundamentals of cyclone design and opera-
tion. Proceedings of the Australian Institute of Mining
and Metallurgy 152, 153:203, 1949.
Barth W. Berechnung und auslegung von zyklonabscheidern
aufgrund neuerer untersuchungen. BWK 8(1):19, 1956.
Barth W, Leineweber L. Beurteilung und auslegung von
zyklonabscheidern. Staub 24(2):4153, 1964.
Broodryk NJ, Shingles T. Aspects of cyclone operation in
industrial chemical reactors. Preprints of Fluidization
VIII. Tours, France, May 1419, 1995, p 1083.
Bryant HS, Silverman RW, Zenz FA. How dust in gas aects
cyclone pressure drop. Hydrocarbon Processing June,
pp 8790, 1983.
Dietz PW. Collection eciency of cyclone separators. AIChE
J 27:888, 1981.
Gauthier TA, Briens CL, Bergougnou A, Galtier P. Uniow
cyclone eciency study. Powder Technol 62:217225,
1990a.
Gauthier TA, Briens CL, Galtier P, Bergougnou MA. Gas
solid separation in a uniow cyclone at high solids load-
ings: eect of acceleration line. In: Basu P, Horio M,
III. Oxford: Pergamon Press, 1990b, pp 639644.
Gauthier TA, Briens CL, Crousle O, Galtier P, Bergougnou
MA. Study of the gas circulation patterns in a uniow
cyclone. Can J Chem Eng 70:209215, 1992.
Homan AC, de Groot M, Peng W, Dries HWA, Kater J.
Advantages and risks in increasing cyclone separator
length. AIChE J 47(11):24522460, 2001.
Homan AC, de Jonge R, Arends H, Hanrats C. Evidence of
the natural vortex length and its eect on the separa-
tion eciency of gas cyclones. Filt Sep 32:799, 1995.
Hugi E, Reh L. Design of cyclones with high solids entrance
loads. Chem Eng Technol 21(9):716719, 1998.
Iinoya K. Memoirs of the faculty of engineering. Nagoya
University 5(2): September, 1953.
Kaliski H. Vergleich von iehkraftentstaubern. Freiberger
Forschungsheft A93:78100, 1958.
Karri SBR. The eect of pressure on cyclone performance.
PSRI Research Report No. 74, June 1, 1998.
Knowlton TM, Bachovchin D. Coal processing Technol
4:122127, 1978.
Koman JL. Gas and Oil Power 48:8994, 1953.
Lapple CE. Ind Eng Chem 32:605617, 1940.
Lapple CE. Dust and mist collection. Air Pollution
Abatement Manual, Manufacturing Chemists
Association, Washington, D.C., 1951.
Lewnard JJ, Herb BE, Tsao TR, Zenz JA. Eect of design
and operating parameters on cyclone performance for
circulating uidized bed boilers. In: Avidan AA, ed.
AIChE, 1993, pp 525531.
Muschelknautz E. Auslegung von zyklonabscheidern in der
technischen praxis. Staub-Reinhaltung der Luft
30(5):187195, 1970a.
Muschelknautz E. Design of cyclone separators in the
engineering practice. Staub-Reinhaltung der Luft
30(5):187195, 1970b.
Muschelknautz E, Krambrock W. Pressure drop and separa-
tion eciency in cyclones. VDI Heat Atlas, 6, and rst
English edition, Ch. Lj, 18, 1996.
Muschelknautz U, Muschelknautz E. Improvements in
cyclones in cfb power plants and quantitative estimations
of their eects on the boilers solids inventory. In: Werther
Wurzburg, Germany, 1999, pp 761767.
OMara R. Combustion 21:3843, 1950.
Parker R, Jain R, Calvert S, Drehmel D, Abbott, J Envir Sci
Technol 15(4):451, 1981.
Patterson P, Munz R. Can J Chem Eng 67:321, 1989.
Rentschler W. Abscheidung und druckverlust des gaszyklons
in abha ngigkeit von der staubbeladung. VDI-
Fortschrittbericht, Reihe 3, Nr. 242, 1991.
Rietema K, Verver CG. Cyclones in Industry. New York:
American Elsevier, 1961.
Rosin P, Rammler E, Intelmann W. Principles and limits of
cyclone dust removal. Zeitschrift Verein Deutscher
Ingenieure 76:433, 1932.
Smellie J. Iron Coal Trades Review 144:169, 1942.
Stairmand CJ. Trans Inst Chem Eng 29:356, 1951.
Ter Linden AJ. Inst Mech Engrs 160:233, 1949.
van Tongren H. Mech Eng 58: February, p 127, 1936.
Whiton LC Jr. Trans Am Soc Mech Engrs 213218, 1941.
Yuu S, Tomosada J, Yuji T, Yoshida K. The reduction of
pressure drop due to dust loading in a conventional
cyclone. Chem Eng Sci 33:15731580, 1978.
Zenz FA. Cyclone separators. In: Chapter 11, Manual on
Disposal of Renery Wastes; Volume on Atmospheric
Emissions, API Publication 931. Washington, DC: Pet.
Inst. Rening, 1975.
Zenz FA. Cyclone design tips. Chem Eng January, p 60,
2001.
23
Dilute-Phase Pneumatic Conveying
George E. Klinzing
University of Pittsburgh, Pittsburgh, Pennsylvania, U.S.A.
1 INTRODUCTION
Pneumatic conveying occurs in almost every industrial
application that uses powders and granular materials.
Most of this has been handled by dilute-phase convey-
ing. Particle sizes ranging from a few microns to 3
inches in diameter have been successfully conveyed,
and the size of the lines employed for conveying go
from
1
4
-inch to 24 inches in diameter. Generally, if a
material is sticky or moist, other methods of transport
such as belt conveyors are preferred over dilute-phase
pneumatic conveying. One guideline in material hand-
ling with dilute-phase pneumatic conveying is to avoid
abrasive materials; however, this guideline often will be
violated to solve a particular conveying problem.
Abrasive materials usage causes high maintenance
cost for pneumatic conveying systems. Pneumatic con-
veying has been applied to such diverse situations such
as the conveying of ice over 1 kilometer in length to
cool gold mines, under reduced gravity conditions to
explore the possibility of using the technology on the
moon, and even for the conveying of live animals.
Pneumatic conveying protects the product from the
environment and the environment from the product
because of its enclosed nature. There have been
dilute-phase pneumatic conveying applications that
extend up to 3 miles in distance, while other uses call
for a distance of a few meters.
Dilute-phase pneumatic conveying is not an energy-
ecient way of conveying, since generally considerable
horsepower is needed to provide the motive air or gas.
Dilute-phase pneumatic conveying, however, is easy
and convenient to put into operation, even if often it
does not work at optimum conditions, mostly owing to
poor understanding of the overall concept of the
process.
There are ve components to a dilute pneumatic
conveying system:
1. Conveying line
2. Air/gas mover
3. Feeder
4. Collector
5. Controls
Basically, dilute-phase pneumatic conveying systems
can be broken down into three categories based on
the physical principle used for conveying.
1. Pressure system
2. Vacuum system
3. Pressure/vacuum system
The modes of pneumatic transport are often classied
as dilute, strand (two-phase), and dense phase ows.
This discussion will concentrate on dilute transport
technology.
2 PARTICLE CHARACTER AND BEHAVIOR
2.1 Size Analysis
In order to convey particles, it is rst necessary to
characterize the material according to size and size
distribution. In the past, the literature was lled
with studies that reported that one conveyed sand
or a particular seed with little mention of the size
and size distribution. The average particle size used
in most calculations is the d
50
, which represents the
diameter at which 50% of the particles are less in size.
For particles that are formed by a breaking or
fracture process, the log normal distribution is most
commonly used to represent the size distribution.
Particles formed by a condensation process can be
described by the regular normal distribution. There
are many dierent types of particle size standards,
and depending on the type of operation, one sizing
standard or measurement technique can be preferred
over another. Microscopes give length mean averages,
while light measuring devices give volume or surface
mean averages. The common sieving operation
provides one with the weight mean. The weight
mean average is the one often employed in most of
the discussion in pneumatic conveying. The calcula-
tions to be employed use the weight mean average for
the size of the particle. In general, unless one has a
particularly unique or odd size distribution, this
average size will be acceptable as an average particle
size in the calculations. For a more detailed
discussion of size analysis, see Chapter 1, Particle
Characterization and Dynamics.
2.2 Shape Analysis
In addition to knowing the size and distribution of
the particles, the shapes of the particles are essential
for understanding pneumatic conveying dierences.
Particles can have various shapes: rounded, angular,
and brous. The latter is the most dicult to address
in choosing a correct particle size for conveying analy-
sis.
The shape of the particle can be given by the
traditional sphericity factor, fractal analysis, or by
Fourier transform representations. The latter is a bit
involved, requiring several coecients for complex
denition. Fractal analysis is receiving more attention
of late in representing the shape of the particles
handled. Chapter 1, Particle Characterization and
Dynamics, presents a more complete description of
shape analysis.
After dening the size and shape, one is now ready
to address the overall force balance of the particlegas
system. A macroapproach analysis will be presented in
this treatment of dilute-phase transport. Using a
kinetic approach with individual particle behaviors is
computationally intense but has shown some success in
depicting the actual physics of the process. The conti-
nuum approach is another procedure that can be com-
putationally intensive but productive. There has been
much discussion over the proper force terms and
viscosity presentations and turbulence models in
these analyses.
2.3 Force Balance
The macroapproach uses Newtons second law initially
for a single particle system and then expands to a
multiparticle system. Some of the concepts of a sin-
gle-phase friction factor will be employed in the multi-
particle system analysis. In addition to the acceleration
and drag forces, the gravity and electrostatic forces
should be considered in gassolid analysis. The varia-
tion of the drag coecient with the kind of particle and
condition of ow is important in the analysis. The
terminal velocity of the particle is often used as
another way to characterize particles, and the larger
the value of the terminal velocity, the greater the size
and/or density of the particle.
Multiparticle systems must consider the same terms
as the single particle analysis with particleparticle and
particlewall interaction essential for a complete
description of the system. The pressure behavior is
also of the utmost importance because of the nature
of gaseous environments. In treating multiparticle sys-
tems, the drag term has often been corrected to include
a modication to the drag on a single particle. One
correction employs the voidage term, which comes
from the classic Richardson and Zaki (1954) analysis
on settling of multiparticle systems (voidage
4.7
where
the voidage is dened below). Pneumatic conveying
analysis must concentrate on the most important fac-
tors in its analysis, the pressure drop loss or the energy
loss, the product ow rate, and the volumetric gas ow
rate. These factors inuence the operational cost of the
overall system over long periods of time.
There are several approaches to developing predic-
tive models for pressure drop and energy losses in
dilute-phase pneumatic conveying. The German litera-
ture, for instance, uses an expression proposed by
Barth (1954) that includes a lift term in the horizontal
analysis, while the U.S. approach is prone to lump this
eect into the frictional term.
If one reviews the basic pressure loss equation,
P
L
,cg ,
p
1 cg f
s
,
p
1 c
u
2
p
2D
2f
g
,cu
2
g
2D
1
Pressure gas solid solids gas
loss gravity gravity friction friction
effect effect
there are a number of parameters that are required to
begin to calculate some values:
1. Voidage
2. Particle velocity
3. Friction factors
In order to complete the picture, these parameters will
be discussed and guidelines and suggestions will be
given on appropriate values and calculations.
2.4 Voidage
The voidage is the amount of void space present in a
pipe carrying solids. The voidage in the pipe can be
expressed as
c 1
4W
s
u
p
,
p
2
For a packed bed, voidage values of 0.4 to 0.6 are
common, while in dilute pneumatic conveying, values
of 0.98 and above can be found. One should also
realize that in pneumatic conveying, especially in the
horizontal direction, there is a gradation of the
concentration of particles from top to bottom, smaller
to larger in value. In most of the calculations presented
here, a uniform voidage will be assumed. The super-
cial gas velocity in a conveying line is the velocity
when no particles are present, while the uid velocity
when particles are present is larger by the factor of
1/voidage.
2.5 Particle Velocity
For determining the particle velocity, some guidelines
are available. For particles less than 40 micrometers in
size, the particle size can be given by the dierence
between the supercial gas velocity and the terminal
velocity:
u
p
u
g
u
t
3
Actually, in our studies we have found that this expres-
sion is good up to about 500 micrometers in particle
size. Several other studies have given models for the
particle velocities. In order to be sure that actual mea-
surement of the particle velocity is carried out, the
HinkleIGT (1978) model can be related to several
systems and particle parameters:
u
p
u
g
1 0.68d
0.92
,
0.5
p
,
0.2
g
D
0.54
4
Yang (1977) has developed an implicit equation to pre-
dict the particle velocity that is described as
u
p
u
g
u
t
1
2f
s
gD
u
2
p
_ _
c
4.7
_ _
0.5
5
f
p
4f
s
6
2.6 Friction Factors
The frictional representation in pneumatic conveying
assumes that the gas can be treated by itself for repre-
senting the gas friction and the solid frictional term for
the solid friction. This linear representation has been
recently challenged by Weber (1991), who provides
some success with a nonlinear combination of these
two terms. For the single-phase frictional term, the
Koo equation (1932) as shown below is suggested,
although the Blasius (1913) could just as well be used.
f
g
0.0014 0.125Re
0.32
7
For dilute-phase transport, the solid frictional term has
been suggested by many investigators. We have found
that the expression of Konno and Saito (1969) has the
widest applicability; it even spills over into dense phase
analysis:
f
s
0.0285gD
0.5
u
1
p
8
Another frictional representation that has been receiv-
ing some attention has been developed by Michaelides
(1987):
f
s

K
4
gD
p
u
p
9
where K depends on the type of particle conveyed. For
coal it is 0.058.
The eect of dierent pipe wall materials has been
probed by Rizk (1973) and others. This eect, how-
ever, appears to be minor and even of little conse-
quence when a ne powder coats the conveying line
wall. For inclined ow Klinzing et al. (1989) has
given a frictional representation that covers all kinds
of inclined ows from horizontal to vertical:
f
s

D
2u
2
p
3,4 C
D
c
4.7
,
g
u
f
u
p
_ _
2
_ _
,
p
,
s
d
g sin 10
where is the angle of inclination.
2.7 Acceleration
The analysis so far has assumed that the ows are
steady and nonaccelerating. When entering a convey-
ing line, going around a bend or connection, and exit-
ing a line, acceleration eects come into play. The
acceleration can be handled with the basic Newton
analysis as well as through empiricism. A combination
of both of these has been suggested as one approach
for analysis. It should also be noted that acceleration
eects are more dominant in horizontal conveying
than in vertical conveying. A model that has proven
useful for acceleration is
L
accel
27.66
d
D
_ _
0.953
j
0.0912
,
,
p
_ _
0.924
11
2.8 Saltation
Saltation is a phenomenon that occurs in horizontal
ow: particles fall out of the suspension that is convey-
ing the material and deposit on the bottom surface of
the pipe. Many studies have been explored to deter-
mine the saltation condition, since it is a parameter
essential in designing conveying systems. The most
commonly used correlation is that of Rizk (1973),
which was developed for plastic pellets and shows no
eect of particle density or pipe size:
j
1
10
o
u
g
gd
_
_ _
12
where
o 1.44d 1.96
1.1d 2.5
Matsumoto et al. (1977) have a useful correlation for
saltation that breaks the analysis into two parts, one
for larger particles and one for smaller ones.
The saltation velocity has been used often as the
design parameter for the velocity at the feed point
for conveying. In addition, the pickup velocity has
also been employed for this purpose.
2.9 Pickup Velocity
The pickup velocity in pneumatic conveying has two
dierent denitions. One is the velocity required to
pick up particles from the bottom of the pipe where
they are at rest because of a ow condition or block-
age. The other denition is the velocity at the feed
point of the solids. Mostly in this discussion attention
will be paid to the former, although the latter will be
mentioned sometimes. A rule of thumb is the relation-
ship that indicates that the saltation velocity is roughly
half the pickup velocity. Thus designing on the pickup
velocity for the conveying velocity is more conserva-
tive. In studying the pickup velocity of particles, one
nds that as the particle size decreases from values of
1000 micrometers, the pickup velocity will decrease
until a minimum point is reached and it begins to
rise with further size decrease. The smaller particles
have larger surface and interparticle forces acting on
them. For larger particles, recent work by Cabrejos
and Klinzing (1994) gives what appears to provide reli-
able saltation and pickup velocities;
u
s
gD
p
u
s0
gD
p 0.0022
,
s
,
_ _
1.25
j
0.5
13
u
gpu
gD
p 0.9428Re
0.175
p
_ _
D
d
_ _
0.25
,
,
s
_ _
0.75
14
where Re
p
du
g
,,j, 25 - Re
p
- 5000;
8 - D,d - 1340, 700 - ,,,
s
- 4240.
A phase diagram showing the interconnectiveness of
these two parameters was also developed and is shown
in Fig. 1.
2.10 Compressibility
Up until this point we have not discussed the obvious
issue of compressibility of the conveying gas in dilute-
phase pneumatic conveying. For short systems (- 300
ft) and low-pressure systems, the change in the gas
density is relatively small, and an average value can
be employed over the system without encountering
much error. For longer distance conveying and high-
pressure systems, compressibility is of paramount
importance in the analysis. It should also be noted
that in most conveying situations the temperature of
the gas is relatively constant so isothermal conditions
can be applied (Gas temperature Product tempera-
ture).
2.11 Bends
Most conveying systems have bends or connections. If
one has a system that is relatively long (> 300 ft) and
has few bends, the presence of the bends has a minor
eect on the overall system pressure loss. If a system
has several bends and is of relatively short length
(- 300 ft), the bend pressure loss is signicant.
Again, for the analysis of bend pressure loss, the linear
combination of the eects of the gas and the eects of
the solids is applied. The classic work of Ito (1959,
1960) for single-phase ow is used to establish the
base for the pressure loss in bends. The work of
Schuchart (1969), although performed on a limited
material, has been employed with success to determine
the solids contribution to the bend pressure loss.
P
bend
P
_ _
solids
210
2R
B
D
_ _
1.15
15
The design suggestion of using the traditional pressure
loss factors for single-phase ow in two-phase ow has
also met with some success in analyzing the pressure
loss around bends. One factor that is important in
bend analysis in pneumatic conveying is to avoid hav-
ing three bends in quick succession in the design. This
arrangement will most often lead to blockages and
unsteady operation.
2.12 Choking Conditions
Choking conditions for vertical ow is likened to the
saltation condition in horizontal ow. The most com-
plete analysis of choking to date has been carried out
by Yang (1975), who correlated a vast amount of exist-
ing literature data. These two expressions must be
solved simultaneously:
2gD c
4.7
choking
1
_ _
u
choking
u
t
_ _
2
0.01
u
p
1 c
choking
_ _
u
choking
u
t
_ _
1 c
choking
_
16
Figure 1 Bifurcation diagram, including pickup and saltation mechanisms of solid particles and dierent ow regimes observed
in fully developed ow of gassolids suspensions in horizontal pipelines.
One should note that if a system has both vertical and
horizontal legs, then saltation will occur sooner than
choking in a system. Only when the pipe diameter is
increased in the vertical ow is there concern of chok-
ing problems.
With this presentation on pneumatic conveying, the
basic design consideration and analysis is reviewed.
Further analysis of such systems is possible. One
theme that has received much attention is the area of
ow classication with several dierent phase diagram
approaches.
The strategy for determining the energy required for
transport is given in Table 1.
3 FLOW CLASSIFICATIONS
The ow classication of Geldart (1973) for uidiza-
tion has been applied also to pneumatic conveying.
Detailed discussion of Geldsarts classication of
powders is presented in Chapter 3, GasSolids
Fluidization. The four materials A, B, C, and D
have a signicant relationship with pneumatic convey-
ing ease and ability. The C type powders, which are
cohesive in nature, are denitely the most dicult to
move, although the piston type ows can be easily
formed with these materials. The free-owing A pow-
ders work well in dilute-phase ow. These materials
cause few problems in the coupling of the feeder unit
with the conveying line. The B materials generally do
not move well in the more heavily loaded systems as
strand ow/two-phase ow. The D materials have been
moved over long distances in dilute-phase conveying in
some mining applications.
The classic phase diagram in pneumatic conveying
is that proposed by Zenz (1949) having the pressure
drop versus gas velocity plotted at constant solids
ow rates. A cross plot of this data can also be used
to explain phase ow behaviors. The Zenz diagram
points clearly to the unsteady regimes and shows that
a minimum in pressure drop is achieved at a particular
conveying velocity pointing to the ideal operational
condition. Using the Zenz diagram one can follow
the ow from the beginning to the end of the pipeline,
since as the ow continues, the gas velocity increases
and the density decreases. One operating condition is
not present (note Fig. 2). If one plots the product of the
density and gas velocity instead of the gas velocity
only, a single operating point for the whole system is
achieved and a more comprehensive picture of the sys-
tem is presented. In pneumatic conveying the concept
of the clustering of particles is often mentioned
although not truly observed yet. Some indirect mea-
surements also point to the presence of this phenom-
enon. Clustering in vertical ows has been studied
more than in horizontal ows.
Our research team has looked at phase relationships
in pneumatic conveying. With the suggestion given by
Matsen (1982), the thermodynamic analogy of van der
Waals equation to pneumatic conveying has been used
with a certain degree of success. In this analogy the
temperature maps as the gas ux, the pressure as the
particle ux, and 1/volume as (1-voidage). A division
between dilute and dense phase has been clearly
demonstrated. Figure 3 depicts this phase behavior
with actual experimental points.
Flow pattern identication has recently been probed
in our research group by observing the pressure uctua-
tions seen by a pressure transducer placed on the wall of
the conveying line. This concept comes from the use of
audio vibrations by Solt (personal communications,
1988) to diagnose the behavior of pneumatic conveying
systems. By analyzing the pressure uctuations with
statistical properties, it has been shown that distinct
ngerprints can be generated by the dierent ow con-
ditions. Rapid determination of the ow conditions can
permit one to employ this information as a control
device for conveying systems. In a similar analysis,
fuzzy logic has been employed to develop a phase dia-
gram that has regions of varying degrees of stability of
the ow condition. The boundaries dividing the regions
are the most unstable. Fuzzy logic can be employed in
such diagrams to establish the ow regimes.
Table 1 Design Strategy for Dilute-Phase
Transport.
1. Select the criteria for the transport velocity:
saltation velocity, pickup velocity, specied
velocity.
2. Determine the length of each horizontal and
vertical section.
3. Determine the number of bends.
4. Determine the pressure drop for each section,
horizontal, vertical, and bends.
5. Calculate the acceleration pressure loss for the
entrance of the particles into the system.
6. Calculate the over-pressure loss by summing the
component parts.
7. Decide if the system should have a stepping
pipeline arrangement if the over-distance is
greater than 300 feet.
8. Select the blower, feeder, and collector for the
system.
Figure 2 General Zenz-type diagram of pressure drop per unit length versus velocity for varying owrates. Note dense, dilute,
and saltation regimes.
Figure 3 Phase diagram-type plot of reduced pressure analog versus the volume analog at constant reduced gas ow analog.
Note similarities of strand ow to liquid phase, ow above strand ow to vapor phase, and homogeneous ow to gas phase.
In order to complete the design of a dilute-phase
pneumatic conveying system, other components of
the systems must be determined.
4 ESSENTIAL COMPONENTS OF DILUTE-
PHASE PNEUMATIC CONVEYING
SYSTEMS
4.1 Air Movers
There are three dierent kinds of air movers that are
commonly used for dilute-phase pneumatic conveying:
1. Fans
2. Blowers
3. Compressors
If the power supply is reliable and the ow is relatively
dilute, fans are reliable units. If there is an upset in the
conveying, fans cannot provide the energy to reentrain
the solids from the bottom of the pipe, and the system
must be cleaned out before restarting. Of course, fans
are the most inexpensive of the three air movers. The
blower is the workhorse of pneumatic conveying, and
its perfection about the early 1950s gave a boost to
pneumatic conveying installations. Blowers can be
used for a variety of ow conditions and provide
sucient energy to restart a ow that has been inter-
rupted. The blower produces air that can be quite high
in temperature that can also heat the associated piping.
If one is conveying plastic particles, limits on the exit
pressure may be necessary to reduce the temperature of
the conveying air so that the plastic particles do not
soften and melt. In order not to exceed the auto-
ignition temperature of the oil in the blower, 232
C
(450
F), a control should be set to shut down the

blower when this condition is approached. When the
need is for pressure over 2 bars, the blower cannot
respond, so compressors are called into play. These
units, however, are expensive and require very good
ltration of the input air to function well. In a dusty
environment, this restriction can be severe. Plant air
that comes from a central compressor system can be
used for conveying. Often practitioners think that this
air is free and do not factor this cost into their opera-
tion, since the compressors are not designed solely for
the conveying system.
4.2 Feeders
There are many varieties of feeders for pneumatic
conveying systems. For vacuum systems, one can
use a simple orice device for free-owing material.
A nozzle arrangement can be used in a vacuum system
much as a vacuum cleaner. This nozzle can be
employed for small projects such as the cleanup of a
powder spill to large operations such as the unloading
of grain from the hold of a ship. In pressure systems,
one can use the uid head pressure as in an airlift
operation as well as a uidized feed arrangement.
The venturi feeder can also be used in pressure
systems. The most common feeder for pneumatic
transport is the rotary feeder. The design of these
rotary feeders varies. The rotary feeder usually oper-
ates between 10 and 30 rpm. The throat of the rotary
feeder should not restrict the ow of solids. Leakage
of air through the rotary feeder requires one to design
for venting of the feeder.
For pressure systems one often nds lock hopper or
gate lock feeders. These units operate similar to the
blow tanks.
The screw feeder is another common type of
feeder for pneumatic conveying. This type of feeder
can often be used as measuring devices. Screw feeders
can operate on a wide range of materials from powders
up to
1
2
inch diameter materials. In general, moderately
abrasive materials can be tolerated by screw feeders.
Pressure tanks can serve as feeders for pneumatic
conveying in a batch or continuous operation.
Pressure tanks can be used for both dilute and dense-
phase conveying operations.
4.3 Diverter Valves
Diverter valves are units commonly employed in pneu-
matic conveying to divide ows or redirect ows in
various units. One constant concern with diverter
valves in dilute-phase conveying is the erosion that
may take place at this high impact point.
4.4 Filters
Filters or other collection devices are required for all
conveying systems. The reserve-jet air ow systems for
cleaning the lter are very common units. For lower
pressure, the air-to-lter cloth ratio is usually approxi-
mately 1.1 to 0.3 CMM/m
2
(3.5 to 1.0 CFM/sq.ft). For
higher pressure reserve ow systems, a higher air-to-
cloth ratio is employed, 2.1 to 0.3 CMM/m
2
(7.0 to 1.0
CFM/sq. ft). For more detailed discussions, see
Chapters 22 and 28, GasSolids and LiquidSolids
Separation.
4.5 Cyclones
Cyclone collectors are simple units with a complex ow
eld that even today baes our understanding. The
cyclones work best for the collection of coarse materi-
als. Cyclones are a source of high wear and breakage of
the material conveyed. Friable materials should not be
employed in cyclone arrangements. See Chapter 22 for
more details.
4.6 Scrubbers
When wet environments are used for collection, the
resultant product is a slurry. Disposal and handling
of this slurry must be dealt with. Often all the scrubber
does is to transfer the handling problem from a dry
phase to a wet phase.
4.7 Piping Systems Components
The material of construction has a minimal eect on
the overall dilute-phase conveying characteristics of
pneumatic conveying. Some manufacturers have had
success in treating the inside of a pipe so as to create
a roughened surface for plastic pellet conveying.
The layout of the piping systems has many impor-
tant factors in pneumatic conveying. One should keep
the ow path as the most direct between two points.
Bends should be eliminated as much as possible. Care
should be taken in the design that the distance after a
feed point before the rst bend is inserted in a
minimum of 3 meters (10 feet) when two-phase con-
ditions are present. If the ow is dilute or dense, this
distance is not crucial. The two-phase condition tends
to cause a sloshing of the solids in the bend in an
unsteady condition. This sloshing behavior causes
plugging and other upsets in the operation of the
pneumatic conveying systems. As noted before, one
should at all costs avoid more than two bends in
quick succession.
Sloping lines are generally not a good idea for pneu-
matic conveying. When the ow is dilute, sloping will
have minor eects on the transport operation. When
two-phase ow is present, the slopes of 30 to 85 degrees
are dangerous because of the recycling of the solids,
which causes instabilities and erosion. When one has a
downward ow, generally one does not have any con-
cerns in operation. For the two-phase ow regime, one
should have horizontal runs of 12 meters (40 feet) after
reaching the bottom to have the phases separate.
Couplings in pneumatic transport are important
hardware units. Considerable vibrations can be placed
on the system from the feeder or air supply as well as
the condition of ow. The piping should be kept
aligned and have electrical conductivity achieved by a
conductive strap over nonconductive sections such as
gaskets. Welded connections are good but dicult to
maintain since repairs require breaking the welds. For
high-pressure systems, anges can be used, but align-
ment of the piping is imperative, since anged connec-
tions are not always properly aligned. Clamps can be
used for low-pressure systems but may provide sources
of leakage or line separation, and they require electri-
cal grounding.
Bends are crucial hardware elements in all pneu-
matic conveying systems. Historically the practice has
been to use long-radius bends for all turns. Long
radius is dened as a radius of 12 pipe diameters or
more. These bends have been a great source of wear
problems that usually occur on the outside of the bend
primarily and secondarily on the inside of the bend.
Wear back additions are common in these types of
bends. Short-radius bends have a radius of 2.25 pipe
diameters. One nds the wear rates of these bends to be
higher than those of long radius. The dead end T-bend
(barrel bends, etc.) has been found to be a successful
wear resistant element in pneumatic conveying. Some
secondary wear is experienced about one meter (3 feet)
after the bend. In using these bends, sticky and very
cohesive materials are not recommended. In general,
the pressure loss around the bends is similar in nature
for all bends with an equivalent straight pipe loss of
approximately 4.6 meters (15 feet).
5 TROUBLESHOOTING
The topic of troubleshooting is extensive and can grow
as additional information is obtained. An articial
intelligence (AI) program has been developed in our
laboratory for troubleshooting (Dhodapkar and
Klinzing, 1993), in which we assimilate the experience
of a number of experts and literature citations; yet this
program is not complete. This program does provide a
beginning for the operations person to attempt to
resolve problems in the conveying system as they
arise. The technique of troubleshooting a system
depends on whether it is brand new or has been work-
ing for some time and develops problems.
For a system that used to meet capacity, one can
look to worn equipment such as an airlock that has
increased leakage or blocked or plugged aeration
media. Plugged venting systems, changes in the solids
characteristics, such as size distributions, and material
feed blocked by foreign material are often-encountered
problems.
In order to increase the inadequate capacity of a
pneumatic transport system, a number of actions can
be taken. For example, consider modifying the air
supply quantity:
1. Stream ow system (reduce the air volume to a
minimum steady conveying).
2. Two-phase ow system (increase the air volume
to reduce slugging).
3. Dense phase system (increase the air volume).
Modifying the air supply pressure: increased pressure
may be required.
Change line congurations:
1. Shorten the line length.
2. Eliminate the bends.
3. Increase the line diameter for stepped line at the
end of the conveying system.
If one experiences excessive line wear, one can
1. T-bends can be used.
2. The material-to-air ratio can be increased.
3. Step pipe diameters through the system can be
carried out to maintain a constant velocity
throughout the system.
When material breakage is present, one can
1. Reduce the conveying velocity
2. Eliminate bends, cyclones, and impact points
3. Reduce the velocity before the terminal point of
the system.
6 LONG DISTANCE CONVEYING
Conveying of solids over long distances up to a few
thousand feet encounters dierent phenomena owing
to the gas expansion of that distance. The gas expan-
sion caused the gas and solid velocities to increase and
thus aect the attrition of the particles and the erosion
of the pipeline, both of which increase to the nth power
with velocity, where n ranges from 2.5 to 5.0. To slow
the particles down, one can step the pipeline to larger
diameters to decrease the velocities. Another technique
developed in our laboratories is to use a ow econo-
mizer, which eectively takes gas from the system in a
prescribed fashion to reduce the gas velocity. Figure 4
shows a schematic of the ow economizer tested in our
laboratory. This unit withdraws a prescribed amount
of air at critical points in the transport line.
7 APPLICATIONS OF COMPUTER ANALYSIS
In order to design a dilute-phase pneumatic conveying
system, one needs to have a mechanism to select and
calculate a number of factors and components. The
rst question is, What type of system would be appro-
priate for the conveying operation? Should one use a
vacuum or pressure system? Is the dilute-phase the
mode most appropriate for transport? Using an expert
system approach, our laboratory has developed a pro-
gram entitled NUSELECT (Pneumatic Conveying
Consultants) to help answer these types of questions.
After selecting the system, one needs to choose the
appropriate feeder to inject the material into the pipe-
line for transport. Again, we have addressed this ques-
Figure 4 Diagram of the ow enhancer.
tion by developing an expert system program entitled
FEEDER for this task. Calculations presented above
should provide the basis for predicting the energy
needed for transport to help complete the design.
Most often a cyclone or reverse jet lter is used for
collecting the product after transport. When a system
is in operation, more than likely there will be occasions
when the system is not functioning properly and
troubleshooting the operation is needed. The expert
system PANACEA provides the basis for beginning
troubleshooting before an expert in the eld is called
in for analysis.
NOMENCLATURE
C
D
= drag coecient
D = diameter of the pipe
f
g
= gas friction factor
f
s
= solids friction factor
g = gravity constant
L = length
P = pressure drop
u
g
= gas velocity
u
p
= particle velocity
W
s
= solids ow rate
Greek
c = voidage
= angle of inclination of the pipe
,
g
= gas density
,
p
= particle density
j = loading
REFERENCES
Barth W. Stromungstechnische probleme der verfarhrens-
technik. Chem Ing Tech 20, No. 1:2932, 1954.
Blasius H. Grenzschichten in u ssigkeiten mit kleiner
reibung. Mitt Forschungsarb 131:140, 1913.
Cabrejos FJ, Klinzing GE. Pickup and saltation mechanisms
of solid particles in horizontal pneumatic transport.
Powder Tech 79:173186, 1994.
Dhodakpar SV, Klinzing GE. In: Roco, M, ed. Expert
Systems in Solids Handling. Butterworth-Heinemann,
1993, pp 743777.
Geldart D. Types of gas uidization. Powder Tech 7:285-292,
1973.
Institute of Gas Technology. Preparation of a coal conver-
sion system technical data book. Project 8979, ERDA
Contract No. EX 76-C-01-2286. October, 1978.
Ito H. Friction factors for turbulent ow in curved pipes.
Trans ASME J Basic Engr 81D:123, 1959.
Ito H. Pressure losses in smooth pipe bends. Trans ASME J
Basic Engr 82D:131, 1960.
Klinzing GE, Rohatgi ND, Myler CA, Dhodapkar SV,
Zaltash A. Pneumatic transport of solids in an inclined
pipe. Can J Chem Engr 67:237244, 1989.
Klinzing GE, Rizk F, Marcus RD, Leung LS. Pneumatic
Conveying of Solids. 2d ed. London: Chapman and
Hall, 1997.
Konno H, Saito SJ. Pneumatic conveying of solids through
straight pipes. Chem Eng Japan 2:211, 1969.
Koo EC, Drew TB, McAdams WH. Friction factors for
clean round pipes. Trans AIChE 28:56, 1932.
Matsen JM. Mechanism of choking and entrainment.
Powder Tech 32:2123, 1982.
Matsumoto S, Kikuta M, Maeda S. Eect of particle size on
the minimum transport velocity for horizontal pneumatic
conveying of solids. J Chem Eng Japan 10, No. 2:273,
1977.
Michaelides EE. Motion of particles in gases: average
velocity and pressure loss. Trans ASME 109:172178,
1987.
Richardson JF, Zaki WN. Sedimentation and uidization.
Part I. Trans Instn Chem Engrs 32:35, 1954.
Rizk F. Dr.-Ing. Dissertation, Technische Hochschule
Karlsruhe, 1973.
Schuchart P. Pressure drop across bend in pneumatic trans-
port. Chem Eng Technol 41:1251, 1969.
Weber M. Friction of the air and the air/solid mixture in
pneumatic conveying. Bulk Solids Handling 11:99, 1991.
Yang W-C. A mathematical denition of choking phenom-
ena and a mathematical model for predicting choking
velocity and choking voidage. AIChE J 21, No.
5:10131015, 1975.
Yang W-C. A unied theory on dilute pneumatic transport. J
Powder Bulk Solids Handling 1, No. 1:8995, 1977.
Zenz FA. Two-phase uid-solids ow. Ind Eng Chem
41:28012806, 1949.
24
Electrostatics in Pneumatic Conveying
George E. Klinzing
1 INTRODUCTION
Electrostatics or, more aptly, triboelectric eects in
solids processing is always encountered except in the
region of the world where the humidity of the ambient
is well over 75% most of the time. Electrostatic eects
are seen when one walks across a deep-piled rug during
winter and then reaches for a metal doorknob: a famil-
iar startling snap most likely occurs. Of course in most
cases such an incident does not trigger an explosion or
re. In solids processing, the danger of an explosion and
re is very serious, and such events have had a dramatic
eect in several industries. An internet article dealt with
the explosion that occurred in a our mill in the 1700s in
Italy (http://www.chemeng.ed.ac.uk/~emju49/SP2001/
webpage/intro/intro.html). The report given at the
time was quite detailed, and led the reader to the con-
clusion that electrostatics was the culprit.
Three requirements are needed for a dust explosion:
fuel, oxidizing agent, and ignition source. The ignition
source must provide a minimum spark energy for igni-
tion, a certain duration of the spark, and a spatial
requirement of the spark. It is particularly noteworthy
that often the carrier gas in pneumatic conveying is
changed to nitrogen in order to reduce the explosion
hazard for powder materials. The use of nitrogen will
increase the cost of operation. Sometimes the reduc-
tion of the percentage of oxygen is all that is needed to
reduce the hazard. For example, in the case of some
coals one can reduce the oxygen content of a gas
stream to 14% as opposed to the 21% normally
present in air. One must be careful in the transport
of material that has its own oxidizer present, such as
smokeless gunpowder. Using nitrogen as the conveyer
has no eect on reducing the hazard of pneumatically
conveying this material.
Another potentially dangerous situation in pneu-
matic conveying with electrostatics present exists
when there is a ammable vapor present in the gas
stream, possibly due to the polymer that is being con-
veyed. Likewise, when polymer-lined vessels are being
employed for material storage, the charge on the mate-
rial will not be dissipated through this lining, and
charge accumulation can build to dangerous levels.
One can liken the charging of materials to a series
similar to the electromotive series employed in chemi-
cal analysis. For example, one nds that glass and
nylon charge positively, while polyethylene and teon
charge negatively. In addition, one nds that bipolar
charging can occur with some materials as they are
conveyed pneumatically.
2 PNEUMATIC CONVEYING AND CHARGING
Although not all electrostatic discharges will cause an
explosion, the resulting damage to electronic equip-
ment, like the discomfort felt by operators who experi-
ence the discharge of electrostatics accumulated in a
conveying line, makes the elimination of all electro-
static buildup a must. In addition, it is well known
that electrostatic charging will increase the pressure
requirements for conveying a material (Marcus et al.,
1990). Estimating this additional energy requirement is
dicult.
Airborne dusts of most polymers are explosive, and
many of themare also electrically insulating and triboe-
lectrically active (Jones and King, 1991). Minor electro-
static discharges or drains are benecial in that they
help to dissipate accumulated charge in the system,
whereas discharges of large accumulated charges are
dangerous and can cause ignition. Of all the solids pro-
cessing operations, the most dangerous with respect to
the handling of ne powders is pneumatic conveying.
Jones (1994) has commented on a number of
electrostatic hazards in powder handling and cites pneu-
matic transport as being the process generating the most
charge with 10 to 10
2
C=kg of material owing.
The particles in this type of transport have thou-
sands of opportunities to interact with both themselves
and the surrounding piping and equipment.
Contacting between surfaces of dierent work func-
tions is the driver for electrostatic generation. The
work potential is an inherent property of the material
and the condition of its surface. Polymers can obtain a
wide range of charging from positive to negative. The
more the charges are separated in a conveying system,
the greater the potential to develop signicant voltages
and associated currents.
Investigators have studied the use of electrostatic
interactions or electrication of solid particles in
measuring various two-phase ow parameters in pneu-
matic transport, namely, the average particle velocity
and the mass ow rate (Gajewski et al.,1990; Gajewski
et al., 1991; Klinzing et al., 1987). The eect of elec-
trication on system pressure drop, choking, and salta-
tion velocities, ow patterns, pressure uctuations, and
particle velocity was also investigated in the past
(Myler et al., 1985; Ally and Klinzing, 1983, 1985;
Smeltzer et al., 1982a,b). It was found that at constant
loading, greater electrostatic eects were seen for small
particles over large particles because of the high parti-
cle number density and thus increased interactions for
the smaller particles. Bipolar charging was detected for
polyethylene powder with ne particles charging nega-
tively and the coarse particles positively (Cartwright et
al., 1985). The material electrication obtained experi-
mentally has also been compared to theoretical values
(Gajewski, 1989). Nifuku et al. (1989) found that the
electrostatic charge increases, peaks, and decreases
with increasing powder concentrations or transport
velocities; charge generation was greatly inuenced
by the powder ow pattern in the pipe.
Higher loadings at low relative humidities also pro-
duced increased electrostatic eects. Little or no
electrostatic charging was generated at ratios of kg of
water per kg of solids, which were greater than 0.1.
Smeltzer et al. (1982) and Nieh and Nguyen (1988),
experimenting with a 2 inch copper pipe and glass
beads, found that when the system moisture content
exceeded a relative humidity of 76%, the charge on the
glass beads became neutralized. The charging exhibited
itself as an increase energy requirement for ow, which
could under certain conditions be represented by an
increase in pressure drop of up to 70% (Ally and
Klinzing, 1983).
Boshung and Glor (1980) showed that the particle
size and the material are the two most important para-
meters associated with charging in pneumatic trans-
port systems.
In solids processing one should be on guard for ne
particles with a high specic surface area of non-
conductive material in an atmosphere of low relative
humidity owing in nonconductive or nongrounded
containments. These factors all add up to a trouble-
some situation.
Adhesion of particles to surfaces due to electro-
statics causes problems by coating piping walls. In
these cases the conveyed materials contact the wall
coating rather than the material of construction of
the pipe and can lead to charge buildup. In addi-
tion, coating of sensors by the adhesion of particles
by electrostatics will impede the process of measur-
ing properties of the gassolid ow systems. It is
well known that powder and granular materials
generate charging more readily than large diameter
particles.
Not all electrostatic charges should be viewed with
suspicion in the handling of powders. One can engineer
and manage the charging to provide avenues to mea-
sure eciently the solid ows and separate powders of
dierent properties.
3 USEFULNESS OF ELECTROSTATICS
After rst having to deal with the hazards of electro-
static generation in pneumatic conveying and their
adverse eects on electronics, these phenomena can
be channeled into a useful tool for measurement. The
charges generated in the conveying operation can be
measured at two distinct points, and these signals can
be cross-correlated to provide a measure of the particle
velocities. In fact, one can use a coil as an antenna
outside a nonconducting pipe to pick up the signals
generated (Myler, 1989). This provides a nonintrusive
technique for measuring the particle velocity. Another
phenomenon that is associated with electrostatics is the
magnetic eld that it generates associated with the
owing of charged particles or current in the pipe. A
polarized light is known to change the degree of polar-
ization as it goes through a magnetic eld. Channeling
a polarized light through a ber optic cable wrapped
around the conveying pipe will expose this light to the
magnetic eld, whose orientation can be easily mea-
sured. The larger the ow rate of solids, the greater
the magnetic eld generated. This concept has been
used to develop a ow meter in our laboratory
(Rader et al., 1995).
Masuda and Matsusaka (1994) have developed a
ow meter to measure powder ow rates using
electrostatic measurements. The method depends on
the impact of the particles with the inside of the tube
wall where a current is generated. The initial charge on
the particles can be eliminated by using two metallic
pipes whose inner surfaces are coated with dierent
materials. The pipe is 6 mm in diameter and carries
y ash. The powder ow rate and the mean particle-
charge value could be measured simultaneously.
4 PRECAUTIONS
Electrostatics and electronic equipment such as com-
puters and A/D converters do not mix. One should be
very careful, when a system is shown to have electro-
static tendencies and the measuring devices employed
have a computer linkage, that the computer should be
isolated from electrostatic charging eects. This isola-
tion is sometimes very challenging, since the charging
eects can travel on the surface of even nonconducting
surfaces to interfere or damage the electronics.
Changing the electric signal to a light signal is some-
times a solution to avoid large discharges interfering
with and damaging computer operations. Sometimes
such conversions will modify the original signal, so
knowledge of this modication is essential before
applying this procedure.
One of the most dangerous procedures when dealing
with electrostatic charging is covering a charged non-
conducting surface with a metal foil and shield. This
may appear as a possible solution to the grounding of
the charge generated, but in eect, what has happened
in applying the foil is that one is constructing a large
capacitor that can store energy and discharge this
energy over a very short period of time, causing or
triggering an explosion.
Another condition that is sometimes ignored until
problems occur is proper grounding of surfaces. If one
constructs a transport system from metal and if in
time, certain regions that would normally serve as a
grounding path become interrupted by paint or rust,
the system can become nonconducting in nature and
interfere with the ow of solids.
The use of nonconducting gaskets in transport line
connection also has the potential to create discharges.
One often places a conducting strap across the anges
to ensure continuity of current ow to a ground. It is
good practice in pneumatic conveying to ensure that all
vessels and units are grounded and that this grounded
path is not interrupted by corrosive or painted sur-
faces. In general it is good practice to pay attention
to proper housekeeping in the handling of powders
and granular material to reduce the potential for
explosive limits to be exceeded.
The most convenient method of eliminating electro-
statics in a system is by humidifying the conveying gas
and ensuring that all possible areas of charge buildup
are grounded. Another method is through the addition
of a small amount of antistatic agent to the solids
inventory. Antistatic agents will operate either to
reduce the generation of the charge or to increase the
charge leakage (Cross, 1987). Unfortunately, this addi-
tion may cause contamination of the material to be
conveyed, and it would have to be separated after the
collector to obtain the pure material again. A quater-
nary ammonium salt called Larostat
1
519 was used in
a 0.150.20 wt% mixture with the particles to be trans-
ported. The low percentage did not interfere with the
characteristics of the gassolid ows. Experimental
tests conrmed and veried this result. The quaternary
ammonium salt is a nonpolymeric cationic compound
and is considered a hydrophilic additive. This additive
increases the surface conductivity by increasing the
water adsorption at low relative humidities. Dahn
(1992) evaluated possible sources of dust explosions
in large-scale conveying systems and recommended
ways to minimize them.
More recently the waxing of the interior walls of a
nonconductive pipe with a conductive wax similar to
that employed in clean room maintenance has proved
eective in reducing charging potentials.
A very practical book on electrostatic eects has
been prepared by Jones and King (1991). This work
gives many procedures for reducing electrostatics in a
plant.
5 SOME EXPERIMENTAL OBSERVATIONS
OF ELECTROSTATICS IN PNEUMATIC
CONVEYING
In order to portray the behavior of electrostatics and
their eects on pneumatic conveying, some experiments
conducted in our laboratories will be presented. These
results will give the avor of the behavior of this phe-
nomenon and the challenges in dealing with such a
system.
5.1 Horizontal System
The conveying air employed was conditioned for
humidity to study the conveying characteristics in the
absence of electrostatic charging. The pressure drop
versus gas velocity plot shows no unusual behavior.
However, in the absence of humidication, the convey-
ing characteristics of PVC particles (137m in dia-
meter) are entirely dierent. The pressure drop
increases by about 100% as the particles start adhering
to the wall. Pressure drop is a measure of power con-
sumption in pneumatic conveying systems and there-
fore a prime concern. The undesired accumulation of
particles inside the pipe also leads to unsteady ow
conditions. It should be noted that the conveying
line, except for the glass viewing section, is made of
copper pipe that has been well grounded. It appears
that the charged particles do not dissipate the charge
completely to the grounded pipe during the short dura-
tion of particlewall collision.
The nature of interparticle forces changes owing to
electrostatic charging. These particles become more
cohesive and coat the inner surface of the pipe. It
can be surmised that the charge on these particles
must be renewed constantly by other particles. If the
solids were shut o during conveying, or both gas and
solids were stopped, the coating of particles on the
sight-glass disappeared. No visual observations could
be made about the existence of particles on the inner
surface of the copper tube. It is unlikely that the par-
ticles could be getting charged because of the glass
section, since the deposition on the wall could be
seen on the entire length of the glass section.
The electrostatic charge on the particles changed the
ow patterns usually observed under similar condi-
tions. Figure 1 (Dhodapkar, 1991) shows the range
of ow patterns observed under conditions of low
humidity (less than 65%). From the experimental
data it may be concluded that the appearance and dis-
appearance of these ow patterns are independent of
the solids ow rate and depend mainly on the gas
velocity.
From Fig. 1 it can be seen that at low gas velocities
(Pattern 5) and large gas velocities (Pattern 1), the pipe
essentially remains clean at the top. At lower gas velo-
cities a settled layer may be seen that is quite expected.
At high velocities a homogeneous ow is observed with
no particles sticking to the wall. A possible explanation
is that the particles are at equilibrium owing to electro-
static forces and uid shear forces. It was found that at
gas velocities higher than 18 m/s, the uid shear is
greater than the electrostatic forces and therefore no
particles can be seen sticking to the wall. In addition,
at gas velocities less than 4 m/s, no particles can be
seen adhering to the wall because of lack of electro-
static charging.
At intermediate velocities one may observe numer-
ous ring-shaped ow patterns that appear to move in
the direction opposite to the direction of the main ow.
In all these cases the particles coat the inner surface of
Figure 1 Flow patterns for PVC with electrostatic charging.
the wall, defying gravity. Comparison of Figs. 1 and 2
(Klinzing et al., 1994) indicates that the unique maxima
observed in Fig. 2 correspond to the formation of ring-
shaped structures on the pipe wall. The intensity of
pressure uctuations were also found to have a maxi-
mum (see Fig. 3) at a supercial gas velocity of 9.3 m/s.
This is a remarkable departure fromthe usual behavior.
5.2 Vertical System
Electrostatics eects were experienced several times in
the vertical system until the addition of multiple
grounding points and Larostat
1
519. Electrostatics
eects were generated by the high number of particle-
particle and particlewall interactions between the
0.0254 m Lucite tube and the glass particles. The nitro-
gen gas used as the transport medium had zero relative
humidity. Prior to eliminating the electrostatics, inter-
esting data was acquired for 97m and 545 m glass
beads.
5.3 Electrostatics with Small Glass Beads
Initially, tests were conducted on 97 m glass beads at
atmospheric pressure with the results of pressure drop
versus gas velocity showing the classical Zenz-type dia-
gram. However, upon repeating the set of experiments,
electrostatics developed in the system. Electrostatic
discharges were evident in the system by visual and
audible observations. The classical state diagram,
which was found in the rst set of experiments, was
altered and a double minimum was found. Figures 4
and 5 show the results of the eects of electrostatics on
the pressure drop. The minimum P=L for the second
Figure 2 Eect of electrostatics charging on the pressure drop behavior of PVC particles in 0.0254 m (I.D.) system
W
s
1:14 kg=min.
set of experiments is shifted to the right, thereby
changing the previous minimum in the gas velocity
that occurred without electrostatics. Operating at
what was thought of as the previous minimum would
have increased the pressure drop signicantly, which
could result in reduced conveying capacity. Three dif-
ferent solids ow rates were investigated at several gas
velocities, with the higher solids ow rates having the
most pronounced results.
The solids were mainly in a homogeneous transport
state during the initial tests, except at the lowest gas
velocity. During the second series of tests, the ow
patterns changed from a homogeneous ow at the
start of the experiments at low gas velocities to a
more clustered or packet ow at the intermediate test
velocities. The particles formed clusters for gas veloci-
ties in the range of 4.56.0 m/s for the particular set of
conditions investigated.
The rst minimum occurring at the lower gas velo-
city of the double minimum increased with increasing
solids ow rate (Figs. 4 and 5). This increase in pres-
sure drop, as the gas velocity was in the range of 4.5
7.6 m/s, occurred owing to the clustering of the part-
icles caused by the electrostatic generation of the glass
beads with the Lucite tubing. This clustering caused
the pressure drop in the system to be equivalent to
that of larger particles. The greater the charging of
the particles, the larger and more dened the clusters
were. This was also evident by the tracking of the
P=L vs. time for a given U
g
and W
s
. The pressure
drop would continue to increase until a discharge, of
the system, as evident by sight and sound, occurs. At
this point the pressure drop would fall back to a cer-
tain level and continue to increase again with time until
there is another discharge. A certain residual charge
resided in the particles and on the pipe.
Figure 3 Eect of electrostatics charging on the standard deviation of the pressure drop for PVC particles in 0.0254 m (I.D.)
system W
s
1:14 kg=min:
Operating at a gas velocity for what seemed to be a
minimum for the pressure drop in the presence of
electrostatics could result in the pressure drop being
increased substantially. This would lead to undersized
equipment for the transport of the material and would
result in less material transported than the system was
actually designed for.
The double minimum condition has also been
reported by Zaltash (1987) and Klinzing et al. (1986).
Klinzing suggests that the smaller particles behave as a
cluster at the dilute-phase regime, and the clusters tend
to break up as the choking regime is approached,
resulting in the particles behaving as smaller particles
and thus causing an oscillation in the pressure drop.
This described ow phenomenon was observed in the
experiments. Figure 6 shows the ow patterns observed
with electrostatic charging.
5.4 Electrostatics with Larger Glass Beads
Electrostatics was also observed when 545 m glass
beads were used as the transport solid. This time the
pressure drop was not observed or investigated as a
function of gas velocity, but rather pressure uctua-
tions were observed with time. Electrostatic discharges
were observed with arcs discharging from a few centi-
meters to a meter in length accompanied by a powerful
snapping noise.
The pressure drop was seen to decrease suddenly
upon discharge, rise sharply prior to another discharge
Figure 4 Pressure drop for 800 g/min with and without electrostatics present.
of the solids, and fall back to the original level
(pressure drop) after discharge. Table 1 shows a cyclic
example of this for the 545 m glass beads at a gas
velocity of 6.6 m/s and a solids feed rate of 1100 g/
min. Standard deviation of the pressure drop was also
found to increase with the increased charging of the
system.
The rise and fall of the pressure signal is similar to
that of a voltage vs. time plot observed during the
charging and discharging of a capacitor (Floyd,
1987). Figure 7 shows the similarity of the curves.
In eect, the transport of the solids causes the parti-
cles to become charged, and the charge accumulates
on the solids until the solids reach their dielectric
breakdown and discharge across an air (or nitrogen)
gap. The collector could be considered a large capa-
citor, since the solids were isolated from any conduct-
ing medium due to the Lucite walls, and therefore it
contained a charge. The pressure drop decreases just
prior to the charging portion of the curve. Since the
particles are large (545 m) in comparison with the 9
7 m glass beads, it is hypothesized that the particles
are weakly held together or attracted by electrostatic
charges, causing a higher pressure drop than if they
were individual entities. The dip in the pressure drop
curve prior to the charging and discharging portion of
the pressure drop curve could be due to the particles
separating (behaving as individual particles) and pos-
sibly losing some of the charge. Then upon losing the
charge or becoming separated, there is an increasing
force that causes the particles to reattract or cluster
together.
Figure 5 Pressure drop for 400 g/min with and without electrostatics present.
5.5 Summary of Testing
Several interesting observations were made and
recorded for two dierent conveying systems that
were electrostatically charged. The usual Zenz-type
diagram is modied to show two minima in the curve
when the systems are charged, rather than a single
minimum. Flow patterns will also change from their
usual appearance and become more wavy or take the
form of rings for horizontal conveying, and form clus-
ters in the case of vertical conveying.
By observing the pressure uctuations (for the ver-
tical system) of a charged system, the pressure drop
versus time curves exhibit the appearance of a capa-
citor charging and discharging. Visual observations in
the vertical column showed that the ow patterns of
the particles become clustered until a discharge
occurred in the system, from which the particles
regain their individuality and the ow becomes
more uniform again. Particles appeared to cluster in
the gas range (4.57.6 m/s) for vertical ow. The par-
ticle packets or clusters tend to break up as the chok-
ing regime is approached, so that the particles achieve
their true identity. As the higher gas velocity is
approached, the particle clusters tend to break up
owing to the turbulence of the ow. This behavior
causes the pressure drop to oscillate with gas velocity
over this range.
NOMENCLATURE
t
i
= Time, s
U
g
= Gas Velocity, m/s
U
t
= Terminal velocity of particle, m/s
P = Dierential pressure drop, Pa
P=L = Pressure drop per unit length, Pa/m
= Standard deviation, Pa
Figure 6 Particle ow patterns observed for vertical ow with electrostatic charging.
Table 1 Transient of the Pressure Drop
P, Pa Time
Initial reading (taken arbitrarily during
the charging cycle
610
701
720
t
1
t
2
t
3
discharge occurred in system
Readings after discharge 595 t
4
Times between readings, i.e., t
i1
t
i
t, are approximately the same
REFERENCES
Ally MR, Klinzing GE. Electrostatic eects in gas-solid
pneumatic transport with loadings to 100. J Powder and
Bulk Solids Tech 7(3):1320, 1983.
Ally MR, Klinzing GE. Inter-relation of electrostatic
charging and pressure drops in pneumatic transport.
Powder Tech 44(1):8588, 1985.
Boshung P, Glor M. Methods for investigating the electro-
static behavior of powders. J Electrostatics 8:22052219,
1980.
Cartwright P, Singh S, Bailey AG, Rose LJ. Electrostatic
charging characteristics of polyethylene powder during
pneumatic conveying. IEEE Transactions on Industry
Applications IA21(2):541546, 1985.
Cross JA. Electrostatics: Principles, Problems and
Applications. Bristol, U.K.: IOP, 1987, pp 376382.
Dahn JC. Electrostatic hazards of pneumatic conveying
of powders. Plant/Operations Progress 11(3):201204,
1992.
Dhodapkar SV. Flow pattern classication in gassolid
suspensions. Ph.D. diss., University of Pittsburgh,
Pittsburgh, PA, 1991.
Floyd TL. Electronics Fundamentals: Circuits, Devices
and Applications. Columbus, OH: Merrill, 1987,
pp 334339.
Gajewski A. Measuring the charging tendency of polystyrene
particles in pneumatic conveyance. Journal of
Electrostatics 23:5566, 1989.
Gajewski JB, Glod B, Kala W. Electrostatic method for
measuring the two-phase pipe ow parameters. IEEE
Industry Applications Society Annual Meeting, Seattle,
WA, Oct. 1990.
Gajewski JB, Glod B, Kala W. Electrostatic ow meter for
measuring the two-phase ow parameters in pneumatic
transport. Results of preliminary tests. Proceedings of
the 8th International Conference on Electrostatics,
Oxford, U.K., 1991, pp 159164.
Jones TB. NEPTIS-3. Electrostatic Problems in Powder
Technology. Nov. Kyoto: Nisshin Engr, 1994, p 79.
Jones TB, King JL. Powder Handling and Electrostatics:
Understanding and Preventing Hazards. Chelsa, MI:
Lewis, 1991, pp 1103.
Klinzing GE, Rohatgi ND, Zaltash A, Myler CA. Pneumatic
transport: a review and generalized phase diagram
approach to pneumatic transport. XIV Encontro Sobre
Escoamento Em Meios Porosos, Octubro 1986, p 14.
Klinzing GE, Zaltash A, Myler C. Particle velocity measure-
ments through electrostatic eld uctuations using
external probes. Particulate Science and Technology
5(1):95104, 1987.
Klinzing GE, Dhodapkar SV, Plasynski SI. NEPTIS-3.
Electrostatic Measurement of Powder Flow Rate in
GasSolids Pipe Flow. Nov. Nisshin Engr, 1994, p 21.
Marcus RD, Leung LS, Klinzing GE, Rizk F. Pneumatic
Conveying of Solids: A Theoretical and Practical
Approach. London, U.K.: Chapman and Hall, 1990,
p 67.
Masuda H, Matsusaka S. NEPTIS-3. Electrostatic
Problems in Powder Technology. Nov. Nisshin Engr,
1994, p 30.
Myler CA. Use of a thermodynamic analogy for pneumatic
transport in horizontal pipes. Ph.D. diss., University of
Pittsburgh, Pittsburgh, PA, 1989.
Myler CA, Zaltash A, Klinzing GE. Gas-solid transport in
a 0.0508-m pipe at various inclinations with and without
electrostatics, Part I, II. J Powder and Bulk Solids 10:1
17, 1986.
Nieh S, Nguyen T. Eects of humidity, conveying velocity,
and particle size on electrostatic charges of glass beads in
a gaseous suspension ow. Journal of Electrostatics
21:99114, 1988.
Figure 7 Pressure drop time series trace.
Nifuku M, Ishikawa T, Sasaki, T. Static electrication
phenomena in pneumatic transportation of coal. Journal
of Electrostatics 23:4554, 1989.
Rader J, Prakash A, Klinzing GE. AIChE Symposium Series
on Fluidization 91(308):154163, 1995.
Smeltzer EE, Weaver ML, Klinzing GE. Individual electro-
static particle interaction in pneumatic transport. Powder
Tech 33(1):3142, 1982.
Smeltzer EE, Weaver ML, Klinzing GE. Pressure drop losses
due to electrostatic generation in pneumatic transport.
Industrial and Engineering Chemistry, Process Design
and Development 21(3):390394, 1982b.
Zaltash A. Application of thermodynamic approach to
pneumatic transport at pipe orientations above the
horizontal. Ph.D. diss., University of Pittsburgh,
Pittsburgh, PA, 1987.
25
Instrumentation and Measurements
Masayuki Horio, Rafal P. Kobylecki
+
, and Mayumi Tsukada
Tokyo University of Agriculture and Technology, Tokyo, Japan
1 INTRODUCTION
Instrumentation is an important issue in uidization
engineering, particularly because of the multiphase,
dynamic, and nonlinear nature of uidized beds. In
commercial plants, instrumentation should be con-
ducted basically for process control, but detection of
unusual behavior and prevention of unwanted losses
are important as well. For bench and pilot plants
instrumentation should be dierent compared to that
for commercial plants, since maximum information
output should be aimed at for safe design and scale-up.
After over 60 years of research and experience on
uidization since the rst FCC erection, there is an
accumulation of a variety of detection and data eva-
luation methods. At the same time, recent progress in
information technologies and in sensor technologies
give us more possibilities of innovation in instrumenta-
tion and monitoring. New instrumentation techniques
are required also by the tighter economic demands for
higher eciency and lower risk as well as by new appli-
cations. Depending on the process conditions, we have
to evaluate the appropriate detection principles and
data processing logics. Figure 1 shows the variety of
measurement demands in uidization engineering,
including both research and development (R&D) and
commercial operations. In R&D we have much wider
choices, including very expensive and sophisticated
systems. As is often said, and as is shown in Fig. 1,
in the scale-up chemical engineers are encouraged to
design their full-scale concept rst to nd the risky
factors in their design. Then they should design scale-
down facilities and measurement systems to obtain
data for sure scale-up. In the beginning they have to
think small in instrumentation, but along with the size
increase their instrumentation concepts also have to be
scaled up. Figure 1 also shows that even in large-scale
or commercial stages, it is recommended to keep
laboratory facilities for plant backup, particularly for
particle characterization.
The following are the items to be monitored or mea-
sured either in commercial processes or in experimental
test rigs: (1) temperature, (2) gas pressure, (3) voidage/
bed density/bed height/solids inventory, (4) solids mass
ow rate/solids mass ux/solids velocity, (5) gas ow
rate/gas velocity, (6) chemical composition of gas and
solids, (7) particle characterization/particle size distri-
bution (PSD), (8) bed structure and ow regime char-
acterization, including bubble and cluster size. Items 1
and 2 are the most fundamental in process monitor-
ing. Table 1 shows the interrelationships among sen-
sors and parameters to monitor a uidized bed.
Among all the sensors and probes, pressure sensors
can provide the most basic information not only on
pressure itself but also on voidage, bed height, and
ow regime or quality of uidization. Pressure moni-
toring is essential in most cases to maintain an appro-
priate loop pressure pattern or a macroscopic pressure
+
Current aliation: Czestochowa Technical University,
Czestochowa, Poland.
distribution in the plant where particles are usually
circulated between columns pneumatically or by grav-
ity ow. In many cases the reaction conditions of each
column are quite dierent. The gas ow for each col-
umn is usually separated by the so-called material seal-
ing, where pressure balance takes a major role.
Additionally, the state of solids and their distribution
in the column can be detected from pressure gradient
and pressure uctuation. As long as the solids are ui-
dized, their weight is suspended by uid, and the solids
holdup can be detected from the static pressure gradi-
ent. In many cases, when a bed is deuidized (e.g., by
sintering), the bed pressure drop suddenly decreases.
On the other hand, the pressure uctuations indicate
good or poor bubbling as well as the ow regime tran-
sitions.
To obtain information of particular interest in the
relevant column, it is required to choose appropriate
sensing as well as data processing systems and to orga-
nize proper measurements to obtain necessary infor-
mation with a high accuracy. Accordingly, in the
present chapter it is intended to provide information
to engineers, rst on available sensors and probes as
well as relevant data necessary to use them, second on
data processing methods, and third on experimental
methods, particularly for uidization measurements
and monitoring. In Sec. 2, a variety of sensors, probes,
and tracers are introduced. In Sec. 3, some information
on data processing and visual imaging is provided. In
Sec. 4, uidized bed diagnostic techniques are
reviewed.
For further information readers can see
Cheremisino and Cheremisino (1984),
Cheremisino (1986), Grace and Baeyens (1986),
which all concentrate on bubbling bed measurements,
and Werther et al. (1990, 1993), Soo et al. (1994), Yates
and Simons (1994), Nieuwland et al. (1996), Louge
(1997), Chen (1999) or Werther (1999), although
many of them are rather limited in scope.
2 DETECTION PRINCIPLES AND
DETECTORS
2.1 Temperature Sensors/Probes
2.1.1 Thermocouples
Thermocouples are the most common temperature
measurement devices. A detailed description of
thermocouples and other temperature measuring
instruments and sensors can be found in Holman
(1994) or Stephenson et al. (1999). Maximum tempera-
ture limit, up to which a thermocouple can be used
without damage, depends strongly on its diameter
and the surrounding environment. The thicker the
thermocouple, the higher the temperature application
limit, but on the other hand the longer the response
time.
2.1.2 Optical Pyrometer
The optical pyrometer is a noninvasive temperature
measurement device utilizing Plancks law to obtain
temperature, T[K], from radiant energy ux,
E(z)[W], of a particular wavelength z:
Figure 1 Measurement demands for scale-up and process improvement.
Table 1 Interrelationships Among Measurement Techniques or Sensors and Process Parameters of Fluidized Beds Applicable for Bench Scale
(B), Pilot Scale (P), and Commercial Scale Tests (C). Case C also Includes Emergency Measurements
Flow regime
& structure
Particle
characteristics Pressure Temperature Voidage
Gas
Velocity
Solids
velocity
Solids
mass
ux
Gas/solids
composition
Pressure sensors B, P, C B, P, C B, P, C B, P, C B, P, C
Thermocouples B, P, C B, P, C
Pyrometers B, P, C
Suction probes B, P B, P, C B, P B, P, C, B, P, C
Gs meters B, P, C B, P B, P, C
Capacitance probes B, P B, P B, P
Optical ber probe B, P, C B, P, C B, P, C B, P B, P, C B, P B, P
Laser sheeting B B B B
LDV/PIV B, P B, P B, P B, P B, P B, P
Tomography B B B, P
Solid/gas tracers B B B B, P B, P
Camera observation B, P, C B, P, C B, P, C
(IR camera)
B, P, C B, P
Gas sampling & injection B, P, C B, P B, P, C
E(z) =
C
1
c(z)
z
5
e
C
2
,zT
1
_ _ (1)
where c(z) is emissivity of an object,
C
1
= 3.74 10
16
Wm
2
, and C
2
= 0.0114 mK.
For shorter wavelengths and for lower tempera-
tures, the above equation is simplied to the formula
called Wiens law:
E(z) =
C
1
c(z)
z
5
exp
C
2
zT
_ _
(2)
The schematic of an optical pyrometer is shown in
Fig. 2. The radiation of a target body, whose tempera-
ture is measured, is viewed through the lens and lters.
The aim of the lters is to absorb the radiation and
reduce its intensity. A lamp is placed in the optical path
of the incoming radiation. By adjusting the lamp cur-
rent, the color of the lament is changed so that it
matches that of the incoming radiation. Temperature
calibration is made in terms of the lamp heating cur-
rent. Additionally, the lter is installed in the eyepiece
to ensure that comparisons are made for essentially
monochromatic radiation, thus eliminating uncertain-
ties resulting from variation of radiation properties
with wavelength. Based on a similar principle, Ross
et al. (1981) determined the burning char temperature
by comparing its monochromatic photograph with
that of a calibrated lament.
To avoid the uncertainties of temperature measure-
ment associated with the uncertainty of emissivity, a
two-color pyrometer is recommended, which analyzes
the body radiation properties for two wavelengths. In
the two-color or ratio pyrometer the radiation is
detected at two separate wavelengths for which emis-
sivities of the surface can be considered nearly the
same. Thus the ratio of the thermal radiation sensor
output calculated for two wavelengths by Wiens law is
obtained as
X
E
1
E
2
=
c(z
1
)z
5
1
e
C
2
,z
1
T
c(z
2
)z
5
2
e
C
2
,z
2
T
(3)
Since c(z
1
) - c(z
2
), Eq. (3) can be rewritten and solved
for T as
T - C
2
1
z
2
1
z
1
_ _
ln X
z
5
1
z
5
2
_ _
1
(4)
2.1.3 Infrared Camera/Thermometer
Another noninvasive tool to measure temperature is
the infrared (IR) thermometer, which usually has an
optical response time below 0.1 s. The measurement
principle is based on determination of the thermal
radiation Q of a target body. The radiation is propor-
tional to the temperature of the body according to the
StefanBoltzmann law
Q =
_
E(z) dz = coT
4
(5)
where o is the StefanBoltzmann constant,
o = 5.67 10
8
W,m
2
K
4
.
MgF
2
, ZnSe, and Sapphire (Al
2
O
3
) glasses are used
as IR transparent media windows, e.g., MgF
2
: wave-
length 35.4 jm, 95% of IR passing through; ZnSe:
0.520 jm, 90%; Al
2
O
3
: 0.24 jm, 85% transparency
for IR.
Since the IR thermometer must adapt itself to the
temperature of the surroundings, it may not measure
the temperature accurately without enough time to
adapt. The measurement results are also aected by
the angle of optical axis to the surface as shown in
Fig. 3. When the angle is larger than 40
, an IR
camera placed at position A gives temperature values
higher than the real ones.
IR cameras have been applied to so-called inverse
heat transfer problems, which concern the identica-
tion of unknown temperature distribution or thermal
Figure 2 Schematic of the optical pyrometer.
resistance distribution inside a body from the observed
surface temperature distribution (Pasquetti and Le
Niliot, 1991; Le Niliot and Gallet, 1998).
2.2 Heat Flux Probes
The overall heat ux in uidized beds consists of a
radiative and a convective component. Assuming, for
simplicity, that the probe surface and the suspension
ow are gray, parallel, and separated by an opaque
gas, one may express the total bed-to-wall heat ux as
q
total
= q
rad
q
conv
=
o(T
4
bed
T
4
wall
)
1,c
bed
1,c
wall
1 ( )
h T
bed
T
wall
( )
(6)
where h is the heat transfer coecient, T is tempera-
ture, c is emissivity, o is the StefanBoltzmann con-
stant, and q is the heat ux.
Since heat transfer in uidized beds is dominated by
surface renewal of bed materials, the bed-to-wall heat
transfer coecient for a vertical surface is much larger
at its leading edge. Accordingly, the heat transfer coef-
cient determined by small probes can be abnormally
large. Furthermore, depending on gas and solids
motion, surface renewal rate, and the time fraction of
coverage by bed material, the heat transfer coecient
varies for dierent segments of the surface, particularly
for horizontal tubes. The requirement to determine the
instantaneous heat transfer coecient is the low ther-
mal inertia of the probe, such as that used e.g., by
Mickley et al. (1961) or Ma and Zhu (1999) and
shown in Fig. 4ab, respectively.
In some experiments it is necessary to separate the
radiative and convective heat uxes. This can be done
by using two probes of dierent emissivities (e.g.,
Botterill et al., 1984) or by separating the convective
component from the total heat ux by using quartz
windows (Basu and Knouche, 1988) or ZnSe windows
(Ozhaynak et al., 1984). Quartz windows have rather
narrow transmittance bands for infrared radiation
(roughly 0.153.5 jm wavelengths compared to 0.520
jm for ZnSe) but are resistant to thermal shocks. To
measure simultaneously the radiative and total heat
ux, Luan et al. (1999) proposed the multifunctional
probe shown in Fig. 4c. They determined the radiative
component of heat ux both by using a ZnSe window
to separate the two uxes and by estimating the radia-
tive ux from comparison of the measurement from
two probes of dierent emissivities. More on heat ux
measurement can be found in Mickley and Fairbanks
(1955), Botterill (1975), Mathur and Saxena (1987),
Leckner and Andersson (1992), Goedicke and Reh
(1992), Werdermann and Werther (1993), Couturier
et al. (1993), and Molerus and Wirth (1997).
2.3 Gas/Solid Pressure and Force Sensors/Probes
In most cases the pressure is detected using diaphragm
sensors classied in terms of sensing element as metal
strain gauge, piezoelectric semiconductor, electric
capacitance, reluctance, and LVDT (linear variable-
dierential transformers) sensors. The structure of a
silicon semiconductor sensor, where the pressure is
detected by a silicon diaphragm, is shown in Fig. 5.
In most cases these sensors can be used at temperatures
lower than 70
C.
2.3.1 Gas Pressure
Gas pressure sensing is the most basic issue in uidiza-
tion and solids transport. The key points of the sensing
system are the volumetric capacity of its tubing and the
Figure 3 Ball surface temperature versus observation angle, d
ball
= 20 mm. (a) Schematic of measurement; (b) horizontal
temperature distribution of the equator; (c) vertical temperature distribution. (Kobylecki and Horio, 2000.)
ability to prevent the tube from solids plugging. Figure
6 shows a simplied pressure measurement system con-
sisting of a leading tube of diameter d
pp
, length L and
volume V at the end where the pressure transducer is
placed.
If laminar friction resistance is assumed for the ow
in the tube, the pressureamplitude ratio is given by
(Holman, 1994)
p
p
0
=
1
1 o,o
n
( )
2
_ _
2
4h
2
o,o
n
( )
2
_ _
0.5
(7)
where h is the damping ratio p is the amplitude of
pressure signal impressed on the sensor, p
0
is the pres-
sure at the tube inlet, o is frequency, and o
n
is the
resonance frequency.
The resonance frequency, o
n
, damping ratio, h, and
phase angle of the signal, , are given by
o
n
=
3r
2
c
2
4LV
_
(8)
h =
2j
,
f
cr
3
3LV
_
(9)
Figure 4 Examples of heat transfer probes. (a) Low thermal capacity instantaneous type of Mickley et al. (1961); (b) miniature
heat transfer probe of Ma and Zhu (1999); (c) probe to measure radiative and total heat ux. (Luan et al., 1999.); (d) radiative
heat ux vs. suspension density. (Luan et al. 1999.) [(a): Reproduced with permission of the American Institute of Chemical
Engineers. Copyright # 1986 AIChE. All right reserved. (b)(d): with permission of Elsevier Science]
= tan
1
2h(o,o
n
)
1 (o,o
n
)
2
_ _
(10)
where r = d
pp
,2 is the tube radius, c is sound velocity
in the uid, and j and ,
f
are the viscosity and density
of the uid, respectively.
Figure 7 shows examples of pressure uctuations for
two tubes of various diameters. If the tube diameter is
too narrow, the probe gives quite poor response.
2.3.2 Manometer
Manometers are classical and still useful gas pressure
measurement devices used also to calibrate semi-
conductor pressure sensors. Applying the manometers
as pressure measurement devices, caution should be
taken to prevent the overowing as well as the
spontaneous oscillation of the liquids. To obtain
pressure distribution in one glance, the manometers
can be assembled and connected to a common-
reservoir.
2.3.3 Solid Pressure
To measure the particle pressure separately from the
gas pressure, the chamber of the sensor covered by the
diaphragm should have one or more bypass tubings to
make the gas pressure inside equal to the outside, so
Figure 5 Silicon semiconductor piezo-resistive pressure sen-
sor. (Chau et al., 1999, with permission of CRC Press.)
Figure 6 Schematic of a pressure transmitting system.
(Holman, 1994.)
Figure 7 Eect of tube diameter, d
pp
, on pressure uctuations. Particles: 68 jm FCC, purge gas velocity: 1 m/s, pressure port
length: 2 m dead volume, 450 mm
3
. (Xie and Geldart, 1997.) (With permission of Elsevier Science)
that the diaphragm measures only the pressure from
solids. The most important design issues are the size of
the probe and its dynamic response in the circumstance
of uctuating gas pressure. Depending on the volume
inside the diaphragm and the gas ow resistance in the
connection tubings, there can be substantial deviation
of the dynamic solid pressure data, both in amplitude
and phase, as is shown in Fig. 8. Fortunately, the
major frequency of gas pressure uctuation in uidized
beds is up to 10 Hz (even in turbulent conditions), and
it should not be dicult to adjust the probe character-
istics at the proper range.
2.3.4 Solid Force Sensors
To determine the bulk solid ow in bubbling ui-
dized beds, force sensors have been used in lab
scale and pilot scale units. Examples can be found
in Miyauchi and Morooka (1969), Tomita and
Adachi (1973), Mann and Crosby (1977), Raso et
al. (1983), Fasching and Smith (1993), and Ito et
al. (1999).
2.3.5 Fluidized Bed Viscometer
Figure 9 shows two typical probes to measure viscos-
ity: a rotating drum and a rotating bar. It shows also
some measurement results obtained with using a rotat-
ing bar. In case the internals are heavily placed in the
bed, the solid-to-internal friction may aect solids cir-
culation. In such cases, solid viscosity also has to be
taken into account in the scaling experiments, as dis-
cussed e.g., by Naruse et al. (1996).
2.4 Light Sources and Optical Sensors
2.4.1 Fundamentals of Light Measurement
The main parameters to describe light-related phenom-
ena are classied as radiant quantities and luminous
quantities; the latter are psychophysical parameters.
Table 2 shows their denitions and units.
Total optical energy input
i
consists of three com-
ponents: reection
r
, transmission
t
, and
a
adsorption:
i
=
r
a
(11)
The commonly used parameters reectance and trans-
mittance are dened as
Reflectance =
i
(12)
Transmittance =
i
(13)
Light emitted to a suspension is scattered according
either to Mies or to Rayleighs rule. Mie scattering
occurs for particles much larger than the wavelength.
In Mie scattering, the spectra of scattered and emitted
light are similar. Rayleigh scattering occurs for part-
icles much smaller than the wavelength, and the spec-
trum of scattered light shifts to the blue side. The
directions of scattered light are dierent for these
two scatterings, as shown in Fig. 10.
Owing to the light scatter, the relative decrease of
light transmission through the suspension is propor-
tional to the suspension volumetric concentration, e
p
,
and the travel path length, dl. Thus for the intensity of
light, I, one can write
Figure 8 Frequency response of a particle pressure sensor to the gas pressure change. (a) Setup to test the dynamic response; (b)
gas pressure (measured by a pressure transducer) and the phase shift. (Campbell and Wang, 1990.) (With permission of Institute
of Physics)
dI
dl
= oIc
p
(14)
For a dilute and monodisperse suspensions, Eq. (14) is
integrated and the solution, known as the Lambert-
Beer law, is obtained as
I
I
0
= exp oc
p
l
_ _
(15)
where o is the absorbance coecient and I
0
is the
intensity of the light source. Determining o by calibra-
tion we can calculate c
p
from the observed I,I
0
.
2.4.2 Light Sources
Light sources that have been applied to uidized bed
measurement are lamps (halogen, mercury, metal
halide), LEDs (light emitting diodes), and lasers of
Figure 9 Viscosity measurements in uidized beds. (a) Viscosity measurement with rotating drum. (Schugerl et al., 1961.) (With
permission of Elsevier Science) (b) T-type rotating bar to measure viscosity. (Naruse et al., 1996.) (c) Viscosity measured with a
bar viscosimeter. (Naruse et al., 1996.) D
t
= 105 mm; bed material: SiO
2
; d
p
= 582 jm and 150 jm.
Table 2 Radiant and Luminous Quantities
Radiant quantities Luminous quantities
Parameter Denition Unit Parameter Denition Unit
Radiant
energy
Q
e
Total energy emitted by a
body (as electro-magnetic
wave or particle)
J Quantity of
light
Q
v
_

v
dt lm s
Radiant
ux
e
dQ
e
,dt W Luminous
ux
v
K
m
_

ez
(z)V(z)dz lm = cd sr
Radiant
exitance
M
e
d
e
,dA W,m
2
Luminous
exitance
M
v
d
v
,dA lm,m
2
Irradiance E
e
d
e
,dA W,m
2
Illuminance E
v
d
v
,dA lx = lm,m
2
Radiant I
e
d
e
,d W,sr Luminous
intensity
I
v
d
v
,d cd
Radiance L
e
d
2
e
dAdcos
W,(m
2
sr) Luminance L
v
d
2
v
dAdcos
cd,m
2
dA is innitesimally small area, d is an innitesimally small solid angle, is the angle between a normal direction to the surface
and a given direction, and V(z) is the spectral luminous efciency.
various types (gas, solid, semiconductor) as listed in
Table 3.
The ranges of output wavelengths of major com-
mercial LEDs and lasers applied to uidized bed mea-
surement are shown in Table 4. An example of an
emission spectrum of an LED in comparison with
those of various lamps is presented in Fig. 11. The
most commonly used gas lasers are He-Ne, CO
2
, and
Ar ion lasers. In the case of solid-state lasers the light is
emitted by atoms, which are xed within a glassy mate-
rial or crystal (e.g., Nd YAG laser, where the laser
medium is an yttrium-aluminum-garnet matrix with
trivalent neodymium ions present as impurities).
Semiconductor lasers made usually of AlGaInP,
GaAlAs, InGaAsP, or lead salt materials use the spe-
cial properties of a p-n junction. More precise data can
be found e.g., in Weber (1999).
If the phase matching condition is satised, the
OPO (optical parametric oscillator) laser can continu-
ously emit light of various wavelengths within the
range indicated by the equation
z
s
z
i
= z
p
(16)
where z
s
, z
i
, and z
p
are the wavelengths of signal, idler,
and pumping waves, respectively.
The conguration and wavelength of an OPO laser
are shown in Fig. 12. This laser emits light of wave-
lengths z
s
and z
i
. The laser is pumped with the phase
matching angle by the light of wavelength z
p
ltered
by a nonlinear crystal.
2.4.3 Photodetectors
The types and various applications of photodetectors
are shown in Fig. 13. They are classied into two major
categories: point sensors and image sensors. The struc-
tures and characteristics of the typical point sensors,
i.e., CdS cells, photodiodes, and photomultipliers, are
shown in Fig. 14. The characteristics of CdS cells and
photodiodes are sensitive to temperature and require
compensation circuits.
2.4.4 Optical Fibers
Optical bers are exible brous light transfer media
used in optical systems. The bers are made of plastics
or quartz glass. Quartz bers are much more expensive
than plastic ones and not so exible. However, they are
temperature resistant (up to roughly 1000
C) and
transmit light more eciently, particularly in the UV
range. Whatever the material, the optical ber has a
center core with a refractive index ,
1
, surrounded by a
layer of another material (called cladding) with a
refractive index ,
2
. ,
2
must be lower than ,
1
in order
to keep the light inside the core area. Plastic core is
usually made of PS (polystyrene) or PMMC (poly-
methylmethacrylate). Plastic cladding is usually made
of teon or silicone. Figure 15 shows the transmission
Figure 10 Distributions of scattering angle. (a) Mie scatter-
ing; (b) intermediate; (c) Rayleigh scattering.
Table 3 Light Sources Applied to Fluidized Bed Measurement
Type Application Reference
Lamp Halogen Visible light source for optical ber probe
Lighting for picturing
Light source for space ltering Kamiwano and Saito (1984)
Mercury UV light source for O
3
detection Akagi and Furusaki (1983)
UV light source for uorescent particles Morooka et al. (1989)
Xenon Lighting for optical ber imaging Hatano and Kudo (1994)
Metal halide Lighting for optical ber microscope Commercially available
Light emitting diode (LED) Visible light source for optical ber probe
to dip into the bed
Yamazaki et al. (1988)
Laser Gas Light source for optical ber probe Horio et al. (1988)
Sheet lighting for visualization Horio and Kuroki (1994)
IR laser for contact-free heating (CO
2
laser) Yamada et al. (1996)
Semiconductor Visible light source for optical ber probe Werther and Hage (1996)
Tunable laser Laser-induced uorescence (LIF)
characteristics of PMMA bers and quartz bers ver-
sus the wavelength.
2.4.5 Optical Fiber Probe
Types of optical ber probes are presented in Table 5,
and some examples of probes are shown in Fig. 16. In
the case of the reection-type optical ber probe, the
volume from which the probe can detect the informa-
tion depends on ber conguration. The intensity dis-
tribution of reected light as a function of the distance
between a ber tip and a at surface is shown in Fig.
17. The size of the core of the optical ber should be
selected depending on the maximum diameter of par-
ticles being investigated. If individual particle passages
are to be detected, the ber core diameter should be
less than or equal to the particle diameter. On the other
hand, if only the passage of solids or the solids con-
centration is to be detected, the ber diameter of a
nonbundle-type probe or the diameter of a ber bundle
probe should be ve to ten times larger than the par-
ticle diameter.
The main diculty in using optical ber probes is
associated with their proper calibration. For solids
velocity measurements Ohki and Shirai (1976)
calibrated their probe (cf. Table 5) by dipping its
top, about 2 mm, into a thin bed of the same mate-
rial placed on a rotating disk (a music record disk
player was used) and obtained a cross-correlation of
the signals received from two bers. The cross-cor-
relation can be successfully applied for the analysis
of velocity data unless the velocity uctuates too
often.
To measure solids concentration, the calibration of
the probe is more dicult, because it requires a
homogeneous suspension. In many cases, the suspen-
sion tends to contain agglomerates or clusters, which
may seriously aect the output. Various ways to cali-
Table 4 Major Commercial LEDs and Lasers
LED Lasers
Color Material
Peak
wavelength
(nm) Type Material Wavelength Power Operation
Red GaP/GaP 700 Gas HeNe 632.8 nm 0.1100 Continuous
GaA1As/GaAs 660 (most cases) mW
InGaAlP/GaAs 644
GaAsP/GaP 635 Argon ion 488 nm, 100 mW Continuous
515.4 nm, 20 W
and others
InGaAlP/GaAs 623 CO
2
10.6 jm, 1W15 kW Continuous
Orange InGaAlP/GaAs 620 9.4 jm, > 1 MW Pulse
and others
InGaAlP/GaAs 612 Excimer 150 W Pulse
GaAsP/GaP 610
Yellow InGaAlP/GaP 590 Solid Nd: YAG 1064 nm > 2 kW Continuous
GaAsP/GaP 587 and pulse
GaP/GaP 570
Green InGaAlP/GaAs 574 Nd:YAG 300950 nm Continuous
Yellow (OPO) and pulse
GaP/GaP 565
Green InGaAlP/GaAs 562 Liquid Organic 2501100 nm Continuous
GaP/GaP 555 dye
+
and pulse
Blue SiC/SiC 470
GaN/sapphire 450 Semiconductor GaInP/GaAs 670680 nm Continuous
Infrared GaAs 910
GaAs:Si 940
+
Tunable laser, excited by Excimer, Nd: YAG or N
2
gas laser, etc.
Sources: LEDs: Noguchi, 1999. Lasers: Rearranged from Muraoka and Maeda, 1995.
brate the probes can be found, e.g., in Matsuno et al.
(1983), Hartge et al. (1986), Lischer and Louge
(1992), Amos et al. (1996), or Zhang et al. (1998).
Matsuno et al. (1983) calibrated their optical ber
probe using free-falling glass beads assuming that
the particle velocity equaled the terminal velocity u
t
.
For solids concentration up to 50%, Hartge et al.
(1986) conrmed a linear relationship between parti-
cle volume concentration and light reection intensity
for the liquidsolid uidized bed. Assuming that a
similar relationship exists for gassolid systems, they
compared the cross section average concentration
from the probe with the ,-ray adsorption or pressure
gradient data to obtain the calibration coecient for
quartz sand particles. Lischer and Louge (1992)
inserted an optical ber into the center of a capaci-
tance wall probe and calibrated the optical probe by
comparing the two signals. Herbert et al. (1994)
applied a method similar to that of Matsuno et al.
(1983) to FCC particles but used particle velocity
Figure 11 Example of emission spectrum of GaAlAs LED and lamps. (a) GaAlAs LED. (Hamamatsu Photonics, 1999c.) (b)
Various lamps. (Hamamatsu Photonics, 2000.) (c) Low-pressure mercury lamp. (Hamamatsu Photonics, 2000.) (d) Metal halide
lamp (150 W). (Hamamatsu Photonics, 2000.)
determined by optical probe measurement. They
found a clear eect of an antielectrostatic additive
(Larostat 519, PPG/Mazer Chemicals) in dispersing
particles, as shown in Fig. 18 (in their original
paper Figs. 4a and 4b should be reversed). They
obtained a power-law relationship between solids
volume fraction and light reection. Zhang et al.
(1998) also reported similar probe calibration.
Rensner and Werther (1993) determined the
eective measuring volume of a single ber optical
probe (d = 0.6 mm) for FCC and quartz sand as a
function of particle concentration. In both cases (i.e.,
quartz and FCC), the 50% transmission length was
less than 1 mm for the solids volume fraction c
p
=
0.002 and less than 0.1 mm for c
p
= 0.2. Concerning
the calibration of particle concentration probes
Amos et al. (1996) did detailed analysis and experi-
ments.
2.4.6 LDVLaser Doppler Velocimetry
Laser Doppler velocimetry (LDV) enables measuring
the local uid velocities from the frequency of light
reection from ne tracer particles (called seed part-
icles) in the uid. LDV can also be used to detect the
velocities of coarse particles or, with seeding, both gas
and particle velocities. The wavelength of laser light
reected from a moving particle diers from that of
a stationary source. This phenomenon is called the
Doppler eect, and its frequency shift o
i
is expressed
as
o
i
= k
s
k
i
( )v
p
(17)
where v
p
is particle velocity and k
s
and k
i
are the wave
number vectors of emitted and scattered light, respec-
tively.
Since in many cases the velocity v
p
is not suciently
large to make a large shift, the interference of scattered
lights from two laser beams at an angle is usually
utilized. For such a situation let k
1
and k
2
denote the
wave number vectors for the two reected lights, which
result in two dierent Doppler shifts, o
1
and o
2
.
Then, subtracting o
2
from o
1
, we obtain the fre-
quency of the interference, f
D
, called the Doppler fre-
quency, which can be expressed as a function of the
angle by
f
D
=
o
2
o
1
2
=
2v
px
z
sin
2
(18)
where v
px
is the x-component of particle velocity (cf.
Fig. 19).
When the light reection from a particle is detected,
the signal starts beating with a frequency f
D
, and the
component of particle velocity can be determined from
Eq. (18). There are several modes of light emitting and
Figure 12 OPO laser conguration and wavelength. (a) Wideband OPO; (b) narrow band OPO; (c) eect of phase matching
angle on idler wavelength of BBO crystal OPO. (Muraoka and Maeda, 1995.) (With permission of Sangyotosho)
Figure 13 Photodetectors and their applications. (Hamamatsu Photonics, 1999a.)
receiving systems, among which a dual-beam forward-
scattering mode is the most popular and commercially
available. It is popular because the signal can be
easily detected and processed, and high-frequency
resolution is not required. In the dual-beam for-
ward-scattering method a laser beam is split in two.
The two beams create interference fringes as illu-
strated in Fig. 19. The distance o between the two
bright peaks is expressed by
o =
z
2 sin(,2)
(19)
where z is laser wavelength and is the angle between
the beams.
Figure 14 Structures and characteristics of typical point sensors. (a) Spectral responses of photoconductive sensors.
(Hamamatsu Photonics, 1999b.) Note: Luminous eciency equals V(z) of Table 2. (b) Spectral responses of photodiodes.
(Hamamatsu Photonics, 1998b.) (c) Spectral response of a photomultiplier. (Hamamatsu Photonics, 1998a.)
Figure 15 Spectral attenuation and transmittance of optical bers. (a) Spectral attenuation of PMMA optical ber (diameter:
0.53 mm, room temperature, 65% RH). (Tolay Industries, 1991.) (b) Spectral transmittance of a 1 m long quartz optical ber.
A: SiO
2
core; B: GeO
2
,SiO
2
core. (Fujikura Ltd., 1988.)
Table 5 Types of Optical Fiber Probes in the Literature
A. Single Tip
Structure Type Fiber diameter Target parameter/s Reference
Reection type
a Single ber d
f
d
p
c
p
Bubble fraction
Hartge et al. (1986, 1988)
Tsukada et al. (1994)
b
Two-ber d
f
> d
p
c
p
Bubble fraction
Matsuno et al. (1983)
Horio et al. (1985)
d
f
d
p
Solid loading
and velocity
Cocco et al. (1995)
c
Three-ber d
f
d
p
v
p
(one-dimensional)
Bubble detection
Ohki and Shirai (1976)
Horio et al. (1980a)
Hernandez et al. (1998)
d
Multiber d
f
d
p
v
p
(two-dimensional)
Patrose and Caram (1982)
Ishida et al. (1980)
e
Space lter d
f
d
p
v
p
(one or two
dimensional)
Fielder et al. (1977)
f
Fiber bundle d
f
d
p
Particle passage Morooka et al. (1989)
v
p
and c
p
Nowak et al. (1991)
Nieuwland et al. (1996)
Tayebi et al. (1999)
Bore
scope
Cluster image
Particle image
Li et al. (1991)
Hatano et al. (1996)
If a particle passes through the interference region
with velocity component v
px
, we have a beat of
scattered light with the following frequency f:
f =
v
px
o
=
2v
px
z
_ _
sin
2
(20)
This expression is the same as Eq. (18) obtained from
Doppler shifts. Equation (20) is usually called the
fringe model for the beat frequency in LDV.
A typical signal obtained from a particle passing
through the measuring volume and having the dia-
meter less than the fringe distance d is shown in
Fig. 19b and in Fig. 20b (where it is shown as
signal A).
In the measurement of soliduid two-phase ow
we need to discriminate the signals from coarse parti-
cles and from nes (i.e., tracer seed). As schematically
shown in Fig. 19b, a ne (seed) particle whose diameter
Structure Type Fiber diameter Target parameter/s Reference
Transmission type
g Two in line d
f
d
p
Gas concentration
Tracer gas passage
Akagi and Furusaki (1983)
h
Bore
scope
Images of clusters
and strands
Takeuchi and Hirama (1991)
Takeuchi et al. (1998)
Two parallel d
f
d
p
Bubble passage Yasui and Johanson (1958)
+
Kai and Furusaki (1985)
Nakajima et al. (1991)
i
Curved ber
type
d
f
d
p
Bubble passage Ohki and Shirai (1976)
Ji et al. (2000)
Bore
scope
Particle image Hatano et al. (1998)
Light receiving only
j Single ber d
f
d
p
Burning char
temperature of
pyrometer
LaNauze et al. (1987)
Linjewile et al. (1994)
Joutsenoja et al. (1999)
* Quartz rods and prisms are used instead of ber probes.
B. Combined type
Probe shape Type Target parameter References
k
l
Pair Velocity (u
b
. v
p
. v
cl
. u
g
)
Correlation
Ohki and Shirai (1976)
Horio et al. (1988)
Greon et al. (1997)
Multi or
array
Distribution of bubbles Yamazaki et al. (1988)
Bubble shape
Jet shape
Lord et al. (1982)
Khattab and Ishida (1986)
m Two parameter
measurement
Gas and solid simultaneous
detection
Hatano et al. (1993)
d
f
: ber diameter, d
p
: particle diameter; : ow direction, : light direction
Table 5 continued
is smaller than o, passing through the interference
region, emits a signal described as gas signal. On the
other hand, a coarse particle emits the signal with a
large oset (described by sand particle signal).
Accordingly, in the uidsolid two-phase ow we can
separate the signals from uid (i.e., from seed) and
from coarse particles.
Figure 20 shows a schematic of an LDV system
applied to two-phase ows and the signal processing
scheme (Tsuji and Morikawa, 1982). The signal has
two components, called pedestal and burst. Pedestal
is the d.c. component and is obtained by low-pass l-
tering of the original signal (cf. signal D in Fig. 20b).
Burst (cf. signal H in Fig. 20b) is the a.c. component
and is obtained by subtracting the pedestal from the
original signal. Figure 20c shows an example of a sig-
nal from LDV.
In Fig. 20 the signal B indicates the scatter fre-
quency, based on which coarse and ne particles
can be discriminated. The discrimination can be
made in the following manner: The signals having L
= 1 and M = 1 come from the coarse particles,
while those having L = 1 and M = 0 from the
nes. The cases when L = 0 are rejected as ambig-
uous signals, but those with L = 0 and M = 1 (i.e.,
from the nes) are accepted because the ne parti-
cles usually do not make a large pedestal. The
threshold values T
1
and T
2
for pedestal and T
d
for burst to obtain logical discriminations should
be adjusted carefully by examining the relationship
between the threshold values and the number of
samples detected.
The seed particles required to measure uid velo-
city by LDV must be suciently small so that the
slip velocity between them and the uid can be
neglected. Additional requirements for the seed par-
ticles are large scattering area, uniform size (at least
half of o), no agglomeration, easy to feed, and non-
toxic, and they must be tolerated in the measure-
ment eld (Nakajima, 1995). Spraying water, teon
powder, silicone oil, or other components, as well as
chemical reactions, have been used as a source of
seed particles. TiO
2
, formed from the reaction
between TiCl
4
and moisture in the air has often
been applied. However, the unwanted by-product of
this reaction is HCl, which is very corrosive. Tsuji
and Morikawa (1982) used NH
4
Cl of mean diameter
0.6 jm.
Recently, the phase Doppler velocimetry (PDV) is
becoming a popular technique to measure particle
velocity and size. Doppler signals from a PDV are
received by an array of photomultipliers, and the
phase shifts between the signals from dierent photo-
multipliers are analyzed to determine particle size and
velocity simultaneously. Figure 21 illustrates the prin-
ciple of PDV. The particle velocity and diameter are
calculated (cf. Bachalo and Houser, 1984, van den
Moortel et al., 1997) as
Figure 16 Examples of optical ber probes. (a) Optical ber probe for simultaneous measurements of solid and gas velocities
(solid: light reection measurement; gas: transmission measurement with O
3
tracer). (Horio et al., 1992a.) (b) Water cooled
optical ber probe. (Werther and Hage, 1996.) (With permission of United Engineering Foundation)
v
p
= f
D
z
2 sin(,2)
(21)
d
p
= H
f o
2l
(22)
Notations in Eqs. (21) and (22) are the same as those
for LDV except for f , , l, and H, which are the
focal length of the transmitter lens, the phase shift
between two photomultipliers, the space between
detectors (say e.g., #1 and #2 or #1 and #3), and the
optical constant, respectively.
The advantage of LDV is that the setup can detect
reversal ows for particles, bubbles, and droplets in a
single and a multiphase ow system. LDV also
enables to measure uctuating velocity, size, and con-
centration of suspended particles. However, it is more
suitable to apply to the risers of rectangular cross
section than to circular ones. The LDVs limitation
in determining particle concentration (roughly up to
1%) can be improved greatly by combining it with an
optical ber probe. Wei et al. (1998) reported mea-
surements of particle velocities for solid concentration
Figure 17 Conguration and characteristics of reection type optical ber probes. (a) Fiber congurations. (b) The intensity of
reected light as a function of the distance to a at target. (Krohn, 1986. Cf. also Reh and Li, 1991.) (With permission of
International Society for Optical Engineering)
Figure 18 Calibration curves for FCC particles. (a) Without Larostat. (b) With 0.5 wt% of Larostat. (Herbert et al., 1994.)
(With permission of Elsevier Science)
up to 0.21. More information on LDV and its appli-
cations can be found e.g., in Watrasiewicz and Rudd
(1976), Drain (1980), Arastopour and Yang (1992),
van de Wall and Soo (1994), and Onofri and
Tadrist (1999).
2.4.7 Space Filters
Space ltering is a method of determining particle velo-
city with the spatial light distribution of a gratelike
structure including interference fringes. Dual beam
mode LDV is one type of space lter; the others
include a reticle type lter and a lter with gratinglike
detectors.
In the case of reticle-type lters, the illuminated
particle passes through the measurement volume and
reects light to the grating, thus being a source of a
periodical signal (Fiedler et al., 1997). From the fre-
quency f of this signal, the particle velocity v
p
is deter-
mined as
v
p
=
fg
[
(23)
where g is the grating constant and [ is the image scale,
i.e., the ratio of image size at grating to that at the
measuring point.
An example of a light-reection-type space lter
system is shown in Fig. 22a. The light from halogen
lamps is focused by a lens and guided by an optical
image ber bundle into the measuring space (bundle
diameter 3 mm, length 0.3 m; the bundle consists of
71,000 optical glass bers, 10 jm diameter each). The
light reected from the moving particles returns
through the same ber and is detected by two linear
image sensors via a lens and a half mirror.
The particle passage through the measuring space
is shown in Fig. 22b,c. Each of the wave patterns (I)
through (IV) in Fig. 22c corresponds to each state (I)
through (IV) in Fig. 22b. In Fig. 22c curve (I) shows
the background timing signal to distinguish the out-
put signal of image sensor 1 from sensor 2. When a
particle passes sensor 1, a line signal of n
1
bits is
detected and it moves from the n
1
bit to the n
2
bit within time interval t. Thus the radial and ver-
tical components of particle velocity, v
r
and v
z
, are
obtained as
v
r
=
X
r
t
(24)
v
z
=
W
t
(25)
where X
r
is the distance between the bit positions n
1
and n
2
and W is the distance between the two image
sensors.
A CCD (charge coupled detector) type space lter,
shown in Fig. 23, was used by Fiedler et al. (1997) to
measure sand particle velocity in a CFB riser. The
authors were able to measure the velocities between
0.12 and 40 m/s, at local solid volume concentration
0.14%. They reported that particles ner than 5 jm
Figure 19 Fringes and output signals of a dual-beam forward-scattering LDV. (a) Interference fringes and a measurement
volume. (b) Typical signal from freeboard of a uidized bed. (Levy, 1986.) (Reproduced with permission of the American
Institute of Chemical Engineers. Copyright # 1986 AIChE. All rights reserved.)
Figure 20 Processing scheme for gassolid two-phase ow measurement (Tsuji and Morikawa, 1982.) and an example of
experimental data. (Lehner et al., 1999.) (a) Optical arrangement and signal processing system. (b) Time chart of the signals.
(c) Experimental data obtained with LDV. Description of the signals of case (b): A: raw signal; B: frequency tracker signal of A;
C: synchronized pulse, which the tracer gives at every signal processing; D: Pedestal component of A after low-pass ltering; E:
square wave corresponding to the pedestal amplitude larger than the threshold T
1
; F: square wave corresponding to the pedestal
amplitude smaller than the threshold T
2
; G: derived from E and F through an exclusive NOR gate; H: alternative signal of A
(Doppler amplitude, burst signal); I: signal produced by comparator; J: square wave corresponding to the envelope of Doppler
amplitude larger than threshold T
d
; K: logical product G and J by AND gate; L: signal obtained from K and C through the D-
ip-op circuit; M: signal obtained from F and C through the D-ip-op circuit. The main characteristics of the optical system:
15 mW HeNe laser of wavelength 632.8 nm used as light source, = 25
, o = 1.2 jm, w
0
= 76.2 jm, l = 250 jm, 41 fringes in
the scattering volume, frequency-to-velocity conversion factor 0.6838 ms
1
MHz
1
. (a,b: Reproduced with permission of
Cambridge University Press; c: with permission of DECHEMA e.v.)
adhered to the optical window, but this did not
decrease the accuracy of measurement: it only caused
an increase in the d.c. level by scattering and reec-
tion.
2.5 Capacitance and Conductivity Sensors/Probes
A capacitance sensor provides information on solids
concentration in the target space between its two elec-
trodes. When a voltage E is applied between the elec-
trodes, an equilibrium charge Q is accumulated in
them. The charge can be determined by integrating
the measured electric current between the electrodes
over time. Capacitance is the ratio Q/E. For a set of
two parallel plates of area A, separated by a homo-
genous dielectric of thickness d, the capacitance is
given by
C =
kA
d
(26)
where k is the dielectric constant or permittivity of the
media between the electrodes [F/m]. k is a function of
frequency and varies with temperature. Usually it is
expressed in terms of the relative dielectric constant
K of the medium as
K =
k
k
vacuum
[] (27)
In gassolids systems the volume fraction of solids is
inferred from the measurements of the eective dielec-
tric constant (or eective permittivity), expressed by
K
eff
= cK
f
(1 c)K
s
(28)
where c is the voidage, K
f
is dielectric constant of the
uid, and K
s
is the dielectric constant of solids (i.e., the
ratio of the permittivity of the system to the permittiv-
ity of free space).
Some data of the relative dielectric constant, K, are
shown in Table 6.
An example of a capacitance probe and its output
signals are shown in Fig. 24. To eliminate the inu-
ence of external capacitance of various parts of the
probe (e.g., cables) that can have much larger capa-
citance than that of the probe itself, a third conduc-
tor called a guard is necessary, as discussed by Acree
Riley and Louge (1989). More on capacitance mea-
surements can be found in Bottcher (1952), Bakker
and Heertjes (1958), Meredith and Tobias (1960),
Lanneau (1960), Werther and Molerus (1973),
Yoshida et al. (1982), Brereton and Stromberg
(1986), Louge and Opie (1990), and Brereton and
Grace (1993). Capacitance tomography measure-
ments are described in Sec. 3.3.
2.6 Acoustic sensors
2.6.1 Microphones
Acoustic emission from uidized beds also provides
information on hydrodynamic conditions. Win et al.
(1995) used a high-sensitivity microphone to detect
the carryover of coarse particles from a multisolid
circulating uidized bed consisting of porous alumina
(coarse particles) and glass beads (the nes) They
detected acoustic signals, which the particles gener-
ated on the cyclone surface. Since coarse particles
had larger mass, they generated sounds of magnitude
Figure 21 Doppler signals corresponding to a single particle passage received by dierent photomultipliers. (van den Moortel et
al., 1997.) (With permission of Elsevier Science)
Figure 22 Schematic of a light reection type space lter (a), tracer particle passage (b), and wave pattern of output signals from
light reection type space lter in a gasliquid system (c). (Kamiwano and Saito, 1984.) (Reproduced with permission of the
American Institute of Chemical Engineers. Copyright # 1984 AIChE. All rights reserved.)
Figure 23 CCD type space lters. Grating constant: 64 jm; image scale (i.e., the size at grating divided by the size at measuring
point): 0.0621; sampling frequency: 156 kHz. (a) CCD type space lter probe. (b) Block diagram for CCD type space lter.
(Fiedler et al., 1997.) With permission of Elsevier Science
much larger than the nes. Dahlin et al. (1995)
applied a knife-edge acoustic sensor connected to a
sensitive transducer to detect particles of various sizes
in a gas cleanup system.
2.6.2 Acceleration sensors
Sensors of this type were used by Fukayama et al.
(2000), who monitored the changes in uidization
conditions inside a 350 MW
e
AFBC by determining
the acoustic signals collected through in-bed boiler
tubes. The signals were detected by the acceleration
sensors placed onto the tubes outside the uidized
bed. The monitoring has been performed for 1.5
years. More details can be found in Fukayama et
al. (2000).
2.7 Electrostatic Sensors/Probes
Electric charge can be measured by a Faraday cage as
shown in Fig. 25a. Fasso et al. (1982) used this cage
to study electrostatic charges and concentration of
particles in the freeboard of a uidized bed of glass
beads. The gassolid suspension was sucked into the
cage, and the solids charge was determined. The mea-
suring cage was put into another Faraday cage to
shield against any external charge. Examples of the
electrostatic probes are shown in Fig. 25b,c. Soo et al.
(1964) applied the probe shown in Fig. 25b to mea-
sure local solids concentration in a gassolid ow.
The authors were able to measure instantaneous con-
centration, but no impact direction could be identi-
ed. Boland and Geldart (1971) used the probe
shown in Fig. 25c to determine static electrication
in a 2D bubbling uidized bed. They described a
method of visualizing particle motion based on elec-
trostatic charging.
2.8 Gas Sensors
To determine oxygen concentration in a uidized bed
combustor, zirconia sensors (commonly known as z-
probes) have been used. The zirconia sensor acts as a
kind of battery, whose output voltage depends on oxy-
gen concentration (the voltage is low at oxidizing and
high at reducing conditions). The schematic of the sen-
sor and an example of its ouput signal are shown in
Fig. 26.
Table 6 Relative Dielectric Constants of Some Common
Materials
Material
Temperature,
C
Frequency,
Hz K
Air 0 - 3 + 10
6
1.001
Water 20 Low 80.37
Alumina 20 10
6
4.58.5
Nylon 66 25 10
6
3.33
Neoprene rubber 24 10
6
6.26
Polyethylene 23 10
6
2.26
Potassium nitrate 20 10
6
5.6
Silicon rubber 25 10
6
3.1
Vycor glass 20 10
6
3.85
Pyrex glass 20 10
6
46
Figure 24 An example of a capacitance probe. (a) High-temperature two-needle capacitance probe of Wiesendorf et al. (1999)
(With permission of American Society of Mechanical Engineers). (b) Output signals from a 12 mW (Chalmers) and a 250 MW
(Gardanne) CFB boiler obtained with the probe shown in (a). (Johnsson et al., 1999.) With permission of DECHEMA e.v.
2.9 Humidity Sensors
The main types of instruments measuring humidity in
gases are various types of hygrometers (gravimetric,
mechanical, condensation, infrared absorbance detec-
tor, electric sensor, thermal conductivity, Al
2
O
3
/sili-
con, P
2
O
5
) and psychrometers.
2.10 Radioactive Sensors
Radioactive particles can be applied as tracers, used to
study solids mixing, particle trajectory, or RTD (resi-
dence time distribution) of gas or solids within a ui-
dized bed, as well as to measure the suspension density.
Inorganic salt scintillators, i.e., crystals of inorganic
salts containing trace quantities of activators to
enhance the emission have been used to detect the
radiation. A popular type is the sodium iodide scintil-
lator activated with thallium (NaI(Tl)). However, this
material is hygroscopic and must be encapsulated.
Other commonly used scintillators contain cesium
iodide activated with thallium or sodium, CsI(Tl),
CsI(Na).
Examples of some radioactive isotopes used for ui-
dized bed tracking are shown in Table 7. As discussed
e.g., by Seville et al. (1995) and Benton and Parker
(1996), the scattering or absorption of the emitted
radiation may lead to incorrect determination of tracer
position. However, since the scattered rays have lower
energy than the correct ones, by removing the scat-
ter data one may avoid this problem. Application of
radioactive particles as tracers is relatively easy and
inexpensive, even for large uidized bed plants.
However, it may sometimes disturb the ow and de-
nitely requires strict safety regulations during the mea-
Figure 25 Faraday cage used by Fasso et al. (1982) (With permission of Elsevier Science) (a) and examples of electrostatic
probes of Soo et al. (1964) (Reprinted with permission from Ind. Eng. Chem. Fundam., 1964, 3, 98106 Copyright # 1964
American Chemical Society) (b) and Boland and Geldart (1971) (With permission of Elsevier Science) (c).
surements (particularly for the particles having long
half-life). More on radioactive particle tracking can
be found in Seville et al. (1995), Stein et al. (1997),
and Larachi et al. (1997). More on radioactive tracers
is also given in Sec. 2.11.
2.11 Tracers
2.11.1 Solid Tracers
To obtain information on solids movement and mix-
ing, the solid tracers are fed into a uidized bed and
detected either in situ, as in the case of isotope tra-
cers, or by sampling and analysis. It is assumed that
the tracers behave like the solids in the ow. Some
examples of commonly used tracers are shown in
Table 8.
In choosing the radioactive tracer one should take
into consideration the following aspects:
1. The purity of radionuclide; to assure that ,-
rays are emitted from disintegration of the
selected radionuclide and not from some of its
impurities.
2. The activity of the source; high counting rate
can be measured more accurately than a low
one.
3. The energy of ,-rays: according to the
LambertBeer law, the attenuation is an expo-
nential function of the linear attenuation coe-
cient and the thickness of the material between
the source and the detector.
4. The half-life: it should be at least an order of
magnitude longer than the duration of the
tracking study. Otherwise, the loss of source
activity may be signicant during the experi-
ments. On the other hand, too long a half-life
time is not recommended, because it may bring
about diculties in the test equipment after the
experiment (e.g., owing to the residual activity
of the tracer).
2.11.2 Gas Tracers
By injection of a tracer gas into the ow (continuous
or batchwise form) and by measuring its concentra-
tion and velocity one obtains information on the
residence time distribution (RTD), mixing or mass
transfer. Examples of applications of gas tracers to
uidized beds are presented in Table 9. Figure 27
presents a model of gas ow in the uidized bed
and an example of recorded data. Gas tracking is
also a useful way to measure mass transfer in ui-
dized beds, as reported by Ebert et al. (1993) and de
Kok et al. (1986).
2.12 Suction Probes for Solids Sampling
There are two kinds of solids suction: isokinetic and
nonisokinetic. In isokinetic sampling, the gas suction
velocity is adjusted to equal the local gas velocity.
This method was mainly applied to aerosol or other
Figure 26 Schematic of a zirconia sensor and an example of the output signal (measurement in a 12 MW CFB boiler).
(Johnsson et al., 1999.) With permission of DECHEMA e.v.
Table 7 Half-Life of Some Positron
Emitting Isotopes
Nuclide Half-life Nuclide Half-life
Rb
82
78 s Zn
62
,Cu
62
9.2 h
O
15
122 s Ga
66
9.7 h
N
13
10 min Cu
64
12.7 h
C
11
20.3 min I
124
4.2 days
Ga
68
68 min Ge
68
,Ga
68
271 days
F
18
110 min Na
22
2.6 years
Ti
45
3.1 h
Source: Benton and Parker, 1996.
Table 8 Various Types of Particles Used as Tracers in Fluidized Beds
Tracer type Tracer Remarks Author(s)
Chemical NaCl tracer FCC reactor, the tracer detected by electric
conductivity of samplewater solution
Bader et al. (1988)
Limestone Diusivity in a large-scale FB boiler Ito et al. (1999)
Phosphor Eect of tracer size on solids mixing in the
0.14 m i.d., 10.4 m high FCC riser
Du et al. (1999)
Color Colored particles Semicircular uidized bed, video recording Yang and Keairns (1982)
Color Colored particles Video recording of particle velocity in 2D
uidized bed
Gbavcic et al. (1990)
Fracture Spheres of dierent
strengths
Kinetic forces acting on particles in a uidized
bed determined from analysis of tracer particles
Kono et al. (1987)
Fluorescent Paint pigment attached
to coal particles
Particle velocity in the cocurrent downward ow
of coal and air
Brewster and Seader (1980)
Fluorescent Tracer detected by LDV, 319 176 mm cross-
section FB
Hamdullahpur et al. (1987)
FCC impregnated with
a uorescent material
Particles illuminated by UV light, reected light
detected by optical bers
Kojima et al. (1989)
Nowak et al. (1991)
Magnetic Ferromagnetic Solids mixing in a bed of FCC, D
t
= 0.15 m i.d.
Bed operated in a bubbling, slugging, and
turbulent regime
Avidan and Yerushalmi
(1985)
Sand coated with ferrite
Char impregnated with
magnetic powder
Solids mixing studied by magnetic separation
Measurement of the ow of dark colored
particles
Yamaki et al. (1994)
Thermal Hot particles Temperature measurements Westphalen and
Glicksman (1995)
Radioactive Co
60
Observation of particle motion in a FB Kondulov et al. (1964)
Na
22
Studies on particle motion in a 0.03 m i.d. gas-
solid uidized bed. Scintillation detector placed
below the uidized column
Borlai et al. (1967)
Na
24
Particle motion in a FB of glass beads
Scintillation counter placed above the bed
Van Velzen et al. (1974)
Radioactive Na
2
CO
3
RTD measurement in CFBs of various sizes Helmrich et al. (1986)
Ga
68
Particle tracking in a CFB Ambler et al. (1990)
Radioactive SiO
2
Studies on RTD in a 82.8 mm i.d. CFB Patience et al. (1991)
Sc
46
oxide Studies on particle motion and solids mixing in a
3D liquidsolid uidized bed
Larachi et al. (1994)
Larachi et al. (1995)
Radioactive tracers of
various sizes and
densities
Measurement of local particle velocities in a cold
CFB
Weinell et al. (1995)
F
18
Particle tracking in a 0.15 m i.d. sand bed Seville et al. (1995)
Au
198
Studies on the hydrodynamics of a CFB tracer Godfroy et al. (1996)
Positron emission Tracer velocity measured 250 times per second Stein et al. (1997)
Irradiated Cu Particle tracking, 3D resolution below 15 mm Stellema et al. (1998)
Mn
56
and cordierite Studies on particles tracking in the 80 MW
th
CFB
boiler. 16 NaI(T1) detectors used, data
collected with the frequency of 20 Hz
Lin et al. (1999a)
Ca
137
tracer combined
with optical ber set
Solid concentration axial prole and
instantaneous local values measured in a 0.4 m
i.d. 15.6 m CFB riser, G
s
up to 50 kg/m
2
s
Schlichthaerle and
Werther (1999)
dilute two-phase ow sampling, as well as to uidized
beds (e.g., van Breugel et al., 1969; Nguyen et al.,
1989).
In case of a CFB, the amount of sampled solids is
not sensitive to the gas suction velocity because of
frequent ups and downs of solids in the riser and
because of high inertia of particles, as reported
e.g., by Monceaux et al. (1986), van Breugel et al.
(1969), and Herb et al. (1992). There is a certain
suction velocity window where the suction velocity
does not aect the solid mass ux, as shown in
Fig. 28, and the nonisokinetic sampling is more prac-
tical.
Examples of application of solids suction are
listed in Table 10. The measurement setup consists
usually of two parts: the suction part, immersed in
the system, and the measurement section, located
outside. More on suction probes and their equipment
can be found in Dry (1987), Rhodes and Laussmann
(1992a, 1992b), Leckner et al. (1991), Kruse and
Werther (1995), and Schoenfelder et al. (1996). A
schematic of a solids suction probe is shown in
Fig. 29.
In the case of measurement in a CFB, because of the
rapid solids up and down motion, the orientation of
the suction tube is also important. At each measure-
Table 9 Various Gas Tracers
Author(s) Tracer type Comments
Cankurt and Yerushalmi (1978) CH
4
Radial gas diusivity estimated from radial proles of tracer concentration.
Adams (1988) CH
4
Studies on gas tracking in a CFB riser
Khattab et al. (1988) O
3
Studies on gas velocity in uidized beds. The tracer was detected by optical
bers through which UV light was transmitted
Horio et al. (1992a) O
3
Optical ber probe applied to measure particle and gas velocities
simultaneously; ozone injection upstream of two optical probes, tracer
gas velocity measured using UV light, optical ber and photomultiplier
Ye et al. (1999) O
3
Ozone decomposition within a CFB. Cast steel used as ozone decomposing
catalyst
Sauer and Wallen (1999) CO
2
Studies on gas mixing in a 2D ABFBC
Yang et al. (1984) He Concentration proles within the bed measured by continuous gas
injection
Bader et al. (1988) He Continuous injection of gas into an FCC bed
Li and Weinstein (1989) He Continuous He injection applied to study gas backmixing in a 0.152 m i.d.
CFB
Martin et al. (1992) He Gas radial velocity proles and dispersions in a CFB studied by gas
chromatograph analysis of concentration of injected He
Shen et al. (1992) He Studies on hydrodynamics and mass transfer in a 2D FB, thermal
conductivity detector used to measure tracer concentration
Rivault et al. (1995) He Tracer injection into the leg of primary cyclone; tracer gas sampled by a
displaceable probe equipped with a lter and analyzed
chromatographically
Lin et al. (1999b) He Studies on gas RTD in a high density FCC unit. Solids mass ux up to
430 kg/m
2
s
Krambeck et al. (1987) SF
6
Studies on tracer adsorption in the FCC bed. Sulphur hexauoride
injected below the bed. Tracer concentration determined 1.8 m above
the bed
Dry et al. (1995) Warm gas
injection,
liquid N
2
injection
Studies on gassolid contact eciency in two CFBs of dierent scale,
102 mm i.d. and 600 mm i.d
Li and Wu (1991) H
2
pulses Studies on axial gas mixing and RTD in the CFB containing FCC
particles (d
p
= 58 jm)
White and Dry (1989) Ar pulses Studies on gas residence time, tracer detected by two rapid response mass
spectrometers, reactor 0.09 m i.d. CFB
ment point the suction should be conducted at both up
and down positions of the tube in order to measure the
local upward and downward solids mass uxes (one
should remember that even in the core of the riser
the solids downux always exists, as reported e.g., by
Herb et al., 1992).
2.13 Gas Sampling
Gas sampling is a basic tool providing information
about the local/total gas composition and concentra-
tion. Examples of gas sampling studies are listed in
Table 11.
In order to separate the gas and solid phase during
uidized bed sampling, a ceramic porous plate at the
inlet of the sampling system is recommended (e.g.,
Werther et al., 1990). Naruse et al. (1994) applied the
probe shown in Fig. 30a,b and were able to sample the
gas separately from the bubble phase and from the
dense phase. The bubble passage was detected from
pressure drop between two sensors, as shown in Fig.
30c.
3 DATA PROCESSING AND VISUALIZATION
SYSTEMS
3.1 Data Processing
Signal processing after the measurements is required:
(1) to reduce noises and (2) to obtain characteristic
values such as mean values, characteristic frequencies,
and phase delay between signals. More information can
be found in, e.g., Kawada and Minami (1994) and
Edgar et al. (1997). Table 12 shows the major algo-
rithms for data processing applied to the uidized beds.
Some examples of data processing are shown in
Figs. 3133. The simplest window function is unity
for all data (irrespective to the time they were taken)
and gives the arithmetic moving average. Figure 31
presents the ammonia IR spectrum processed by a
polynomial equation as a window function. The num-
ber of averaged samples 2m1 should be chosen
appropriately to obtain the necessary information out
of the data set.
In the case of stationary signals, the averaging
enhances the S/N ratio. Such averaging is used in
ordinary chemical analyzers, e.g., FTIR, ICP, EDS,
EPMA. If the number of averaging is too large, the
Figure 27 Path of meandering gas in a uidized bed (a) and the variation of tracer concentration (b). (From Jovanovic et al.,
1980.)
Figure 28 Eect of gas suction velocity on solids sampling.
(Leckner et al., 1991.) (With permission of American Society
of Mechanical Engineers) U
o
= 3.4 m,s. G
s
= 13 kg,m
2
s.
averaging process becomes time-consuming. However,
by that averaging the S/N ratio can be improved with-
out knowing the frequency of noises. Figure 32 shows
a schematic of the change of the signal vs. the number
of averaging.
Filtering in the frequency domain is the most pop-
ular method of signal denoising, where a conventional
resistorcapacitor circuit is used as an analog lter to
eliminate the noises, i.e., signal components having
frequencies much higher than that of the main signal.
A schematic of signal processing by low-pass ltering is
shown in Fig. 33.
3.1.1 Autocorrelation
The autocorrelation function, dened in Table 12,
becomes periodic for periodic signals. Figure 34
shows examples of the autocorrelation functions
and power spectra for a bubbling bed and a slugging
bed.
Table 10 Examples of Recent Applications of Solids Suction Probes
Author(s) Remarks
Verloop et al. (1993) Dust sampling from a PFBC
Wang et al. (1995a) Nonisokinetic solids sampling from a 16.1 cm i.d. CFB riser operated at various temperatures
Wang et al. (1995b)
Nowak et al. (1995) Solids suction to measure local solids mass uxes and PSD in a multisolid FB
Mattison and Lyngfelt (1995) Application of a water-cooled sampling probe to measure the amount of CaS in a 12 MW
CFB boiler
Kozinski et al. (1995) Solids sampling from a CFB sludge waste combustor to detect solids PSD and composition.
Dahlin et al. (1995) Application of a particle sampling probe to evaluate hot gas cleanup system
Junfu et al. (1999) Application of a water-cooled solids mass ux meter to measure local G
s
at the furnace
outlet of a 75 t/h CFB boiler
Brunier et al. (1999) 13 mm i.d. suction probe successfully applied to a CFB boiler
Figure 29 High temperature solids sampling probe of Mattisson and Lyngfelt (1995). (With permission of American Society of
Mechanical Engineers)
Table 11 Examples of Gas Sampling in Fluidized Bed Systems
Author(s) Remarks
Swift et al. (1975) Flue gas sampling to measure the concentration of Hg, F, and the solids fraction
Yang et al. (1984) Continuous He injection into a FB. Gas concentration proles measured
Bader et al. (1988) Detection of He injected into an FCC riser
Adams (1988) Application of CH
4
tracer to a CFB riser
Atkinson and Clark (1988) Detection of bubbles in a FB. Bubble passage determined using a dual static pressure probe
Naruse et al. (1994)
Naruse et al. (1995)
Gas sampling probe, shown in Fig. 30c, applied to sample gas from bubble and
emulsion phases separately. Reactor: 0.1 m i.d. 0.7 m FBC; bed material: sand, fuel: coal
Kassman et al. (1995) Gas sampling system applied to measure the concentration of NH
3
and HCN inside a
12 mW CFB boiler
Hansen et al. (1995) Particle probe, z-probe, gas probe and alkali metal sampling probe used to monitor a
20 MW
e
CFB combustor
Mann et al. (1995) Alkali-sampling probe applied to characterize sorbents for a PFBC.
Hayrinen et al. (1999) Alkali concentration measurements in a 10 MW PCFB using laser uorescence and plasmas
spectroscopy sensors
Kassman et al. (1999) Gas sampling from a 12 MW CFB boiler
Lin et al. (1999c) Simultaneous measurement of N
2
O and NO
x
concentration, as well as particle
temperature in the vicinity of a burning char particle
Figure 30 Gas sampling system (a), probe details (b), and bubble passage determined from pressure drop (c). (Naruse et al.,
1994.) (With permission of Combustion Institute)
Table 12 Data Processing Algorithm and Application for Fluidization
Processing Denition/explanation
Examples
Signal Remarks Reference
Moving average
y
i
=
m
j=m
w(j)x(i j)
m
j=m
w(j)
(i = m1. n m)
w(j): window function
Arithmetic
moving
average
y
i
=
1
2m1
m
j=m
x(i j)
Pressure S/N improvement,
equivalent to simple
average for random
noise
Hatano et al.
(1999)
Filtering in
frequency region
x(t) Fourier Fourier inverse
X(o)W(o) y(t)
Eective when
noise frequency
is much higher
than the signals
X(o) =
_
o
o
x(t)e
jot
dt: Fourier transform
W(o): filter function
Low-pass lter
W(o) =
1(o _ o
1
) (o
1
: cut off
0(o > o
1
) frequency)
_
S/N improvement,
elimination of
signal from
individual particle
Averaging
y =
1
N
N
i=1
x
i
Pressure S/N improvement, for
signal only repeatable
measurement; frequency
response information for
signal and noise is not
necessary
Root mean
square (RMS) x
rms
=
_
x
2
x
1
( xx x)
2
dx
x
2
x
1
_
Pressure
uctuation
Phase transition Yerushalmi
and Cankurt
(1979)
Probability
distribution
function (PDF)
p(x) =
1
x
lim
To
n
i=1
t
1
T
Light
LDV
Light
Determination of threshold
value to distinguish:
Bubble and emulsion phases
Gas and particle signals in
pneumatic transport
Cluster and lean phase
Horio et al. (1985)
Tsuji and
Morikawa (1982)
Horio et al.
(1988)
Autocorrelation
function
R(t) =
lim
To
_
T
0
x(t)x(t t) dt
Pressure Slugging condition Broadhurst and
Becker (1976)
Pressure
Light
Light
Heat transfer
Bubbling condition
Bubbling regularity
Freeboard turbulence
Dynamics of CFB
Fan et al. (1981)
Greon et al. (1997)
Li et al. (1993)
Cross-correlation
function
R
xy
(t) =
lim
To
_
T
0
x(t)y(t t)
Pressure
Pressure
Bubbling condition
Pressure wave velocity
Fan et al. (1981)
Roy and
Davidson (1990)
Light Particle velocity Horio et al. (1980a)
3.1.2 Cross-correlation
Cross-correlation, dened in Table 12, provides infor-
mation about the statistical mean delay time between
two signals if there is any correlation between them.
Figure 35 shows an example of a bed pressure uctua-
tion signal in a bubbling uidized bed. From cross-
correlation shown in Fig. 35b the obtained mean
delay time is 0.12 s.
The degree of interaction between two signals can
also be determined from cross-power spectrum, coher-
ence (van den Schaaf et al., 1999b), or cross-covariance
function (Greon et al., 1997). In case of wide solid
velocity uctuations with periodic ups and downs, as
in the case of particles or clusters, the cross-correlation
cannot give the right solid velocity; a peak search algo-
rithm with a time lag determination for every peak of
the original signal is required to obtain the proper
results.
3.1.3 Power Spectrum
The dominant frequency of a uctuating signal is
obtained from a Fourier transform power spectrum.
Figure 36 shows an example of a pressure uctuation
Processing Denition/explanation
Examples
Signal Remarks Reference
Power spectrum
S
x
(o) =
lim
To
1
T
X
T
(o)
2 Pressure
Pressure
Bubbling condition
Solid structure in a CFB
Fan et al. (1981)
van den Schaaf
et al. (1999a)
Pressure Phase transition Horio and
Morishita (1988)
Heat transfer Dynamics of a CFB Li et al. (1993)
Pressure
(Maximum
entropy
method,MEM)
Analysis of long range
force
Mehrabi et al.
(1997)
Fractal analysis Self similarity, evaluated by
fractal dimension
Image
Pressure
Cluster shape
Chaos analysis
Ito et al. (1994)
Fan et al. (1990)
Ross-Pence (1997)
Solid
momentum
probe signal
Chaos analysis Bai et al. (1996)
Chaos analysis Evaluated by Kolmogorov
entropy, correlation
dimension, Lyapunov
exponent, presence of the
chaotic attractor, Hurst
coecient, etc.
Pressure, light,
,-ray porosity,
hot wire prove
signal, etc.
Quantication of the
dynamics and phase
transition in the BFB,
CFB, mixing chamber, etc.
Schouten and van
den Bleek (1992),
Marcocchella
et al. (1997),
Ji et al. (2000),
Huilin et al.
(1995)
Briens et al.
(1997)
Wavelet
transform
Analysis of time-varying
signal simultaneously from
both time and frequency
perspective
cf. Table 15.13 cf. Table 15.13 cf. Table 15.13
W
x
(t. a) =
1
a
_
_
o
o
x(t) (t t),a ( ) dt
Table 12 continued
power spectrum in which dierent dominant frequen-
cies are observed.
3.1.4 Maximum Entropy Method
The maximum entropy method (MEM) is developed
to obtain the maximum spectrum information from
the limited number of data. It enables us to esti-
mate the power spectrum without an FFT (fast
Fourier transform) using a distinct Fourier trans-
form (DFT). The main problems in the FFT
method are the so-called spectrum leaks from
other frequencies, i.e., in addition to the true
range of frequencies, the power spectrum also con-
tains components at other unwanted frequencies,
which leads to errors in spectral analysis. To
demonstrate the spectrum leaks associated with
FFT, suppose that an original continuous signal is
the one shown in Fig. 37a. Its Fourier transform
power spectrum is shown in Fig. 37b and has one
sharp peak. From the limited number of data (c)
the FFT is obtained as (d), which is still similar.
However, from another set of limited data but half
a period longer than the data (c), we obtain a spec-
trum with a few small peaks. This is the spectrum
leak. To cope with this, a windowed Fourier trans-
form shown in (g) with a lenslike window has to be
applied to improve the spectrum to (h).
As already mentioned the MEM gives a frequency-
leak-free spectrum without FFT. Furthermore, for
signals having sharp spectrum peaks, the MEM gives
a sharper spectrum than FFT, as shown in Fig. 38.
3.1.5 Wavelet Transform
By a wavelet transform it becomes possible to detect
unusual as well as periodical signals. Several kinds of
wavelet transforms have been proposed, including con-
tinuous and discrete transforms. Examples are sum-
marized in Table 13.
Gabor wavelet transform, illustrated in Fig. 39a, is
one of the continuous transforms, in which the follow-
ing wavelet, ((t t),a), is used:
t t
a
_ _
=
1
2
_
o
exp 0.5
t t
ao
_ _
2
_ _
exp
jo
0
(t t)
a
_ _
(29)
The wavelet transform of a signal x(t), W
x
(t. a), is
dened by
W
x
(t. a) =
1
a
_
_
o
o
x(t)
t t
a
_ _
dt (30)
where a is a scale dened by using the representative
frequency of the signal o
0
, as
a
o
0
o
(31)
(t
+
) satises the following relationships:
Figure 31 The moving average with second and third polynomial adaptive window. Left: original; center: 7 points averaging;
right: 21 points averaging. (Minami, 1986.)
Figure 32 Eect of averaging on the shape of signal: (a) rst, (b) second, (c) Nth, (d) nal. (Minami, 1986.)
Figure 33 Elimination of high-frequency noises by low-pass ltering. (a) Original signal, (b) Real part of Fourier transform, (c)
Processed signal. (Minami, 1986.)
Figure 34 Autocorrelation function and power spectrum. Upper row: original pressure signal; middle row: autocorrelation (no.
of lags 20 = 0.45 s; sampling interval: 22.5 ms; data points: 1000; D
t
= 0.1 m, u
0
= 0.1 m/s; bed material: silica sand, d
p
= 0.183
mm); bottom row: power spectrum. (a) Smoothly uidized bubbling bed (L,D
t
= 1); (b) slugging bed (L,D
t
= 2). (Broadhurst
and Becker, 1976.)
Figure 35 Pressure uctuation signal from a slugging uidized bed and their cross-correlation function. (Roy et al., 1990.)
A
t
= 0.12 0.12 m
2
; L
mf
= 1.6 m; silica sand bed, d
p
= 240 jm. (a) Original signal from pressure probes (upper row: 1.5 m
above the distributor; lower row: 0.07 m above the distributor. --- is the theoretical signal from Kehoe and Davidson, 1973). (b)
Cross-correlation function. (With permission of Elsevier Science)
Figure 36 Power spectrum density of pressure uctuations in a bubbling uidized bed of sand (van den Schaaf et al., 1999a). D
t
:
0.80 m; u
0
: 0.44 m/s; d
p
: 390 jm; settled bed height = 2.19 m. (With permission of American Society of Mechanical Engineers)
_
o
o
(t
+
) dt
+
= 0 (32-1)
_
o
o
(t
+
)
2
dt
+
= 1 (32-2)
The second exponential function in Eq. (29) is used
in ordinary Fourier transforms, where no window is
applied. In the Gabor wavelet transform a window
around t = t is superposed by the rst part of Eq.
(29) (i.e., by the error function). On the other hand,
in the classical Gabor transform, not the Gabor wavelet
transform, a windowed Fourier transform with a xed
width window is applied, as illustrated in Fig. 39b.
3.2 Visual Observation
3.2.1 Photo and Video Imaging
For visual observation through a column wall of a cold
test rig, particle adhesion to the wall by static electricity
should be reduced. By putting charge absorbers into the
bed (Ilias et al., 1988; Chang and Louge, 1992), by coat-
ing the riser surface by a conductive material (Myler et
al., 1986), or by covering the inside wall with cellophane
tape (Horio and Kuroki, 1994; Tsukada et al., 1997),
the adhesion trouble can be avoided.
Observation of a high-temperature uidized bed can
be easily done in a lab scale gold mirror furnace con-
sisting of a quartz tube, a gold-coated Pylex tube, and
a coil heater between them. If temperature dierence is
created in the bed by some reactions it can be well
observed from the outside as, e.g., done by Horio et
al. (1986c). To observe the inside of a hot uidized bed,
an endoscopic system can be used (Zevenhoven et al.,
1999b). More information can be found in Horio et al.
(1980b), Yang and Keairns (1982), Prins (1987), Yang
and Chitester (1988), Takeuchi and Hirama (1991),
Zou et al. (1994), and Hull and Agarwal (1995).
3.2.2 Laser Imaging
At low solids concentration the laser sheet imaging
shown in Fig. 40a is useful in observing uidized
beds. However, the original images on the light source
Figure 37 Spectrum leak associated with FFT. (Minami, 1986.)
side are brighter than those on the other side, owing to
the expansion of laser beam. This can be corrected in
the following manner: suppose L and L
/
are the gray
scales of a pixel on the original and modied picture,
respectively, which are located at a distance r from the
rod lens. L
/
= Lr
0
,r gives the corrected scale, where r
0
is the distance of a reference pixel in the image (cf. Fig.
40b).
Figure 38 Spectrum of mercury lamp obtained by FFT and MEM. (Kawata et al., 1983.)
Table 13 Application of Wavelet Transform to Fluidized Bed Measurement
Type of wavelet
transform Measurement Derived information Reference
Orthonormal wavelet
transform
Pressure
uctuation
Decomposition of the signals into long-term and short-
term correlation components; one represented by the
self similarity Hursts parameter of fractional Brownian
motion originated by bubble, and another represented
by the intensity parameter of Gaussian white noise
originated by gas jetting, small bubble formation, and
turbulence of uidized beds, etc.
Hee et al. (1997)
Morlet wavelet Heat ux Self-similar bifurcation and trifurcation phenomena Ross and Pence
(1997)
Discrete wavelet
transform based on
Mallet and Zhang
Optical ber
probe signal
Phase separation (identication of the transition from
the dense phase to the dilute phase)
Ren and Li
(1998)
Discrete analog of
wavelet transform
(orthogonal wavelet
basis functions by
dilating and translating
in discrete steps
Pressure
uctuation
Signal denoising Roy et al.
(1999)
Mallats pyramidal
algorithm used for
computing one-
dimensional orthogonal
wavelet transform
Pressure
uctuation
Signal ltered by a wavelet; peak frequency and value
corresponded to the bubble frequency and diameter,
respectively
Lu and Li
(1999)
Gabor wavelet Signal from an
acceleration
sensor
Continuous transform with a Gabor wavelet [cf. Eq. (29)]
applied to a commercial scale AFBC
Fukayama
et al. (2000)
At higher solid uxes, the core of the riser can be
observed using a hood or an endoscope as shown in
Fig. 40c,d. Examples of laser sheet images are shown in
Fig. 40e.
3.2.3 Particle Image Velocimetry (PIV)
If particle images are distinct in a time series of
visualized images, one can track the particle and
determine its velocity and trajectory. This is the
PIV technique. Rix et al. (1996) measured particle
movement in the freeboard using PIV. Reese and
Fan (1997) tracked local velocities of bubbles and
particles in gasliquid and gasliquidsolid uidized
beds using a continuous mode argon ion laser as a
light source, and a CCD camera for image record-
ing. Assuming linear movement during the time
interval between consecutive images, particles and
bubbles were tracked and their velocities determined.
More information on this technique can be found in
Chen and Fan (1992) and in Fan (1995). An experi-
mental setup used by Chen and Fan (1992) is shown
in Fig. 41.
3.3 Computer Tomography (CT) and Image
Reconstruction
Tomographic visualization is based on numerical
solution of a set of linear algebraic equations corre-
sponding to the transmission intensity of high-
energy electromagnetic radiation such as x- or ,-
rays or other beams (visible light, positrons, neu-
trons, ultrasound, magnetic resonance), and/or eld
parameters such as impedance, inductivity, or capa-
citance.
3.3.1 Image Reconstruction Principle
As for an example, let us suppose that for the construc-
tion of a tomographic image we use light of initial
intensity I
0
. According to the LambertBeer law, the
intensity of light after transmission through an object
of thickness L can be written as
I
L
= I
0
e
j,L
(33)
where j is the mass attenuation coecient, , is the
medium density, and accordingly j, is the linear
attenuation coecient of the medium the beam passes
through. L is the thickness, and the term j,L is the
absorptance of the medium.
Now, let us introduce cells from 1 to m in the objec-
tive cross section as shown in Fig. 42a. The cross sec-
tion is scanned by number of prex beams i n, and
the average passage length for cell j was I
ij
(for those
cells where beam i does not pass I
ij
= 0). The rst
medium fraction in cell j expressed by c
j
is the main
unknown variable. Then the total absorbance I
Li
,I
0
can be written as
I
Li
I
0
= exp
j
j
1
,
1
c
j
j
2
,
2
(1 c
j
)
_ _
l
ij
_ _
(34)
By rearranging the above expression we obtain
j
j
1
,
1
j
2
,
2
( )l
ij
c
j
= ln
I
Li
I
0
_ _
j
2
,
2
L
i
(35)
Thus we have a set of linear algebraic equations to
obtain c
j
:
(a
ij
)(c
j
) = (b
i
) (36)
where
a
ij
= j
1
,
1
j
2
,
2
( )l
ij
(37)
b
i
= ln
I
Li
I
0
_ _
j
2
,
2
L
i
(38)
Now let us study a more general case in which the
distribution of attenuation coecients, described by a
Figure 39 Examples of Gabor wavelet and Gabor function. (a) Gabor wavelet (dilated window); (b) Gabor function (xed
window width). (Sheng, 1996.) (With permission of CRC Press)
function f (x. y), is projected through a line L as
dened in Fig. 42b. The measured value P(l. ) ( is
the projection angle and l is the length, as in Fig. 42b)
can be written
P(l. ) =
_
L
f (x. y) ds (39)
When f (x. y) is a continuous function with a con-
tinuous rst derivative, Radon (1917) reported that
f (x. y) can be reconstructed from an innite set of
line integrals P(l. ) by
f (x. y) =
1
4
2
_
2
0
_
o
o
l
l x cos y sin
oP(l. )
ol
dl d
(40)
where (l xcos y sin ) is the perpendicular dis-
tance of the point (x. y) from the line l.
In reality the reconstruction is always done as an
approximation of Radons solution for a certain num-
ber of projections. As shown in Fig. 42b, for each angle
of projection , the value of P(l. ) depends on the
Figure 40 Laser sheet technique, image modication, and laser sheet picturing in dense suspension. (a) Scanning laser sheet
imaging (Tsukada et al., 1997); (b) horizontal proles of gray scales of the images before and after image modication (r
0
= 0.5
m, cellophane tape correction by interpolation) (Tsukada et al., 1997); (c) hood system of Kuroki and Horio (1994); (d)
endoscopic system of Werther and Rudnick (1996, courtesy of Prof. Werther); (e) 3D image obtained from scanning laser
sheet images (Horio and Ito, 1997)upper: turbulent bed (u
0
= 0.58 m/s; G
s
= 0.045 kg,m
2
s), and lower: fast bed (u
0
= 1.1
m/s, G
s
= 0.087 kg/m
2
s).
Figure 41 PIV setup of Chen and Fan (1992). With permission of Elsevier Science.
Figure 42 An algebraic approach to image reconstruction in CT (a) and the geometry for CT image reconstruction (b).
(Adopted from Kumar and Dudukovic, 1997.) With permission of Elsevier Science.
distance on line l. The back-projected function, b(x. y),
for a certain is
b(x. y) =
_
P(l. )o(x cos y sin l) dl (41)
where o is the Dirac delta function.
The complete back-projection, being an approxima-
tion of the function f (x. y), is the sum of b(x. y) corre-
sponding to all angles , at which the projections were
made:
f (x. y)
approx
=
_
0
b(x. y) d
=
_
0
_
P(l. ) o(x cos y sin l) dl d
(42)
Normalization of f (x. y)
approx
is achieved by multiply-
ing by (1,2).
The quality of image reconstruction depends on the
number of projections and detectors, the speed of mea-
surement and of the data acquisition system, and the
homogeneity of the measurement volume. More on
image reconstruction can be found, e.g., in Kak and
Slaney (1988); Kalender (2000); and Kumar and
Dudukovic (1997).
The rst reported application of tomography was
probably that of Grohse (1955), who investigated the
variation in density of a bed of silicon powder as a
function of uidizing velocity. At almost the same
time Bartholomev and Casagrande (1957) used
Co
60
as ,-ray source to monitor a 20.4
//
i.d. FCC
unit and detected the ,-rays by a GeigerMuller
tube.
In capacitance tomography, the measurement
setup consists of an array of electrodes installed
around the external surface of the monitored system,
and measurements of capacitance between various
pairs of electrodes are performed, so that data is
received on the capacitance in various directions
within the system. Then, from the algorithm
described above, the volume fraction of solids or
gas within the system can be calculated. Examples
of various sources for uidized bed CTs are presented
in Table 14. An example setup and results are shown
in Fig. 43.
Positron emission tomography (PET) uses radioac-
tive tracers that decay via the emission of a positron.
It results in the production of two back-to-back ,-
rays, from which the tracer position can be deter-
mined. More information can be found, e.g., in
Hawkesworth et al. (1991). Another noninvasive
method is nuclear magnetic resonance (NMR) imag-
ing. The measurements are conducted based on the
paramagnetic properties of atom nuclei. During the
experiments, external signals (i.e., radio frequency
pulses and magnetic eld gradient pulses) interact
with the nucleus spin system positioned in a static
magnetic eld. NMR imaging enables us to observe
dense ows, but it can only be performed with water-
containing systems. Details can be found, e.g., in
Pangrle et al. (1992), Shattuck et al. (1997), and
Sederman et al. (2001).
An advantage of tomography is its capability of
monitoring a relatively large volume at the same
moment. However, safety considerations concerning
particularly the emission of x-rays or ,-rays during
tomographic scanning have to be kept in mind. Using
x-rays instead of ,-rays enables adjusting the energy of
the emitted radiation by changing the voltage of cath-
ode of an x-ray generator. A disadvantage of the tomo-
graphic imaging is the diculty of transforming and
analyzing the huge amount of data from the scanner,
but faster computers should solve this diculty.
3.3.2 Neutron Imaging
The interiors of uidized bed systems can also be
observed by application of the neutron imaging tech-
nique. This method can provide fast monitoring of
relatively large cross-sectional areas at high resolu-
tion. The disadvantage of neutron imaging is the
necessity of taking special precautions concerning
the application of neutrons and the necessity of
using a special neutron source. By this method
Onodera et al. (1998) observed the ow pattern
around the tube and bubble movement within a ui-
dized bed. Tasdemir et al. (1999) used a panoramic
camera and image segmentation to study mixing in a
0.2 0.2 m
2
cross-section uidized bed consisting of
three types of particles of various colors. Ozawa et al.
(1998) applied neutron radiography visualization to a
rectangular 30 60 100 cm uidized bed. Since they
used sand as the bed solids (sand is not able to cap-
ture the neutrons itself), 0.3% of specially prepared
sand (i.e., coated with cadmium) was added to cap-
ture the neutrons. More on neutron imaging can be
found, e.g., in Fredd et al. (1997) and Umekawa et al.
(1999a).
3.4 On-Line Particle Size Distribution Measurement
Techniques for in-situ determination of particle size
and shape have been developed in the eld of granula-
tion (e.g., Tanino et al., 1993). There are three dierent
methods of monitoring the granule size distribution. In
the easiest method, the granules are sampled by com-
pressed air and captured on the adhesive tape. Then
the particle image on the adhesive tape is taken with a
CCD camera, and nally the sampling tube is cleaned
by air, the spent part of the adhesive tape is wound up,
and new samples are sucked into the tube. In another
system, laser scattering of sampling frequency up to 50
Hz is applied for size determination. Since the light
scattered by the particles in the measurement volume
has a scatter angle unique for the particle diameters,
the particle size can be determined by a coaxial multi-
circular detector. Another type of instrument provides
real images taken by a CCD camera in front of the
probe. The measurement system consists of illumina-
tion, purging, and telescope devices, and the image
processing for determining the mean particle size can
be done relatively fast. More information is given, e.g.,
in Fluidization Handbook (1999).
4 FLUIDIZED BED DIAGNOSTICS
Combined with the proper knowledge of uidization
systems, the instrumentation and measurement tech-
niques can provide us with much meaningful informa-
tion. Since information in other chapters should be
Table 14 Examples of Fluidized Bed CTs
Author(s) Source Type Remarks
Weimer et al. (1985) ,-ray Monitoring the voidage along the height of a FB using Cs
137
as
radiation source
Seville et al. (1986) ,-ray Studies on voidage distribution above the gas distributor in a
0.146 m i.d. uidized bed of sand
Seo and Gidaspow (1987) ,-ray, x-ray Voidage measurement in a FB using simultaneous densitometry
Weinstein et al. (1984) x-ray Application of tomography in a CFB, using an optical densitometer
to analyze the images
Weinstein et al. (1992) x-ray Studies on the distribution of solid fraction in a fast uidized bed
system (d = 0.15 m)
Contractor et al. (1992) x-ray Observation of images from uidized beds using x-ray source and
phototransistors
Kantzas (1994) x-ray Images of density distribution and gas holdup obtained for
uidized bed of glass beads and polyethylene uidized by
nitrogen
Holobo et al. (1995) x-ray CT scanner used to study variability of gas voidage in a
polyolen/air uidized bed of 0.1 m i.d
Durand et al. (1995) x-ray Imaging technique combined with video camera to study the
hydrodynamics of a polyethylene FB
Fiorentino and Newton
(1998)
x-ray Identifying of scale-up issues for predicting large-scale BP reactor
performance
Grassler and Wirth
(1999)
x-ray Application of a 60 kV CT scanner to a cold 0.19 m i.d., 15 m
high CFB, results obtained with 0.2 mm resolution
Kai et al. (2000) x-ray Dynamic imaging of cross-sectional voidage distribution in every
4 ms by a fast scanning system consisting of 18 x-ray sources
and 122 detectors
Huang et al. (1989) Capacitance Studies on voidage distribution in the uidized bed
Halow and Nicoletti
(1992)
Capacitance Studies on voidage distribution in the uidized bed, bed diameter
0.15 m, sensing electrodes installed round the riser in four rings,
each containing 32 electrodes
Dyakowski et al. (1997) Capacitance Application to CFB, good agreement with comparison of voidage
calculated from pressure transducers
Rhodes and Wang (1999) Capacitance Studies on distribution of solids volume fraction in a 0.09 m i.d
riser, four sets of sensors applied, each sensor contained 12
electrodes
Figure 43 CT scanning system of Halow et al. (1993) (a) (with permission of Elsevier Science) and pseudo 3D image of bubbles and their interface structure (Kai et
al., 1999.) (b)
referred to for this purpose, only fundamental remarks
are presented in this section.
4.1 Diagnostics of Bubbling Bed
4.1.1 Bed Diagnostics by Pressure Measurements
The relationship between bed pressure drop p and
supercial gas velocity, as well as the pressure distribu-
tion along the bed determined by multiple pressure
sensors on the column wall, can provide fruitful infor-
mation on the uidizing conditions. As illustrated in
Fig. 44, the longitudinal pressure distribution curve
can indicate: (1) the bed surface location and uidized
solids holdup, (2) segregation of bed materials, and/or
(3) the height of the not completely uidized region
above the distributor. The p vs. u
0
relationship can
of course also indicate (1) the minimum uidization
velocity of the solids in the bed, (2) coexistence of
deuidized coarse/heavy particles, and (3) existence
of channels or rat holes.
Pressure uctuation can provide information on
bubbling conditions and slugging. For slugging, the
frequency f
s
, of pressure uctuation is the frequency
of slugs passing by the pressure probe. Suppose slug
length is approximately equal to the column diameter,
we can write
f
s
=
(u
0
u
mf
)(D
2
t
,4)
V
s
=
u
0
u
mf
D
t
(43)
where D
t
is column diameter and V
s
is slug volume.
On the other hand, the major frequency f
n
of spon-
taneous oscillation of the system composed of the
plenum chamber capacity, the distributor resistance,
and the bed inertia, particularly when the distributor
resistance is small, is given by (Moritomi et al., 1980)
o
n
= 2f
n
=
P
D
A
t
( ), ,
p
1 c
mf
( )L
mf
V
d
_ _
_
rad,s
(44)
where A
t
, L
mf
, P
D
, V
d
, c
mf
, and ,
p
are bed cross-
sectional area, bed height at the minimum uidization
condition, time averaged plenum chamber pressure,
plenum chamber volume, bed voidage at the minimum
uidization condition, and particle density, respec-
tively.
Pressure uctuation can be also used to monitor the
gassolid contact mode and to adjust the particle size
distribution. Figure 45 is a good example in a catalytic
reactor (Ikeda, 1972).
Bed pressure drop can be measured either directly
by a pressure tap placed right above the gas distri-
butor such as a vertical pressure probe, a wall pres-
sure tap and/or a distributor buried type, or
indirectly from the pressure of the plenum chamber
by subtracting the distributor pressure drop measured
separately. Pressure taps on the freeboard wall are
also necessary in case the gas outlet pressure drop
cannot be negligible. To avoid gas dynamic pressure
the hole of a pressure tap must face parallel to the
main gas ow. To prevent particle entering, one
needs to use some ltering material such as a wire
gauze (Figs. 46a and 46c) or to slant the pressure tap
tube (Fig. 46b). Figure 46d shows immersion type
probes, which can be traversed to measure the pres-
sure distribution.
4.1.2 Deuidization Velocity
Deuidization is caused by the accumulation of large
or heavy particles in the bed owing to the increasing
size of the solids because of agglomeration, coating, or
the feed of a large size fraction. Deuidization is
serious in commercial-scale operations (e.g.,
Anthony, 1995; Souto et al., 1996; Skrifvars et al.,
1997) and its symptom has to be detected as early as
possible. The deuidization can be detected by moni-
toring the pressure drop and its uctuation in the bed.
When deuidization takes places, the pressure drop
suddenly decreases, as reported by Siegell (1984).
However, much before that, the accumulation of
large particles at the bottom of the bed can be detected
by the decrease of the pressure drop and the longitu-
dinal pressure gradient. The temperature of deuidiza-
tion depends strongly on gas velocity, as reported, e.g.,
by Siegell (1984).
4.1.3 Bubble Fraction and Visible Bubble Flow
Distribution
The local bubble fraction can be determined by
either an optical or a capacitance probe. Suppose
that a probe output as shown in Fig. 47 is obtained.
By introducing a threshold value, which has to be
carefully determined by the real eye observation
and/or material balance, the time can be divided
into bubble passage periods T
bi
(i = 1. 2. . . .) and
emulsion phase passage periods T
ei
. The sum of
bubble passage periods divided by the total observa-
tion period (T
bi
T
ei
) corresponds to the bubble
fraction c
b
:
c
b
=
T
bi
(T
bi
T
ei
)
(45)
The local visible bubble velocity u
vb
is given by mul-
tiplying the voidage by bubble velocity u
b
measured by
the two needle probes or by a borescope or other image
sensors so that
u
vb
= u
b
c
b
=
u
bi
T
bi
(T
bi
T
ei
)
(local time average)
(46)
By traversing the point probe across the bed one can
obtain the bubble ow distribution.
4.1.4 Bubble Size Calculation from Pierced Length
The length of the part of a bubble pierced by a probe
l
bi
(pierced bubble length) is:
Figure 44 Determination of uidization regime from pressure data. P
+
: theoretical pressure drop = (bed weight-buoyancy
force)/(total cross section).
Figure 45 Pressure uctuations versus time. (Ikeda, 1972.)
l
bi
= u
bi
T
i
(47)
However, l
bi
does not immediately give the bubble dia-
meter. To obtain the real bubble diameter, D
b
, from a
series of data for l
bi
, let us rst assume (as a simplest
case) that the bubbles are spherical and uniform in
size and that they are uniformly distributed radially
across the bed. The probability, (D
b
. l
b
) dl
b
, of
having a pierced bubble length within the interval of
l
b
l
b
dl
b
for bubbles of diameter D
b
is 8r[dr[,D
2
b
,
where the radius, r, is dened by r
2
= (D
2
b
l
2
b
),4
(accordingly we have r dr = l
b
dl
b
,4). Then, since
8r[dr[,D
2
b
= 2l
b
dl
b
,D
2
b
, and since l
b
is less than D
b
,
the probability distribution density function for bubble
size (D
b
. l
b
) is given by
(D
b
. l
b
) =
2l
b
D
2
b
for 0 - l
b
- D
b
(48-1)
and
(D
b
. l
b
) = 0 for l
b
- 0 or D
b
- l
b
(48-2)
Since (D
b
. l
b
) has to satisfy the normalization condi-
tion, we can write
_
D
b
0
(D
b
. l
b
)dl
b
= 1 (49)
Thus the time averaged pierced length is given by
l
b
=
_
D
b
0
D
b
. l
b
( )l
b
dl
b
=
2
3
D
b
(50)
Finally by rearranging the above expression, the bub-
ble size is given by
D
b
=
3
2
N
i=1
l
3
bi
N
_ _
1,3
(51)
where N is the number of bubble samples.
For a more general case, where the bubble size is
distributed (Tsutsui and Miyauchi, 1979), let us
assume a distribution density function for bubble
diameter, (D
b
). In this case, (D
b
. l
b
) introduced
above in Eq. (48) becomes a conditional probability
to get a pierced length l
b
from a bubble of size D
b
.
What we need is the probability density distribution
Figure 46 Various types and arrangements of pressure measurement tubes.
Figure 47 Type of signal from capacitance probes. (Lanneau 1960.) With permission of the Institution of Chemical Engineers.
of pierced length l
b
for the case where D
b
is distrib-
uted,
+
(l
b
).
+
(l
b
) can be related to (D
b
) and
(D
b
. l
b
) by
+
(l
b
) dl
b
=
_
o
l
b
(D
b
. l
b
) dl
b
(D
b
) dD
b
(52)
Dividing both sides by dl
b
, we obtain
+
(l
b
) =
_
o
l
b
(D
b
. l
b
)(D
b
)dD
b
(53)
Dierentiating it by l
b
, we obtain
d
+
dl
b
= (l
b
. l
b
)(l
b
) (54)
Finally, substituting the above expression into the nor-
malization condition
_
o
0
(D
b
) dD
b
= 1 (55)
we obtain
(D
b
) =
1
(D
b
. D
b
)
d
+
(D
b
)
dD
b
_ _
_
o
0
1
(l
b
. l
b
)
d
+
dl
b
_ _
dl
b
(56)
where
+
(l
b
) is measured and (D
b
. l
b
) can be calcu-
lated for a bubble of particular shape. Then the aver-
age bubble size is given by
D
b
=
_
o
0
(D
b
)D
3
b
dD
b
_ _
1,3
(57)
If bubbles are not spherical, the same procedure
can be applied by changing (D
b
. l
b
). Cluster size mea-
surements can be done on the same principle, but they
have a more complicated structure.
4.1.5 Solid Circulation and Mixing
Solid tracers and probes discussed in Secs. 2.11, 12
can be applied to determine solid circulation and mix-
ing. If real solids sampling is conducted, bed sectio-
nizing is a useful toolafter a certain uidization
period the gas is cut o with a sequencer and the
partitioning device is dropped, so that it can quickly
reach the bed bottom. Then the bed particles are
sampled section by section by a vacuum cleaner
(Horio et al., 1986a).
4.1.6 Gas Sampling from the Bed
To sample gas from uidized beds it is important to
know if the sample is taken from the bubble phase,
emulsion phase, and/or cluster or lean phase. There
are elaborated probes with an pressure sensor to detect
the phase from which the samples are taken, as dis-
cussed in Sec. 2.13. Another option is the eective
removal of solids from the gassolid suspension by
using a small cyclone. Ceramic or metal lters can
also be applied. However, some precaution must be
taken in order not to accumulate too much chemically
active solid on the lter.
4.2 Diagnostics of Circulating Fluidized Bed
4.2.1 Solid Holdup Distribution
Solid holdup in a CFB can be determined from the
time-averaged pressure gradient, op,oz, based on Eq.
(58) applicable except for the acceleration region (the
latter should not be too high in circulating uidized
beds, cf. Horio, 1997):
op
oz
= ,
p
(1 c)g (58)
where z (m) is height, while g (m,s
2
), ,
p
(kg,m
3
), and c
are gravity acceleration, particle density, and bed voi-
dage, respectively.
4.2.2 Solid Circulation Rate
Overall solids mass ux is one of the key parameters to
determine the regime of circulating uidization. It also
has a great inuence on the pressure prole along the
riser, as well as voidage and mixing. Table 15 sum-
marizes the previous methods of measuring overall
solids mass ux.
Solids circulation rate can be measured directly at
the downcomer by devices such as those shown in
Fig. 48. Then the overall solid mass ux is obtained
by
G
s
=
w
s
A
kg,m
2
s (59)
where w
s
is the solids mass ow rate and A is riser cross
section.
The solids circulation rate can also be determined
from the observed local time-averaged solids velocity,
v
p
, and solids concentration, (1 c), by the following
integration over the bed cross section:
W
s
=
_ _
A
,
p
(1 c)v
p
dA (60)
If the local pressure is measured in parallel with
overall solids circulation, the cross-sectional average
solids concentration (1 c) can be determined know-
ing the value of G
s
, particle density ,
p
, and velocity v
p
.
Otherwise, local solids ux can be determined by suck-
ing solids from the riser as described in detail in Sec.
2.12. More on the techniques for measuring solid mass
uxes can be found, e.g., in Burkell et al. (1988).
4.3 Diagnostics of Particle Characteristics
Although particle characterization is not an ordinary
instrumentation issue, some supporting laboratory
scale equipment helps even the commercial operation
by dening the realistic nature of the particles in each
occasion. Thus it is recommended to have a small, say
5 cm in diameter, cold test unit with a owmeter and
pressure sensors.
4.3.1 Minimum Fluidization Velocity
Minimum uidization velocity is determined from bed
pressure drop vs. gas velocity relationship for decreas-
ing gas velocity, as already described in Fig. 44a. It
should be recognized that the sample must be taken
carefully so that it represents the real solids in the
bed. In many cases the bed material is already quite
dierent from the one fed to the bed by reaction, attri-
tion, or agglomeration. Bed pressure drop has to be
measured for decreasing velocity from a completely
uidized condition. When bed pressure drop is mea-
sured for increasing gas velocity, pressure drop vs. gas
velocity curves show higher pressure because of
changes of particle interaction and bed structure and
do not give a reproducible and generally applicable
value of minimum uidization velocity.
Table 15 Various Solids Mass Flux Measurement Methods
Method Author(s) Remarks
Installation of a buttery
or slide valve(s) in the
downcomer
Lasch et al. (1988)
Bader et al. (1988)
Hartge et al. (1988)
Beaud and Louge (1995)
Measurements based on periodical opening/closing of the valve(s)
Dicult to apply to hot FB due to valve requirements and its
damage possibility at high temperatures
Disc valve Ye et al. (1999) Measurements of solids mass ux in a CFB of cast steel
Three-way valve
technique
Arena et al. (1988, 1993) Measurements based on switching over the three-way valve
(placed outside the hot reactor) and accumulating solids in a
vessel through the certain time period. No continuous
measurements possible
Two ring sensors
installed in the CFB
Dybeck et al. (1995) Measurement method based on eect of electrostatic induction of
charged particles on a surrounding metallic ring
recirculating line Qi and Farag (1995) Measurements of solids uxes in a 0.14 m i.d. CFB riser of glass
beads and FCC. The solids were collected in each probe for a
given time period and then weighed. Local particle concentration
was measured using the capacitance probe
Study on ve ways of
solids mass ux
measurement systems
Burkell et al. (1988) Study on application of closing valve in the return leg, observation
of an identiable particle, using a device to record the force
imparted by returning solids form the cyclone, measuring the
pressure drop across the constriction in the return loop, and
estimating solids mass ow from the heat balance on a
calorimetric section in the standpipe
Semicontinuous solids
mass ux meter
Horio et al. (1992b) Application of load cell. The solids were accumulated in the device
through a certain time period. Dicult application to hot units
due to load cell requirements
Patience et al. (1990) Idea based on measurements of pressure drop and gas velocity.
Assuming that gassolid slip velocity is related to the terminal
velocity of individual particles, the mass ux was calculated.
Davies and Harries (1992) Idea based on application of weighing chamber and an
electronic balance
Kobylecki and Horio
(2002)
Scoop-like device measuring continuously the mass uxes of
various solids including group C
Whether a bed is completely uidized or not can be
examined from the bed pressure drop at the velocity
slightly higher than the minimum uidization velocity.
If it is, the pressure drop should equal the bed weight
divided by the bed cross section when gas density is
suciently low that the buoyancy force acting on par-
ticles can be neglected (cf. P
+
in Fig. 44).
If the particles have a wide size distribution, the
pressure drop vs. gas velocity curve shows a rather
vague transition from a uidized condition to the
xed bed condition, as already shown in Fig. 44b.
From this gure one can obtain gas velocities corre-
sponding to the beginning of uidization u
bf
and to the
completion of uidization u
cf
.
4.3.2 Agglomerating Behavior of Cohesive Powders
As shown in Fig. 49, channel formation and its col-
lapse and transition to the agglomerating uidization
can be detected from the cyclic response of the bed
pressure drop. However, it should be noted here that
in the cohesive powder uidization the wall eects are
quite important, and the scale eect related to them
cannot be neglected.
Very ne particles are uidized forming agglomer-
ates. Minimum uidization velocity of such agglomer-
ates u
a.mf
can be measured in the same manner as in the
case of noncohesive particles, i.e., from the decreasing
velocity period (period after #7 in Fig. 49). A particle
recycling device, such as a cyclone or a lter, is neces-
sary for ne powder experiments. From u
a.mf
, the
apparent agglomerate size d
a
can be determined from
minimum uidization velocity correlations such as
Wen and Yus (1966) by using the particle density
separately determined by mercury porosimetry or
other methods.
The agglomerate size of Geldarts group C powders
can be roughly determined by sieve analysis. In such
cases, Ro-tap sieving may not be advisable because its
action also helps agglomeration or attrition. If direct
determination of agglomerate size is critical for opera-
tion, as, e.g., in spray granulation, the in situ particle
size determination methods already introduced in Sec.
3.4 should be helpful.
4.3.3 Expansion Characteristics
High bed expansion is an important feature of powders
suitable for good gassolid contacting. To quantify the
expansion characteristics, the bed contraction mea-
surement test is recommended as a standard method
in the laboratory. In this test the gas supply to the bed
is stopped after a certain uidization period and the
change of bed height is recorded. During the measure-
ments some attention has to be paid to the eects of
plenum chamber volume and the distributor resistance
on the result. Figure 50 shows the observed bed con-
traction curves from which the emulsion phase voidage
can be obtained.
For Geldarts group A powders, the bed expands
homogeneously even above the minimum uidization
velocity, u
mf
, until the minimum bubbling velocity u
mb
.
The u
mb
is another good index of high emulsion phase
expansion, as shown in Fig. 51. This gure also illus-
trates how u
mb
is determined.
4.3.4 Regime Transition Velocities
In addition to u
mf
and u
mb
, other regime transition
velocities such as u
c
, u
k
, and u
tr
should be important
for high-velocity operation. Figure 52 shows the pres-
sure uctuation response to gas velocity change, from
which u
c
and u
k
are determined. These characteristic
Figure 48 Solids mass ow meters: (a) Horio et al. (1992b), (b) Kobylecki and Horio (2002).
transition velocities should be determined with a col-
umn where slugging can be avoided. Furthermore, to
avoid criticism as for the early studies (Geldart and
Rhodes, 1986), the axial solid holdup should be
always monitored, so that the dense bed is kept in
the column, preventing the pressure tap from expo-
sure to the lean phase, which causes the decrease in
pressure uctuation at higher gas velocities and gives
ctitious values for u
c
and u
k
. For this issue readers
are recommended to check Horio (1997, pp 3436). A
careful evaluation of the regime transition shows that
u
c
, u
k
, and u
tr
are independent and unique character-
istic velocities.
5 SUMMARY
Instrumentation and measurement techniques for
uidized bed processes are reviewed, starting from
Figure 49 Pressure drop response and uidization behavior of Al
2
O
3
powder (5.0 jm) bed. (Nishii et al., 1993.)
Figure 50 Fluidized bed collapse of group A powder (glass ballotini, 26 jm), reported by Geldart (1986) (#John Wiley & Sons.
Reproduced with permission) (a), and collapsing of uidized bed reported by Tung and Kwauk (1982) (b).
fundamental knowledge on sensor elements, probe
conguration, and tracers to more sophisticated
image analysis systems. Data to evaluate sensing sys-
tems are included as much as possible for readers con-
venience. In the last part of the chapter a practical
approach to the diagnostics of uidization systems as
well as powders are discussed.
ACKNOWLEDGMENTS
The authors thank Ms. Malgorzata Kobylecka for her
support in preparing the manuscript.
REFERENCES
Abrahamsen AR, Geldart D. Behaviour of gas-uidized beds
of ne powders, Part II. Power Technol 26:4755, 1980.
Acree Riley C, Louge M. Quantitative capacitance measure-
ments of voidage in dense gassolid ows. Particulate Sci
Technol 7:5159, 1989.
Adams C. Gas mixing in fast uidized beds. In: Basu P,
Large J, eds. Circulating Fluidized Bed Technology II.
Oxford: Pergamon, 1988, pp 299306.
Akagi H, Furusaki S. Direct measurement of gas con-
centration in bubbles in a uidized bed. Prep of the
Meeting of Soc Chem Eng Japan. Shinshu, Japan, 1983,
pp 4647.
Ambler P, Milne B, Berruti F, Scott D. Residence time dis-
tribution of solids in a circulating uidized bed: experi-
mental and modeling studies. Chem Eng Sci
45:21792186, 1990.
Figure 51 Expansion curve for group A powder (60 jm ballotini). (Abrahamsen and Geldart, 1980.) With permission of
Elsevier Science.
Figure 52 Pressure uctuation vs. gas velocity. (Yerushalmi
and Cankurt, 1979.) (With permission of Elsevier Science)
Amos G, Rhodes M, Benkreira H. Calculation of optic bers
calibration curves for the measurement of solids volume
fractions in multiphase ows. Powder Technol
88:107121, 1996.
Anthony E. Fluidized bed combustion of alternative solid
fuels: status, successes and problems of the technology.
Prog Energy Combust Sci 21:239268, 1995.
Arastopour H, Yang Y. Experimental studies on dilute gas
and cohesive particle ow behavior using a laserDoppler
anemometer. In: Potter O, Nicklin D, eds. Fluidization
723730.
Arena U, Cammarota A, Massimilla L, Pirozzi D. The
hydrodynamic behavior of two circulating uidized bed
units of dierent sizes. In: Basu P, Large J, eds.
Pergamon, 1988, pp 223230.
Arena U, Cammarota A, Marzocchella A, Massimilla L.
Hydrodynamics of a circulating uidized bed with second-
ary air injection. Proc 12
th
FBC Conf San Diego, 1993,
pp 899905.
Atkinson C, Clark N. Gas sampling from uidized beds: a
novel probe system. Powder Technol 54:5970, 1988.
Avidan A, Yerushalmi J. Solids mixing in an expanded top
uid bed. AIChE J 31:835841, 1985.
Bachalo W, Houser M. Phase Doppler spray analyzer for
simultaneous measurements of drop size and velocity
distributions. Optical Engineering 23(5):583590, 1984.
Bader R, Findlay J, Knowlton T. Gas/solid ow patterns in a
30.5 cm diameter circulating uidized bed. In: Basu P,
Bai D, Shibuya E, Masuda Y, Nakawaga N, Kato K. Flow
structure in a fast uidized bed. Chem Eng Sci 51:957
966, 1996.
Bakker P, Heertjes P. Porosity measurements in a uidised
bed. British Chem Eng 3(May):240246, 1958.
Bartholomew RN, Casagrande RM. Measuring solids con-
centration in uidized systems by gamma-ray absorption.
Ind Eng Chem 49(3):428431, 1957.
Basu P, Knouche F. Radiative heat transfer from a fast ui-
dized bed. In: Basu P, Large J, eds. Circulating Fluidized
Bed Technology II. Oxford: Pergamon, 1988, pp 245253.
Beaud F, Louge M. Similarity of radial proles of solid
volume fraction in a circulating uidized bed. Prep. of
Fluidization VIII, Tours, France, 1995, pp 97104.
Benton D, Parker D. Non-medical applications of positron
emission tomography. In: Chaouki J, Larachi F,
Dudukovic M, eds. Non-Invasive Monitoring of
Multiphase Flows. New York: Elsevier, 1996, pp 161
184.
Boland D, Geldart D. Electrostatic charging in gas uidized
beds. Powder Technol 5:289297, 1971/72.
Borlai O, Hodany L, Blickle T. Proc Int Symp Fluidization.
Drinkenburg A, ed. Amsterdam: University Press, 1967,
pp 433441.
Bottcher C. Theory of Electric Polarization. New York:
Elsevier, 1952.
Botterill J. Fluid-Bed Heat Transfer. New York: Academic
Press, 1975.
Botterill J, Teoman Y, Yuregir K. Factors aecting heat
transfer between gas-uidized beds and immersed sur-
faces. Powder Technol 39:177189, 1984.
Brereton C, Grace J. Microstructural aspects of the behavior
of circulating uidized beds. Chem Eng Sci 48:25652572,
1993.
Brereton C, Stromberg L. Some aspects of the uid dynamic
behavior of fast uidized beds. In: Basu P, ed. Circulating
Fluidized Bed Technology. Oxford: Pergamon, 1986, pp
133144.
Brewster B, Seader J. Nonradioactive tagging method for
measuring particle velocity in pneumatic transport.
AIChE J 26:325330, 1980.
Briens C, Mirgain C, Bergougnou M, Pozo M, Loutaty R.
Evaluation of gassolids mixing chamber through cross
correlation and Hursts analysis. AIChE J 43:14691479,
1997.
Broadhust T, Becker H. Measurement and spectral analysis
of pressure uctuations in slugging beds. In: Kearins D,
ed. Fluidization Technology. Vol. 1. Washington:
Hemispher Publishing Co., 1976, pp 6385.
Brunier E, Yemmou M, Evrard R. Flux particles measure-
ments in industrial CFB. In: Werther J, ed. Proc 6
th
Conference on CFB. Wurzburg, Germany, 1999, pp
825830.
Burkell J, Grace J, Zhao J, Lim C. Measurement of solids
circulation rates in circulating uidized beds. In: Basu P,
Campbell C, Wang D. A particle pressure transducer suitable
for use in gas-uidized beds. Meas Sci Technol 1:1275
1279, 1990.
Cankurt N, Yerushalmi J. Gas backmixing in high velocity
uidized beds. In: Fluidization. Cambridge: Cambridge
Chau K, Goehner R, Drubetsky E, Brady H, Bayles W Jr,
Pedersen P. Pressure and sound measurement. In:
Webster J, ed. The Measurement, Instrumentation, and
Sensors Handbook. New York: CRC Press LLC, 1999,
Chap. 26.
Chen JC. Experiments that address phenomenological
issues of fast uidization. Chem Eng Sci 54:55295539,
1999.
Chen R, Fan LS. Particle image velocimetry for characteriz-
ing the ow structure in three-dimensional gasliquid
solid uidized beds. Chem Eng Sci 47(13/14):36153622,
1992.
Cheremisino N. Review of experimental methods for
studying the hydrodynamics of gassolid uidized beds.
Ind Eng Chem Process Des Dev 25:329351, 1986.
Cheremisino N, Cheremisino P. Hydrodynamics of Gas-
Solids Fluidization. New York: Gulf, 1984.
Cocco R, Cleveland J, Harner R, Chrisman R. Simultaneous
in-situ determination of particle loadings and velocities in
a gaseous medium. AIChE Symp Ser 91(308):147153,
1995.
Contractor R, Pell M, Weinstein H, Feindt H. The rate of
solid loss in a circulating uid bed following a loss of
circulation accident. In: Potter O, Nicklin D, eds.
1992, pp 243248.
Couturier M, Steward F, Poolpol S. Experimental
determination of heat transfer coecients in a 72 MWth
circulating uidized bed boiler. Proc 12th FBC Conf, San
Diego, CA, USA, 1993, pp 12151222.
Dahlin R, Pontius D, Haq Z, Vimalchand P, Brown R,
Wheeldon J. Plans for hot gas cleanup testing at the
power systems development facility. Proc 13th FBC
Conf, Orlando, FL, USA, 1995, pp 449459.
Davies C, Harris B. A device for measuring solids ow-
rates: characteristics and application in a circulating
uidized bed. In: Potter O, Nicklin D, eds.
1992, pp 741748.
De Kok J, Stark N, van Swaaij W. The inuence of solids
specic interfacial area on gassolid mass transfer in gas
uidised beds. In: Ostergaard K, Sorensen A, eds.
1986, pp 433440.
Drain L. The Laser Doppler Technique. New York: John
Wiley, 1980.
Dry R. Radial particle size segregation in fast uidized beds.
Dry R, White R, Close R, Joyce T. The eect of CFB scale-
up on gassolid contact eciency. Prep of the
Du B, Wei F, Wang Z. Eect of particle size on solids mixing
in a FCC riser. In: Werther J, ed. Proc. 6th Conference on
CFB. Wurzburg, Germany, 1999, pp 405410.
Durand D, Llinas R, Newton D. Optimisation of the
bubbling in a polyethylene uid bed process through
modelling of experimental test rig data. Prep
Fluidization VIII, 1995, pp 923935.
Durst F, Melling A, Whitelaw J. Principles and Practice of
Laser-Doppler Anemometry, 2d ed. New York: Academic
Press, 1981.
Dyakowski T, Edwards R, Xie C, Williams R. Application of
capacitance tomography to gassolid ows. Chem Eng Sci
52(13), 1997.
Dybeck K, Nagel R, Schoenfelder H, Werther J, Singer H.
The D ring sensora novel method for continuously
measuring the solid circulation rate in CFB reactors.
Prep Fluidization VIII, Tours, France, 1995, pp 916.
Ebert TA, Glicksman LR, Lints M. Determination of particle
and gas convective heat transfer components in a circulat-
ing uidized bed. Chem Eng Sci 48:21792188, 1993.
Edgar T, Smith C, Shinskey G, Gassman G, Schafbuch P,
McAvoy T, Seborg D. Section 8: process control. In:
Perry R, Green D, Maloney J, eds. Perrys Chemical
Engineers Handbook, 7
th
ed. New York: McGraw-Hill,
Inc., 1997.
Fan LS. Bubble dynamics in liquidsolid suspensions.
AIChE Symp Ser 91(308):143, 1995.
Fan LT, Ho T, Hiraoka S, Walawender W. Pressure uctua-
tions in a uidized bed. AIChE J 27:388396, 1981.
Fan LT, Neogi D, Yashima M, Nassar R. Stochastic analysis
of a three-phase uidized bed: fractal approach. AIChE J
36(10):15291535, 1990.
Fasching G, Smith N. A particle impact probe for hydrody-
namic studies in uidized beds. Powder Technol 75:2127,
1993.
Fasso L, Chao B, Soo S. Measurement of electrostatic
charges and concentration of particles in the freeboard
of a uidized bed. Powder Technol 33:211221, 1982.
Fiedler O, Werther J, Labahan N, Kumpart J, Christofori K.
Measurement of local velocities and velocity distributions
in gassolid ows by means of the spatial lter method.
Fiorentino M, Newton D. Application of x-ray imaging to
uidized bed scale-up. In: Fan LS, Knowlton T, eds.
1998, pp 589597.
Fredd C, Fogler H, Lindsay J. Neutron transmission tomo-
graphy applied to reactive dissolution through percolating
porous media. In: Chaouki J, Larachi F, Dudukovic M,
eds. Noninvasive Monitoring of Multiphase Flows. New
York: Elsevier Science, 1997, pp 185211.
Fujikura Ltd., Optical Fiber Communication System,
Catalog No. T1160-8, 1988, p 51.
Fukayama Y, Horio M, Matsumoto K, Shimohira K. Int
Meas Cond IVI IMEKO World Congress, Vienna,
Austria, 2000.
Gbavcic Z, Vukovic D, Zdanski F. Tracer particle movement
in a two-dimensional water-uidized bed. Powder
Technol 62:199201, 1990.
Geldart D. Single particles, xed and quiescent beds. In:
Geldart D, ed. Gas Fluidization Technology. New
York: John Wiley, 1986, pp 1132.
Geldart D, Rhodes M. From minimum uidization to pneu-
matic transporta critical review of the hydrodynamics.
In: Basu P, ed. Circulating Fluidized Bed Technology.
Godfroy L, Larachi F, Kennedy G, Chaouki J. Simulation
measurement of 3-D position and velocity of a single
radioactive particle in a CFB riser at high velocity. Proc
CFB 5, Beijing, China, 1996, pp 633638.
Goedicke F, Reh L. Particle induced heat transfer between
walls and gassolid uidized beds. AIChE Symp Ser
89(296):123136, 1992.
Grace J, Baeyens J. Instrumentation and experimental tech-
niques. In: Geldart D, ed. Gas Fluidization Technology.
New York: John Wiley, 1986, pp 415462.
Grassler T, Wirth KE. X-ray computed tomographya non-
intrusive technique to characterize local solids distribution
in a circulating uidized bed (CFB). In: Werther J, ed.
Proc 6
th
Conference on CFB. Wurzburg, Germany,
1999, pp 885890.
Greon J, Berben J, Mudde R, van den Akker H. Local bub-
ble properties in 40 and 80 cm diameter uidized beds.
AIChE Symp Ser 93(317):147151, 1997.
Grohse E. Analysis of gas uidized solid systems by x-ray
absorption. AIChE J 1:358365, 1955.
Halow J, Nicoletti P. Observations of uidized bed coales-
cence using capacitance imaging. Powder Technol.
69(3):255277, 1992.
Halow J, Fasching G, Nicoletti P, Spenik L. Observations of
a uidized bed using capacitance imaging. Chem Eng Sci
48:643659, 1993.
Halow J. Electrical capacitance imaging of uidized beds. In:
Chaouki J, Larachi F, Dudukovic M, eds. Noninvasive
Monitoring of Multiphase Flow. New York: Elsevier
Science, 1997, pp 263307.
Hamamatsu Photonics K.K. CdS Photoconductive Cells.
Catalog No. KCDS 0001E03, 1999b.
Hamamatsu Photonics K.K. Light Sources. Catalog No.
TLSO 0002J04, 2000.
Hamamatsu Photonics K.K. Optosemiconductors.
Condensed Catalog, Catalog No. KOTH 0001E03, 1999a.
Hamamatsu Photonics K.K. Photodiodes. Catalog No. KPD
0001J07, 1998b.
Hamamatsu Photonics K.K. Photomultiplier Tubes and
Related Products. Catalog No. TPMO 0003J02, 1998a.
Hamamatsu Photonics K.K. Solid State Emitters. Catalog
No. KLED 0001E06, 1999c.
Hamdullahpur F, Pegg M, MacKay G. A laser-uorescence
technique for turbulent two-phase ow measurements. Int
J Mult Flow 13(3):379385, 1987.
Hansen P, Bank L, Dam-Johansen K. In-situ measurements
in uidized bed combustors: design and evaluation of four
sample probes. Proc 13
th
Conf FBC. Orlando, FL, USA,
1995, pp 391398.
Hartge EU, Li Y, Werther J. In: Basu P, ed. Circulating
Fluidized Bed Technology. Oxford: Pergamon Press,
1986, pp 153160.
Hartge EU, Rensner D, Werther J. Solids concentration and
velocity patterns in circulating uidized beds. In: Basu P,
Hatano H, Kudo N. Microscope visualization of solid parti-
cles in circulating uidized beds. Powder Technol 78:115
119, 1994.
Hatano H, Ogasawara M, Horio M, Hartge EU, Werther J.
Local ow of gas and solids in a circulating uidized bed.
Proc 6
th
SCEJ Symp Fluidization, 1993, pp 107114.
Hatano H, Matsuda S, Takeuchi H, Pyatenko A, Tsuchiya
K. Local interactive patterns of dispersed and swarm par-
ticles in a circulating uidized bed riser. Ind Eng Chem
Res 35:43604365, 1996.
Hatano H, Takeuchi H, Sakurai S, Masuyama T, Tsuchiya
K. Motion of individual FCC particles and swarms in a
circulating uidized bed riser analyzed via high-speed
imaging. AIChE Symp Ser 94(318):3136, 1998.
Hatano, H, Matsuda S, Tsuchiya K. Continuous measure-
ment of velocity and direction of particles using the lag-
time method. AIChE Symp Ser 95(321):9599, 1999.
Hawkesworth M, Parker D, Fowles P, Crilly J, Jeries N,
Jonkers G. Nucl Instr Meth A310:423434, 1991.
Hayrinen V, Hernberg R, Oikari R, Gottwald U,
Monkhouse P, Davidsson K, Lonn B, Engvall K,
Petersson J, Lehtonen P, Kuivalainen R. On line determi-
nation of alkali concentrations in pressurized uidized-
bed coal combustion by combined measurement tech-
niques. Werther J, ed. Proc 6
th
Conf on CFB.
Wurzburg, Germany, 1999, pp 873878.
He Z, Zhang WD, Hei KM, Chen BC. Modeling pressure
uctuation via correlation structure in a gassolids ui-
dized bed. AIChE J 43(7):19141920, 1997.
Helmrich H, Schugerl K, Janssen K. Decomposition of
NaHCO
3
in laboratory and bench scale circulating ui-
dized bed reactors. In: Basu P, ed. Circulating Fluidized
Bed Technology. New York: Pergamon, 1986, pp 161
166.
Herb B, Dou S, Tuzla K, Chen J. Solid mass uxes in
1992.
Herbert P, Gauthier T, Briens C, Bergougnou M.
Application of ber optic reection probes to the mea-
surement of local particle velocity and concentration in
gassolid ow. Powder Technol 80:243252, 1994.
Hernandez D, Lu J, Gauthier D, Flamant G. A new optical
probe working in high temperature uidized beds. Proc
9
th
Conf Fluidization. New York: Engineering
Holman J. Experimental methods in engineering, 6
th
ed. New
York: McGraw-Hill, 1994.
Holobo J, Kantzas A, Kalogerakis N. Utilization of com-
puter assisted tomography in the determination of the
spatial variability of voidage in a gassolid uidized
bed. Prep 8
th
Int Conf Fluidization. Tours, France,
1995, pp 295302.
Horio M. Hydrodynamics. In: Grace J, Avidan A, Knowlton
T, eds. Circulating Fluidized Beds. New York: Chapman
& Hall, 1997, pp 2185.
Horio M, Ito M. Prediction of cluster size in circulating
uidized beds. J Chem Eng Japan 30(4):691697, 1997.
Horio M, Kuroki H. Three-dimensional ow visualization of
dilutely dispersed solids in bubbling and circulating ui-
dized beds. Chem Eng Sci 49(15):24132421, 1994.
Horio M, Mori S, eds. Fluidization Handbook, Tokyo:
Baifukan, 1999, pp 219221.
Horio M, Morishita K. Flow regimes of high velocity uidi-
zation. Japanese J Multiphase Flow 2:118136, 1988.
Horio M, Taki A, Hsieh Y, Muchi I. Elutriation and particle
transport through the freeboard of a gassolid uidized
bed. In: Grace J, Matsen J, eds. Fluidization. New York:
Plenum Press, 1980a, pp 509518.
Horio M, Kiyota H, Muchi I. Particle movement on a per-
forated plate distributor of uidized bed. J Chem Eng
Japan 13(2):137142, 1980b.
Horio, M, Hayashi H, Morishita K. A novel in-bed heat
transfer tube capable of decreasing hw linearly with load
turndown. Proc 8
th
Int Conf Fluidized Bed Combustion.
US DOE/METC, 1985, pp 655663.
Horio M, Takada M, Ishida M, Tanaka N. The similarity
rule of uidization and its application to solid mixing and
circulation control. In: Ostergaard K, Sorensen A, eds.
Fluidization V. New York: United Engineering
Foundation, 1986a, pp 151158.
Horio M, Nonaka A, Hoshiba M, Morishita K, Kobukai Y,
Naito J, Tachibana O, Watanabe K, Yoshida N. Coal
combustion in a transparent uidized bed. In: Basu P,
ed. Circulating Fluidized Bed Technology. Oxford:
Pergamon Press, 1986b, pp 255262.
Horio M, Morishita K, Tachibana O, Murata N. Solid dis-
Basu P, Large J eds. Circulating Fluidized Bed
Technology II. Oxford: Pergamon Press, 1988, pp
147154.
Horio M, Mori K, Takei Y, Ishii H. In: Potter O, Nicklin D,
eds. Proc Fluidization VII. New York: United
Engineering Foundation, 1992a, pp 757762.
Horio M, Ishii H, Nishimuro M. On the nature of turbulent
and fast uidized beds. Powder Technol 70:229236,
1992b.
Huang S, Plaskowski A, Xie C, Beck M. Tomographic ima-
ging of two-component ow using capacitance sensors. J
Phys E Sci Instrum 22, 1989.
Huilin L, Gidaspow D, Bouillard J. Dimension measure-
ments of hydrodynamic attractors in circulating uidized
beds. AIChE Symp Ser 91(308):103111, 1995.
Hull A, Agarwal P. Bubble behaviour in a tube bank: mod-
eling and experimental investigation using digital image
analysis. Prep Fluidization VIII. Tours, France, 1995,
pp 343350.
Ikeda Y. Recent issues in uidized bed reactor developments.
Chem Eng 10:4145, 1972 (in Japanese).
Ilias S, Ying S, Mathur G, Govind R. Studies on a swirling
circulating uidized bed. In: Basu P, Large J, eds.
Ishida M, Shirai T, Nishiwaki A. Measurement of the velo-
city and direction of ow of solid particles in a uidized
bed. Powder Technol 27, 1980.
Ito M, Tsukada M, Horio M. Three dimensional meso-scale
structure of gassolid suspensions in circulating uidized
beds, determined by the scanning laser sheet image
analysis. Proc 1
st
Particle Technology Forum, Vol. 1,
1994, pp 428433.
Ito O, Kawabe R, Miyamoto T, Orita H, Mizumoto M,
Miyadera H. Direct measurement of particle motion in
a large-scale FBC boiler model. Proc 15
th
Conf
Fluidized Bed Combustion, Savannah, Georgia, 1999.
Ji H, Ohara H, Kuramoto K, Tsutsumi A, Yoshida K,
Hirama T. Nonlinear dynamics of gassolid circulating
uidized bed system. Chem Eng Sci 55:403410, 2000.
Johnsson F, Sterneus J, Leckner B, Wiesendorf V, Hartge
EU, Werther J, Monat D, Briad P. Fluid dynamics of
the bottom zone of CFB combustor. Werther J, ed.
Proc 6
th
Conf CFB. Wurzburg, 1999a, pp 191196.
Joutsenoja T, Heino P, Hernberg R. Pyrometric temperature
and size measurements of burning coal particles in a ui-
dized bed combustion reactor. Combust Flame
118:707717, 1999.
Jovanovic G, Catipovic N, Fitzgerald T, Levenspiel O.
The mixing of tracer gas in uidized beds of large
particles in uidization. In: Grace J, Matsen J, eds.
Fluidization III. New York: Plenum Press, 1980, pp 325
332.
Junfu L, Jiansheng Z, Guangxi Y, Qing L, Xiaoma L, Jiling
L, Xiaoxing Z, Yi L, Long Y, Xudong L, Zhengyi L. The
progress of the water cooled separator CFB boiler in
China. Proc 15
th
Conf Fluidized Bed Combustion,
Savannah, Georgia, 1999.
Kai T, Furusaki S. Behavior of uidized beds of small par-
ticles at elevated temperature. J Chem Eng Japan
18:113118, 1985.
Kai T, Takahashi T, Misawa M, Tiseanu I, Ichikawa N,
Takada N. Observation of rising bubbles in a uidized
catalyst bed using a fast X-ray CT scanner. Proc, 5th
SCEJ Symp. on Fluidization, Tsukuba, 1999, pp. 435440.
Kai T, Misawa M, Takeuchi T, Tiseanu I, Ichikawa N,
Takeda N. Application of fast x-ray CT scanner to visua-
lization of bubbles in uidized bed. J Chem Eng Japan
33:906909, 2000.
Kak A, Slaney M. Principles of Computerized Tomographic
Imaging. Institute of Electrical and Electronics Engineers,
1988.
Kalender W. Computed Tomography. Publicis MCD Verlag,
2000.
Kamiwano M, Saito F. Measurement method of ow
velocity of liquid and irregular solid particles using an
image sensor with an image ber. AIChE Symp
Ser 80(241):122128, 1984.
Kantzas A. Computation of holdups in uidized and trickle
beds by computer-assisted tomography. AIChE J
40(7):12541261, 1994.
Kassman H, Abul-Milh M, Amand LE. Measurement of
NH
3
and HCN concentrations in a CFB boiler, a
comparison between a conventional absorption and
FTIR technique. Proc 13
th
FBC Conf. Orlando, FL,
USA, 1995, pp 14471454.
Kassman H, Karlsson M, Amand LE. Inuence of air-
staging on the concentration proles of NH
3
and HCN in
the combustion chamber of a CFB boiler burning coal.
Proc 15
th
Conf Fluidized Bed Combustion, Savannah,
Georgia, USA, 1999.
Kawata S, Minami S. Image data processing for scientic
measurement. CQ Publishing, Tokyo, 1994.
Kawata S, Minami K, Minami S. Supersolution of Fourier
transform spectroscopy data by the maximum entropy
method. Appl. Optics 22:35933598, 1983.
Kehoe P, Davidson J. Pressure uctuation in slugging
uidized beds. AIChE Symp Ser 69(128):3440, 1973.
Khattab I, Ishida M. Visualization of gas voids behavior
above the distributor in a three-dimensional gas uidized
bed. J Chem Eng Japan 19(5):367374, 1986.
Khattab I, Kuroda C, Ishida M. Radial and vertical
distributions of the interstitial gas velocity in a uidized
bed. J Chem Eng Japan 21(3):282287, 1988.
Kobylecki R, Horio M. On the microscopic aspect of inter-
particle heat transfer. Xu D, Mori S, eds. Proc 7
th
China
Japan Symposium on Fluidization. Xian: Xian
Publishing House, China, 2000, pp 5964.
Kobylecki R, Horio M. A simple solid mass ow meter for
circulating uidized beds. J Chem Eng Japan 35:456467,
2002.
Kojima T, Ishihara K, Yang G, Furusawa T. Measurement
of solids behavior in a fast uidized bed. J Chem Eng
Japan 22:341346, 1989.
Kondulov N, Kornilaev A, Skachko I, Akhromenkov A,
Kruglov A. An investigation of the parameters of moving
particle in a uidized bed by a radioisotopic method. Int
Chem Eng 4, 1964.
Kono H, Soltani-Ahmadi A, Suzuki M. Kinetic forces of
solid particles in coarse particle uidized beds. Powder
Technol 52:4-58, 1987.
Kozinski J, Rink K, Lighty JS. Combustion of sludge
waste in FBC, distribution of metals and particle
sizes. Proc 13
th
FBC Conf, Orlando, FL, USA,
1995, pp 139147.
Krambeck F, Avidan A, Lee C, Lo M. Predicting uid bed
reactor eciency using adsorbing gas tracers. AIChE J
33:17271734, 1987.
Krohn D. Intensity modulated ber optic sensors overview.
In: SPIE, Vol. 718, Fiber Optic and Laser Sensors IV,
1986, pp 211.
Kruse M, Werther J. 2D gas and solids ow prediction in
circulating uidized beds based on suction probe and
pressure prole measurements. Chem Eng Proc 34:185
203, 1995.
Kumar S, Dudukovic M. Computer assisted gamma and
x-ray tomography: applications to multiphase ow sys-
tems. In: Chaouki J, Larachi F, Dudukovic M, eds.
Non-Invasive Monitoring of Multiphase Flows. New
York: Elsevier, 1997, pp 47103.
LaNauze R, Jung K, Dent D, Joyce T, Tait T, Burgess J.
Measurement of the temperature of burning particles in
uidized beds. 9
th
Int Conf FBC. ASME, Boston, MA,
USA, 1987, pp 707712.
Lanneau K. Gas-solids contacting in uidized beds. Trans
Instn Chem Engrs 38:125143, 1960.
Larachi F, Kennedy G, Chaouki J. A -ray detection system
for 3-D particle tracking in multiphase reactors. Nuclear
Instrum Methods Phys Res 338, 1994.
Larachi F, Lord E, Chaouki J, Chavarie C, Behie L.
Phenomenological study of solids mixing in a binary
liquid uidized bed. Prep Fluidization VIII, Tours,
France, 1995, pp 385392.
Larachi F, Chaouki J, Kennedy G, Dudukovic M.
Radioactive particle tracking in multiphase reactors: prin-
ciples and applications. In: Chaouki J, Larachi F,
Dudukovic M, eds. Non-Invasive Monitoring of
Multiphase Flows. New York: Elsevier Science, 1997,
pp 335397.
Lasch B, Caram H, Chen J, Korenberg J. Solid circulation in
cyclonic uidized bed combustors. In: Basu P, Large J,
Leckner B, Andersson B. Characteristic features of heat
transfer in circulating uidized bed. Powder Technol
70:303314, 1992.
Leckner B, Golriz M, Zhang W, Andersson B, Johnsson F.
Boundary layersrst measurements in the 12 MW CFB
research plant at Chalmers University. Anthony E, ed.
Proc 11
th
Int Conf FBC. ASME, New York, USA,
1991, pp 771776.
Lehner P, Richtberg M, Wirth KE. Measurement of solids
velocities in a co-current gas/solids ow with a laser
Doppler anemometer and a capacitance probe system.
Werther J, ed. Proc 6
th
Conf CFB. Wurzburg,
Germany, 1999, pp 843848.
Le Niliot C, Gallet P. Infrared thermography applied to the
resolution of inverse heat conduction problems: recovery
of heat line sources and boundary conditions. Rev Gen
Therm 37:629643, 1998.
Levy Y. LDA technique for measurements in freeboards of
Li J, Weinstein H. An experimental comparison of gas back-
mixing in uidized beds across the regime spectrum.
Chem Eng Sci 44:16971705, 1989.
Li Y, Wu P. A study on axial gas mixing in a fast uidized
bed. In: Basu P, Horio M, Hasatani M, eds. Circulating
1991, pp 581586.
Li H, Xia Y, Tung Y, Kwauk M. Microvisualization of two-
phase structure in a fast uidized bed. In: Basu P, Horio
Technology III. Oxford: Pregamon Press, 1991, pp
183188.
Li H, Qian R, Huang W, Bi K. An investigation on instan-
taneous local heat transfer coecients in high-tempera-
ture uidized beds II. statistical analysis. Int J Heat
Mass Transfer 35:43974406, 1993.
Lin J, Grace J, Bi H, Zhu J. Gas mixing in a high density
CFB riser. Werther J, ed. Proc 6
th
Conf CFB. Werther J,
ed. Wurzburg, Germany, 1999b, pp 399404.
Lin M, Kulasekaran S, Ignowski J, Linjewile T, Agarwal P.
Emissions of N
2
O in uidized bed combustion of coal.
Proc 15
th
Georgia, USA, 1999c.
Lin W, Weinell C, Hansen P, Dam-Johansen K.
Hydrodynamics of a commercial scale CFB boiler
study with radioactive tracer particles. Werther J, ed.
Proc 6
th
Conf CFB. Wurzburg, Germany, 1999a,
pp 861866.
Linjewile T, Hull A, Agarwal P. Optical probe measurements
of the temperature of burning particles in uidized beds.
Fuel 73:18801886, 1994.
Lischer D, Louge M. Optical ber measurements of particle
concentration in dense suspensions: calibration and
simulation. Applied Optics 31:51065113, 1992.
Lord W, McAndrew G, Sakagami M, Valenzuela J,
Glicksman L. Measurement of bubble properties in
uidized beds. Proc 7
th
Combustion. US DOE/EPRI/, 1982, pp 7688.
Louge M. Measurement techniques. In: Grace J, Avidan
A, Knowlton T, eds. Circulating Fluidized Beds. New
York: Blackie Academic and Profesional, 1997,
pp 312358.
Louge M, Opie M. Measurements of the eective dielectric
permittivity of suspensions. Powder Technol 62:8594,
1990.
Lu X, Li H. Wavelet analysis of pressure uctuation signals
in a bubbling uidized bed. Chem Eng Sci 75:113119,
1999.
Luan W, Lim C, Brereton C, Bowen B, Grace J.
Experimental and theoretical study of total and radiative
heat transfer in circulating uidized beds. Chem Eng Sci
54:37493764, 1999.
Ma Y, Zhu J. Experimental study of heat transfer in a co-
current downow uidized bed (downer). Chem Eng Sci
54:4150, 1999.
Mann U, Crosby E. Flow measurement of coarse particles in
pneumatic conveyers. Ind Eng Chem Proc Des Dev
16:913, 1977.
Mann M, Swanson M, Yagla S. Characterization of alkali
and sulfur sorbents for pressurized uidized bed
combustion. Proc 13
th
FBC Conf, Orlando, FL, USA,
1995, pp 333340.
Martin M, Turlier P, Bernard J, Wild G. Gas and solid
behavior in cracking circulating uidized beds. Powder
Technol 70:249258, 1992.
Marzocchella A, Zijerveld R, Schouten J, van den Bleek C.
Chaotic behavior of gassolid ow in the riser of a labora-
tory-scale circulating uidized bed. AIChE J
43:14581468, 1997.
Mathur A, Saxena S. Total and radiative heat transfer to an
immersed surface in a gas-uidized bed. AIChE J
33:11241135, 1987.
Matsuno Y, Yamaguchi H, Oka T, Kage H, Higashitani K.
The use of optic ber probes for the measurement of
dilute particle concentrations: calibration and application
to gas-uidized bed carryover. Powder Technol
36:215221, 1983.
Mattisson T, Lyngfelt A. The presence of CaS in the com-
bustion chamber of a 12 MW circulating uidized bed
boiler. Proc 13
th
FBC Conf, ASME, New York, 1995,
pp 819829.
Mehrabi A, Rassamdana H, Sahimi M. Characterization of
long-range correlations in complex distributions and pro-
les. Physical Review E 56:712722, 1997.
Meredith R, Tobias C. Resistance of potential ow through a
cubic array of spheres. J Appl Phys 31(7):12701273,
1960.
Mickley H, Fairbanks D. Mechanism of heat transfer to
Mickley H, Fairbanks D, Hawthorn R. The relation between
the transfer coecient and thermal uctuations in
uidized bed heat transfer. Chem Eng Prog Symp Ser
57(32):5160, 1961.
Minami S (ed.). Wave Data Processing for Scientic
Measurement. Tokyo: CQ Publishing, 1986.
Miyauchi T, Morooka S. Circulating ow and its eects on
chemical reaction in uid bed contactor. Kagaku Kogaku
33:369376, 1969.
Molerus O, Wirth KE. Heat Transfer in Fluidized Beds. New
York: Chapman and Hall, 1997.
Monceaux L, Azzi M, Molodtsof Y, Large J. Overall and
local characterization of ow regimes in a circulating
uidized bed. In: Basu P, ed. Circulating Fluidized Bed
Technology. Toronto: Pergamon, 1986, pp 185192.
Moritomi H, Mori S, Araki K, Moriyama A. Periodic
pressure uctuation in a gaseous uidized bed. Kagaku
Kogaku Ronbunshu 6:392396, 1980.
Morooka S, Kusakabe K, Ohnishi N, Gujima F, Matsuyama
H. Measurement of local nes movement in a uidized
bed of coarse particles by a uorescent tracer technique.
Muraoka K, Maeda M. Laser Application for Gas Plasma
and Gas Measurement. Tokyo: Sangyotosho, 1995.
Myler C, Zaltash A, Klinzing G. Gas-solid transport in a
0.0508 m pipe at various inclinations with and without
electrostatics i: particle velocity and pressure drop. J
Powder Bulk Solid Technol 10:512, 1986.
Nakajima K. Laser Doppler velocimeter. Kikai-no-Kenkyu
47:6571, 1995.
Nakajima M, Harada M, Asai M, Harada R, Jimbo G.
Bubble fraction and voidage in an emulsion phase in the
transition to a turbulent uidized bed. In: Basu P, Horio
Technology III. Oxford: Pergamon Press, 1991, pp 7984.
Naruse I, Ohtake K, Koizumi K, Kuramoto K, Lu G. Gas
exchange between the bubble and emulsion phases during
bubbling uidized bed coal combustion elucidated by con-
ditional gas sampling. 25
th
Symposium on Combustion.
Comb Inst 1994, pp 545552.
Naruse I, Kuramoto K, Lu GQ, Ohtake K. Inuence of
bubble and emulsion phase on N
2
O formation in
bubbling uidized bed coal combustion by conditional
sampling. Prep Fluidization VIII, Tours, France, 1995,
pp 499506.
Naruse K, Kashima N, Hosoda S, Toyoda S, Oshita T,
Tsukada M, Horio M, Shinozaki S, Sugiyama T. On the
eect of uidized bed viscosity on the solid circulating rate
across a tube bank. J Soc Powder Technol Japan
33:468475, 1996.
Nguyen T, Nguyen A, Nieh S. An improved isokinetic sam-
pling probe for measuring local gas velocity and particle
mass ux of gassolid suspension ows. Powder Technol
59:183189, 1989.
Nieuwland J, Meijer R, Kuipers J, van Swaaij W.
Measurements of solids concentration and axial solids
velocity in gassolid two-phase ows. Powder Technol
87:127139, 1996.
Nishii K, Shimizu Y, Horio M. Characterization of cohe-
sive powder behavior in uidized beds. Proc 6
th
Int
Symp Agglomeration. Nagoya, Japan, 1993, pp 740
745.
Noguchi M. How to Use Visual LED Lamps. Transistor
Technology, Supplemental Issue, Hardware Design
Series No. 8, CQ Pub, 1999, pp 816.
Nowak W, Mineo H, Yamazaki R, Yoshida K. Behavior of
particles in a circulating uidized bed of a mixture of
two dierent sized particles. In: Basu P, Horio M,
Hasatani M, eds. Circulating Fluidized Bed
Technology III. Oxford: Pergamon Press 1991, pp 219
224.
Nowak W, Win K, Matsuda H, Hasatani M, Kruse M,
Werther J. Local solids uxes and particle size distribu-
tions in a multi-solid uidized bed. Prep Fluidization
VIII, Tours, France, 1995, pp 209216.
Ohki K, Shirai T. Particle velocity in uidized bed. In:
Keairns D, ed. Fluidization Technology. Washington:
Onodera T, Okura Y, Umekawa H, Ozawa M, Takenaka N,
Matsubayashi M. Large particles and a bubble movement
in uidized-bed. Proc Specialists Meeting Neutron
Radiography Techniques for Application and
Utilization, Nov 1998.
Onofri F, Tadrist L. Experimental analysis of spatio-tem-
poral instabilities of the dilute gassolids ow in a CFB.
th
Conf CFB. Wurzburg, Germany,
1999, pp 837842.
Ozawa M, Umekawa H, Matsuda T, Takenaka N,
Matsubayashi M. Flow pattern and heat transfer in
tube banks of a simulated uidized bed heat exchangers.
JSME Intern J, Ser B 41(3):720726, 1998.
Ozawa M, Umekawa H, Takenaka N, Matsubayashi M.
Quantitative ow visualization of uidized-bed under
normal and down-ow-mode operations by neutron
radiography. Experiments in Fluids, 1999.
Ozhaynak T, Chen J, Frankeneld T. An experimental
investigation of radiation heat transfer in a high
temperature uidized bed. In: Kunii D, Toei E, eds.
Fluidization IV. New York: United Engineering
Pangrle B, Walsh E, Moore S, DiBiasio D. Magnetic reso-
nance imaging of laminar ow in porous tube and shell
systems. Chem Eng Sci 47(3):517526 1992.
Pasquetti R, Le Niliot C. Boundary element approach
for inverse heat conduction problems: application to
a bidimensional transient numerical experiment.
Numerical Heat Transfer, Part B 20:169189, 1991.
Patience G, Chaouki J, Grandjean B. Solids ow metering
from pressure drop measurement in circulating uidized
beds. Powder Technol 61:9599, 1990.
Patience G, Chaouki J, Kennedy G. Solids residence time
distribution in a circulating uidized bed reactors. In:
Basu P, Horio M, Hasatani M, eds. Circulating
1991, pp 599604.
Patrose B, Caram HS. Optical ber probe transit
anemometer for particle velocity measurements in
Prins W. Fluidized bed combustion of a single carbon
particle. Professors thesis, Twente University, The
Netherlands, 1987.
Qi C, Farag I. Experimental measurement of lateral particle
ux over the riser cross-section. In: Large JF, Laguerie C,
eds. Fluidization VIII. New York: Engineering
Radon J. Ueber die bestimmung von funktionen durch ihre
integralwerte langs gewisser manningfaltigkeiten. Berichte
Saechsische Akad Wiss 69, 1917.
Raso G, Tirabasso G, Donsi G. An impact probe for local
analysis of gassolid ows. Powder Technol 34:151159,
1983.
Reese J, Fan LS. Particle image velocimetry: application for
the characterization of the ow structure in three phase
uidized beds. In: Chaouki J, Larachi F, Dudukovic M,
eds. Non-Invasive Monitoring of Multiphase Flows. New
York: Elsevier Science, 1997.
Reh L, Li J. Measurement of voidage in uidized beds by
optical probes. In: Basu P, Horio M, Hasatani M, eds.
Circulating Fluidized Bed Technology III. New York:
Ren J, Li J. Wavelet analysis of dynamic behavior in ui-
dized beds. In: Fan LS, Knowlton T, eds. Fluidization
IX. New York: United Engineering Foundation, 1998,
pp 629636.
Rensner D, Werther J. Estimation of the eective measuring
volume of single-bre reection probes for solid volume
concentration measurements. Part Part Syst Charact
10:4855, 1993.
Rhodes M, Laussmann P. A simple nonisokinetic sampling
probe for dense suspensions. Powder Technol 70:141151,
1992a.
Rhodes M, Laussmann P. Characterizing nonuniformities in
gasparticle ow in the riser of a circulating uidized bed.
Powder Technol 72:277284, 1992b.
Rhodes M, Wang X. Studying fast uidization using electri-
cal capacitance tomography. Werther J, ed. Proc 6
th
Conf
CFB. Wurzburg, Germany, 1999, pp 891896.
Rivault P, Nguyen C, Laguerie C, Bernard J. Countercurrent
stripping dense circulating beds eect of the baes. Prep
of the Fluidization VIII, Tours, France, 1995, pp 491498.
Rix S, Glass D, Greated C. Preliminary studies of elutriation
from gas-uidised beds using particle image velocimetry.
Chem Eng Sci 51(13):34793489, 1996.
Ross K, Pence D. Wavelet and fractal analysis of uidized
bed heat transfer, HTD 351. Proc ASME Heat Transfer
Div 1:245253, 1997.
Ross I, Patel M, Davidson J. The temperature of burning
carbon particles in uidised beds. Trans Inst Chem Eng
59:8388, 1981.
Roy R, Davidson J, Tuponogov V. The velocity of sound in
Roy M, Kumar V, Kulkarni B, Sanderson J, Rodes M, van
der Stappen M. Simple denoising algorithm using wavelet
transform. AIChE J 45:24612466, 1999.
Sauer S, Wallen V. Mixing in experiments and modelling of
ABFBC. Proc 15
th
Savannah, Georgia, USA, 1999.
Schlichthaerle P, Werther J. Axial pressure proles and solids
concentration distributions in the CFB bottom zone.
th
Conf CFB. Wurzburg,
Germany, 1999, pp 185190.
Schoenfelder H, Kruse M, Werther J. Two dimensional
model for circulating uidized-bed reactors. AIChE J
40:18751888, 1996.
Schouten J, van den Bleek C. Chaotic hydrodynamics of
uidization: consequences for scaling and modeling of
uid bed reactors. AIChE Symp Ser 88(289):7084, 1992.
Schugerl K, Merz M, Fetting F. Rheological characteristics
of gassolid uidized bed system. Chem Eng Sci 15:138,
1961.
Schuurmans H. Measurements in a commercial catalytic
cracker unit. Ind Eng Chem Process Des Dev
19:267271, 1980.
Sederman A, Alexander P, Gladden L. Structure of packed
beds probed by magnetic resonance imaging. Powder
Technol 117:255269, 2001.
Seo Y, Gidaspow D. An x-ray-,-ray method of measurement
of binary solids concentrations and voids in uidized
beds. Ind Eng Chem Res 26:16221628, 1987.
Seville J, Morgan J, Clift R. Tomographic determination of
the voidage structure of gas uidized beds in the jet
region. In: Ostregaard K, Sorensen A, eds. Fluidization
V. New York: United Engineering Foundation, 1986,
pp 8794.
Seville J, Simons S, Broadbent C, Martin T, Parker D,
Beynon T. Particle velocities in gasuidized beds. Prep
Shattuck M, Behringer R, Johnson G, Georgiadis J.
Convection and ow in porous media. Visualization by
magnetic resonance imaging. J Fluid Mech 332:215245,
1997.
Shen Z, Briens C, Kwauk M, Bergougnou M. Improved
thermal conductivity detector technique for helium mea-
surements in a uidized bed. Powder Technol 69:249253,
1992.
Sheng Y. Wavelet transform. In: Poularikas A, ed. The
Transforms and Applications Handbook. New York:
CRC Press, 1996, pp 747827.
Siegel R, Howel J. Thermal Radiation Heat Transfer. Tokyo:
McGraw-Hill Kogakusha, 1972.
Siegell J. Hight-temperature deuidization. Powder Technol
38:1322, 1984.
Skrifvars B, Sris G, Backman R, Widegren-Dafgard K,
Hupa M. Energy and Fuels 11:843848, 1997.
Soo S, Trezek G, Dimick R, Hohnstreiter G. Concentration
and mass ow distributions in a gassolid suspension.
I&EC Fundam 3(2):98106, 1964.
Soo S, Slaughter M, Plumpe J. Instrumentation for ow
properties of gassolid suspensions and recent advances.
Part Sci Tech 12:112, 1994.
Souto M, Rodriguez J, Conde-Pmpido R, Guitian F,
Gonzalez J, Perez J. Fuel 75(6):675680, 1996.
Stein M, Martin T, Seville J, McNeil P, Parker D. Positron
emission particle tracking: particle velocities in gas ui-
dized beds, mixers and other applications. In: Chaouki J,
Larachi F, Dudukovic M, eds. Non-Invasive Monitoring
of Multiphase Flows. New York: Elsevier Science, 1997,
pp 309334.
Stellema C, Mudde R, Kolar Z, Gerritsen, de Goeij J, Van
den Bleek C. Single particle tracking in interconnected
uidized beds. In: Fan LS, Knowlton T, eds.
Fluidization IX. New York: United Engineering
Stephenson R, Moulin A, Welland M, Burns J, Sapo M,
Reed R, Frank R, Fraden J, Nicholas J, Pavese F, Stasiek
J, Madaj T, Mikielewicz J, Culshaw B. Temperature
measurement. In Webster J, ed. The Measurement,
Instrumentation and Sensors Handbook. New York:
CRC Press LLC, 1999.
Swift W, Vogel G, Panek A, Jonke A. Trace-element mass
balances around a bench-scale combustor. Proc 4
th
FBC
Conf, McLean, VA, USA, 1975, pp 525543.
Takeuchi H, Hirama T. Flow visualization in the riser of a
circulating uidized bed. In: Basu P, Horio M, Hasatani
Takeuchi H, Pyatenko A, Hatano H. Gross behavior of
parabolic strands in a riser. In: Fan LS, Knowlton TM,
eds. Fluidization IX. United Engineering Foundation,
1998, pp 173180.
Tanino T, Yawata Y, Otani S, Inoue H, Sato K, Takeda T,
Mizuta T. Particle size control of anti-acid drug granules
in uidized-bed granulation. Proc 6
th
Int Symp
Agglomeration, Nagoya, Japan, 1993, pp 548553.
Tasdemir A, Sozmen N, Vural H. Determination of mixing
quality in uidized beds by image processing technique.
Proc 15
th
Georgia, USA, 1999.
Tayebi D, Svendsen H, Grislingas A, Mejdell T, Johannessen
K. Dynamics of uidized-bed reactors, development and
application of a new multi-ber optical probe. Chem Eng
Sci 54:21132122, 1999.
Tolay Industries, Inc., Polymer Optical Fiber, Catalog No.
PE0103-10, 1991.
Tomita M, Adachi T. The eect of bed diameter on the
behavior of bubbles in gassolid uidized beds. J Chem
Eng Japan 6:196201, 1973.
Tsuji Y, Morikawa Y. LDV measurements of an air-solid
two-phase ow in a horizontal pipe. J Fluid Mech
120:385409, 1982.
Tsukada M, Nakanishi D, Horio M. Eect of transport
velocity in a circulating uidized bed. In: Avidan A,
ed. Circulating Fluidized Bed Technology IV. New
York: AIChE, 1994, pp 209215.
Tsukada M, Ito M, Kamiya H, Horio M. Three-dimension
imaging of particle clusters in dilute gassolid suspension
ow. Can J Chem Eng 75:466470, 1997.
Tsutsui T, Miyauchi T. Fluidity and its inuence on behavior
in a uidized bed with ne particles. Kagaku Kogaku
Ronbunshu 5:4046, 1979.
Tung Y, Kwauk M. Dynamics of collapsing uidized
beds. In: Kwauk M, Kunii D, eds. Fluidization
Science and Technology. Beijing: Science Press and
Gordon and Breach Science Publishers, 1982, pp 155
166.
Umekawa H, Ozawa M, Onodera T, Okura Y, Takenaka N,
Matsubayashi M. Visualization of large-particle move-
ment in uidized-bed by neutron radiography. Proc 6
th
World Conf Neutron Radiography. Osaka: Gordon and
Breach Science Publishers, 1999.
Van Breugel J, Stein J, Vriens R. Isokinetic sampling in a
dense gassolid stream. Proc Instn Mech Engrs
184(113C):1823, 1969/1970.
Van de Wall R, Soo S. Measurement of particle cloud density
and velocity using laser devices. Powder Technol 81:269
278, 1994.
Van den Moortel T, Santini R, Tadrist L, Panataloni J.
Experimental study of the particle ow in a circulating
uidized bed using a phase Doppler particle analyser: a
new postprocessing data algorithm. Int J Multiphase
Flow 23(6):11891209, 1997.
Van den Schaaf J, Schouten JC, Johnsson F, van den Bleek
CM. Multiple modes of bed mass oscillation in gassolids
uidized beds. Proc 15
th
Combustion. ASME, 1999a.
Van den Schaaf J, Johanson F, Schouten JC, van den Bleek
CM. Fourier analysis of nonlinear pressure uctuations in
gassolid ow in CFB risersobserving solids structures
and gas/particle turbulence. Chem Eng Sci 54:55415546,
1999b.
Van Velzen D, Flamm H, Langenkamp H, Casile A. Can J
Chem Eng 52:156161, 1974.
Verloop W, Boersma D, Van den Akker H, Hein K. The uid
dynamics of particles in the freeboard of a pressurized
uidized bed combustor. Proc 12
th
FBC Conf, San
Diego, CA, USA, 1993, pp 5362.
Wang X, Rhodes M, Gibbs B. Solids ux distribution in a
CFB riser operating at elevated temperatures. Prep. of the
Fluidization VIII, Tours, France, 1995a, pp 4148.
Wang X, Gibbs B, Rhodes M. Flow structure in a CFB
risereect of operating temperature. Proc 13
th
FBC
Conf, Orlando, FL, USA, 1995b, pp 663670.
Watrasiewicz B, Rudd M. Laser Doppler Measurements.
Boston: Butterworths, 1976.
Weber M. Handbook of Laser Wavelength. Boston: CRC
Press, 1999.
Wei F, Lin H, Cheng Y, Wang Z, Jin Y. Proles of particle
velocity and solids fraction in a high-density riser. Powder
Technol 100:183189, 1998.
Weimer A, Gyure DC, Clough D. Application of a gamma-
radiation density gauge for determining hydrodynamic
properties of uidized beds. Powder Technol
44:179194, 1985.
Weinell C, Dam-Johansen K, Johnsson J. Local up- and
downward particle velocities in circulating uidized
beds. Prep Fluidization VIII, Tours, France, 1995,
pp 7380.
Weinstein H, Shao M, Wasserzug L. Radial solid density
variations in a fast uidized bed. AIChE Symp Ser
80(241):117121, 1984.
Weinstein H, Feindt H, Chen L, Gra R. The measurement
of turbulence quantities in a high velocity uidized bed.
In: Potter O, Nicklin D, eds. Fluidization VII. New York:
United Engineering Foundation, 1992, pp 305312.
1966.
Werdermann C, Werther J. Solids ow pattern and heat
transfer in an industrial-scale uidized bed heat exchan-
ger. Proc 12
th
FBC Conf, San Diego, CA, USA, 1993,
pp 985990.
Werther J. Measurement techniques in uidized beds.
Powder Technol 102(1):1536, 1999.
Werther J, Hage B. A bre-optical sensor for high-tempera-
ture application in uidized bed combustion. In: Large J,
Laguerie C, eds. Fluidization VIII. New York: United
Werther J, Molerus O. The local structure of gas-uidized
beds. I. A statistically based measuring system. Int J
Multiphase Flow 1:103122, 1973.
Werther J, Rudnick C. Modeling the uid mechanics of a
circulating uidized bed based on a local ow structure
analysis. In: Werther J, Markl H, eds. In-situ Measuring
Techniques and Dynamic Modeling of Multiphase Flow
Systems SFB 238 Progress Report 19941996, Verlag des
SFB 238, Hamburg, 1996.
Werther J, Hartge EU, Rensner D. Messtechniken fu r gas/
feststo-wirbelschichtreaktoren. Chem Ing Tech
62(8):605613, 1990.
Werther J, Hartge EU, Kruse M. Radial gas mixing in the
upper dilute core of a circulating uidized bed. Powder
Technol 70:293301, 1992.
Werther J, Hartge EU, Rensner D. Measurement techniques
for gassolid uidized bed reactors. Int Chem Eng
33(1):1827, 1993.
Westphalen D, Glicksman L. Lateral solid mixing measure-
ments in circulating uidized beds. Powder Technol
82(2):153167, 1995.
White C, Dry R. Transmission characteristics of gas in a
circulating uidised bed. Powder Technol 57:8994, 1989.
Wiesendorf V, Hartge EU, Werther J, Johnsson F, Sterneus
J, Leckner B, Montat D, Briand P. The CFB boiler in
Gardannean experimental investigation of its bottom
zone. Proc 15
th
Savannah, Georgia, USA, ASME, New York, 1999.
FBC99-0151
Win KK, Nowak W, Matsuda H, Hasatani M, Bis Z,
Krzywanski J, Gajewski W. Transport velocity of coarse
particles in multisolid uidized bed. J Chem Eng Japan
28(5):535540, 1995.
Xie H, Geldart D. The time response of pressure probes.
Yamada J, Kurosaki Y, Morikawa T. Radiation emitted
from uidized particles adjacent to a heat transfer surface
in a uidized bed. Kikai-Gakkai-Ronbunshu B
62:234240, 1996.
Yamaki T, Kusakabe K, Morooka S. A probe for detecting
the movement of dark-colored particles. Powder Technol
78:189190, 1994.
Yamazaki M, Fukuta K, Li YH, Tokumoto J. Distribution
of porosity of emulsion phase and its eect on conver-
sion in a uidized bed. J Chem Eng Japan 21(1):4756,
1988.
Yang WC, Chitester D. Transition between bubbling and
turbulent uidization at elevated pressure. AIChE Symp
Ser 84 (262):1021, 1988.
Yang WC, Keairns D. Solid entrainment rate into gas and
gas-solid two phase jets in a uidized bed. Powder
Technol 33:8994, 1982.
Yang G, Huang Z, Zhao L. Radial gas dispersion in a fast
uidized bed. In: Kunii D, Toei E, eds. Fluidization IV.
New York: United Engineering Foundation, 1984,
pp 145153.
Yasui G, Johanson L. Characteristics of gas pockets in ui-
dized beds. AIChE J 4(4):445452, 1958.
Yates J, Simons S. Experimental methods in uidization
research. Int J Multiphase Flow 20(suppl):297330,
1994.
Ye H, Kikuchi R, Nicolai R, Reh L. Characterization of
ozone decomposition in a semiindustrial CFB reactor
using neutral networks: an experimental study. Werther
J, ed. Proc 6
th
Conf CFB. Wurzburg, Germany, 1999, pp
879884.
Yerushalmi J, Cankurt N. Further studies of the regimes of
Yoshida K, Sakane J, Shimizu F. A new probe for measuring
uidized bed characteristics at high temperatures. Ind Eng
Chem Fundam 21:8385, 1982.
Zevenhoven R, Kohlmann J, Laukkanen T, Tuominen
M, Blomster AM. Near-wall particle velocity and con-
centration measurements in circulating uidised beds
in relation to heat transfer. Proc 15
th
Conf
Fluidized Bed Combustion, Savannah, Georgia, USA,
1999.
Zhang H, Johnston P, Zhu J, de Lasa H, Bergougnou M. A
novel calibration procedure for a ber optic solids con-
centration probe. Powder Technol 100(2/3):200272,
1998.
Zou B, Li H, Xia Y, Ma X. Cluster structure in a circulating
26
LiquidSolids Fluidization
Norman Epstein
University of British Columbia, Vancouver, British Columbia, Canada
1 ORIGINS
In seeking the origins of uidization in general (Bro tz,
1952; Leva, 1959; Grace, 1992; Kwauk, 1996), and of
liquid uidization in particular (Di Felice, 1995), it is
common to point to the great tome of Agricola (1556),
in which a hand jigging operation for ore dressing is
described and illustrated. Jigging involves sorting of
particles (classication by density) by rapidly alternat-
ing surges of liquid (usually water) through a screen
supporting the particles. The upward surges represent
a series of short-lived uidizations which are each fol-
lowed by hindered settling, the short duration of which
prevents the development of a signicant drag to resist
the motion of the particles. This accelerating motion is
then primarily determined by the normalized buoyed
density of the particles, r
p
r
_ _
,r
p
, acting under the
inuence of gravity with little sensitivity to their shape
and even size (Brown et al., 1950). In modern jigging
operations the screen may be xed and liquid pulsation
eected mechanically, or the screen itself may be given
a reciprocating motion; in Agricolas time, the same
eect was created by hand oscillation of the screen.
The resemblance between jigging and what we now
call uidization, i.e., the continuous upward (or down-
ward) ow of uid relative to a mobile swarm of
particles, is thus small.
Liquid uidization in the latter sense was used for
both sorting (classication by density) and sizing (clas-
sication by size) for more than a century (Richards,
1893) and was commonly referred to as teetering in
the mineral dressing literature (Gaudin, 1939; Richards
and Locke, 1940; Taggart, 1953). A liquid-uidized
bed was referred to as a teeter column or as being
in the teeter condition (Hancock, 1936). It was only
after the term uidization was coined to describe
mobile swarms of particles contacted upwardly by
gases (ca. 1940) that the same term began to be used
for analogous liquidsolid contacting, and the term
teeter bed was replaced by liquid-uidized bed.
2 PARTICULATE FLUIDIZATION
2.1 Introduction
Let us start with the most conventional of liquid-
uidized beds shown schematically in Fig. 1. The
cylindrical vertical column is commonly circular in
cross section, though it may also be rectangular, e.g.,
square. The function of the calming or homogenizing
section, usually located upstream of the distributor, is
to produce, in conjunction with the distributor, as
radially uniform a liquid velocity distribution in the
uidized bed as possible, thus eliminating or at least
minimizing any tendency toward channeling or bulk
circulation (gulf streaming). The system shown in
Fig. 1 is a semibatch (or semicontinuous) operation,
i.e., continuous ow of liquid for an unchanging
batch of the uidized solids, which in this case are
bottom-restrained. We shall see later that many other
operating modes are possible under the rubric of
liquidsolid uidization.
We shall also rst consider nonporous uidized par-
ticles that are all uniform in density, size, and shape. In
their pioneer study of individual particle motion in
such a liquid-uidized bed, Handley et al. (1966)
found that, in the absence of gulf streaming, uniform
uidization resulted and was characterized by a
homogeneous random motion of the particles and by
a constant mean uid-velocity and momentum distri-
bution throughout the bed, but with larger particle
velocity uctuations (viz. axial/radial -2.3, as later
veried by Carlos and Richardson, 1968a; Latif and
Richardson, 1972; Kmiec, 1978) and displacements in
the vertical direction than in the horizontal, i.e.,
D
ia
D
ir
. We shall therefore begin by assuming,
after Couderc (1985), that liquidsolid uidization
results in stable operation and beds which are homo-
geneous, with a spatially uniformly distributed concen-
tration of solid particles. This idealized assumption of
particulate uidization will be subsequently sub-
jected to signicant qualication.
2.2 Buoyancy and Drag
Consider now a representative single particle of
volume V and density r
p
in such a uidized swarm of
monodispersed particles. Macroscopically steady-state
uidization requires that downward gravitational
force = upward buoyant force upward drag, i.e.,
Vr
p
g = B F
D
(1)
There is a major disagreement in the literature on the
correct expression to use for the buoyancy B. The con-
ventional formula for B advocated, for example, by
Clift et al. (1987), Jean and Fan (1992), and Clift
(1993), is simply the buoyant force under static (no-
ow) conditions,
B = Vrg (2)
The counterargument, presented among others by
Gibilaro et al. (1984, 1987) and Astarita (1993), posits
that
B = Vr
B
g (3)
Figure 1 Semibatch liquid uidization loop. (After Couderc. 1985.)
where r
B
is the uidized bed suspension density given
by
r
B
= r
p
1 e ( ) re (4)
The grounds for this position is that under uidized
bed conditions, the total pressure gradient, dP,dz,
which develops to support the suspension, is given by
dP
dz
= r
B
g (5)
so that the eective or dynamic buoyancy in the ui-
dized state is considerably greater than under static
conditions. Combination of Eqs. (1) and (2) gives
F
D
= V r
p
r
_ _
g (6)
while combination of Eqs. (1), (3), and (4) results in
F
D
= V r
p
r
_ _
eg (7)
Note that, by virtue of the greater value in Eq. (1) of B
given by Eq. (3) than by Eq. (2), F
D
via Eq. (7) is less
than F
D
via Eq. (6) by a factor of e(e
mf
- e - 1).
Advocates of Eq. (6) argue that it is the drag force
F
D
that causes the dissipation of mechanical energy
and that this dissipation must include the energy
required to convert the original static buoyancy of
the particle in the unuidized state to the subsequent
dynamic buoyancy in the uidized state. Partisans on
both sides of the argument agree that the frictional
pressure gradient, dp
f
,dz, of the uidized bed must
be given by the specic weight of the suspension
corrected for hydrostatic head, i.e.,
dp
f
dz
= r
B
g rg (8a)
= [r
p
(1 e) re]g rg (8b)
= (1 e)(r
p
r)g (8c)
Since V is the volume of a representative particle in a
homogeneously uidized bed, the representative
volume of liquid associated with this particle (in a
unit cell so circumscribed that the liquid-to-solid
ratio is the same as in the bed as a whole) is
Ve
_
(1 e) and the correspondingly representative
volume of bed is simply V
_
(1e) [= {Ve,(1 e)] V].
To get from Eq. (6) to Eq. (8), one writes that
dp
f
dz
=
drag force on representative particle
volume of bed associated with
representative particle
=
F
D
V,(1 e)
= (1 e)(r
p
,)g
However, to get from Eq. (7) to Eq. (8), one must write
that
dp
f
dz
=
drag force on representative particle
volume of liquid associated with
representative particle
=
F
D
Ve,(1 e)
= (1 e)(r
p
r)g
If, as is commonly assumed, dp
f
,dz represents fric-
tional force per bed volume (particles plus liquid)
rather than simply frictional force per volume of liquid
in the bed, then that is an additional argument in favor
of Eqs. (2) and (6) over (3) and (7).
We will here adopt the buoyancy convention incor-
porated in Eqs. (2) and (6) rather than (3) and (7).
However, since the latter convention has made signi-
cant inroads into the literature, the reader is referred to
Table 1 of Khan and Richardson (1990) for useful drag
coecient relationships based on the above alternate
conventions, as well as on alternate denitions of the
characteristic liquid velocity, and to conversions by
Jean and Fan (1992) of several equations incorporating
buoyancy as dened by Eq. (3) [derived by Foscolo
et al. (1983, 1989), Foscolo and Gibilaro (1984, 1987)
and Gibilaro et al. (1985a, 1986)] to the corresponding
equations based on buoyancy as dened by Eq. (2).
2.2 Hydrodynamic Representation
After the column of Fig. 1 has been lled with liquid
and the monodispersed particles introduced as a xed
packed bed, but before any velocity is imparted to the
liquid, the total pressure drop, P, across height H,
measured from immediately above the distributor
to a plane well above the xed or any subsequently
uidized bed, is given by
P = P
1
P
2
= rgH (9)
where P
1
and P
2
are each determined independently,
e.g., by a pressure gauge or transducer. The dynamic
pressure drop, p, due to any motion of the liquid,
as measured for example by a dierential manometer,
is then zero. Once ow is imparted to the liquid, the
total pressure drop increases and, at steady conditions,
is given by
P = p rgH
so that
p = P rgH (10)
Assuming the cross-sectional area of the column is
unchanging with bed level, e.g., the bed is untapered,
then, neglecting minor bed entrance, bed exit, and wall
friction eects, the pressure drop, p, due to the
liquid motion is equivalent to the frictional pressure
drop, p
f
, due to ow through the bed of particles.
A typical plot of p
f
as a function of liquid velocity
is shown in Fig. 2.
AB corresponds to an immobile xed bed, the pres-
sure drop across which increases as the velocity
increases. From B to C, across the blunt maximum
caused by interlocking of particles followed by their
rearrangement, the bed, if it has been densely packed,
loosens up with each incremental velocity increase;
nally, at and beyond C, it continues to expand in
the mobile uidized state, during which the pressure
drop remains constant. The constant frictional pres-
sure drop is then given by
p
f
= L 1 e ( ) r
p
r
_ _
g (11)
which is simply the integral form of Eq. (8). As D is
approached, the bed level disappears; beyond D, all
particles are carried out of the column by the ascend-
ing liquid, and in the ideal case where wall eects,
including radial gradients of longitudinal velocity, are
absent, the liquid velocity at D is U
0
, the free settling
terminal velocity of the particles. If the ow is reversed
before the bed disappears, the path described is DCE
because when deuidization occurs, the bed settles at
its random loose (Oman and Watson, 1944; Eastwood
et al., 1969) or minimum uidization (Leva, 1959) voi-
dage, e
mf
. The point C is easier to determine and to
reproduce by the velocity-decreasing rather than the
velocity-increasing route, although once the packed
bed is in the fully expanded or random loose con-
dition, the same path is followed (i.e., no hysteresis)
on both increasing and decreasing the liquid velocity.
The supercial velocity at point C is the incipient or
mf
.
Plots corresponding to Fig. 2 of bed height L, voi-
dage e, and liquidsolids bulk density, each as a func-
tion of supercial velocity, are shown in Figs. 3, 4, and
5, respectively. For a total mass M of particles, the
volumetric particle concentration in either a xed or
a uidized bed is given by
1 e =
M
r
p
AL
(12)
Therefore, for a uidized bed, the product of bed
height and particle concentration is constant and can
be obtained from either Eq. (11) or Eq. (12):
L 1 e ( ) =
p
f
(r
p
r)g
=
M
r
p
A
= constant (13)
Figure 2 Frictional pressure drop as a function of liquid supercial velocity for monodispersed particles.
Thus as L increases, e increases accordingly, as shown
in Figs. 3 and 4. The minimum uidization voidage,
e
mf
, is typically about 0.4 for monosize smooth spheres
in the absence of signicant wall eect (d
p
,D
c
_ 0.01)
but varies considerably with particle geometry (Leva,
1959; Eastwood et al., 1969). Empirically, the slope
of log e vs. log U for liquid uidization is commonly
constant for xed particle and liquid properties up
to e = e
c
, where in most instances, 0.8 _ e
c
_ 0.9
(Wilhelm and Kwauk, 1948; Garside and Al-
Dibouni, 1977; Riba and Couderc, 1977; Chong et
al., 1979). The characteristics of this slope (= 1,n)
will be examined in Sec. 4. The value of the uidized
bed (CD) suspension density in Fig. 5 is given by Eq.
(4) as well as by
r
B
=
P
Lg
(14)
which is obtained by integrating Eq. (5) from z = 0 to
z = L.
3 MINIMUM FLUIDIZATION
Let us now address the problem of determining the
mf
, for hard, non-
porous particles of uniform size, shape, and density,
knowing e
mf
, the minimum uidization voidage. This
problem is essentially the same for liquid uidization
as it is for gas uidization. At the point of incipient
uidization, we will equate the frictional pressure
gradient as given by Eq. (8) at e = e
mf
with that
for a single phase Newtonian uid ow through the
corresponding loose packed bed, as given by the
widely accepted Ergun (1952) equation:
dp
f
dz
= (1 e
mf
)(r
p
r)g
=
150U
mf
m(1 e
mf
)
2
f
2
d
2
p
e
3
mf
1.75U
2
mf
r(1 e
mf
)
fd
p
e
3
mf
(15)
in which the product of sphericity and equivolume
sphere diameter d
p
has been substituted for the equiva-
lent surface-volume sphere diameter d
sv
. Algebraic
manipulation and rearrangement of this equation
results in
Figure 3 Bed height as a function of liquid supercial
velocity.
Figure 4 Bed voidage as a function of liquid supercial
velocity.
Figure 5 Liquidsolids bulk density as a function of liquid
supercial velocity.
Re
2
mf

150
1.75
1 e
mf
f
Re
mf

fe
3
mf
Ar
1.75
= 0 (16)
where Re
mf
= d
p
U
mf
r
_
m and Ar = d
3
p
r(r
p
r)g,m
2
.
The physically realistic solution of quadratic Eq. (16) is
Re
mf
= (C
2
1
C
2
Ar)
1,2
C
1
(17)
where
C
1
=
150(1 e
mf
)
2f(1.75)
= 42.86
(1 e
mf
)
f
(18)
and
C
2
=
fe
3
mf
1.75
= 0.5714fe
3
mf
(19)
Thus knowing d
p
, r
p
, r, and m, knowledge of both e
mf
and f is also required in order to solve Eq. (17) for
Re
mf
and hence U
mf
. Except in the case of monosized
smooth spheres for which f = 1 and e
mf
- 0.4, the
values of f and e
mf
are more often than not unavail-
able, though e
mf
is relatively easy to measure
(Eastwood et al., 1969). Starting with Wen and Yu
(1966), based on assumptions of f and e
mf
or on equa-
tions relating f and e
mf
, or based more directly
on empirical ts to experimental data on U
mf
, many
investigators have proposed dierent combinations of
values for the parameters C
1
and C
2
. Several of these
are listed in Table 1.
There is evidence that a uidized bed at a given
porosity oers somewhat less resistance to uid ow
than the corresponding xed bed at the same porosity
(Happel and Epstein, 1954; Happel and Brenner, 1957;
Richardson and Meikle, 1961a; Barnea and Mednick,
1975), owing to the greater degree of freedom to
arrange themselves possessed by the mobile uidized
than by the immobile xed particles, though this eect
is undoubtedly less pronounced at e=e
mf
than
at higher voidages. Furthermore, the constants 150
and 1.75 in the Ergun equation are not universally
accepted, e.g., Carman (1937), based on considerable
experimental data, proposed 180 instead of 150 for the
viscous ow constant. Therefore the empirical t basis
above is, for the specic conditions tted, more reliable
than the others. The values of C
1
and C
2
in Table 1
have in most cases been subjected by their proponents
to this criterion and are simultaneously based, in some
cases, on assumptions about e
mf
and f.
Since the dierences in results for Re
mf
obtained in
applying the various combinations of C
1
and C
2
in
Table 1 can be signicant (Bin, 1986), some discrimi-
nation is warranted. If good estimates of both e
mf
and
f can be made (cf. Leva, 1959), then Eq. (17) should be
used in conjunction with Eqs. (18) and (19). If e
mf
is
Table 1 Parameters C
1
and C
2
in Eq. (17)
Investigation C
1
C
2
Applicability claims
Wen and Yu, 1966 33.7 0.0408 Various particle shapes
Bourgeois and Grenier, 1968 25.46 0.0382 Spheres only
Richardson, 1971 25.7 0.0365 Spheres only
Saxena and Vogel, 1977 25.28 0.0571 Various particle shapes
Babu et al., 1978 25.25 0.0651 Various particle shapes
Grace, 1982 27.2 0.0408 Various particle shapes
Foscolo et al., 1983 25.74 e
4.8
mf
,0.336 Spheres
Thonglimp et al., 1984 31.6 0.0425 Various particle shapes
Chitester et al., 1984
Lin and Fan, 1997
28.7 0.0494
High pressure; spheres only
for liquid uidization
Nakamura et al., 1985 33.95 0.0465 High temp. & press., spheres only
Lucas et al., 1986 29.5 0.0357 Round granular particles,
0.8 _ f _ 1
32.1 0.0571 Sharp jagged particles,
0.5 _ f - 0.8
25.2 0.0672 Other particles, e.g., rings,
0.1 - f - 0.5
Chen, 1987 33.67f
0.1
0.0408/f
0.45
All particle shapes, but must
estimate f = d
sv
,d
p
= d
2
p
,d
2
s
Chyang and Huang, 1988 33.3 0.0333 Granular particles
Tannous et al., 1994 25.83 0.043 Many particle shapes
unknown, but a reasonable estimate of f is available,
then Eq. (17) should be applied with C
1
and C
2
eval-
uated from the expressions of Lucas et al. (1986) given
in Table 1. Alternately, the previous procedure can be
used, after measuring e
mf
(as described immediately
below) or estimating it from one of the simpler and
more creditable of several proposed empirical approx-
imations in the literature for relating e
mf
to f, such as
that of Chen (1987),
e
mf
=
0.415
f
0.483
(20)
for 0.2 - f _ 1 and d
p
> 0.05 mm (Wen and Yu,
1966). Except for spheres (f = 1), Eq. (20) shows e
mf
to be somewhat smaller than the corresponding ran-
dom loose voidages of Brownell et al. (1950), but
larger than e
mf
(f) from empirical equations proposed
by others (Wen and Yu, 1966; Limas-Ballesteros, 1980,
as stated by Couderc, 1985). If neither U
mf
nor f are
known, then Eq. (17) should be used with the values of
C
1
and C
2
given for the three classes of particles listed
beside Lucas et al. (1986) in Table 1.
Measurement of e
mf
is best eected by uidizing the
particles involved with a liquid, e.g., water, in the same
column for which the subsequent uidization opera-
tion will be carried out and then rapidly shutting o
the liquid ow, the volume of particles (M/r
p
) divided
by the volume of collapsed bed (AL
mf
) representing
1 e
mf
. If that is not feasible, a simpler procedure
for reproducing the random loosed voidage, which at
least for granular particles (f > 0.8) is a good approx-
imation of e
mf
, is to invert a covered cylindrical vessel
partially lled with the particles and then quickly right-
ing it. If this vessel is smaller in diameter D
c
than the
eventual uidizing column, a correction for wall eect
may be required if D
c
,d
p
- 100 (Eastwood et al.,
1969). Enough particles should be used by either
method so that L
mf
_ 2D
c
.
A second route to U
mf
is by putting U = U
mf
(or
Re = Re
mf
) and e = e
mf
in one of the many uidized
bed expansion equations relating e to U (or Re) that
have been proposed in the literature, a subject dis-
cussed below. This could provide a reasonable second
estimate of U
mf
but, because the expansion equations
are usually correlations based on the whole range of e
from e
mf
to e
c
or even to unity, they are likely to be less
accurate at the e
mf
extremity than an equation such as
Eq. (17) which is tailored to this extremity.
A third alternative for U
mf
is the use of empirical
equations other than Eq. (17), which have usually been
developed for more restrictive conditions. Thus, for
liquid uidization of spheres, based on an empirical
t for a wide variety of solidliquid combinations
over the range of Re
mf
= 101000, Riba et al. (1978)
have proposed
Re
mf
= 0.0154Ar
0.66
r
p
r
r
_ _
0.04
(21)
The fact that this equation does not require any speci-
cation of f or e
mf
is distinctly in its favor, since equa-
tions such as (17)(19) above are very sensitive to e
mf
,
and even for spheres (f = 1) an exact specication of
e
mf
can be elusive (Eastwood et al., 1969). The density
ratio, (r
p
r)/r, can safely be eliminated from this
equation since, for all the solidliquid combinations
investigated, the factor (r
p
r/r)
0.04
was never greater
than the scatter of the experimental data. (The index
on Ar could then be rounded o to 2/3). Equations for
U
mf
by all three routes are listed in Table 1.6 of
Couderc (1985).
Both e
mf
and hence also U
mf
(see Fig. 4) can be
signicantly reduced by pulsing the uidizing liquid
(El-Temtamy and Epstein, 1986). On the other hand,
neither is aected signicantly by changes in solids wett-
ability (Mitra and Epstein, 1978). For non-Newtonian
liquids that can be represented by the power-lawmodel,
an approach analogous to Eq. (17) has been successful
in predicting U
mf
for liquidsolid uidization of mono-
size spheres (Miura and Kawase, 1997, 1998).
For solids of mixed sizes, assuming any hysteresis
eects have been eliminated by rst uidizing the solids
and then gradually reducing the liquid velocity to zero,
there is no longer a sharp transition from a xed to a
uidized bed as in Fig. 2. Instead, on raising the velo-
city, there is a gradual transition from the fully xed
bed to the fully uidized bed (Couderc, 1985), as illu-
strated in Fig. 6. The initial deviation from the fully
xed bed line occurs at U
bf
, the velocity for beginning
uidization; the fully uidized region starts at U
tf
, the
velocity for total uidization; and the intersection of
the extrapolated xed bed line with the extrapolated
uidized bed line happens at U
mfa
, the apparent mini-
mum uidization velocity (Obata et al., 1982; Casal
and Puigjaner, 1983). U
bf
and U
tf
can be calculated
approximately from Eq. (17), with d
p
taken as the
smallest and the largest particle size, respectively,
while U
mfa
may be estimated from the same equation,
with d
p
evaluated as the Sauter or reciprocal mean
diameter (Jean and Fan, 1998),
d
p
=
1
N
i=1
v
i
,d
pi
_ _
(22)
which, for particles of xed shape, is the equivalent of
the surface-volume mean diameter. Figure 6 would
also apply to particles of mixed shapes and densities
(as well as mixed sizes), for which estimation of U
bf
and U
tf
would require that the particle properties in
Eq. (17) be those of the particles with the lowest and
the highest individual values of U
mf
, respectively, while
estimation of U
amf
would require that r
p
in Eq. (17) be
the mean particle density,
r
p
=
N
i=1
M
i
N
i=1
(M
i
,r
pi
)
(23)
In the case of an upwardly diverging tapered or conical
bed (e.g., y = 58308) of uniform particles, the bottom
of the bed is uidized before the top because of its
higher liquid velocity (Kwauk, 1992). The result is
that, on increasing the velocity through the initially
xed packed bed, a very much higher and sharper
peak pressure drop is reached than in Fig. 2 before
the pressure drop falls again and full uidization is
achieved (Peng and Fan, 1997). The plot of frictional
pressure drop vs. liquid velocity then resembles that of
a gas-spouted bed (Mathur and Epstein, 1974), as do
the corresponding uidparticle mechanics, and, as in
the case of both uidization and spouting, reducing the
velocity produces a hysteresis eect whereby there is no
longer (or at most barely) a discernible maximum in
the plot of p
f
vs. U. Peng and Fan (1997) identied
four transition velocities between an immobile packed
bed and full uidization: (1) the minimum velocity of
partial uidization, U
mpf
, corresponding to the point at
which the pressure drop is a maximum as the velocity
is increased; (2) the minimum velocity of full uidiza-
tion, U
m
, above which the bed expands at a constant
pressure drop; (3) the maximum velocity of partial
deuidization, U
mpd
, below which the bed is no longer
totally uidized as the velocity is decreased; (4) the
maximum velocity of full deuidization, U
mfd
, below
which a xed packed bed is fully restored on reducing
the velocity. In general, U
mfd
- U
mpf
- U
mff
= U
mpd
,
and equations for each of these velocities have been
developed by Peng and Fan based on integration of
Eq. (15) using the conical boundary conditions appro-
priate to each.
4 FLUIDIZED BED EXPANSION
4.1 Introduction
Crucial to the design of a uidized bed is a quantitative
knowledge of the bed expansion as a function of the
liquid supercial velocity, i.e., of either L or e [the two
being interrelated by Eq. (13)] as a function of U. The
chaotic behavior of a uidized bed, in contrast, say, to
viscous ow through an ordered (e.g., simple cubic)
array of immobilized spheres, renders measurements
from the former inherently less reproducible than
from the latter.
The equations that have been developed for predict-
ing the frictional pressure drop due to ow through a
xed, including an expanded, packed bed (of which
there are many in the literature) yield a relationship
between dp
f
,dz, e, and U that, when combined with
Eq. (8) to eliminate dp
f
,dz, give an equation relating
e to U. Because, as mentioned earlier, particles in a
uidized bed, in contrast to those in a rigid array,
are relatively free to move around and arrange them-
selves so as to minimize or at least decrease the resis-
tance to ow (Richardson and Meikle, 1961), therefore
for a given value of e, U for a uidized bed will usually
be larger than U for a xed bed under the same pres-
sure gradient. Consequently, the use of a packed bed
equation will in most cases underestimate U for a given
e, or overestimate e and L for a given U (Happel and
Epstein, 1951). If, for example, we remove the mini-
mum uidization (subscript mf) restriction from Eqs.
(15)(19), then Eq. (17) becomes
Figure 6 Frictional pressure drop vs. liquid supercial velo-
city for particles of mixed sizes, shapes, and/or densities.
(After Couderc, 1985.)
Re =
42.86
2
1 e ( )
2
f
2
0.5714fe
3
Ar
_ _
1,2
42.86 1 e ( )
f
(24)
For a bed of smooth spheres (f = 1) with d
p
= 0.5 mm
and r
p
= 2000 kg/m
3
, contacted by water at 208C,
Ar = 1225. If the bed is immobilized at e = 0.6, then
by Eq. (24), Re = 3.95. If, however, in the absence of
wall eect (k = 1), Eqs. (34), (33), and (32) below are
used to compute n, Re
0
, and Re, respectively, then for
the same spheres uidized by the same liquid to a
voidage of 0.6, n = 3.09, Re
0
= 27.9, and Re = 5.76.
Thus the required uidization velocity is in this case
almost 50% greater than that predicted by Eq. (24). (It
should be noted that for voidages greater than about
75%, the Ergun (1952) equation, from which Eq. (24)
originates, starts progressively to break down even as a
predictor of pressure drop through an immobilized
bed, and it fails entirely to approach the correct limit
as e approaches unity. Several other immobilized bed
equations in the literature have been proposed to
remedy this deciency.)
The relative freedom of particle movement charac-
teristic of uidization is shared also by the initial con-
stant rate stage of sedimentation, of which there is an
enormous literature, including a large number of
studies in which the two operations are jointly or inter-
changeably correlated. Assuming an equivalent distri-
bution of particles and its uctuations, it is easily
shown that the liquid supercial velocity required to
uidize a bed of particles to a given voidage is equal
(though opposite in sign) to the initial hindered settling
velocity at the same voidage of these particles relative
to the walls of a sedimentation vessel with an imper-
vious bottom. In the former case, the relative velocity
between the moving liquid and the stationary parti-
cles is U,e 0 = U,e. In the latter case, the downward
motion of the particles at velocity V
p
and volumetric
ux V
p
(1 e) results in an upward movement of
the displaced liquid at a linear velocity equal to
V
p
1 e ( )
_
e. The relative velocity between liquid
and particles is therefore V
p
1 e ( )
_
e V
p
=
V
p
_
e. Assuming, after Mertes and Rhodes (1955)
and Lapidus and Elgin (1957), that for monodisperse
uidparticle systems, the relative or slip velocity is
a unique function of voidage (or vice versa) irrespective
of the ow direction of either uid or particles, it fol-
lows that U = V
p
. Thus plots of V
p
vs. e have fre-

quently been interpreted as equivalent to those of U vs.
e. Some caution should be exercised in assuming this
equivalence, as there is some evidence that the design
of the distributor (which is absent in sedimentation),
through its eect on the uniformity of the liquid ow,
may exert some inuence on the expansion behavior of
a liquid uidized bed (Adler and Happel, 1962; Jean
and Fan, 1989); in addition, the upward moving liquid
in the uidization column is more likely to develop a
boundary layer induced velocity prole radially
(Neuz il and Hrdina, 1965) than the backowing liquid
in the sedimentation vessel, especially at low particle
concentrations (high voidages). Sedimentation follow-
ing agitation of the suspension and/or vertical rotation
of the sedimentation vessel may give rise to particle
settling velocities somewhat dierent from sedimenta-
tion initiated by rst uidizing the particles with super-
cial liquid velocity U and then allowing them to settle
without liquid ow (Jean and Fan, 1989), V
p
for the
latter then being closest to U. In general, the dier-
ences between the velocityvoidage relationships for
sedimentation as opposed to uidization can be con-
sidered second-order discrepancies, especially when
compared with the larger dierences between either
of these operations and ow through an immobilized
packed bed, so that for most engineering purposes,
results for constant rate hindered settling can be
applied with little or no modication to liquid uid-
ization.
4.2 Theoretical Models
Analytical solutions to the bed expansion problem,
without resort to empiricism, are almost entirely lim-
ited to spheres in creeping (i.e., viscous or Stokes) ow,
Re
0
< 0.2. Many of these are discussed by Happel and
Brenner (1965) and later by Jean and Fan (1989).
It is instructive, after Saman (1973), to compare
the expansion characteristics that are predicted for
spheres in a regular periodic array, in a random
array in which the particles are held rigidly, and in a
random array of freely moving particles. The rst case
corresponds to a geometrically ordered immobilized
packing, the second to a randomly immobilized pack-
ing, and only the third resembles a homogeneously
uidized bed. Unfortunately, the rigorous statistical
solution of the NavierStokes equation in creeping
ow (Happel and Brenner, 1965) for this third case,
by Batchelor (1972), is limited to very dilute suspen-
sions (e.g., c = 1 e - 0.05). This solution is
U
U
0
= 1 6.55c (25)
As an example of a regular periodic array, also dilute,
we can cite the analytical solution of Hasimoto (1959)
for a simple cubic lattice:
U
U
0
= 1 1.76c
1,3
(26)
For both a face-centered cubic and a body-centered
cubic array, the coecient of c
1,3
in Eq. (26) changes
only slightlyto 1.79 (Hasimoto, 1959).
Rigid random arrays have generally been simulated
by cell models that have not been limited to dilute
suspensions. An early example of a cell model is that
of Brinkman (1947), who considered ow past a single
sphere in a porous medium of permeability k. The ow
is described by an equation that collapses to Darcys
(1856) law (in its post-Darcy form, which includes vis-
cosity) for low values of k and to the creeping ow
version of the NavierStokes equation for high values
of k. His solution is
U
U
0
= 1 0.75c 1
8,c ( ) 3
_
_ _
(27)
which, though it breaks down for e _ 1,3 (i.e., c _ 2,3,
for which U,U
0
_ 0), nevertheless gives reasonable
answers in the uidization region, 0.4 - e - 1, and
even agreement with the lower values of U/U
0
reported
in the literature for liquid-uidized beds at low Re and
e=0.51 (Verschoor, 1951). To render Eq. (27) com-
parable to Eqs. (25) and (26), we rewrite it for the case
of c 0, for which it collapses to
U
U
0
- 1 1.5
2
_
_ _
c
1,2
- 1 2.12c
1,2
(27a)
A more rigorous cell model, inasmuch as it involves a
solution of the NavierStokes equation for creeping
ow rather than a modication of that equation as in
the case of Brinkman, is that of Happel (1958). In this
case the basic cell is that of a single sphere surrounded
by a concentric spherical envelope of uid, the volume
of which bears the same ratio to the volume of the cell
as bed voidage does to unity. The crucial feature of this
model is that the outer surface of the uid envelope
is frictionless (zero shear stress), so that it is often
referred to as the free surface model. The solution is
U
U
0
=
3 4.5c
1,3
4.5c
5,3
3c
2
3 2c
5,3
(28)
which at low concentrations reduces to
U
U
0
- 1 1.5c
1,3
(28a)
A comparable cell model is that of Kuwabara (1959),
the only dierence being that the spherical cell surface
is in this case assumed to be at zero vorticity rather
than at zero shear stress. The coecient of c
1/3
in
Eq. (28a) for dilute suspensions becomes 1.8 in
Kuwabaras solution, instead of 1.5.
It is notable that in the equations for both the
ordered arrangement of spheres and the cell models,
1 (U,U
0
) for dilute suspension is directly propor-
tional to c
1/3
(or c
1/2
in the case of Brinkmans
model) rather than to c as in Eq. (25), a fact that
would appear to render those equations inadequate
for uidization or sedimentation purposes (Batchelor,
1972; Saman, 1973). If we compare the actual values
of 1 (U,U
0
) predicted by the above equations at
c = 0.01, we nd that 1 (U,U
0
) = 0.0655, 0.379,
0.212, and 0.323 from Eqs. (25), (26), (27a), and
(28a), respectively, which, combined with experimental
ndings that values of 1 (U,U
0
) for uncharged set-
tling spheres at c = 0.01 are even lower than 0.0655
(Buscall et al., 1982; Tackie et al., 1983), conrms
this inadequacy, especially for dilute suspensions.
Two attempts have been independently made to
modify the free surface model of Happel (1958) so
that it includes the additional degrees of freedom of
movement characteristic of uidization or sedimenta-
tion as opposed to a xed array of spheres. Jean and
Fan (1989) incorporated a vertical line-up of particle
pairs into doublets to account for possible alignment
(which reduces ow resistance) in a uidized bed, and
they also incorporated a normal distribution function
of e (= 0.41.0), with the standard deviation as a sys-
tem parameter, to account for voidage nonuniform-
ities. The standard deviation varied with distributor
design and liquid velocity, and by nonlinear regression
of experimental results at low Re
0
was found to reach a
maximum near an average bed voidage of 0.75, which
agrees roughly with Al-Salims (Eisenklam, 1967) nd-
ings at moderate Re
0
and with Trupps (1968) ndings
at high Re
0
(in the Newton region, Re
0
> 500) that
the root-mean-square of the local voidage uctuations
(and their frequency too, for Al-Salims data) are a
maximum at an average e - 0.70. The values of
U,U
0
thus predicted by Jean and Fan fall into closer
line with experimental values at low Re
0
than does the
unmodied Eq. (28). More recently Smith (1998) mod-
ied Eq. (28) by allowing the quantity of liquid in each
spherical cell to vary randomly, subject only to the
constraints that the minimum local voidage = 0.2595,
corresponding to a rhombohedral close packing (which
is the densest possible for monosize spheres), and that
the overall voidage be maintained at the specied value
(between 0.5 and 1). Improved agreement with empiri-
cal results was again obtained and, unlike the solution
of Jean and Fan (1989) or even Eq. (25) above, the
present solution predicts sucient particle clustering
at low particle concentrations that U,U
0
exceeds
unity at c _ 0.03, approaching 1.07 as c approaches
zero. Values of U,U
0
exceeding unity to as high as
1.07 have been observed between c = 0.001 and
c = 0.03 by Kaye and Boardman (1962), who
explained their sedimentation results by the assump-
tion of cluster formation, which at low particle con-
centrations overrides the small return ow eect
associated with hindered settling. For even smaller
concentrations (c - 0.001), the values of U,U
0
mea-
sured by Kaye and Boardman, unlike the theoretical
values of Smith, reduce as expected to unity. The
minor defect in Smiths theory is inconsequential to
uidization practice, in which volumetric particle
concentrations are invariably larger than 0.001.
The zero vorticity cell model has been extended
numerically to intermediate Reynolds numbers for
spheres (LeClair and Hamielec, 1968), and both the
zero vorticity and the free-surface cell models (unmo-
died) to spheroids at low Reynolds numbers (Epstein
and Masliyah, 1972). However, these extensions are
more applicable to immobilized packed beds than to
uidized beds, and only in the absence of turbulence,
which for unexpanded, xed packed beds of spheres
develops at Re > 110150 (Jolls and Hanratty, 1966).
Grbavc ic et al. (1991) have developed an intuitive
variational method of predicting U for any given value
of e that is not restricted to any given ow regime. The
method, which has been applied successfully to water
uidization of glass spheres over the range of Re
0
=
180920 (Grbavc ic et al., 1991), is succinctly summar-
ized by Jean and Fan (1998), as
U = U
mf
e
3
1 e ( )a
/
mf
e
3
mf
1 e
mf
( )a
/
_ _
1,2
(29)
where
a
/
a
/
mf
= 1 B
2
( )
1
l
B
1 l
B
e e
mf
1 e
mf
_ _
B
1
_ _
2
_ _
1,2
(29a)
B
1
= 1
U
2
mf
U
2
e
3
mf
_ _
2
_
_
_
_
1,2
(29b)
B
2
=
(1 B
2
1
)
1,2
(1 B
2
1
)
1,2
B
1
(29c)
and
l
B
= 1 B
2
1
_ _
1,2
B
1
(29d)
This method, which requires prior knowledge of e
mf
,
U
mf
, and U
0
, predicts e all the way from e
mf
to unity
for particles of any shape. e
mf
and U
mf
can be deter-
mined as discussed in Sec. 3, while equations for U
0
will be given below. Outside the creeping ow region,
this method is the only serious theoretical comp-
etitor to the empirical methods discussed below for
predicting bed expansion.
4.3 Empirical Equations
Perhaps the simplest and certainly the most widely
used, if not always the most accurate, of the empirical
equations proposed in the literature for predicting the
expansion of a liquid uidized bed is that of
Richardson and Zaki (1954), which was anticipated
by several prior investigators (Hancock, 1937/38;
Lewis et al., 1949; Jottrand, 1952; Lewis and
Bowerman, 1952). In its most primitive form this
equation is simply
d log U
d log e
= n = constant (30)
i.e., a plot of log U vs. log e, or of U vs. e on loglog
coordinates, for any given combination of liquid and
monodispersed solids, can be well approximated by a
straight line of slope n, at least from e equal to or
somewhat greater than e
mf
to e=e
c
, where e
c
0.850
0.15 (see Figs. 4 and 7), i.e., for relatively concen-
trated suspensions. Commonly, Eq. (30) has been
integrated with the boundary condition U = U
0
at
e = 1, so that
U
U
0
= e
n
= 1 c ( )
n
(31)
which is the most commonly written form of the
RichardsonZaki equation. Unfortunately, plots of log
U vs. log e for concentrated suspensions, when linearly
extrapolated to log e = 1, often yield intercepts equal
to log U
e
, where U
e
_ U
0
(Fig. 7). Therefore Eq. (31)
must be modied to
U
U
0
=
Re
Re
0
_ _
= ke
n
(32)
where k = U
e
,U
0
_ 1.
To use Eq. (32) in order to predict U for a given
value of e (or vice versa) requires knowledge of U
0
, n,
k, and preferably also e
c
. For smooth spheres (f = 1),
to which we shall initially restrict this discussion, the
free-settling velocity U
0
is most conveniently calculated
from the empirical correlation of Turton and Clark
(1987),
Re
0
=
d
p
U
0
r
m
= Ar
1,3
_
18
Ar
2,3
_ _
0.824
0.321
Ar
1,3
_ _
0.412
_
1.214
(33)
which applies up to Re
0
= 260,000 (Haider and
Levenspiel, 1989).
The index n for spheres was originally correlated by
ve empirical equations relating this index to Re
0
and
the wall eect ratio, d
p
,D
c
(Richardson and Zaki,
1954). However, the wall eect on n has not been con-
rmed by subsequent investigators (Chong et al., 1979;
Di Felice and Parodi, 1996), and several investigators
have proposed a single equation relating n to either
Re
0
(Wallis, 1969; Garside and Al-Dibouni, 1977;
Rowe, 1987) or Ar (Khan and Richardson, 1989),
the advantage of the latter being that Ar, unlike Re
0
,
can be calculated directly without further ado if the
particle and liquid properties are known. The equation
of Khan and Richardson, which has been carefully
adjusted to the available data, is
4.8 n
n 2.4
= 0.043Ar
0.57
(34)
This equation, as well as two of the others, reect the
fact that the upper limit of n, for spheres in the Stokes
region (Re
0
< 0.2, Ar < 4), is somewhere between 4.6
and 4.9, while the lower limit, for spheres in the
Newton region (Re
0
> 500, Ar > 85,000), falls
between 2.3 and 2.4. This much variation in n indicates
that, whereas U
0
exceeds U
mf
in the Stokes regime by
about two orders of magnitude, it does so by only one
order of magnitude in the Newton regime. The equa-
tion of Garside and Al-Dibouni yields values of n
about 0.3 higher than Eq. (34) for all values of Ar
and Re
0
, possibly due to the inclusion, in their deter-
minations of n, of data points for which e exceeds e
c
and which showed a slope n
/
exceeding n, as illustrated
in Fig. 7.
Greater uncertainty is associated with the prediction
of k. In the original paper by Richardson and Zaki
(1954), it was concluded that k = 1 for sedimentation,
while for uidization k was simply a wall eect factor
given by
k = 10
d
p
,D
c
(35)
and independent of Re
0
. Subsequently, however, Khan
and Richardson (1989) have dropped any distinction
between sedimentation and uidization as far as k is
concerned, and have proposed, on the basis of a regres-
sion analysis of experimental data covering the range
Re
0
= 0.017000 and d
p
/D
c
= 0.0010.2, that
k = 1 1.15
d
p
D
c
_ _
0.6
(36)
Equation (36), about which there is considerable data
scatter (coecient of correlation = 0.867), shows a
much larger wall eect than Eq. (35). To complicate
the issue still further, Rapagna` et al. (1989) have found
experimental values of k that are in some cases even
lower than those given by Eq. (36) over the range
Re
0
= 501000, for which they proposed the following
correlation, ignoring d
p
/D
c
, again with considerable
data scatter:
k = 1.09 0.11 log Re
0
(37)
On the other hand, based on the work of Fan et al.
(1985), in the absence of wall eect, k - 1 for
Re
0
< 35 (Di Felice, 1995), which agrees qualitatively
Figure 7 Expansion characteristics of a stable liquid ui-
dized bed. Concentrated suspensions follow bold solid line
of slope n, which may at e = e
c
change to either slope n
/
> n
all the way to e = 1 (bold dotted line), or slope n
//
- n to
e - 1 (bold dashed line).
with the trend displayed by Eq. (37). In fact, the data
of Rapagna` et al. for Re
0
_100 are better correlated by
the equation
k =
7
Re
0
_ _
0.8 (38)
which extrapolates to k = 1 at Re
0
= 35. It is therefore
recommended that, for Re
0
= 1001000, k be taken as
the lower of the values given by Eqs. (36) and (37); for
Re
0
= 35100, as the lower of Eqs. (36) and (38); and
for Re
0
< 35, as the value given by Eq. (36), i.e., unity
in the absence of wall eect. For Re
0
> 1000, it is not
known whether the trend of k with Re
0
shown by Eq.
(37) continues, but Eq. (36) is still presumably applic-
able to account approximately for any wall eect.
It is possible in many cases to reduce the uncertainty
associated with an assumed value of k by integrating
Eq. (30) with the boundary condition U = U
mf
at
e = e
mf
, so that
U
U
mf
=
e
e
mf
_ _
n
(39)
Dividing Eq. (31) by Eq. (38) in order to solve for k, we
nd
k =
U
mf
U
0
e
n
mf
(40)
If e
mf
and U
mf
are known or determined by the
methods described in the previous section, and n is
determined from Eq. (34), then k can be estimated
from Eq. (40). Agreement, say, within 5% of a pre-
vious estimate should bolster condence in that
estimate. Disagreement, however, even assuming that
e
mf
, U
mf
, and n are correct, could be due to an unsa-
tisfactory original estimate of k, but it could also be
due to the fact that the primary straight line relation-
ship between log U and log e sometimes starts at a
value of e somewhat in excess of e
mf
(Wilhelm and
Kwauk, 1948; Richardson and Zaki, 1954).
The critical voidage, e
c
, above which there is a
change of slope in the plot of log U vs. log e for con-
centrated suspensions, varies between 0.85 and unity
when the slope change is positive, e.g., at Re
0
= 50
1000, according to Rapagna` et al. (1989), and between
0.85 (or even 0.9, Chong et al., 1979) and 0.7 when the
change of slope is negative, which occurs at Ar _ 1600
(Re
0
_ 35) according to Fan et al. (1985). Most liquid
uidization applications occur at e - e
c
.
Referring to Fig. 7 it can be seen that, discounting
those situations in which the slope of log U vs. log e
remains unchanged all the way to U = U
e
= U
/
0
(i.e.,
e
c
- 1), the region in which e exceeds e
c
bifurcates on
loglog coordinates into two routes. In the rst route,
exemplied by the data of Rapagna` et al. (1989), the
dilute suspension data follow a straight line of slope
n
/
> n, which these investigators correlated very
roughly by the equation
n
/
= 6.4 0.61 log Re
0
(41)
Thus in the absence of wall eects,
U
U
0
= e
n
/
= 1 c ( )
n
/
(42)
which for dilute solutions is well approximated by
U
U
0
1 n
/
c (42a)
Equation (42a) is of the same form as Eq. (25) above
derived by Batchelor (1972) for dilute suspensions in
viscous ow (Re
0
_ 0.2). Note that as Re
0
decreases,
the value of n
/
in Eq. (41) appears to be approaching
the 6.55 of Eq. (25), and a value of n
/
- 6.5 was actu-
ally found by Di Felice and Parodi (1996) in some of
their viscous sedimentation experiments. In the pre-
sence of a measurable wall eect (d
p
,D
c
> 0.01), the
intercept of the dilute suspension line is U
/
0
, which for
viscous ow (Di Felice and Parodi, 1996) is related to
U
0
by either of the two empirical equations proposed
by Francis (1933) to describe the settling velocity of a
ball falling axially through a liquid in a cylindrical
vessel. The simpler of the two equations, which give
almost identical results up to d
p
/D
c
= 0.3, is
U
/
0
U
0
= 1
d
p
D
c
_ _
2.25
(43)
Equation (43) yields answers very close to those
derived theoretically by Ladenburg (1907) and Faxe n
(Emersleben and Faxe n, 1923) up to d
p
,D
c
= 0.1 and
0.2, respectively, and agrees fortuitously also with
the value of k(= U
e
,U
0
) given by Eq. (35) up to
d
p
,D
c
= 0.2. The agreement of U
/
0
by these equations
with the dilute suspension intercept at e = 1 has been
demonstrated by Garside and Al-Dibouni (1977) for
d
p
,D up to 0.06 and Re
0
up to 1200, i.e., well beyond
viscous ow. The corresponding equation of Munroe
(198889), obtained by uidizing single spheres in a
cylindrical vessel in the range of Re
0
= 1503400,
shows a much smaller wall eect. An empirical equa-
tion for U
/
0
,U
0
proposed by Di Felice (1996), which
spans the whole range of Re
0
from the Stokes to the
Newtonian region, including the data of Fidleris and
Whitmore (1961) in the intermediate region, is
U
/
0
U
0
=
1 d
p
,D
c
_ _
1 0.33d
p
,D
c
_ _
_ _
a
(44)
where the single parameter a is given by
3.3 a
a 0.85
= 0.1 Re
0
(44a)
A more accurate two-parameter model was subse-
quently developed by Kehlenbeck and Di Felice
(1999).
The other dilute suspension route illustrated in Fig.
7 involves an initial slope decrease to n
//
- n, which
usually occurs at Re
0
_ 35 (Chong et al., 1979; Fan
et al., 1985; Di Felice, 1995). The slope n
//
has been
correlated with Ar by Fan et al. (1985), as has the
corresponding intercept U
/
e
- U
e
( ), obtained by linear
extrapolation to e = 1. The nal termination of the
actual (as opposed to the extrapolated) data points
for the very dilute suspensions is at e = 1, U = U
/
0
(Ben a et al., 1963). It should be noted that the rather
abrupt changes of slope on loglog coordinates illu-
strated in Fig. 7 do not occur when the same data
are plotted on arithmetic coordinates, a fact that is
important for the variational method of Grbavc ic et
al. (1991) summarized earlier.
Liquid uidized beds of nonspherical, rough or very
ne particles have been shown by many investigators
(Jottrand, 1952; Lewis and Bowerman, 1952;
Richardson and Zaki, 1954; Whitmore, 1957;
Richardson and Meikle, 1961a; Fouda and Capes,
1977; Chong et al., 1979; Cleasby and Fan, 1981) to
expand according to Eq. (32), with k depending not
only on d
p
,D
c
and possibly on Re
0
but also on spheri-
city f (Dhamarajah and Cleasby, 1986). The depen-
dence of k on d
p
,D
c
is again better represented by
Eq. (36) than by Eq. (35) (Dhamarajah and Cleasby,
1986). The value of n in Eq. (32) is always greater than
that given by Eq. (34). This positive deviation increases
with decreasing sphericity, decreasing particle size, and
decreasing Re
0
. Thus at Re
0
< 0.1, n for particles with
d
p
> 280 mm increased from 4.8 to 5.8 as f decreased
from unity to 0.7 (Chong et al., 1979) and increased
from 6.9 to 9.5 as d
p
of methacrylate powder decreased
from 194 to 65 mm (Whitmore, 1977); while n decreased
from about 3.55.5 (over a wide sphericity spread) at
Re
0
= 10 (Ar -310) to about 2.42.6 (over a similarly
wide spread in f) at Re
0
= 1000 (Ar - 330,000)
(Dhamarajah and Cleasby, 1986). The last observation
indicates that the particle shape eect on n decreases
markedly as Re
0
increases. However, no reliable quan-
titative correlations between n and f, d
p
, or Re
0
have
yet been developed. Chianese et al. (1992) scored a
modicum of success in predicting their experimental
values of n (5.414.90 for Re
0
= 6.631.0) for water
uidization of sodium perborate crystals (f = 0.52)
by means of Eq. (40), assuming k = 1, evaluating e
mf
from one of the two equations relating e
mf
to f pro-
posed by Wen and Yu (1966) [though Eq. (20) above
gives even better agreement with their experimentally
measured values of e
mf
(-0.56)], U
0
by means of the
isolated sphere correlations of Clift et al. (1978) assum-
ing equivolume spheres, and U
mf
by means of Eq. (21)
above (despite its proposed restriction to spheres and
to a range of Re
mf
considerably higher than those
obtained in this study), using the surface-volume dia-
meter d
sv
= fd
p
as d
p
in the Reynolds and Archimedes
numbers. A more recent and convenient empirical
equation for determination of U
0
for nonspherical iso-
metric particles is that of Haider and Levenspiel
(1989),
Re
0
=
d
p
U
0
r
m
= Ar
1,3
18
Ar
2,3
2.335 1.744f
Ar
1,6
_ _
1
(45)
which is recommended for Re
0
< 25,000 and
0.5 _ - 1.
Another approach to nonspherical particles and/or
rough spheres, originating with Steinour (1944) and
adopted by many other sedimentation workers,
assumes that such particles immobilize liquid around
their surface irregularities and thereby behave as
smooth spheres with an eective volumetric concentra-
tion Kc, where the hydrodynamic volume factor K is
the volume of the liquid envelope plus solid divided by
the solid volume. The eective voidage then becomes
1 Kc, the eective particle density r
eff
=
[r
p
r(K 1)],K, and the eective particle diameter
d
eff
= d
p
K
1,3
. By this means, Gasparyan and Ikaryan
(1962) attempted to reconcile their hindered settling
results for nonspherical particles with the original
empirical equations of Richardson and Zaki (1954)
for spheres. The same approach was later applied to
liquid uidization by Fouda and Capes (1977), who
determined n and k for their experiments from the
original empirical equations of Richardson and Zaki
(1954) for spheres, U
0
from tables provided by
Heywood (1962), and K iteratively by tting the equa-
tion
U
U
0
= k(1 Kc)
n
(46)
to their experimental data. They found values of K that
ranged from 1.2 for crushed silica to values sometimes
exceeding 3 for at mica particles. Subsequently they
discovered that all their results (for crushed silica,
crushed steel, aluminum squares, and mica) could be
correlated by (Fouda and Capes, 1979; Epstein, 1979)
K(1 e
b
) = 0.603 0.016 (standard deviation)
(47)
where e
b
is the voidage of the bed after the liquid ow
has been shut o and the bed has settled. Assuming
that this settled bed voidage is equivalent to e
mf
(Eastwood et al., 1969), then the combination of Eqs.
(47) and (20), with a slight correction to render the
result compatible with smooth spheres (for which
f = 1 and K = 1), results in
K =
0.603
1 0.415,f
0.483
0.03 (48)
Equation (46), in conjunction with Eq. (48) and the
previous equations recommended for n, k, and U
0
[assuming spheres of diameter d
e
and density r
e
, or
alternately U
0
by Eq. (45)], is thus a coherent method
for dealing with the expansion of nonagglomerating
nonspherical particles, albeit one that requires further
testing, qualication and elaboration. Note that a high
value of n in Eq. (32) is accompanied by a high value of
K in Eq. (46), in which the value of n is that from Eq.
(34) for unocculated spheres, i.e., the lowest possible
for the given Ar or Re
0
.
The increase of n as d
p
decreases and hence as spe-
cic surface increases could in part be due to the elec-
trokinetic Dorn eect, which retards sedimentation in
ionic liquids (Tackie et al., 1983; Masliyah, 1994), and
is most certainly associated with the occulation of
colloidal particles (Dixon, 1977). Equation (46) is
again applicable to colloidal suspensions, with K now
denoting the volume of the ocs divided by the volume
of the particles contained by the ocs, and U
0
the
terminal free settling velocity of an average oc rather
than of the particles within the oc (Michaels and
Bolger, 1962). The values of K in Eq. (46) and of n
in Eq. (32) are now typically an order of magnitude
greater than for larger nonagglomerated particles
(Dixon, 1977).
Another approach to bed expansion is based on a
postulate by Wen and Yu (1966), anticipated by the
drag force measurements of Rowe and Henwood
(1961), Rowe (1961), and Richardson and Meikle
(1961b) on immobilized beds of spheres at various voi-
dages and Reynolds numbers. This postulate states
that the drag force F
D
on a given sphere in a uidized
swarm of monosized spheres to the drag force F
DS
on a
single isolated sphere past which the same uid ows at
the same supercial velocity, U, is independent of the
Reynolds number, Re = d
p
Ur,m and depends only on
the voidage, e. Thus
F
D
F
DS
= f (e) = e
b
(49)
To arrive at an expression for f(e), we rst consider the
Stokes regime, for which
F
DS
F
D0
=
3pmUd
p
3pmU
0
d
p
=
U
U
0
(50)
where F
D0
is the drag force exerted on an isolated
sphere at its terminal settling velocity U
0
. But from
Eq. (34), in the Stokes regime (Ar < 4, Re
0
< 0.2),
n = 4.8. Assuming k = 1 in Eq. (32), it follows that
F
DS
F
D0
= e
4.8
(51)
for creeping ow. Now consider the Newton regime,
for which
F
DS
F
D0
=
0.44pd
2
p
rU
2
,8
0.44pd
2
p
rU
2
0
,8
=
U
2
U
2
0
(52)
From Eq. (34), in the Newton regime (85,000
< Ar < 1.4 10
10
, 500 < Re
0
< 200,000), n = 2.4.
Therefore, again assuming k = 1 in Eq. (32), it follows
that
F
DS
,F
D0
= (e
2.4
)
2
= e
4.8
(53)
for the high Reynolds number regime. Since
F
D
= V(r
p
r)g (6)
it follows that F
D
= F
D0
, the drag force at the terminal
velocity being balanced by the same gravitational force
modied by the same buoyancy as given by Eq. (6).
Therefore
F
D
,F
DS
= f (e) = e
4.8
(54)
at both ends of the Reynolds number spectrum. Wen
and Yu (1966) assumed that the same voidage func-
tion, which they took as e
4.7
instead of e
4.8
, would
also apply to intermediate values of Re
0
. Therefore
F
D
F
DS
=
C
D
C
DS
= e
4.7
(55)
But for Re - 1000 (Schiller and Naumann, 1933),
C
DS
=
24
Re
_ _
(1 0.15 Re
0.687
) (56)
Substituting Eq. (56) into Eq. (55) and rearranging,
C
D
Re
2
e
4.7
= 24Re 3.6Re
1.687
(57)
In a gravitational eld characteristic of uidization, for
which F
D
is given by Eq. (6), in which V = pd
3
p
,6,
C
D
Re
2
=
pd
3
p
,6
_ _
r
p
r
_ _
g
pd
2
p
rU
2
_
8
d
p
Ur
m
_ _
2
=
4d
3
p
r
p
r
_ _
rg
3m
2
=
4Ar
3
(58)
Eq. (57) is therefore equivalent to
Are
4.7
= 18Re 2.7Re
1.687
(59)
which is explicit in e if Ar and Re are known, and can
be solved for Re by rapid iteration if Ar and e are
specied.
It turns out that Eq. (59) applies well at both low
and high Reynolds numbers, but underpredicts Re
(and hence U) at intermediate values of Re, particu-
larly at the lower values of e (Khan and Richardson,
1990; Di Felice, 1995). It shares this characteristic with
an equation similar in form obtained by equating the
relationship for pressure gradient proposed later by
Foscolo et al. (1983) with the pressure gradient as
given by Eq. (8), viz.,
Are
4.8
= 17.3Re 0.336Re
2
(60)
and with another empirical equation similar in form
proposed still later by Hartman et al. (1992):
Are
4.73
= 20.4Re 1.44Re
1.80
(61)
To remedy the deciency of Eq. (60) at intermediate
Reynolds numbers, the equation was modied empiri-
cally by Gibilaro et al. (1986) by raising both the vis-
cous and the inertial terms on the right-hand side to a
common exponent and raising the sum of these terms
to the reciprocal of this exponent, which is expressed as
a function of voidage. The modied equations are
Are
4.8
= [ 17.3Re ( )
a
0.336Re
2
_ _
a
]
1,a
(62)
a = 2.55 2.1 tanh 20e 8 ( )
0.33
_ _
3
(63)
Equation (59) can be modied more simply by rewrit-
ing it as
Are
b
= 18Re 2.7Re
1.687
(64)
since b has been empirically, though roughly, tted by
Di Felice (1994) as
b = 4.7 0.65e
1.5log Re ( )
2
,2
(65)
[In this equation for b, the 3.7 given by Di Felice, who
couples his result with F
D
as given by Eq. (7), is here
changed to 4.7, since we have adopted the convention
for F
D
incorporated in Eq. (6)]. Knowing Ar (from
uid and particle properties) and specifying U and
hence Re, b can rst be determined from Eq. (65)
and then e from Eq. (64). If Ar and e are specied,
then Re can be determined iteratively by simulta-
neously satisfying both Eqs. (64) and (65).
Equations (59)(65) are all applicable to monosized
spherical particles. Additional empirical equations or
graphical methods (e.g., Barnea and Mizrahi, 1973)
for spheres relating Ar, e, and Re over various ranges
of Ar or Re, both of the same form as the above
equations and some of slightly dierent form, as well
as some, such as the logistic equation of Garside
and Al-Dibouni (1977), with the primary variables
assembled in dierent dimensionless groups, are tabu-
lated by Couderc (1985), Khan and Richardson (1989),
and Jamialahmadi and Mu ller-Steinhagen (1999).
Included also in the same or in dierent (Couderc,
1985) tables are relationships for nonspherical part-
icles, some of which show no shape inuence in the
actual equations, a notable exception being the equa-
tion of Limas-Ballesteros et al. (1982a) for relatively
high Reynolds numbers. Many of these equations
are critically discussed by Di Felice (1995), who inter-
changes the denitions of Ar [= d
3
p
(r
p
r)rg,m
2
)] and
Ga (= d
3
p
r
2
g,m
2
), following in this respect other
authors (e.g., Wen and Yu, 1966), some of whom per-
form this interchange inconsistently (e.g., Khan and
Richardson, 1989). It should also be noted that
GaMv [=Ga(r
p
r)/r] in Coudercs tables, or ArM
in Di Felices review, is equivalent to Ar.
For both upwardly diverging and downwardly
diverging tapered beds with an included angle of 58,
Di Felice et al. (1991a) found that liquid deceleration
or acceleration eects could be neglected, so that the
voidage for monodisperse solids at any level could then
be estimated by applying the expansion equations
above, e.g., Eqs. (32)(34) and (36)(38) for spheres,
at the liquid velocity prevailing at that level. For an
upwardly diverging bed with an included angle of 2.78,
Scott et al. (1978) correctly predicted the bed expan-
sion and the decreasing frictional pressure drop with
increasing U by combining the force balance given by
Eq. (11) with Eq. (59) for e, both applied in small
increments of bed height. Other investigators (Kola r,
1963; Koloini and Farkas, 1973; Maruyama et al.,
1984; Webster and Perona, 1988, 1990) of upwardly
diverging beds, however, found this procedure some-
what inaccurate even for cone angles as low as 0.58
(and as high as 25.18), usually overpredicting the
degree of expansion that actually occurs, and they
have recommended empirical equations for specic
cases, e.g., empirical determination of k and n in Eq.
(32) for a given cone angle and liquidparticle system.
The moderate bed contraction often observed com-
pared to that for the equivalent cylindrical column
can be explained by the additional bulk mixing gen-
erated in a tapered bed (Webster and Perona, 1988,
1990), a contraction eect of bulk mixing discussed
below in Sec. 9. The bulk mixing and consequent bed
contraction were probably attenuated in the study of
Di Felice et al. (1991a) by their use of a ball distribu-
tor, described later.
When particles of mixed size, shape, and/or density
are uidized by a liquid, they may, as will be discussed
later, segregate completely according to species (a given
size, shape, and density), segregate incompletely (i.e.,
mix partially), or mix completely. Irrespective of the
degree of segregation or mixing, however, the overall
bed expands as if it were simply the sum of the N indi-
vidual species, each acting independently of the other
(Lewis and Bowerman, 1953; Richardson and Zaki,
1954; Homan et al., 1960; Cleasby and Bauman,
1976; Epstein et al., 1981; Yu and Shi, 1985), so that
1 e =
N
i=1
c
i
,c
t
1 e
i
_ _
1
(66)
where e
i
is the voidage when species i is uidized alone
at the same supercial liquid velocity as the mixture,
and c
i
,c
t
(= v
i
) is the volume fraction of uid-free
solids which is species i. This experimentally veried
serial model works best in the absence of bulk cir-
culation within the uidized bed (Epstein et al., 1981)
but is in contradiction to theories (Gibilaro et al.,
1985b; Chiba, 1988) that predict some bed contraction
when particle mixing occurs, and to a few experiments
with fully mixed binary particle species where some
bed contraction was measured (Chiba, 1988). The
experimental diculty of detecting the predicted con-
traction probably arises from the smallness of this
eect, illustrated in the next section by the small dif-
ference between the binary line ABD and the closest
monocomponent lines below them in Fig. 9.
5 PARTICLE MIXING AND SEGREGATION
5.1 Deductions from Pressure Gradient
Measurements
When two or more species of particles are uidized by
a liquid in the same column, they may mix completely,
mix partially (i.e., segregate partially) or segregate
completely, depending on their relative sizes, shapes,
and densities. A common situation, e.g., in crystalliza-
tion, is that of particles having uniform shape and
density but diering in size. A constant axial pressure
gradient throughout the bed then denotes perfect mix-
ing of the particles, whereas variation of this gradient
signies segregation, the larger the variation the larger
the degree of segregation (Neuz il, 1964; Scarlett and
Blogg, 1967). For multisized solids in a liquid uidized
bed at supercial velocity U, the voidage at height
z above the distributor may be obtained from the
pressure gradient at that level by means of Eq. (8),
rewritten as
e = 1
dp
f
,dz
g(r
p
r)
(67)
By means of the bed expansion equations developed
above, e.g., Eqs. (32)(34) and (36)(38), we can relate
this value of e for the given r
p
, r, m, and U to a given
value of d
p
, namely d
p
(z); and in general,
e = f (d
p
) (68)
The mass dm
z
of particles in the bed over a dierential
height dz is given by
dm
z
= r
p
A(1 e)dz (69)
Therefore the mass m
z
of particles between the distri-
butor and the height z is given by
m
z
=
_
m
z
0
dm
z
= r
p
A
_
z
0
(1 e) dz (70)
Substituting for e according to Eq. (67),
m
z
=
r
p
A
g(r
p
r)
_
z
0
dp
f
dz
_ _
dz (71)
The total mass M of particles in the bed is then given
by
M =
r
p
A
g(r
p
r)
_
L
0
dp
f
dz
_ _
dz (72)
The cumulative fraction of particles larger than d
p
(z),
all of which are presumed to be located below level z,
becomes
m
z
M
=
_
z
0
dp
f
,dz ( ) dz
_
L
0
dp
f
,dz ( ) dz
=
_
z
0
1 f (d
p
)
_ _
dz
_
L
0
1 f (d
p
)
_ _
dz
(73)
The cumulative particle size distribution of the solids
can then be generated both as m
z
,M vs. z and as m
z
,M
vs. d
p
(the latter of which will coincide with the size
distribution of the solids as determined by other
means, e.g., screen analysis, only if perfect stratication
by size occurs). The corresponding plot of d
p
vs. z has
been dubbed the perfect classication model by Al-
Dibouni and Garside (1979). Even where some particle
mixing occurs, this model gives a good representation
of d
p
vs. z, where d
p
is either the mass average (Garside
and Al-Dibouni, 1974) or the mass median (Al-Dibouni
and Garside, 1979) local particle diameter at level z.
For the particular case of a binary solid mixture,
correct estimate of a local mean particle diameter
allows one to determine the local composition of the
solids. Thus Di Felice et al. (1987) employed Eqs. (62)
and (63) to determine the local mean diameter as a
function of bed level in the transition zone between
the fully segregated large particles at the bottom and
the fully segregated small particles at the top (Fig. 8c),
using the axial voidage proles that Juma and
Richardson (1983) obtained for their two binaries
(from the measured pressure proles) by means of
Eq. (67). Assuming, after Wen and Yu (1966), that
the computed diameters represented Sauter means,
they were thus able to calculate the concentrations of
large and small particles locally, and their answers
agreed reasonably well with the measured local con-
centrations of the two solid species. A more approxi-
mate estimate of these local concentrations was made
by Epstein and Pruden (1999) from the local voidages,
applying locally the serial model previously described
(in Sec. 4) for the overall bed.
In mineral processing, in addition to size, the part-
icle density and possibly the shape may also vary.
Using the measured variation of (dp
f
,dz) with z,
the procedure of Di Felice et al. (1987), just described,
can also be applied to determine the changing compo-
sition with bed level of two spherical particle species
diering in both diameter and density, by means of an
additional equation for the volume-average density,
r
p
=
c
1
r
p1
c
2
r
p2
c
1
c
2
(74)
used in Eq. (67) to related e to the measured frictional
pressure gradient. Thus Eqs. (67), (62), (63), and (74)
are solved iteratively along with the equation for the
Sauter mean diameter,
d
p
=
c
1
c
2
c
1
,d
p1
c
2
,d
p2
(75)
to obtain e, a, r
p
, d
p
, c
1
, and c
2
, where c
1
c
2
= 1 e
(Gibilaro et al., 1986a; Di Felice et al., 1987).
5.2 Binary MixturesOverview
To simplify the analytical description of the phenom-
ena involved in particle mixing and segregation, let us
continue to focus on a binary mixture of particles, i.e.,
two species (components). In the case of a size dier-
ence alone, the larger particles will always separate to
some degree below the smaller, unless the size dier-
ence is small enough and/or complicating factors such
as bulk circulation or hydrodynamic instability are
large enough to mix the two particle species com-
pletely. Under no circumstances, however, will the
larger particles segregate above the smaller (in conven-
tional as opposed to e.g., inverse uidization). If, in
addition, there is a shape dierence, it is that species
with the larger magnitude of the product fd
p
that will
normally segregate below. If there is also a particle
density dierence, then if that dierence is in the
same direction as that of fd
p
, the segregation will be
accentuated; if, however, the density dierence is in the
opposite direction, then depending on the magnitude
of this dierence, the segregation may be attenuated
throughout the entire voidage range as the liquid velo-
city is increased, it may be reversed over the entire
voidage range, or it may be reversed at the lower voi-
dages and attenuated at the higher. The last situation is
the interesting bed (or layer) inversion phenomenon,
which will be discussed later. First we will discuss the
one-way segregation phenomenon, as exemplied by a
mixture of particles of a given material having two
dierent sizes.
Figure 8 Mixing and segregation patterns for liquid uid-
ization of binary-sized particles of equal density. (From
Gibilaro et al., 1985.)
5.3 Binary Mixtures of Equal Density Particles
For simplicity, let us consider spheres of density r
p
and
of diameters d
L
and d
S
. where d
L
> d
S
, uidized by a
liquid having a supercial velocity, U, between U
mfL
and U
0S
. In the absence of the smaller particles, the
uidized bed suspension density will be given by
r
BL
= r
p
1 e
L
( ) re
L
(76)
while in the absence of the larger particles, the suspen-
sion density is given by
r
BS
= r
p
1 e
S
_ _
re
S
(77)
Subtracting Eq. (77) from Eq. (76), we obtain
r
BL
r
BS
= r
p
r
_ _
1 e
L
( ) r
p
r
_ _
1 e
S
( )
(78)
and the dimensionless density ratio, g = (r
BL
r
BS
)/
(r
p
r), is then given by
g = e
S
e
L
(79)
where e
S
and e
L
are the voidages displayed by the
respective monocomponent beds when they are each
separately uidized by the same liquid at the same
supercial velocity, U, as the binary bed. These voi-
dages can be separately measured or they can be pre-
dicted by the bed expansion equations elaborated
earlier. Assuming no major disturbances due to bulk
circulation or instability eects, the value of g can
function as a crude qualitative measure of the degree
of particle segregation as follows (Epstein and Pruden,
1999): g - 0.015 0.005: little segregation, good par-
ticle mixing (Fig. 8b). 0.015 0.005 - g - 0.045
0.015: partial segregation with no interface between
layers (Fig. 8d). 0.045 0.015 - g _ 0.1: segregation
with fuzzy interface and some intermixing (Fig. 8c).
g > 0.1: clean-cut segregation with sharp interface
and little intermixing (Fig. 8a). The last criterion is
consistent with the generalization of Di Felice (1995)
that complete segregation is to be expected when the
size ratio is greater than about two. The fuzzy inter-
face that occurs with the third criterion (Fig. 8c) is a
transition zone, the depth of which extends over the
whole bed length for the second criterion (Fig. 8d).
The implications of Eq. (79) can be made manifest
by writing the RichardsonZaki (1954) equation, Eq.
(32) above, for the two monocomponent beds repre-
sented by Eq. (79), so that
U = k
S
U
0S
e
n
S
S
= k
L
U
0L
e
n
L
L
(80)
But in general (Pruden and Epstein, 1964),
U
0
o d
3m ( ),m
p
= a
/
d
(3m),m
p
(81)
where the exponent m varies from 1 in the Stokes
regime to 2 in the Newton regime and the coecient
a
/
varies correspondingly from (r
p
r)g/18m to
[(r
p
r)g/0.33r]
1/2
. Therefore Eq. (80) becomes
k
S
a
/
S
d
3m
S
( ),m
S
S
e
n
S
S
= k
L
a
/
L
d
3m
L
( ),m
L
L
e
n
L
L
(82)
If the size dierence between the two species of part-
icles is suciently small that k
S
- k
L
, n
S
- n
L
, a
/
S
- a
/
L
,
and m
S
- m
L
, then Eq. (82) can be simplied and
rearranged to
e
S
= e
L
d
L
d
S
_ _
3m ( ),mn
(83)
Therefore from Eq. (79),
g = e
L
d
L
d
S
_ _
3m ( ),mn
1
_ _
(84)
Equation (84) shows that the degree of segregation (or
stratication) by size increases with an increase in
1. d
L
/d
S
, i.e., in size dierence, as would be
expected.
2. c
L,
, i.e., in bed expansion, as has been fre-
quently observed (Wakeman and Stopp, 1976;
Juma and Richardson, 1979; Epstein and
Pruden, 1999), though not by Garside and
Al-Dibouni (1974), who found this trend to
be interrupted by strong mixing in the voidage
range of e = 0.650.80.
3. (3 m),mn, which is roughly equivalent to
3 b,n ( ),b, where n is given by Eq. (34) and
b by Eq. (65), and varies from 0.43 to 0.22 as
one moves from the Stokes to the Newton
regime. Thus as the extent rst of inertial
eects and then of turbulence increases, the
degree of segregation decreases (Epstein and
Pruden, 1999), even without complicating eects
such as bulk circulation and hydrodynamic
instability.
Another approach to the problem of binary particle
stratication is based on the more traditional concept
of axial dispersion according to Ficks law of diusion
in competition with segregation, as formulated by
Kennedy and Bretton (1966). Assuming semibatch
liquid uidization that results in partial segregation
of the type represented by Fig. 8c or 8d, then at any
bed level within the partially mixed region, the volu-
metric particle mixing ux for species i must equal the
particle segregation ux for that species, i.e.,
D
i
dc
i
dz
= c
i
U
pi
(85)
where D
i
is the axial dispersion coecient (or disper-
sivity) of species i and U
pi
the segregation (or classi-
cation) velocity of that species through a swarm of
species j. Kennedy and Bretton proposed that the seg-
regation velocity U
pi
is the dierence between the local
interstitial (or slip) velocity of the uid in the bed
and the interstitial velocity that would be required to
maintain particles exclusively of that size in suspension
at the voidage existing in the local mixed-particle
environment, so that
U
pi
=
U
e

U
i
e
(86)
For a monocomponent system or for the fully segre-
gated regions of either a partially (Fig. 8c) or fully (Fig.
8a) segregated bed, U
pi
= 0, i.e., U = U
i
, assuming i
signies the particles in the segregated region involved.
Substituting for U
i
according to Eq. (32), the result is
D
i
dc
i
dz
= c
i
U
e
_ _
k
i
U
0i
e
n
i
1
_ _
i=L.S
(87)
For the large particles, both sides of this equation are
negative, while for the small particles, both are posi-
tive. The exact solution of Eq. (87) presents problems
because of the diculty in satisfactorily dening the
relevant boundary conditions (Richardson and
Aatin, 1997). However, a number of techniques
have been proposed to arrive at solutions from which
D
i
can be determined from measured axial concen-
tration proles of c
L
, c
S
, and/or e (= 1 c
L
c
S
).
Thus Al-Dibouni and Garside (1979), for partially
segregated beds with no transition zone (Fig. 8d),
rewrote and integrated Eq. (87) as
_
z
0
U
e
k
i
U
0i
e
n
i
1
_ _
dz = D
i
_
c
i
c
i0
dc
i
c
i
= D
i
ln
c
i
c
i0
(88)
and determined D
i
for each species from the best
straight line through the moderately scattered plot of
the left-hand-side of Eq. (88) vs. log c
i
,c
i0
, the slope of
which equals 2.303D
i
. Juma and Richardson (1983),
looking at transition zones in partially segregated
beds (Fig. 8c), owing to measurements of pressure gra-
dients and hence voidages as a function of bed level,
determined that the square-bracketed portion of Eq.
(87), viz. U
pi
, could be represented as
U
pi
= a
i
b
i
z. i = L. S (89)
and they determined the values of a
L
, b
L
, and a
S
, b
S
,
accordingly. Eq. (89) could then be integrated as
D
L
_
c
L
1e
L
dc
L
c
L
= D
L
ln
c
L
(z)
1 e
L
=
_
z
0
a
L
b
L
z ( ) dz
= a
L
z
b
L
z
2
2
(90)
D
S
_
1e
S
c
S
dc
S
c
S
= D
S
ln
1 e
S
c
S
(z)
=
_
H
z
a
S
b
S
z ( ) dz
= a
S
H z ( )
b
S
H
2
z
2
_ _
2
(91)
where the voidage at z = 0 (the bottom of the transition
zone) and below is e
L
and the voidage at z = H (the top
of the transition zone) and above is e
S
. D
L
and D
S
best-
tted to the experimental measurements according to
Eqs. (90) and (91) were then obtained and adjusted to
best-match the prole of c
L
c
S
(= 1 e) vs. z.
In contrast to the procedures and the empirical
results of the above two investigations, Gibilaro et al.
(1985b) assumed that D
L
= D
S
, so that Eq. (87)
applied to the small particles could be divided by the
same equation applied to the large particles, whence
dc
S
dc
L
=
c
S
U k
S
U
0S
e
n
S
( )
c
L
U k
L
U
0L
e
n
L
( )
(92)
where e = 1 c
L
c
S
. A boundary condition for this
equation is c
L
= 0, e = e
S
, c
S
= 1 e
S
, U = k
S
U
0S
e
n
S
S
at z = o, where e
S
is the monocomponent voidage at
the given supercial velocity U (or alternatively c
S
= 0,
e = e
L
, c
L
= 1 e
L
, U = k
L
U
0L
e
n
L
L
at z = o). By
LHo spitals rule,
dc
S
dc
L
c
L
0
=
1 e
S
( )n
S
k
S
U
0S
e
n
S
1
S
U k
L
U
0L
e
n
L
S
(93)
which yields the initial gradient for the downward
marching solution of Eq. (92) as c
S
vs. c
L
, and thence
to an integration of Eq. (87) in parallel with Eq. (92)
from c
L
= 0 to c
L
= 1 e
L
, thus generating curves of
c
L
vs. z and the corresponding c
S
vs. z, for any assumed
value of D
i
. Thus, by this method, curves are generated
without advance knowledge of the voidage or concen-
tration proles. The nal matching of the results to the
experimental concentration proles requires, in addi-
tion to the appropriate choice of D
i
, satisfaction of
the integral conditions stipulating the total volume
uidized of each solid component:
V
i
= A
_
L
0
c
i
dz i = L. S (94)
a requirement that is equivalent to the specication of
Kennedy and Bretton (1966) and Al-Dibouni and
Garside (1979) that the overall average concentration
of each particle species be satised. The results can
represent all cases shown in Fig. 8. A more involved
method, by Patwardhan and Tien (1984), solves Eq.
(85) to generate both the voidage proles and the par-
ticle concentration proles, but even with D
i
as an
adjustable parameter, the success for the concentration
proles was only moderate.
Van der Meer et al. (1984) and, subsequently, both
Di Felice et al. (1987) and Asif and Petersen (1993)
assumed that U
pi
in Eq. (85) varied linearly with c
i
from zero in a monocomponent bed of i particles in
which e = e
i
, c
i
= 1 e
i
to U
pi0
in a monocomponent
bed of j particles in which e = e
j
c
i
= 0, so that
U
pi
= U
pi0
1
c
i
1 e
i
_ _
(95)
Substituting this equation into Eq. (85), taking i = L
and the upward direction as positive,
D
L
dc
L
dz
= c
L
U
pL0
1
c
L
1 e
L
_ _
(96)
that is,
dx
L
x
L
1 x
L
( )
=
U
pL0
D
L
dz (97)
where x
L
= c
L
,(1 e
L
). Integrating Eq. (97) between
the limits x
L
= 0.5 [i.e., c
L
= 0.5(1 e
L
)], z zz
L
, and
x
L
, z, the result is
ln
x
L
1 x
L
=
U
pL0
(z zz
L
)
D
L
(98)
from which a plot of z zz
L
vs. ln x
L
,(1 x
L
) yields a
slope equal to D
L
/U
pL0
, where U
pL0
is taken as the
measured or predicted velocity of a single large particle
moving through a bed of the smaller particles uidized
by the same supercial velocity U as the binary under
investigation (Martin et al., 1981; Prudhoe and
Whitmore, 1964; Grbavc ic and Vukovic , 1991; Van
der Wielen et al., 1996; Di Felice, 1998; Mostou
and Chaouki, 1999). From Eq. (98) it follows that
x
L
=
1
1 e
Pe
L
Z

ZZ
L ( )
(99)
where Pe
L
= LU
pL0
,D
L
and Z

ZZ
L
= (z zz
L
),L.
Similarly it can be shown that
x
S
=
1
1 e
Pe
S
Z

ZZ
S ( )
(100)
where Pe
S
= LU
pS0
,D
S
and Z

ZZ
S
= (z zz
S
)
_
L. But,
in accord with Di Felice et al. (1987), c
L
and c
S
are
linearly related, so that
c
S
1 e
S
= 1
c
L
1 e
L
(101)
Equation (101), rst stated as such by Van Duijn and
Rietema (1982), is simply the serial model, Eq. (66),
applied locally to a binary. Thus
x
S
= 1 x
L
=
e
Pe
L
Z

ZZ
L ( )
1 e
Pe
L
Z

ZZ
L ( )
=
1
1 e
Pe
L
Z

ZZ
L ( )
(102)
Comparing Eqs. (102) and (100), Pe
L
= Pe
S
and
ZZ
L
=

ZZ
S
. Asif and Petersen (1993) accordingly use
Pe
i
= LU
pi0
,D
i
and

ZZ,L as the adjustable parameters.
Methods of determining D
i
by the Fickian approach
above have been supplemented by stochastic or ran-
dom walk methods (Yutani et al., 1982; Yutani and
Fan, 1985; Dorgelo et al., 1985; Kang et al., 1990),
and a plethora of D
i
values have already been reported
in the literature. These vary from as low as 0.24 cm
2
/s
to as high as 515 cm
2
/s (Asif and Petersen, 1993), a
range which brackets the smaller spread in values of
D
ia
(0.533 cm
2
/s) obtained for monocomponent self-
dispersion, i.e., by observing the transient mixing of a
single species of particles initially divided into two
layers, one colored and the other uncolored (Bro tz,
1952; Kennedy and Bretton, 1966; Carlos and
Richardson, 1968b). Attempts to correlate D
i
with sys-
tem properties and operating conditions have met with
limited success. The most recently proposed dimen-
sionally homogeneous correlation, for a large number
of binary systems of spheres from the literature cover-
ing the ranges Re = 0.51000 and e = 0.500.95, by
Asif and Petersen (1993), is equivalent to
D
i
=
(7.9 1.1)d
2
p
(r
p
r)g
erU
U U
mf
U
0
_ _
2.140.05
16.4% (103)
in which U
mf
presumably applies to the larger part-
icles, so that both are fully uidized. When D
i
= D
s
,
this equation thus gives recognition to the properties of
both particle species, directly via d
p
and r
p
and indir-
ectly via U
0
for the smaller particles, and indirectly via
U
mf
for the larger. When, however, D
i
= D
L
, only the
properties of the larger particles are manifested in Eq.
(103), in violation of experiments (e.g., Juma and
Richardson, 1983) that indicate that both species
play a role in the determination of D
i
.
The data on D
i
in the literature, including those that
form the basis of Eq. (103), show an important diver-
gence. Many, e.g., those of Juma and Richardson
(1983), Van der Meer et al. (1984), and Dorgelo et al.
(1985), show no maximum in D
i
with respect to U or e,
while others, e.g., those of Al-Dibouni and Garside
(1979), Yutani et al. (1982), and Kang et al. (1990),
show a denite maximum at e 0.7, 0.70.8, and
0.60.7, respectively. Equation (103) shows no maxi-
mum D
i
within the uidization voidage range,
although at least two of the data sets (Yutani et al.,
1982; Kang et al., 1990) on which it is based do so.
Interestingly, Handley et al. (1966) found that the tur-
bulent particle velocity components and the turbulent
uid velocity uctuations (which were about one-half
the former) both rise to a maximum, at e _ 0.70 and
e _ 0.65, respectively. Trupp (1968) then showed that
the root mean square voidage uctuation in a liquid
uidized bed of heavy spheres in the Newton region
(Re
0
> 500) occurred at e = 0.7, and following his
reasoning and subsequently that of Al-Dibouni and
Garside (1979), the collision frequency of such a bed
should be expected to be proportional to U(1 e),
since average particle velocities were shown to vary
directly as U. Invoking Eq. (32) for U, the collision
frequency is proportional to kU
0
e
n
(1 e), which
shows a maximum with respect to e at e = n,(n 1).
In the Newton region, n = 2.4, so that the maximum
occurs at e = 0.706. At lower Reynolds numbers, n is
higher, and therefore the critical voidage can be
expected to be correspondingly higher. Trupps results
were vindicated by Bordet et al. (1972) and were con-
sistent with the subsequent observations of Kmiec
(1978).
The divergence between data that show a maximum
D
i
within the uidization region and those that do not
is a puzzle yet to be solved. The extreme sensitivity of
these data to gulf-streaming (Al-Dibouni and Garside,
1979) and to the slightest deviation of the uidization
column from verticality (Van der Meer et al., 1984) are
undoubtedly among the factors contributing to this
puzzle, as well as to the unsatisfactory data scatter
and state of data correlation in the literature.
Uncertainties arising from the various experimental
methods of measuring the extent of particle mixing
are also contributing factors (Di Felice, 1995).
5.4 Binary Mixtures of Equal Size Particles
Although the case of multisized particles of xed den-
sity is far more common in industrial situations than
that of particles of uniform size having dierent
densities, it will nevertheless be instructive briey to
consider spheres of diameter d
p
and of density r
ph
and r
pl
, where r
ph
> r
pl
, uidized by a liquid having
supercial velocity, U, between U
mfh
and U
0l
. In the
absence of the lighter particles, the uidized suspension
density will be given by
r
Bh
= r
ph
(1 e
h
) re
h
(104)
while in the absence of the heavier particles, the sus-
pension density is given by
r
Bl
= r
pl
(1 e
l
) re
l
(105)
Subtracting Eq. (105) from Eq. (104),
r
Bh
r
Bl
= (r
ph
r)(1 e
h
) (r
pl
r)(1 e
l
)
(106)
which is equivalent to
g
h
=
r
Bh
r
Bl
r
ph
r
= 1 e
h
r
pl
r
r
ph
r
1 e
l
( ) (107)
where the nondimensionalized bulk density dierence
g
h
and Eq. (107) are now assumed to be qualitatively
indicative of the degree of segregation by particle den-
sity in the same way as g and Eq. (79) are indicative of
the degree of segregation by particle size. Substituting
the generalization (Pruden and Epstein, 1964)
U
0
o (r
p
r)
1,m
= b(r
p
r)
1,m
(108)
into Eq. (32) for each particle species, and making the
assumption that the density dierence between the two
species of particles is small enough that the indices m
and n, and the coecients k and b, are essentially the
same for each of the two particle species, then since U
is given,
(r
ph
r)
1,m
e
n
h
= (r
pl
r)
1,m
e
n
l
(109)
or
e
h
= e
l
r
pl
r
r
ph
r
_ _
1,mn
(110)
By means of Eqs. (83) and (79) for a constant density
binary of dierent particle sizes, and Eqs. (110) and
(107) for a xed size binary of dierent particle densi-
ties, Table 2 indicates what happens to the nondimen-
sionalized bulk density dierence in each case as one
moves from a very dilute system (e.g., e
S
= 1 = e
l
) to a
very dense system (e.g. e
S
= 0.5 = e
l
). It is seen that,
irrespective of ow regime, g decreases as e decreases,
while g
h
does the reverse. This tendency by , was
already rationalized in the discussion (item 2) of Eq.
(84). Substituting Eq. (110) into Eq. (107), we obtain
g
h
= 1
r
pl
r
r
ph
r
e
l
r
pl
r
r
ph
r
_ _
1,mn
r
pl
r
r
ph
r
_
_
_
_
(111)
Since mn equals 4.8 both in the Stokes and the Newton
regimes and is in any case always in excess of unity, it
follows that the expression in square brackets is always
positive, so that g
h
(in contrast to g) always increases
as the voidage decreases. Thus we can explain the
empirical rule, long known in the ore dressing industry
(Gaudin, 1939), that classication by size, i.e., sizing,
is best performed under dilute (free settling) con-
ditions, while classication by density, i.e., sorting,
prevails under concentrated (hindered settling or
teeter bed) conditions. We shall see below how this
rule manifests itself in the bed inversion phenomenon.
Note also in Table 2 [or in Eqs. (84) and (111)] that,
in contrast to segregation by size, which is attenuated
as one moves from the Stokes to the Newton regime,
i.e., as Re
0
increases, there is no such eect in the case
of segregation by density.
Axial dispersivities for particle species of dierent
density, whether or not the sizes are the same, can be
determined by methods similar to those for xed den-
sity, mixed size binaries, but their experimental deter-
mination has received much less attention in the
literature. Equation (103), to the extent that it is
applicable at all, is restricted to binaries of equal
density particles.
5.5 Binary Mixtures of Particles Differing in Size
and Density: Bed Inversion
Consider a binary mixture of spheres of diameters d
lL
and d
hS
(<d
lL
) and densities r
plL
and r
phS
(>r
plL
),
respectively. For the larger, less dense particles (species
lL) subjected alone to liquid uidization,
r
BlL
= r
plL
(1 e) re (112)
The voidage e can be expressed in terms of the super-
cial liquid velocity U, explicitly by means of the
RichardsonZaki relation, Eq. (32), or by Eq. (64)
with (65), and implicitly (requiring iteration) by
means of Eq. (62) with (63), applied to the given
solidliquid system. A plot of r
BlL
vs. U from U
mL
to U
0lL
can thus be generated. Similarly, for the
smaller, denser particles (species hS) subjected alone
to uidization by the same liquid at the same temp-
erature and pressure,
r
BhS
= r
phS
(1 e) re (113)
Again, e can by the same means as for Eq. (112) be
expressed in terms of U, so that a plot of r
BhS
vs. U
from U
mfhS
to U
0hS
can then be obtained. If the two
curves do not intersect between U
mL
and U
0hS
, then no
bed inversion will occur, and the component that has a
consistently higher bulk density will always tend to seg-
regate toward the bottom of the bed. The reduced bulk
density dierence, r
B
,(r
plL
r), then, provides an
indication of the degree of segregation in accord with
g = (r
BL
r
BS
),(r
p
r) for pure sizing (Epstein and
Pruden, 1999). If, however, the two curves do intersect
between U = U
mf1L
(> U
mfhS
) and U = U
0hS
(- U
01L
),
as in Fig. 9, then bed inversion will occur.
The intriguing subject of bed inversion has received
much attention in the literature, both experimental
(Hancock, 1936; Cleasby and Woods, 1975; Van
Duijn and Rietema, 1982; Moritomi et al., 1982;
Epstein and LeClair, 1985; Gibilaro et al., 1986a;
Jean and Fan, 1986; Di Felice et al., 1988; Qian et
al., 1993; Funamizu and Takakuwa, 1996) and theore-
Table 2 Nondimensionalized Bulk Density Differences of the
Two Particle Species in Two Types of Binaries
r
p
= constant
d
L
,d
S
=
2
_
d
p
= constant
(r
ph
r),(r
pl
r) =
2
_
e
S
e
L
g e
l
e
h
g
h
Eq. (83) (Eq. (79) (Eq. (110) Eq. (107)
Stokes regime: 1.0 0.866 0.134 1.0 0.930 0.070
m = 1. n = 4.8 0.5 0.433 0.067 0.5 0.465 0.181
Newton regime: 1.0 0.930 0.070 1.0 0.930 0.070
m = 2. n = 2.4 0.5 0.465 0.035 0.5 0.465 0.181
tical (Van Duijn and Rietema, 1982; Patwardhan and
Tien, 1985; Epstein and LeClair, 1985; Moritomi et al.,
1986; Gibilaro et al., 1986a; Jean and Fan, 1986;
Syamlal and OBrien, 1988; Funamizu and
Takakuwa, 1996; Richardson and Aatin, 1997; Asif,
1997, 1998; Epstein and Pruden, 1999; Howley and
Glasser, 1999). The earlier work incorrectly assumed
that the inversion velocity was entirely independent of
the overall solids composition of the binary and that
therefore inversion occurred at the intersection point C
on Fig. 9 (Hancock, 1936; Van Duijn and Rietema,
1982; Epstein and LeClair, 1985). According to this
view, an idealized inversion phenomenon, i.e., one
not subject to ow disturbances that induce mixing,
would show only the three segregation patterns (a),
(c), and (e) of Fig. 10. In the light of further careful
experimentation (e.g., Moritomi et al., 1982; Jean and
Fan, 1986), backed by theory (Gibilaro et al., 1986a), it
was found that the relative proportions of the two
solids species in the binary inuenced the inversion
velocity, and that this uid velocity was preceded and
followed, respectively, by the two additional segrega-
tion patterns (b) and (d) of Fig. 10.
To gain further insight into this phenomenon, apply
Eq. (62) with (63) or, alternately, (64) with (65) to var-
ious binary mixtures of the two solids species uidized
by the given liquid, assuming after Wen and Yu (1966)
and Gibilaro et al. (1986a) that d
p
is the Sauter mean
particle diameter given by Eq. (75), evaluating r
p
as the
volume-average density by Eq. (74), and noting that
e = 1 c
1
c
2
= 1 c
lL
c
hS
(114)
Assuming knowledge of r, m, d
lL
, r
lL
, d
hS
, and r
hS
,
then for any given value of U, Eq. (62) or Eq. (64)
becomes a relationship between c
lL
and c
hS
. Thus
plots of c
hS
vs. c
lL
can be generated with U as a para-
meter. From the families of c
hS
and c
lL
curves, one can
determine plots of r
B
vs. c
lL
(or r
B
vs. c
hS
) by substi-
tuting Eqs. (74) and (114) into Eq. (4), so that
r
B
= c
lL
r
plL
c
hS
r
phS
(1 c
lL
c
hS
)r (115)
Figure 9 The bed inversion phenomenon of a binary as manifested by the bulk densities of the two layers as a function of liquid
velocity. The dotted line ABD merging with the species hS line to the left of A and the species lL line to the right of D represents
the bottom layer. Assuming v
IL
= 0.4, the two thickened solid lines represent the top layer below and above the inversion
velocity, U
B
, respectively, while B represents conditions at the singular inversion point. r
BlL
and r
BhS
are the bulk densities of the
monocomponent species lL and species hS beds, respectively. (r
B
)
max
refers to the bulk densities of the bottom layer along the
dotted line ABD, and c
lL
/(c
lL
c
hS
) are the corresponding compositions of the binary solids in this layer. (After Epstein and
Pruden, 1999.)
For the lower values of U between U
mL
and U
0hS
the
plots of r
B
vs. c
lL
rise monotonically over the entire
range of c
lL
, while for the higher values of U between
the same limits, they fall monotonically. However, for
a range of intermediate values of U, the plots of r
B
vs.
c
lL
show maxima (Gibilaro et al., 1986a), the locus of
which is represented by curve ABD on Fig. 9. (For
consistency the monocomponent species lines are best
generated by the same equation, either (62) or (64), as
used to generate ABD.) This curve represents the range
of possible inversion velocities and the corresponding
bulk densities over which the bottom layer is a
binary mixture of solids, the solids composition,
c
lL
,(c
lL
c
hS
), of which is given by the nonlinear
scale at the top of Fig. 9. It also represents the range
of possible inversion velocities and the corresponding
bulk densities of the mixed bed at the inversion point,
depending on the overall solids composition, v
lL
. For
example, assuming an overall solids composition,
c
lL
,(c
lL
c
hS
) = v
lL
= 0.4, given by point B on Fig.
9, then on increasing the liquid supercial velocity
from U
mL
to U
0hS
, we can observe ve stages to the
inversion process:
1. From U = U
mL
to U = U
A
, the bottom layer
contains pure solids species hS and the top pure species
lL, both layers expanding with increasing U.
2. From U = U
A
to U = U
B
, the expanding
bottom layer composition is a mixture of the two
species, with c
lL
,(c
lL
c
hS
) increasing from 0 to 0.4
and r
B
decreasing from r
BA
to r
BB
as U increases,
while the contracting top layer continues to contain
pure lL.
Figure 10 Bed inversion phenomenon. Blackened circles represent higher density, smaller spheres (species hS); open circles
represent lower density, larger spheres (species lL). (Based on experimental observations of Moritomi et al., 1982, and Jean and
Fan, 1986, after Di Felice, 1995.)
3. At U = U
B
, the inversion velocity, the entire
bed attains a uniform composition equal to that of
the overall solids composition, x
lL
= 0.4, and a bulk
density equal to r
BB
.
4. From U = U
B
to U = U
D
, the contracting bot-
tom layer solids composition continues to be a mixture
of the two species, with c
lL
,(c
lL
c
hS
) increasing from
0.4 to unity and r
B
decreasing from r
BB
to r
BD
as U
increases, while the expanding top layer now contains
pure species hS.
5. Finally, from U = U
D
to U = U
0hS
, the bottom
layer now contains pure species lL and the top layer
continues with pure hS, both layers again now expand-
ing with increasing U.
Deviations from this idealized complete segrega-
tion model due to particle dispersion, e.g., the pene-
tration by the smaller, denser particles of the upper
zone in Fig. 10a, have been addressed by Di Felice
(1993) and by Asif (1997).
Jean and Fan (1986) have adopted another
approach to predicting the bed inversion velocity.
Recall from Eqs. (85)(87), as applied to the present
situation, that the segregation velocity of species hS
with respect to species lL is given by
U
phS
=
U
e
_ _
k
hS
U
0hS
e
n
hS
1
(116)
The inversion condition is then taken as
d(U
phS
)
dU
= 0 (117)
Accurate agreement with the experimentally measured
value of inversion velocity was thus obtained when the
corresponding voidage in Eq. (116) was measured;
moderately good prediction of this velocity could be
eected by formulating this voidage in terms of the
previously discussed serial model, Eq. (66), applied to
the binary solids mixture:
1
1 e
=
v
lL
1 e
lL
1 v
lL
1 e
hH
(118)
with the monocomponent voidages, e
lL
and e
hH
, each
expressed as functions of U by means of Eq. (32) and
their respective values of both kU
0
(= U
e
) and n.
6 INVERSE FLUIDIZATION
Solids having density smaller than that of a given
liquid may be uidized downward by that liquid, the
distributor of which is located at the top rather than
the bottom of the uidizing column. Since gravity now
acts in the same direction as the drag, while buoyancy
acts in the opposite direction, Eq. (8) must be modied
to
dp
f
dz
= (1 e)(r r
p
)g = (1 e) r
p
r
g (119)
The rst mention of such inverse uidization occurs in
the appendix to a thesis by Page (1970), who sum-
marizes the results of a contemporaneous report by
Field and Riley (1970), the original of which has
since been destroyed. These results and more detailed,
subsequent hydrodynamic studies (Fan et al., 1982;
Garnier et al., 1990; Karamanev and Nikolov, 1992b)
show that both minimum uidization velocity and bed
expansion can, with minor qualications, be predicted
by the same methods as for conventional uidization.
For free rising single spheres in water with r
p
< 300
kg/m
3
, the Newton regime starts at Re
0
130 because
of the additional turbulence then generated by wake
shedding, which produces a spiral particle trajectory
and an increase in the constant drag coecient, C
D
,
from Newtons theoretical value of 0.5 and the
actual value of about 0.44 to a measured value of
0.95 (Karamanev and Nikolov, 1992a). Thus, for
r
p
< 300 kg/m
3
and Re
0
> 130 (i.e., Ar > 12,000),
F
D0
= 0.95
p
4
_ _
d
2
p
1
2
_ _
rU
2
0
=
p
6
_ _
d
3
p
r
p
r
g (120)
whence
U
0
= 1.18
d
p
r
p
r
g
r
_
(121)
which gives values of U
0
about 32% lower than the
usual equation for the Newton regime [with 0.44
instead of 0.95 in Eq. (120) and 1.74 instead of 1.18
in Eq. (121)]. Therefore for very light spheres
(r
p
< 130 kg/m
3
), provided Ar > 12,000, U
0
in the
RichardsonZaki relation, Eq. (32), should be
obtained from Eq. (121) rather than from Eq. (33).
The minimum velocity for inverse uidization of such
spheres is slightly lower than predicted by Eq. (17),
owing presumably to the higher drag exerted on
them (Karamanev and Nikolov, 1992b).
In operating an inverse liquid uidized bed, extra
measures must be taken to rid the liquid of undesired
bubbles since, unlike in conventional uidization, the
larger bubbles tend to move in a direction (upward)
opposite to that of the exit-bound liquid (Karamanev
and Nikolov, 1992b). Inverse liquid uidization, in
which the particles are coated with biolm, has thus
far found its principal application as a vehicle for bio-
chemical reactions (Chavarie and Karamanev, 1986;
Nikolov and Karamanev, 1990), with particular
emphasis on ferrous ion oxidation (Karamanev
and Nikolov, 1988) and wastewater treatment
(Karamanev and Nikolov, 1996).
7 VERTICAL MOVING FLUIDIZED SYSTEMS
(CONTINUOUS FLUIDIZATION)
7.1 Monodisperse Solids
Continuing with our assumption of relatively homo-
geneous particulate uidization, but switching from
semibatch operation, i.e., from an unchanging batch
of bottom- or top-restrained (the latter in inverse ui-
dization) solids to a continuous vertical ow of both
solids and liquid, we consider rst the case of mono-
sized constant density solids. Studies by Mertes and
Rhodes (1955) and by Lapidus and Elgin (1957) have
shown that, irrespective of the ow direction of either
liquid or solids, the relative or slip velocity, U
r
,
between the liquid and the solids in the nonaccelerat-
ing region of a given liquidsolids system is always
the same unique function of the voidage, so that in
general,
U
r
=
U
e

U
p
1 e
= f
/
(e) (122)
in which the upward direction is taken as positive.
Equation (122) has been veried, i.e., f
/
(e) has been
shown to be the same for cocurrent upward ow
(Struve et al., 1958), countercurrent ow (Price et al.,
1959), and cocurrent downward ow (Quinn et. al.,
1961), as well as for both unfed, i.e., semibatch, uidi-
zation (U
p
= 0) and constant-rate bottom-restrained
sedimentation (U = 0) (Mertes and Rhodes, 1955).
The RichardsonZaki (1954) equation, Eq. (32), can
therefore be generalized as
U
e

U
p
1 e
= kU
0
e
n1
(123)
and all the other relationships between U and e for
semibatch uidization can similarly be applied to mov-
ing uidized systems by substituting U
r
e for U in these
equations or plots, with U
r
given by Eq. (122). The
useful graphical plot of Barnea and Mizrahi (1973) is
already expressed in terms of U
r
, based as it is on data
for sedimentation, semibatch uidization, and various
modes of fully continuous operation.
In the case of countercurrent operation (Lapidus
and Elgin, 1957; Price et al., 1959), ooding, i.e.,
backup of solids, will occur when
oU
o(1 e)[
U
p
= 0 (124)
Applying this criterion to Eq. (123), we arrive at
U
p
= nkU
0
e
n1
(1 e)
2
(125)
and
U = kU
0
e
n
1 n(1 e) [ ] (126)
Thus, for a given solids ux U
p
, ooding will occur
when the in situ voidage falls to that given by Eq.
(125), which occurs when the liquid ux U rises to
that given by Eq. (126); and if U
p
is increased, the
ooding will occur at a lower value of e and a corre-
spondingly lower value of U.
7.2 Multidisperse Solids
For solids containing more than one particle size or
particle density, the situation is not nearly as clear-
cut, except in the case of dilute (e > 0.94) suspensions
in creeping ow, for which the theory of Batchelor
(1982) and Wen (Batchelor and Wen, 1982) has been
veried for sedimentation of bidisperse submicron
spheres by Al-Naafa and Selim (1992). For more
concentrated suspensions, e.g., 0.06 < e < 0.50,
Finkelstein et al. (1971) found that, in both counter-
current and cocurrent downward ow, both species of
binary-sized solids moved at the same rate, provided
that the solids residence time was short (<1 minute)
and the size ratio did not exceed 3. For more common
situations, however, in which N solids species have left
their initial acceleration zone, Richardson and Shabi
(1960) have suggested that the relative or slip velocity,
U
ri
, of each particle species (e.g., each narrow cut of
sieved solids of xed density) depends on the total
surrounding voidage in a manner similar to that of
the same species acting monodispersely in the same
liquid, viz.,
U
ri
=
U
e

U
p.i
c
i
= k
i
U
0i
e
n
i
1
. i = 1. 2. 3. . . . . N
(127)
Lockett and Al-Habbooby (1973) provided some
experimental backing for Eq. (127), for both counter-
current ow of three sized (i.e., r
p
= constant) binaries
and one sized ternary, and for batch sedimentation
of several sized binaries, but only when the voidage
exceeded 0.6, below which the inapplicability of Eq.
(127) was attributed to interactions between particle
species. Subsequently they claimed Eq. (127) to be
applicable also to particles of dierent densities
(Lockett and Al-Habbooby, 1974), but the small dif-
ference in density of the equal-sized binary particle
species on which they based this claim, and experi-
ments by later workers, render such a conclusion gen-
erally inadmissible. These experiments and some
theory addressed themselves both to the problem of
species interactions and that of density dierences
among species.
Mirza and Richardson (1979), based on batch sedi-
mentation experiments with several sized binaries
down to e 0.5, concluded that the species interaction
eect could be accounted for by the factor e
0.4
applied
to the right-hand side of Eq. (127), so that
U
ri
= k
0i
U
0i
e
n
i
1
e
0.4
= k
0i
U
0i
e
n
i
0.6
(128)
At the same time, Masliyah (1979) addressed the den-
sity dierence problem. For a monodisperse system,
i.e., a single particle species, both Eq. (123) and Eq.
(127) are equivalent to
U
r
= kU
0
e
n1
(129)
Since from Eq. (4),
e =
r
p
r
B
r
p
r
(130)
it follows that
U
r
= kU
0
e
n2
r
p
r
B
r
p
r
(131)
By using the form of the governing momentum
equation given by Wallis (1969), Masliyah showed
analytically that Eq. (131) can be generalized for
each species i of a multispecies particle system to
U
ri
= k
i
U
0i
e
n
i
2
r
pi
r
B
_ _
, r
pi
r
_ _
(132)
where for N particle species,
r
B
= r
p
_ _
avg
1 e ( ) re (133a)
=
N
i=1
r
pi
c
i
r 1
N
i=1
c
i
_ _
(133b)
Subsequently, Patwardhan and Tien (1985) adopted
Masliyahs model but assigned an eective voidage,
e
ei
, to each particle species to account for the dierent
environment experienced by that species in the pre-
sence of the others than in their absence. Thus Eq.
(132) becomes
U
ri
= k
i
U
0i
e
n
i
2
ei
r
pi
r
B
r
pi
r
(134)
where, by a heuristic cell model approach, e
ei
was
formulated as
e
ei
= 1 1
d
avg
d
i
_ _
{(1 e)
1,3
1]
_ _
3
(135)
with
d
avg
=
N
i=1
c
i
d
i
N
i=1
c
i
=
N
i=1
c
i
d
i
1 e ( )
(136)
Note that for a monodisperse system, e
ei
= e and Eq.
(134) reverts back to Eq. (131). Both Eqs. (132) and
(134) were able to map features of the binary batch
sedimentation data of Richardson and Meikle
(1961a) for spheres of dierent sizes and densities
having the same value of U
0
.
A dierent approach was adopted by Selim et al.
(1983a,b), who considered the buoyancy eect induced
by the smaller (or less dense) particles on the motion of
the larger. Thus particles i are treated as moving in a
uid of viscosity equal to that of the pure liquid and
density equal to that of a mixture of the pure uid and
of all particles smaller than d
pi
(or less dense than r
pi
).
For example, for a ternary mixture of spheres having
diameters d
1
, d
2
, and d
3
and densities r
p1
, r
p2
, and r
p3
,
respectively, such that the relative ranking of U
0
in the
given liquid is U
01
> U
02
> U
03
,
U
r1
= k
1
U
/
01
e
n
1
1
(137)
where U
/
01
is the terminal free settling velocity of the
fastest moving particles in a uid of density equal to
r
2.3
=
c
2
r
p2
c
3
r
p3
er
1 c
1
(138)
and
U
r2
= k
2
U
/
02
e
n
2
1
(139)
where U
/
02
is the terminal free settling velocity of the
intermediate-velocity particles in a uid of density
equal to
r
3
=
c
3
r
p3
er
1 c
1
c
2
(140)
while
U
r3
= k
3
U
03
e
n
3
1
(141)
Thus only the smallest particles will behave as pre-
dicted by Eq. (127). This model was found to agree
well with both multisized sedimentation data
(0.5 < e < 0.85) and the binary-sized countercurrent
results of Lockett and Al-Habbooby (1973).
All ve models discussed above have been subjected
to comparison with data on gravity separation of bidis-
perse suspensions in which light and heavy particles of
the same size are vertically separated from each other
in a liquid of intermediate density (Law et al., 1987),
and with data on sedimentation of polydisperse con-
centrated suspensions (Al-Naafa and Selim, 1989). The
unequivocally inferior model that emerges from a com-
bination of these comparisons is that of Lockett and
Al-Habbooby (1973), while the equivocally superior
one is that of Selim et al. (1983a,b).
The model of Selim et al. (1983a,b) can be thought
of as intermediate between the overall voidage
model (Di Felice, 1995) of Lockett and Al-
Habbooby (1973) and the pseudouid model of Di
Felice et al. (1991a). In the latter model, the applica-
tion of which has thus far been limited to binaries, if
we assume that U
01
> U
02
, then
U
r1
=
U U
p2
e c
2
U
p1
c
1
= k
1
U
//
01
1 c
1
( )
n
1
1
(142)
That is, the larger and/or heavier particles are assumed
to move through a pseudouid having the properties,
and owing with the supercial velocity (or volumetric
ux), of the combined pure liquid and smaller and/or
lighter particles. The free settling terminal velocity,
U
//
01
, is thus based on a pseudouid density of
r
2
=
c
2
r
p2
er
1 c
1
(143)
as in the model of Selim et al., while the pseudouid
viscosity on which U
//
01
is based can be estimated from
one of several empirical or semiempirical equations
(e.g., Trawinski, 1953; Hetzler and Williams, 1969;
Saxton et al., 1970; Rigby et al., 1970) for the eective
viscosity of uidized beds, or from one of several
theoretical or semitheoretical equations (cf. Table 8
of Barnea and Mizrahi, 1973) for the apparent viscos-
ity of suspensions. Di Felice (1998) has most recently
recommended the semitheoretical equation of Thomas
(1965), which applies over the entire voidage range of
0.41 and, for the liquid suspension of the smaller and/
or lighter particles, is given by
m
2
= m 1 2.5c
/
2
10.05(c
/
2
)
2
0.0027e
16c
/
2
_ _
(144)
where c
/
2
= c
2
,(c
2
e). The pseudouid model has
scored its greatest success with single foreign spheres
(c
1
= 0) settling in a uidized bed of smaller and/or
lighter spheres (Di Felice, 1998), though it has also
yielded good predictions for the larger solid species
of a sized binary in both batch and continuous sedi-
mentation and uidization (Di Felice et al., 1989,
1991b). It has, however, not yet succeeded in satisfac-
torily modeling the movement of the smaller species
(Di Felice, 1995).
A subtle mutation of Masliyahs (1979) model,
arrived at quite independently, has recently been
proposed by Asif (1997) and by Galvin et al. (1999).
If Eq. (130) is substituted into Eq. (131), the result is
U
r
= kU
0
r
p
r
B
r
p
r
_ _
n1
(145)
By more heuristic approaches than that of Masliyah,
both Asif and Galvin et al. generalized this empirical
equation as applying to each species of a multispecies
system, so that
U
ri
= k
i
U
0i
r
pi
r
B
r
pi
r
_ _
n
i
1
(146)
which is not equivalent to Eq. (132) except in the case
of a monodisperse suspension. Substituting for r
B
in
Eq. (145) its value given by Eq. (8a) results in
U
ri
= k
i
U
0i
1
(dp
f
,dz)
(r
pi
r)g
_ _
n
i
1
(147)
from which U
ri
can be determined for each species by
measurement of frictional pressure gradient, assuming
knowledge of k
i
U
0i
(= U
ei
) and n
i
for each. Reasonable
agreement with Eq. (147) was obtained by Galvin et al.
for measured values of U
ri
of each particle species in a
liquid uidized bed operating with a continuous feed
consisting of three species of particles diering in size,
shape, and density, separating into overow and raked
underow streams (a teetered bed separator). Galvin
et al. disclaim any possibility of Eq. (146) or (147)
applying to situations in which one of the particle spe-
cies is less dense than the suspension (i.e., r
pi
< r
B
),
and it is likely that the same limitation applies to Eqs.
(132) and (134) in the case of countercurrent ow of
upward liquid and denser solids.
Externally circulating liquid-uidized beds and
the simplied one-dimensional hydrodynamics thereof
made a brief appearance in the literature a decade ago
with respect to particle mixing (Di Felice et al., 1989)
and fermentation (Pirozzi et al., 1989), but more
detailed experimental investigations of their hydrody-
namic complexities have not been reported on until
quite recently (Liang et al., 1996, 1997; Liang and
Zhu, 1997; Zheng et al., 1999). Nonuniformities in
ow structure appear to be much attenuated when
compared with externally circulating gas-uidized
beds.
8 LIQUID DISPERSION
8.1 Axial Dispersion
The dispersion of liquid in the axial or longitudinal
direction within a particulately liquid-uidized bed is
qualitatively and quantitatively intermediate between
that of liquid ow through a packed bed and liquid
ow through an otherwise empty tube. Thus, generally
speaking, the axial dispersion coecient, D
a
, increases
in magnitude as the uidizing liquid velocity increases,
encompassing for dierent conditions values below
1 cm
2
/s to values as high as 100 cm
2
/s, and the corre-
sponding axial Peclet number, Pe
a
(= d
p
U,eD
a
).
decreases from values of 0.42 for packed beds to
1.00.01 or even lower (Scott et al., 1993) for liquid-
uidized beds. Axial liquid dispersion in a uidized bed
above the entry region is contributed to by particle
motion, uid turbulence, particle wake-shedding
(Letan and Elgin, 1972), and voidage waves
(Kramers et al., 1962) but only minimally by longitu-
dinal velocity gradients in the radial (or transverse)
direction, especially if these are abated by uniform
liquid distribution of the inlet liquid and by suciently
high values of D
c
/d
p
(e.g., >50), assuming e < 0.9.
The experimental determination of axial dispersion
coecients has been eected mainly by measurement
of the outlet response to a controlled tracer input. As
with particle dispersion, liquid axial dispersivities,
when plotted against liquid velocity or voidage, in
some cases rise monotonically from minimum uidiza-
tion to particle transport (e.g., Bruinzeel et al., 1962;
Chung and Wen, 1968; Tang and Fan, 1990) and in
some cases show a maximum, usually at e = 0.7 0.1
(e.g., Cairns and Prausnitz, 1960a; Mehta et al., 1976;
Kikuchi et al., 1984). Usually this maximum is fol-
lowed by a minimum (Kramers et al., 1962) and some-
times even a tendency towards a second maximum at
e - 0.9 (Mehta et al., 1976). Even in the absence of a
maximum, Kramers et al. have noted a sharp rise in
the rate of increase of D
a
with e at e - 0.7 for water
uidization of 1 mm glass spheres. They attributed this
rise to the fact that above this voidage the velocity of
horizontal voidage disturbances (Slis et al., 1959),
(1 e)dU,de = n(1 e)U
e
e
n1
, subsequently referred
to as continuity waves (Wallis, 1969), exceeds the
interstitial liquid velocity, U,e = U
e
e
n1
(where
n = RichardsonZaki index), so that eddies produced
by individual particles are then supplemented by eddies
produced by these voidage uctuations. Note that
when
n(1 e)U
e
e
n1
= U
e
e
n1
(148)
the corresponding voidage is
e =
n 1
n
(149)
so that for nonocculating spheres in Newtonian
liquids, the critical voidage will vary from 3.8/4.8
(=0.79) in the Stokes region to 1.4/2.4 (=0.58) in
the Newton region. In the early study of Cairns and
Prausnitz (1960a), water uidization of both 1.3 and
3.0 mm lead spheres showed maxima (at e = 0.60.7),
which suggests that the transition from particulate to
aggregative uidization (see Sec. 9) characteristic of
the latter is an additional factor contributing to the
maxima.
An empirical equation for D
a
of liquid uidized
beds, correlating a large range of experimental
data [d
p
= 0.514.3 mm, r
p
= 128011,300 kg/m
3
, m =
0.000560.0619 kg/(m)(s), and rheological index =
0.8581.0 for some non-Newtonian power law liquids
(Wen and Fan, 1973)] with a standard deviation of
8%, is given by (Krishnaswamy et al., 1978)
1
2D
a
e
UL
_ _
0.5
_ _
e
0.25
= 0.74 (150)
The presence of bed height L = L
mf
(1 e
mf
),(1 e) [ ]
in the correlation is, however, at odds with the results
of Wen and Fan (1973), which explicitly show no
dependence of D
a
on L
mf
. Furthermore, data on D
a
limited to the lower Reynolds number region, obtained
subsequently by Kikuchi et al. (1984), were poorly pre-
dicted by Eq. (150) and by seven other empirical
equations in the literature. For 0.15 - Re - 100,
0.41 - e - 0.93, and 8 10
5
- e
d
- 0.44 m
2
/s
3
, these
investigators recommended
D
a
n
= 500e
0.43
d
exp 20.5(0.75 e)
2
_ _
38% (151)
where e
d
, the energy dissipation rate per unit mass of
liquid, is given by
e
d
=
(r
p
r)(1 e)gU
er
=
(r
p
r)(1 e)gU
e
e
n1
r
(152)
Note that the square-bracketed term in Eq. (151) is a
maximum at e = 0.75 while e
d
, like the continuity
wave velocity given by the left-hand side of Eq.
(148), and other related hydrodynamic phenomena
that are proportional to (1 e)e
n1
, including the
bulk turbulent intensity (Joshi, 1983; Kang and Kim,
1986), all show a maximum with respect to voidage at
e = (n 1),n, which is thus considered a crucial turn-
ing point for these phenomena (Kang and Kim, 1988).
8.2 Radial Dispersion
Radial dispersivities of liquid in liquid-uidized beds
are typically one or two orders of magnitude smaller
than the corresponding axial dispersivities, so that the
radial Peclet number, Pe
r
(= d
p
U,eD
r
), usually falls
within the range of 110 (Couderc, 1985; Jean and
Fan, 1998), D
r
increasing with d
p
at any given value
of the interstitial velocity, U/e (Kang and Kim, 1986),
and with r
p
at any given value of e (Cairns and
Prausnitz, 1960b). Two studies of radial dispersion
have shown minima in Pe
r
at e - 0.7 (Hanratty et
al., 1956; Cairns and Prausnitz, 1960b), which
Hanratty et al. attributed to a measured maximum in
the scale of turbulence at the same voidage, while
Cairns and Prausnitz observed, the most active local
particle motion occurs in the range of e = 0.7. A third
study (Kang and Kim, 1986) showed similar minima in
Pe
r
at e - (n 1)/n, which they attributed to the
maxima in e
d
and bulk intensity of turbulence detailed
in the paragraph above. The experimental values of n
for this study were such that both (n 1),n and the
measured critical values of e were closer to 0.6 than to
0.7.
9 NONHOMOGENEITIES AND INSTABILITY
A frequently encountered but underreported disrup-
tion of an otherwise relatively homogeneous
liquid-uidized bed occurs when nonuniformity of
the liquid velocity prole above the distributor gives
rise to bulk circulation of the solids. This eect can
also manifest itself as a convective instability asso-
ciated with a uniform distributor having an insucient
pressure drop compared to that of the uidized bed
(Medlin et al., 1974; Medlin and Jackson, 1975;
Agarwal et al., 1980). Most commonly it occurs with
upow of solids at the center of the column and down-
ow near the walls (e.g., Latif and Richardson, 1972).
The net eect, usually unnoticed and therefore under-
reported, is that the bed expansion or overall voidage
is smaller than in the absence of the bulk circulation,
i.e., smaller than would be predicted by, e.g., Eq. (32),
for the same supercial velocity (Epstein et al., 1981).
As shown roughly by Hiby (1967), and more rigor-
ously by Masliyah (1989), this phenomenon can be
explained by the incremental increase of voidage in
the solids upow region (channeling) and the corre-
sponding incremental decrease in the downow region,
the result of which is a net decrease of total drag on the
uidized solids. Masliyah demonstrated experimentally
that the same contraction eect can be brought about
by mechanical stirring of a liquid-uidized bed.
An arguably ubiquitous nonhomogeneity, at least
for U,U
mf
> 1.5 0.3 (Ham et al., 1990), which has
received much theoretical and experimental attention
in the literature, e.g., by Jackson (1963, 1985),
Anderson and Jackson (1968, 1969), and Homsy and
coworkers (e.g., El-Kaissy and Homsy, 1976;
Didwania and Homsy, 1981; Ham et al., 1990), are
the horizontal striations or bands of low particle con-
centration, dubbed parvoids by Hassett (1961a,b),
which rise wavelike through a liquid uidized bed
and become more pronounced as they rise. The growth
rate of these voidage waves, which may sometimes be
accompanied by short-lived bubblelike formations
(Cairns and Prausnitz, 1960b; Hassett, 1961a), tend
to increase with increasing voidage for a given system,
and to increase with increasing bubble size for a given
voidage. This type of disturbance, which is associated
with liquid uidization of relatively low-density solids,
e.g., water uidization of 1 mm glass beads (Jackson,
1994), is relatively innocuous, i.e., it is minimally dis-
ruptive of the operating characteristics of a nondis-
turbed bed and its eects are incorporated in the
previously written empirical equations, e.g., in the
equations for predicting bed expansion. Jackson
(1994) has pointed out that this type of instability
should not be confused with the disruptively persistent
bubbling and consequent large-scale particle circula-
tion (Jean and Fan, 1998) characteristic of aggregative
uidization, which is a property of most gas-uidized
beds but for liquid uidization is usually restricted to
relatively high density or large particles, e.g., water
uidization of 3 mm lead spheres (Cairns and
Prausnitz, 1960b). For aggregative uidization, most
of the previously written empirical equations, includ-
ing those for predicting bed expansion, no longer
apply. These two types of instability are not explicitly
distinguished from each other in Table 1 of Di Felice
(1995), in which are recorded the nonhomogeneous
behavior observed in 26 dierent studies of liquid
uidization from the literature.
A straightforward quantitative criterion which pur-
ports to distinguish between particulate and aggrega-
tive uidization of both gas and liquid uidized beds
has been deftly developed by Foscolo and Gibilaro
(1984, 1987) and Gibilaro et al. (1986b). The sources
of the instability are the previously mentioned voidage
disturbances or continuity waves, the kinematic velo-
city of which is given by
U
e
= kU
0
n(1 e)e
n1
(153)
After Wallis (1969), the countervailing stabilizing force
is provided by the particle concentration gradients that
give rise to the so-called elastic waves, the dynamic
velocity of which is derived as
U
E
=
3.2gd(1 e)(r
p
r)
r
p
_
(154)
On equating U
e
with U
E
and rearranging the result,
one arrives at
gd(r
p
r)
k
2
U
2
0
r
p
_ _
0.5
0.56n(1 e)
0.5
e
n1
=
ve. stable (particulate)
0. stability limit
ve. unstable (aggregative)
_
_
(155)
The rst term in Eq. (155) is equivalent to (Arr/r
p
)
0.5
/
kRe
0
. Since in general (Epstein and LeClair, 1985),
U
0
=
c
m
(r
p
r)
1,m
g
1,m
d
(3m),m
r
(m1),m
m
(2m),m
(156)
where m varies from 1 in the Stokes region to 2 in the
Newton region, and c
m
varies correspondingly from
1/18 (=0.0556) to [4/(3)(0.44)]
1/2
(=1.74), Eq. (155) at
the stability limit can be rewritten as
r
(2m2),m
m
(42m),m
k
2
c
2
m
g
(2m),m
d
(63m),m
(r
p
r)
(2m),m
r
p
_ _
1,2
0.56n(1 e)
0.5
e
n1
= 0 (157)
Thus aggregative uidization is favored by high values
of r
p
and d and low values of m and r (and the reverse
in each case for particulate uidization), which agrees
qualitatively with experimental observations.
Quantitatively, three situations arise (Gibilaro et al.,
1986b): (1) The left-hand side (LHS) of Eq. (155) is
positive for all values of e, in which case particulate
uidization prevails from minimum uidization to par-
ticle transport, e.g., for 0.1 mm copper spheres ui-
dized by water. (2) The LHS of Eq. (155) is zero at
e = e
b1
and e = e
b2
, i.e., positive for e up to e
b1
, nega-
tive between e
b1
and e
b2
, and positive again at e > e
b2
,
where e
mf
- e
b1
- e
b2
- 1. Thus particulate uidiza-
tion prevails from e
mf
to the minimum bubbling voi-
dage e
b1
, aggregative uidization from e
b1
to the
maximum bubbling voidage e
b2
, and particulate u-
idization again in the very dilute phase regime from
e
b2
to a voidage of unity, e.g., for 0.5 mm copper
spheres uidized by water, for which e
b1
- 0.5 and
e
b2
- 0.95. (3) The LHS of Eq. (155) is zero for
e = e
b1
- e
mf
and e = e
b2
- 1 and negative between
e
b1
and e
b2
, in which case vigorous bubbling starts as
soon as the bed is uidized and ceases only at e
b2
, the
maximum bubbling voidage, above which dilute phase
particulate uidization prevails, e.g., for 2 mm copper
spheres uidized by water (Chen et al., 1999), for
which e
b2
- 0.97.
Gibilaro et al. (1990) have subsequently corrected
Eq. (155) for the added or virtual mass eect due to the
uid acceleration that accompanies particle accelera-
tion. The correction involves adding r/2 to both r
p
and r wherever they occur in the equation, as well as
multiplying the second term in the equation by a factor
exceeding unity that approaches unity as r/r
p
approaches zero. However, since this correction only
becomes important as r/r
p
approaches unity, i.e.,
under conditions when the LHS of Eq. (155) will in
most cases be positive whether or not the correction
is applied (and will hence indicate particulate uidiza-
tion, in agreement with experiment), and given the
experimental uncertainty involved in pinpointing
regime transitions, this added complication can be
safely dispensed with, at least for purposes of predict-
ing aggregative vs. particulate uidization.
The predictions of Eq. (155) have reasonably well
matched the relatively few available regime transition
data on liquid uidization (Gibilaro et al., 1986b), as
well as the much more numerous data on gas uidiza-
tion (Foscolo and Gibilaro, 1984, 1987; Chen et al.,
1999; these references also include some liquid uidiza-
tion comparisons). However, Batchelor (1988) has
questioned the basis of Eq. (154) and has proposed
other stabilizing mechanisms, while Jackson (1994)
has also criticized Foscolo and Gibilaro for not prop-
erly distinguishing between the stability of a uniform
bed against small perturbations and the phenomenon
of aggregative uidization. The alternative stability cri-
terion developed by Batchelor (1988) is similar in form
to Eq. (155), but it contains two quantities that are
more dicult to evaluate. Until the challenging task
of solving this problem by two-dimensional nonlinear
stability analysis is accomplished (Jackson, 1994), Eq.
(155) can serve as a useful semitheoretical or semi-
empirical guideline.
10 DISTRIBUTOR DESIGN
The simplest conguration for a liquid-uidized bed
distributor, common in the early mineral dressing
literature (Richards and Locke, 1940; Taggart, 1953)
on teeter-bed classiers, is a tapered tube with a
small included angle (e.g., 20 108) which diverges
smoothly into the cylindrical column above, and into
which the uidizing liquid is introduced either axially
(through an optional solids-retaining screen) or con-
centrically. This type of distributor is not often used
now, even for laboratory purposes, as it will usually
violate two conditions that are considered important
for many practical applications, namely, that the fric-
tional pressure drop, p
d
, across the distributor be
suciently large with respect to the frictional pres-
sure drop, p
f
, across the uidized bed, and that
the longitudinal liquid velocity distribution above the
distributor be uniform in the lateral direction.
A liquid distributor in current practice consists most
commonly of a plate with evenly spaced perforations,
but other distributors used include a sintered metal or
glass plate, or several layers of metal gauze, or any
number of other uniformly (or relatively so) porous
materials; and each is normally preceded by a calming
section (Fig. 1), the function of which is to equalize the
liquid ow as much as is feasible before it gets to the
distributor. This section, the bottom of which is a wire
screen and the length of which can vary from less than
to several times the column diameter, is usually lled
from bottom to top with rings, saddles, spheres, or
other immobilized packings, or alternately with tubu-
lar or planar ow straighteners. Bascoul et al. (1988)
found that by only partially lling their calming section
with the same particles as used in the uidized bed
above, the calming section, which then acted as a ui-
dized rather than a xed bed, produced greater unifor-
mity of ow from the distributor and thus much
reduced channeling. The action of this distributor
was then similar to that of an upwardly diverging
conical bed of polydisperse solids, in which the well-
agitated condition of the uidized coarser solids in the
higher velocity bottom zone serves to disperse the
uidizing liquid as it enters the upper zone of ner
solids, thus performing the function of a normal liquid
distributor for the latter, the lower liquid velocity of
which provides the additional advantage of preventing
excessive entrainment (Kwauk, 1992). The ball distri-
butor, developed by Hiquily et al. (1979) for gas
uidization, though anticipated by Adler and Happel
(1962) and Grimmett and Brown (1962) for liquid
uidization, consists only of a screen or porous plate
topped by 50150 mm of spheres, e.g., 23 mm lead
shot, which are much heavier than the particles to be
uidized immediately above them. The simplicity and
ecacy of this type of distributor has resulted in its
wide adoption for liquid uidization (Epstein et al.,
1981; Rapagna` et al., 1989; Masliyah, 1989; Di
Felice, 1993).
The requirement of a minimum ratio of distributor
to uidized-bed frictional pressure drop, coupled with
a modicum of distributor uniformity, arises from the
need to suppress the convective instabilities that give
rise to large-scale particle circulation, usually upward
in the central regions and downward near the walls.
This eect in a two-dimensional column has been stu-
died by Jackson (1985) and coworkers (Medlin et al.,
1974; Medlin and Jackson, 1975; Agarwal et al., 1980),
who found both theoretically and experimentally that
the value of p
d
/p
f
required for convective stability
decreases with increasing bed depth for a given voidage
but increases with bed width W up to some limiting
value of W above which no further increase of p
d
/p
f
is required. This limiting value of W increases with bed
depth. The analysis and experimental results, which
were limited to water uidization of 0.8 mm glass
beads from e
mf
= 0.42 to e = 0.48, indicate in their
most conservative interpretation that p
d
/p
f
should
exceed 0.2 to assure convective stability for large-
diameter (e.g., 1 m) beds, while lower values of this
ratio will be sucient for smaller bed diameters, espe-
cially with deep beds. However, Latif and Richardson
(1972), working at much higher voidages (e = 0.55
0.95, U/U
mf
= 2.38.6) with 6.2 mm soda glass
spheres uidized by dimethyl phthalate (r = 1183 kg/
m
3
, m = 0.0105 Pa s) in a 102 mm diameter cylindrical
column that showed large-scale circulation patterns
that occupied an increasing volume of the bed from
the lower part upward as the voidage was increased,
reported that their multilayered metal gauze distribu-
tor delivered a uniform ow and that under all the
experimental conditions the pressure drop across the
distributor was considerably greater than that across
the uidized bed. These results imply an even higher
p
d
/p
f
requirement at high voidages than the con-
servative value recommended above based on studies
at e close to e
mf
. It should be noted, though, on the
reasonable assumption of Darcys law for ow through
a porous distributor, that p
f
is essentially invariant
with velocity, so that a distributor that satises the
criterion p
d
,p
f
= 0.2 at minimum uidization will
yield considerably higher values of p
d
,p
f
at higher
voidages, e.g., at U = 5U
mf
, p
d
,p
f
would equal
unity.
Even if the above large-scale convective circulation
is avoided by suciency of p
d
, signicant inhomo-
geneities in ow and particle distribution can result
from small imperfections in construction of liquid dis-
tributors of various designs (Volpicelli et al., 1966;
Agarwal et al., 1980; Asif et al., 1992). Typically one
may nd, close to the distributor, high-velocity, high-
voidage channeling streams interspersed with relatively
stagnant, low-voidage regions. Particles with densities
very close to that of the uidizing liquid, e.g., water-
uidized polystyrene (r
p
= 1050 kg/m
3
, d
p
= 3 mm)
particles, have been found to be most vulnerable to
the design of a perforated plate distributor (Asif et
al., 1991), but even for somewhat higher density par-
ticles, e.g., polypropylene (r
p
= 1610 kg/m
3
, d
p
= 3
mm), the liquid axial dispersivity near the distributor
tends to be much greater than higher up in the uidized
bed, unless the hole density of the perforated plate, for
a given fractional free area F, is suciently high, e.g.,
1.3 (2 mm diameter) holes/cm
2
for F = 0.04 (Asif et al.,
1991, 1992). For water uidization of 2 and 3 mm glass
(r
p
= 2460 kg/m
3
) spheres, however, this vulnerability
declines to the point that distributor eects are found
to be negligible over a wide range of hole density (Asif
et al., 1992), presumably owing to the stabilizing eect
of the higher density particles. According to Briens et
al. (1997), channeling in liquid uidized beds can arise
not only from imperfections or poor design of the
liquid distributor but also when the particles to be
uidized are very angular, e.g., biobone particles.
These investigators found that, while channeling can
be gradually reduced by increasing the liquid velocity
from U
mf
upward at the expense of increased pumping
cost, particle attrition, and particle entrainment, a
method of reducing channeling in their 0.1 m diameter
bed without incurring these undesirable consequences
was simply to introduce secondary liquid through
horizontal jets issuing from a sparger located 90 mm
above their perforated plate distributor.
11 MASS (AND HEAT) TRANSFER BETWEEN
PARTICLES AND LIQUID
Steady state heat transfer and mass transfer (of solute
A) between the surface of the solid particles and the
liquid in a vertical uidized bed can be represented
dierentially by the energy balance,
rc
p
UdT
dz
=
(l l
eddy
)d
2
T
dz
2
h
p
S
v
T
s
T ( ) (158)
and the corresponding solute balance,
UdC
A
dz
=
D
a
d
2
C
A
dz
2
k
p
S
v
C
AS
C
A
( ) (159)
respectively, where S
v
= 6(1 e),fd
p
= S,AL, T =
bulk temperature, and C
A
= bulk concentration of A
at the bed level under scrutiny, T
S
and C
AS
are the
corresponding surface temperature and concentration
(e.g., the saturation concentration in the case of a dis-
solving solute), and D
a
, discussed earlier, includes both
the molecular diusivity D
A
and the longitudinal eddy
diusivity of solute A in the liquid. As it turns out, the
heat transfer coecient in a typical liquid-uidized bed
is high enough that thermal equilibrium between the
liquid and the surface of the solids is reached within a
very few particle layers of the distributor. This makes it
very dicult (e.g., Holman et al., 1965), but at the
same time of minor practical importance, to make reli-
able measurements of the particle-liquid heat transfer
coecient h
p
. Because the molecular diusivity of
mass, D
A
, of solute A in a liquid is some two or
three orders of magnitude smaller than the molecular
diusivity of heat, l/rc
p
, (i.e., Sc Pr), particle liquid
mass transfer is much slower and therefore much easier
to measure. We shall therefore focus on mass transfer.
If we assume plug ow of the liquid, D
a
= 0 and Eq.
(159) can be rearranged to
dC
A
C
AS
C
A
=
k
p
S
UAL
_ _
dz (160)
which, when integrated between the limits z = 0,
C
A
= C
A1
and z = L, C
A
= C
A2
, results in
ln
C
AS
C
A1
C
AS
C
A2
_ _
= k
p
S,UA (161)
and is equivalent to
m
A
= Q C
A2
C
A1
( ) = k
p
S C
A
( )
l.m.
(162)
For the opposite extreme of perfect mixing, dC
A
,dz =
0 and C
A
= C
A2
for all values of z except z = 0, so that
m
A
= Q C
A2
C
A1
( ) = k
p
S C
AS
C
A2
( ) (163)
Since (C
A
)
1.m.
> (C
AS
C
A2
), and since the real
situation falls between the two, it follows that determi-
nation of k
p
assuming plug ow will tend to under-
estimate k
p
while its determination assuming perfect
mixing will tend to overestimate k
p
. Arters and Fan
(1986) have solved Eq. (159) for nite D
a
, assuming
appropriate boundary conditions, and for two-phase
particleliquid uidization, k
p
, based on Eq. (162), is
much closer to the answer by this solution than k
p
by
Eq. (163). Measured axial solute proles are also much
closer to those for plug ow than those for perfect
mixing. All this provides justication for applying the
plug ow model to the determination of k
p
, a proce-
dure used by most investigators of particleliquid mass
transfer. Values of k
p
obtained from empirical correla-
tions generated by these investigators should then be
applied using Eq. (162).
The most frequently applied method in the litera-
ture for measuring the total mass transfer rate, m
A
,
between a cluster of suspended particles and the
surrounding liquid is by uidizing a total mass M of
slightly soluble particles (e.g., benzoic in water), which
slowly dissolve into the liquid, so that over a nite time
period, t, m
A
can be determined as M,t, and the
slightly altered particle dimensions averaged from
before and after the run. This procedure in the case
of slow dissolution is often more accurate than deter-
mining m
A
as Q(C
A2
C
A1
) because of the large error
associated with measuring the small change in C
A
.
Besides solids dissolution methods, used especially by
investigators from Toulouse (e.g., Couderc et al., 1972;
Damronglerd et al., 1975; Laguerie and Angelino,
1975; Vanadurongwan et al., 1976; Tournie et al.,
1979), and also by many others (e.g., McCune and
Wilhelm, 1949; Evans and Gerald, 1953; Fan et al.,
1960), other methods include adsorption (e.g., Ganho
et al., 1975), ion exchange (e.g., Rahman and
Streat, 1981) and crystallization under mass-transfer-
controlled (e.g., low Re) conditions (Laguerie and
Angelino, 1975). Some investigators (e.g., Riba and
Couderc, 1980) use an electrochemical method that,
like the earlier heat and mass transfer experiments of
Rowe and Claxton (1965) for a xed array of
spheres, focuses on the transfer from a single xed
particle within the xed or uidized array of parti-
cles. This method, as representative of what happens
when all the particles are involved in transferring
mass, must be treated with some caution, particularly
for dense suspensions at low Reynolds numbers, as it
neglects entirely any interaction of the diusional
boundary layers from the various particles. For
heat transfer, focus on a single particle is even less
representative of what happens when all the particles
are absorbing or emitting heat because of the thicker
and therefore more interactive thermal boundary
layers. The early notion that for a xed or uidized
bed of spheres, Nu
p
(=h
p
d/l) or Sh
p
(=k
p
d/D
A
)
cannot decline to less than 2 as Re approaches zero
(as in the case of an isolated sphere) has been laid to
rest by Cornish (1965) and by Nelson and Galloway
(1975), who showed that for dense assemblages of
particles both Nu and Sh decline to zero as Re
approaches zero.
The early correlations of low ux particleliquid
mass transfer in uidized beds were patterned after
those for xed beds, in which, for spheres in the
absence of free convection,
Sh
p
= fctn Re. e. Sc ( ) (164)
(For forced-convection heat transfer, Nu
p
replaces Sh
p
and Pr replaces Sc, but such a conventional analogy
between heat and mass transfer in liquiduidized beds
has been questioned by Briens et al., 1993.) However,
in uidized beds, unlike xed beds, e is not an inde-
pendent variable but is related to Re and the properties
of both liquid and particles as incorporated in the
Archimedes number, Ar, and exemplied by equations
such as (61), (62), or (64), whereby
e = fctn Re. Ar ( ) (165)
Combining Eqs. (164) and (165), we obtain
Sh
p
= fctn Re. Ar. Sc ( ) (166)
Correlation for spheres then commonly takes the
exponential form
Sh
p
= K
/
Re
a
Ar
b
Sc
c
(167)
Actually, Ar = Ga (r
p
r)/r, and the Toulouse
group that initiated correlation along these lines
always tted their data to
Sh
p
= K
/
Re
a
Ga
b
r
p
r
r
_ _
b
/
Sc
c
(168)
but even for their spread of (r
p
r),r = 0.271.14,
the largest dierence between b and b
/
in any such
correlation developed by this group was only 0.023
(Tournie et al., 1979), while the one study that
extended the upper range of (r
p
r)/r to 7.07 found
that b b
/
= 0 (Nikov and Delmas, 1987). Thus Eq.
(167) takes precedence over Eq. (168) (Epstein, 1992).
The index on Re in Eq. (167) has been generally
found to be zero (Nikov and Karamanev, 1991,
1992) or very close to zero, e.g., +0.004 by Tournie
et al. (1979) for Re = 1.61320 and 0.07 by Riba and
Couderc (1980) for Re - 15025,000 and e = 0.45
0.90. Within the experimental error it is conveniently
taken as zero, which means that the eect of any
increase in liquid supercial velocity for a given bed
is compensated for by a corresponding increase in voi-
dage. We are then left with an equation similar in form
to that recommended by Calderbank (1967) for mass
transfer into liquids from both solid spheres and small
gas bubbles, namely,
Sh
p
= K
/
0
Ar
b
Sc
c
(169)
for which Calderbank proposed and rationalized
b = 1,3, which is in remarkable agreement with values
of b = 0.323, 0.33, 0.333, and 0.33 found experimen-
tally by Tournie et al. (1979), Riba and Couderc
(1980), Nikov and Delmas (1987), and Nikov and
Karamenev (1991), respectively, the last for inverse
uidization. The range of Ar encompassed by these
studies was 6701.1 10
8
. Since both Sh
p
and Ar
1/3
are directly proportional to d, it follows that k
p
is
independent of particle size over this range, a result
reported explicitly by more than one author (e.g.,
Riba and Couderc, 1980; Panier et al., 1980; Nikov
and Delmas, 1987).
There is more disagreement on the value in Eq.
(169) of the exponent c, which can vary from 1/3
assuming boundary layer theory to assuming pene-
tration theory, but the empirical values of which
usually come closer to the former than to the latter.
The lower the value assigned to c, the higher the value
of the dimensionless constant K
/
0
. Thus Tournie et al.
(1979) for Sc = 3051595 found c = 0.400 and
K
/
0
= 0.245, modied slightly to K
/
0
= 0.228 by Arters
and Fan (1986); Riba and Couderc (1980) found c =
0.34 and K
/
0
= 0.267 for Sc = 5507700; while Nikov
and Delmas (1987) for Sc = 86019,900 and Nikov
and Karamanev (1991) for Sc = 9382181 both
found c = 1,3 but K = 0.34 and 0.28, respectively, in
substantial agreement with Calderbank (1967), who
proposed c = 1,3 and K
/
0
= 0.31 0.03. Thus experi-
mental uncertainties at this stage prompt us to write
Sh
p
= 0.23 0.31 ( )Ar
1,3
Sc
0.400.33
(170)
In Eq. (170), application of the upper and lower values
of Sc along with the corresponding lower and upper
values of the coecient K
/
0
should give values of Sh
p
that bracket reality.
Some investigators (e.g., Kikuchi et al., 1983) have
proposed a dimensionless correlation for Sh in which
the key dimensional variable is e
d
, the energy dissipa-
tion term given by Eq. (152). The underlying theory
has been criticized by Arters and Fan (1986) on the
grounds that it assumes all energy input to the system
is dissipated as turbulence when in fact viscous dissipa-
tion is far from negligible, and the actual correlation of
literature data by Kikuchi et al. showed considerable
scatter.
Equation (170) applies to low ux mass transfer
between liquid and uidized spheres, all of which and
the total surface of which are contributing to the mass
transfer. For nonspherical particles, Limas-Ballesteros
et al. (1982b) found that with both Sh
p
and Ar based
on the equivolume sphere diameter, d
p
, Sh
p
was mod-
ied by the factor f
1.35
over the sphericity range 0.71.
For spheres with a surface area partly active and partly
inert, Panier et al. (1980) found electrochemically that,
for dierent ratios and geometries of the active and
inert parts, Sh
p
(in which k
p
is based only on the active
area) increased as the fraction f
A
of active area
decreased, the modication factor being f
0.25
A
over
the range f
A
= 0.0570.98. A qualitatively similar
mass transfer intensication eect occurs when inert
particles of higher density and smaller diameter are
mixed with active particles; the higher the fraction of
inert particles the greater the observed increase in Sh
p
(Yang and Renken, 1998). Finally, when the solute ux
due to mass transfer from or to the surface of the
particles is not completely overshadowed by the main-
stream mass ux (or mass velocity), e.g., for dissolu-
tion of very soluble solids, Sh
p
by Eq. (170) must be
corrected owing to distortion of the mainstream velo-
city and concentration proles, rapid change in the
area of the solidliquid interface, and sharp physical
property variations with concentration near the inter-
face. Any analogy, conventional or otherwise, between
heat and mass transfer breaks down in these circum-
stances. Chhun and Couderc (1980) found that for
dissolution experiments with B
/
= surface mass ux/
mainstream mass ux = 0.0130.674,
Sh
p
Sh
p.low flux
= 1 B
/
_ _
0.56
(171)
which signies a reduction of Sh
p
as B
/
increases. The
reverse eect, viz. an increase in Sh
p
, occurs when the
mass transfer is from the liquid to the solid surface
(Bird et al., 1960), e.g., as in adsorption.
12 HEAT AND MASS TRANSFER BETWEEN
SUBMERGED SURFACES AND LIQUID
12.1 Heat Transfer
Heat transfer between a liquid uidized bed and a sub-
merged surface in the bed, most commonly the bed
wall itself, has been the subject of many experimental
studies, the conditions, scope, and results of which
have been summarized by Haid et al. (1994) in their
Tables 1 and 2. A typical set of results is shown in Fig.
11, where it is seen that the increase of the heat transfer
coecient, h, with supercial liquid velocity, U, that
occurs in the xed packed condition continues in some-
what attenuated form when the bed is uidized, even-
tually achieving a maximum and then declining to
merge with the single-phase uid ow line at U - U
0
.
The increase in h for a liquid-uidized bed relative to
single-phase liquid ow at the same value of U can be
by as much as a factor of seven (Richardson et al.,
1976; Klaren and Halberg, 1980; Bremford et al.,
1996).
Using 2665 data points from 43 publications,
dierent weighting procedures and dierent numbers
of dimensionless parameters, Haid et al. (1994) and
Haid (1997) arrived at six empirical equations, of
which the one with the lowest standard deviation
(11.5%) and the smallest average relative error
(32.0%), considerably smaller than that obtained for
the same data using 38 other proposed correlations
from the literature, was
Nu =
hd
p
l
_ _
= 0.0413Re
0.79
Pr
0.67
1 e ( )
0.12
e
1.66
d
p
D
h
_ _
0.10
r
p
r
r
_ _
0.04
(172)
where D
h
is the hydraulic diameter of the column (or
annulus, in the case of a vertically immersed cylindrical
heater or cooler). The range of parameters (some inter-
related, e.g., e depends upon Re for any given value of
Ar) covered by Eq. (172) was Re = 0.0209400,
Pr = 1.657700, e = 0.400.99, Ar = 3.8567,000,000
and d
p
/D
h
= 0.00130.210. If Eq. (32) is combined
with Eq. (172), then
Nu o e
0.79n1.66
1 e ( )
0.12
(173)
from which it follows that Nu is a maximum at e =
(0.79n 1.66)/(0.79n 1.54). Thus for spheres in the
Stokes regime (Re
0
< 0.2), n = 4.8 and e
max
= 0.95,
while in the Newton regime (Re
0
> 500), n = 2.4
and e
max
= 0.66. Experimentally measured values of
e
max
generally fall between 0.62 and 0.82, with the lar-
gest cluster in the vicinity of 0.73 (Jamalahmadi et al.,
1997). Note that Eq. (172) is inapplicable at e = 1.
Combination of Eqs. (172) and (32) is also equi-
valent to
h o d
1
p
Re
0.72
0
d
0.17
p
(174)
For spheres in the Stokes regime, Re
0
= Ar/18 o d
3
p
,
whence from Eq. (174) it follows that h o d
1.33
p
, while
in the Newton regime, Re
0
=
3Ar
_
o d
3,2
p
, whence
h o d
0.25
p
. Thus the larger the particles the larger the
value of h, but the smaller the inuence of d
p
on h.
Both eects are in agreement with experimental results
(Haid et al., 1994; Jamialahmadi et al., 1996).
Equation (172) also implicitly indicates that, at any
given bed voidage, h increases with particle density,
primarily because the value of Re required for expan-
sion to the given voidage will increase with particle
density and secondarily (and very slightly) through
the term (r
p
r,r)
0.04
. This eect too is in agreement
with experiment (Jamialahmadi et al., 1995).
Figure 11 Heat transfer coecient between wall and liquid for packed bed, uidized bed, and single-phase ow as a function of
supercial liquid velocity. (After Haid et al., 1994; data points from Kato et al., 1981.)
An attempt at a more mechanistic, rather than
purely empirical, approach to the heat transfer
problem in turbulent ow has been made by
Jamialahmadi et al. (1995, 1996, 1997), who divided
the total heat transfer surface area into two
partsA
p
, the portion aected by particle contact at
any instant, and A
c
, the remaining portion. From A
p
(assuming a bed being heated), heat is transferred by
transient conduction to the adjacent liquid layer, which
is then transported in the wake of departing particles
into the liquid bulk and replaced by cooler liquid. This
turbulence promoting disruption of both the viscous
sublayer and the thermal boundary layer is the prin-
cipal source of heat transfer enhancement to single-
phase liquid ow. Some heat, but a much smaller frac-
tion than in the case of gas uidization, is also assumed
to be transferred via conduction into and transport of
particles, i.e., by particle convection. From A
c
, heat is
transferred in parallel by normal single-phase forced
convection, depending upon the type and congura-
tion of surface involved. Crucial to this model is
knowledge of the area ratio A
p
,(A
p
A
c
), which in
turn depends on the fraction of the total number of
uidized particles that are in contact with the heat
transfer surface at any instant, n
p
/N
p
. At this juncture,
Jamialahmadi et al. were forced to resort to empiri-
cism, i.e., n
p
/N
p
had to be expressed as exponential
functions of d
p
/D
h
, 1 e, and e e
mf
, with the coe-
cient and three exponents evaluated from the experi-
mental data, which required assigning dierent values
of each for wall-to-bed, vertically immersed plate,
vertically immersed cylindrical and horizontally
immersed cylindrical heating; and with dierent values
again for aggregative than for particulate uidization.
They thus generated a total of 4 4 2 = 32 empirical
constants, plus another 24 for improved prediction of
h
max
(Jamialahmadi et al., 1997). The criterion recom-
mended for e
max
, corresponding to maximum heat
transfer, was that the ratio U
e
/U
E
, which is obtained
on dividing Eq. (153) by Eq. (154) and is proportional
to (1 e)
0.5
e
n1
, be maximized, the result being
e
max
=
n 1
n 0.5
(175)
Equation (175) yields values at n = 2.44.8 of
e
max
= 0.740.88, whereas Eq. (149), obtained either
from Eq. (148) or by simply maximizing U
e
given
by Eq. (153), yields e
max
= (n 1),n, from which
e
max
= 0.580.79, in better agreement with the experi-
mentally observed ranges of the wall-to-bed maximum
for both heat and mass transfer (Kang and Kim, 1988).
One factor taken into account by the equations of
Jamialahmadi et al. (1995, 1996, 1997) but not by Eq.
(172) is particle convection, the heat transfer coecient
of which is written as directly proportional to
l
p
r
p
c
pp
f
_
, where f is the collision frequency of the
particles. The proportionality constant for cylindrical
particles is assigned twice the value for that of equal
sized spheres to account for the somewhat higher value
of h measured for the former than for the latter
(Richardson et al., 1976), an eect implicitly accounted
for in Eq. (172) by use of the equivolume sphere dia-
meter to characterize the particle size of cylinders. The
particle conductivity, l
p
, and heat capacity, c
pp
, are
not however accounted for in Eq. (172). As it turns
out, though, a 391-fold increase in l
p
r
p
c
pp
as between
lead glass and copper-coated aluminum spheres of
equal diameter (1.98 mm) and density (2890 kg/m
3
)
uidized by water showed only a small gain in h for
the latter, and only at e < e
max
(-0.75), the gain
increasing to a maximum of 18% as e decreased to
0.45 (Wasmund and Smith, 1967; Haid et al., 1994);
while a 255-fold increase in l
p
r
p
c
pp
as between copper
and lead spheres of equal diameter (=4.5 mm) ui-
dized by water produced a maximum net gain in h
for the copper at e _ e
max
(-0.72) of only 14%, both
experimentally and as tted by Jamialahmadi et al.
(1995). Given that the column diameter for the latter
runs was only 23.8 mm, so that the inordinately high
wall eect ratio, d
p
,D
c
= 0.19, could have given rise to
an unrepresentatively high degree of particle convec-
tion, Molerus and Wirth (1997) have questioned the
generality of this result and have recommended con-
servatively that, until further evidence proves other-
wise, the thermal properties of the particles can be
ignored in calculating the heat transfer coecient for
liquid-uidized beds, as in Eq. (172).
12.2 Mass Transfer
Turning to mass transfer between xed submerged sur-
faces and liquid-uidized beds, it has been noted by
several authors (King and Smith, 1967; Briens et al.,
1993; Schmidt et al., 1999) that the conventional ana-
logy between heat and mass transfer is inapplicable.
Thus, substituting Sc for Pr and Sh (= k
m
d
p
,D
A
) for
Nu in Eq. (172) results in overprediction of Sh by
factors varying from about 5 to 2 as measured Sh
rises from 2 to 30 (Schmidt et al., 1999). Schmidt et
al. attribute this result to the presence of particle con-
vection in heat transfer and its absence in mass trans-
fer, but in the light of the discussion immediately
above, factors of 25 cannot be accounted for by this
explanation. King and Smith (1967) have argued that
the analogy breaks down because for mass transfer the
entire resistance to transfer is close to the submerged
surface, while for heat transfer there is also an appreci-
able resistance in the bed, but this observation is
merely a consequence of the fact that Sc often (as in
water) exceeds Pr by orders of magnitude. This fact in
and of itself should not preclude the applicability of,
for example, the ChiltonColburn (1934) analogy,
which is applicable to heat and mass transfer from a
pipe surface to owing liquids despite the same discre-
pancy between Sc and Pr, i.e., between a thin mass
transfer boundary layer and the corresponding thicker
heat transfer boundary layer. A more credible explana-
tion for the apparent breakdown of the analogy is
provided by Briens et al. (1993), who invoke the lm-
penetration model of Toor and Marchello (1958) in
conjunction with the large dierence between Sc and
Pr. However, application of this model depends on
knowledge of two parameters, the lm thickness and
the fractional rate of surface renewal, which requires a
measurement of Sh and Nu for every operating condi-
tion. Though theoretically sound, the method does not
lend itself at present to quantitative prediction of Sh
for xed immersed surfaces.
Most investigators have correlated their mass trans-
fer results in the Beek (1971) form of the Chilton
Colburn (1934) equation,
j
m
e =
k
m
U,e
Sc
2,3
= a
//
Re
1 e
_ _
b
//
(176a)
which is equivalent to
Sh = a
//
Re
1b
//
(1 e)
b
//
e
1
Sc
1,3
(176b)
Values of a
//
and b
//
from dierent studies, mainly
using the electrochemical technique of Lin et al.
(1951), have been tabulated in several papers (Storck
and Coeuret, 1980; Lee et al., 1997; Schmidt et al.,
1999). The coecient a
//
in the case of some studies
(e.g., Jottrand and Grunchard, 1962), is itself a func-
tion of e and/or Sc, but it is more usually a constant. A
plot by Lee et al. (1997) of ten correlations from the
literature as j
m
e vs. Re shows a fourfold variation of
j
m
e at Re/(1 e) = 1 (i.e., a
//
varies from 0.3 to 1.2),
but a much smaller variation (j
m
e = 0.0230.031) at
Re/(1 e) = 1000. The data of Lee et al. for mass
transfer from an axially immersed cylindrical surface
in an aqueously uidized bed of glass beads
(d
p
= 0.300.77 mm) at Sc = 1542, e = 0.550.80,
and Re/(1 e) = 4.2157 showed mass transfer
enhancement factors of 27 relative to single-phase
ow at the same supercial velocity and yielded a
representative correlation, well within the extremities
of the others, namely Eq. (176a) with a
//
= 0.44 and
b
//
= 0.394. If one combines Eq. (32) with this result,
then
Sh o e
0.606n1
(1 e)
0.394
(177)
from which it follows that Sh is a maximum at
e = (n 1.65)/(n 1). Thus e
max
for spheres varies
from 0.83 in the Stokes regime (n = 4.8) to 0.54 in
the Newton regime (n = 2.4). Experimentally, values
of e
max
that have been reported vary from 0.58 to
0.75 (Schmidt et al., 1999), a range about 0.05 lower
than that for the corresponding heat transfer process.
As e approaches unity, Eq. (176) breaks down.
Most reported mass transfer studies are for a single
or a relatively narrow range of Sc, so that the 1/3
power of Sc inherent in the ChiltonColburn equation
is assumed rather than tested. Recently the 1/3 power
was veried by an extensive series of runs encom-
passing Sc = 1517021, for oxygen transfer from an
axially immersed cylindrical membrane in a water-
uidized bed of spheres with d
p
= 0.3253 mm, r
p
=
250011,343 kg/m
3
, Re/(1 e) = 0.91652, Ar = 32
481,365, and e = 0.41. The correlating equation for
Sh, which contains a minor term to describe molecular
diusion not caused by the uidized bed itself and a
major term to account for turbulence and uidization,
is (Schmidt et al., 1999)
Sh = 0.14Re
1,3
Sc
1,3
0.13(1 e)(e e
mf
)
Sc
Re
_ _
1,3
Ar
2,3
(178)
The correlation index (or coecient of determina-
tion) R
2
for this equation was 0.98, as compared to R
2
= 0.8 for the same data correlated by Eq. (176a).
Neglecting the rst term on the right-hand side of
Eq. (178) relative to the second, assuming e
mf
= 0.4,
and again invoking Eq. (32), we nd that
Sh o (1 e)(e 0.4)e
n,3
(179)
from which it follows that for n = 4.82.4, e
max
= 0.590.65, somewhat narrower than, but within,
the range of reported measurements.
13 APPLICATIONS (as abridged by Wen-Ching
Yang)
Some applications of liquid-uidized beds, such as par-
ticle classication, are over a century old, while many,
such as uidized-bed electrolysis and bioreactors, are
of more recent vintage. None have achieved the indus-
trial and commercial prominence of gas uidization
applications exemplied by catalytic cracking or coal
combustion. The brief discussion below will be limited
to identifying the salient features of some applications,
or broad categories thereof, which have been reported
on in literature accessible to this author, and few spe-
cic examples will be cited. The reader seeking more
information on a given application should consult the
literature on the given unit operation or process indus-
try involved.
Particle classication by size (sizing), density (sort-
ing), or even shape (shaping?) depends on the segrega-
tion characteristics of liquid-uidized beds described in
Sec. 5. For particles of xed known density and shape,
axial particle size distribution at equilibrium in semi-
batch uidization can be estimated from the axial pres-
sure prole, using the perfect classication model
discussed in that section. For narrowly sized solids of
varying density, particle density measurements from
successive bottommost suspension samples in conjunc-
tion with frictional pressure drops measured before
and after the samplings will yield the axial density dis-
tribution of the solids, by appropriate use of Eqs. (11)
and (12) (Galvin and Pratten, 1999). The particle stra-
tication in a rising stream of liquid, which occurs once
equilibrium is achieved within a semibatch uidized
bed, can be maintained intact for continuous separa-
tion of the particles involved via a discharging under-
ow and overow, provided the feed rate is kept
suciently low. An annularly compartmentalized unit
of some elegance for eecting continuous classication
has been patented by Delachanal (1963). Traditionally,
liquid-uidized or teeter bed classiers were devoted
mainly to mineral separations, but recently they have
extended their domain to other materials (Galvin et al.,
1998, 1999). In the actual separations, density always
dominated over size.
Sorting of particles, particularly of bidisperse
mineral mixtures, is more cleanly eected by sink-
and-oat separation. This is most simply done by
using a liquid nonsolvent intermediate in density
between that of the two particle species. With overow
and underow discharge streams, and a continuous
feed suspension introduced at an intermediate posi-
tion, the sharpness of separation decreases as the
feed rate, the feed solids concentration, and the under-
ow/overow ratio are increased (Nasr-El-Din et al.,
1988, 1990). In the absence of an acceptable liquid, a
homogeneous suspension, viz., a water-uidized bed of
narrowly cut ne sand (e.g., 325 400 mesh), could
serve instead. By adjusting the upward liquid ow
carefully, the required suspension density can be
achieved. Coarse coal particles can thus be eectively
separated into a clean fraction of low ash content that
oats and denser fractions of higher ash content that
sink (Needham and Lynch, 1945).
Backwashing of downow granular lters (or xed
bed ion exchange columns) by water uidization of the
lter medium or media (or ion exchange resin) is a well
entrenched industrial procedure, especially for munici-
pal-water sand lters, where what is removed in the
backwash is primarily the ltered solids. The voidage
at which maximum removal occurs is about 0.650.70,
and since typically n = 3.13.4 for the sand used as
lter media, the voidage range is close to that predicted
by Eq. (149), which we have already seen as generating
the critical or optimum voidage for several other
hydrodynamic or hydrodynamically related phenom-
ena. However, since the curve of removal vs. e is
quite at near the maximum, more practical bed
expansions of 4050% (rather than 100%, correspond-
ing to e = 0.7) will produce almost as much lter
cleaning as operating at the optimum (Amitharajah,
1978). Improved cleaning will normally be eected by
an upward air scour before or during the water uidi-
zation, and this is especially essential to the satisfactory
functioning of wastewater lters, which receive
heavier, more variable, and stickier suspended-solids
loads than potable water lters (Cleasby et al., 1975;
Cleasby and Lorence, 1978). Wastewater is therefore
not recommended as the backwash liquid. In the case
of petroleum renery wastewaters, the backwash
includes much oil in addition to the suspended solids
(Brody and Lumpkins, 1977; Kempling and Eng,
1977), and the stickiness of other wastewater deposits
is also probably contributed to by nonaqueous, immis-
cible liquid contaminants. Dual-media lters generally
perform somewhat better than single-medium, and no
worse than triple-media (Cleasby and Lorence, 1978).
Typically, in a dual-media lter, anthracite coal sits on
top of denser silica sand in a coal : silica size ratio
between 2:1 and 4:1 (Cleasby and Sejkora, 1975) or
even 6:1 (Brody and Lumpkins, 1977), and in the
case of a triple-media lter a third component of
even greater density and smaller size, e.g., garnet
sand (Cleasby and Woods, 1975), lies below the silica
sand.
In-situ uidized washing of soils is based on the same
principle as lter backwashing, but for soil washing a
downward water jet that spreads out and then reverses
direction has been proposed (Niven and Khalili,
1998a). As in the case of granular lter backwashing,
the use of an air scour, i.e., airwater uidization,
greatly increases the cleaning eciency (Niven and
Khalili, 1998b).
Another long-standing and industrially well-estab-
lished process involving liquid uidization is that of
seeded crystal growth in a bed uidized by a moder-
ately supersaturated (metastable) solution of the solute
to be precipitated. Because stratication by size of the
growing crystals accompanies the process, the units
involved are usually called classifying crystallizers, of
which the most common is the Krystal (or Oslo or
Jeremiassen) crystallizer (Svanoe, 1940; Bamforth,
1965; Perry et al., 1984). Intensive study of the operat-
ing characteristics and design of such crystallizers has
been undertaken by Mullin (1993) and coworkers (e.g.,
Mullin and Garside, 1967; Mullin and Nyvlt, 1970;
Garside et al., 1972), as well as by many others (e.g.,
Bransom, 1960), most recently by Tai et al. (1999).
Most studies involve the crystallization of inorganic
salts from aqueous solution, but the uidized-bed crys-
tallization of acetylsalicylic acid (aspirin) from abso-
lute alcohol has also been reported (Glasby and
Ridgway, 1968). Crystal growth rates in lean beds
(e>0.98) have been found to be substantially the
same as those in the industrially more common dense
beds (e _0.8), and in both types of beds, the crystal
growth rate is almost always signicantly smaller than
the corresponding mass-transfer controlled dissolution
rate (Garside et al., 1972; Phillips and Epstein, 1974;
Jira-Arune and Laguerie, 1979; Tai et al., 1987).
Fluidized-bed crystallizers can be operated in both
the batch mode and the continuous mode, with bottom
discharge of the enlarged crystal product. Models to
describe the behavior of both batch (Shiau et al., 1999)
and continuous (Frances et al., 1994) uidized-bed
crystallizers have been formulated.
Leaching, i.e., physical or chemical dissolution of a
soluble component embedded in the inert matrix of a
granular solid, and washing for removal of the leaching
or other residual solution within the intraparticle pores
and the interparticle voids of the leached or other gran-
ules, can be accomplished quite eectively by liquid
uidization of the solids with the required solvent.
The principal applications, actual or potential, are
for extraction of mineral or metal values from ores,
and of vegetable oils from seeds. A thorough discus-
sion of uidized bed leaching/washing is provided by
Kwauk (1991/92). In the leacher/washer illustrated in
Fig. 12, the slurry feed is hydraulically distributed into
an enlarged settling head, where much excess liquor is
removed. Solid particles then fall countercurrently
against a rising stream of liquor into the leaching/
washing region, in which a reasonably well-dened
interface usually separates an upper dilute-phase zone
from a lower dense-phase zone. The leaching solvent
or washing liquid is sparged in at the bottom of the
dense-phase region, below which the solid slurry
is compressed, densied, and discharged. Compared
to more conventional leaching/washing equipment
(Treybal, 1980), such uidized leachers/washers are
deemed by Kwauk (1991/92) to have the advantages
of complete hydraulic operation with no mechanical
parts, continuous (though imperfect) countercurrent
contacting in a single column, low solvent-to-solids
ratios, low space requirements, and ease of automa-
tion. Other more complicated uidized leaching/wash-
ing setups, some with countercurrent staging, are also
Figure 12 A uidized leacher/washer. (From Kwauk, 1991/
92.)
described by Kwauk, as well as by Slater (1969), who
included pulsed uidization in his review of counter-
current liquidsolids contactors.
Adsorption and ion exchange are both operations
that are conventionally carried out batchwise in xed
packed beds but can gain certain advantages when
performed in the uidized state. Though physical
adsorption involves only two resistances in series
(external mass transfer of solute to particle macro-
surface and internal diusion within particle pores to
microsurface), while ion exchange involves ve (exter-
nal mass transfer and internal diusion of ions
inwardly, exchange of ions, internal diusion, and
external mass transfer of exchanged ions outwardly),
the two operations commonly employ similar types of
equipment that are subject to similar design procedures
(Treybal, 1980). Adsorption from liquids is mainly
applied industrially to the decontamination of water,
aqueous solutions, and petroleum products, while ion
exchange is principally, though not exclusively, used
for water purication and metal recovery from hydro-
metallurgical leach liquors. Though both operations
must usually be coupled with regeneration (elution)
of the adsorbent or ion exchange resin, and often
with washing as well, it is the original loading (adsorp-
tion, exhaustion) process that is most subject to
improvement by liquid uidization and that has
received most attention industrially and academically
in this respect. Since the voidage is greater for the
uidized bed, the same supercial velocity (upow in
uidization, downow in xed bed) for a given batch
of solids will result in a lower pressure drop (Himsley
and Farkas, 1977). The advantage of continuous
operation can be achieved by adopting a setup similar
to that of Fig. 12 (Slater and Lucas, 1976; Koloini and
Zumer, 1979). If the loaded adsorbent or ion exchange
particles are denser than the unloaded ones, as in the
case of copper being exchanged with hydrogen ions
(Selke and Bliss, 1951), the loaded particles move to
the bottom of the bed, thus enhancing the counter-
currency of the operation. Further enhancement in
that direction is provided by staging, which can have
the desirable result of reducing the required solids
inventory appreciably (Slater, 1982). Detailed reviews
of the various staging methods proposed and of those
adopted industrially, especially for uranium recovery
from its leach liquor, have been published by Slater
(1969), Streat (1980), and Slater (1981). The simplest
and most recently revived proposal for a multistage
uidized bed adsorber is a sieve-plate column without
downcomers or pulsation or controlled cycling; see
Gru newald and Schmidt-Traub (1999). The authors
report that stable operation requires, among other
conditions, that the lowest plate have a smaller free
hole area than the others.
Flocculation in order to achieve clarication of tur-
bid liquids is a process similar to adsorption that can
be eected by the liquid uidization of some seeded
ocs, which then enhance subsequent occulation of
the suspended colloidal particles that cause the turbid-
ity (Svarovsky, 1990). One continuous process of this
kind involves liquid-uidized microsand coated (acti-
vated) with an alginate occulant that serves to occu-
late and retain aluminum or ferric salts from incoming
suspensions and is then externally washed to remove
the oc from the sand and then recycled (Sibony,
1981).
Electrolysis in a uidized bed both with inert and
with electrically conductive particles (extended elect-
rodes) had among its earliest proponents the team of
Le Go et al. (1969). Electrolytic recovery of metals
from dilute streams with consequent purication of
these streams can be much enhanced and rendered
economical by inert particle uidized bed electrolysis.
Inert glass beads, typically 0.6 mm in diameter, are
continuously uidized in the electrolytic cell by the
dilute solution, and the overow stream is recirculated
by a pump through the liquid distributor at the bottom
of the cell. Planar metallic mesh electrodes, anodes
interdispersed with cathodes, are vertically immersed
within the full depth of the uidized bed with mesh
apertures much greater than the bead size, so that
the uidized beads can move freely through and
around the electrodes. Since for electrolysis of dilute
aqueous solutions (<5 kg metallic ions/m
3
) ion trans-
port becomes mass transfer controlled, and since the
mass transfer enhancement factor between the
immersed electrodes and the liquid owing to the pre-
sence of the uidized beads can, as discussed in Sec. 12,
be as much as 7, ion transport is augmented accord-
ingly. To achieve maximum mass transfer enhance-
ment, the operation is carried out at bed expansions
beyond e
mf
of 50100%, i.e., at e - 0.60.7, which is in
accord with the ndings summarized in Sec. 12.
Reports by Lopez-Cacicedo (1981) and Boyanov et
al. (1988) recommend connecting several electrolytic
cells in series as a cascade.
Fluidized bed electrodes, since the initial reports by
Goodridge and coworkers (e.g., Backhurst et al., 1969;
Goodridge et al., 1971), have received much more
attention (see Goodridge and Wright, 1983 and
Salas-Morales et al., 1997, for incomplete but useful
reference lists). A uidized bed electrode (FBE) con-
sists of a bed of electrically conducting (metallic or
metal-coated, depending on desired particle density,
which inuences uidization velocity) particles ui-
dized by electrolyte ow, to which DC current is fed
by one or more connecting rods or plates known as
current feeders, and which are often separated from
one or more immersed counterelectrodes by a dia-
phragm of porous or ionically conductive material
that is desirably long lasting. Provided the bed expan-
sion from the static condition is small, e.g., 525%,
some electrical contact is maintained between the par-
ticles and the current feeder and between the particles
themselves, so that the surfaces of the particles then act
as a large extension of the feeder surface, thereby
greatly increasing the current density based on the
counterelectrode surface area. Though the many trials
and proposed applications of FBEs include their use
for fuel cells, organic and inorganic synthesis, indus-
trial wastewater treatment, and electrowinning (or
electrodeposition) of metals, the qualied successes of
this technology have been mainly in the area of elec-
trowinning, especially from dilute solutions, with the
concomitant reduction of metallic components in the
solutions involved. Figure 13 is a schematic view of a
continuous uidized bed electrowinning cell, in the
cathode compartment of which metal particles are ui-
dized by the process stream or wastewater to be trea-
ted. Dimensions and congurations of actual cells
dier greatly, while the metals that have been subjected
to FBE deposition are also various (Van der Heiden et
al., 1978). In fact, as Salas-Morales et al. (1997) have
most recently rearmed, the intermittency of contact
of the FBE particles with each other and with the cur-
rent feeder render them periodically subject to chemi-
cal attack by a strongly acid electrolyte, even when the
bed expansion is small, so that the FBE has been in the
main unsatisfactory for electrowinning from concen-
trated electrolytes. These investigators have reviewed
the prior literature on, and have therefore recom-
mendedas a device which performs between the
extremes of a xed bed electrode and a FBEa
spouted bed electrode, in which the 98% of the parti-
cles that at any moment reside within the annulus as a
moving packed bed are in good electrical contact with
each other and with the current feeder. A similar
recommendation has been made by Hadz ismajlovic et
al. (1996) and by Dweik et al. (1996).
Liquid-uidized bed heat exchangers (FBHX), in
addition to enhancing the clean wall-to-liquid heat
transfer coecient as discussed in Sec. 12, have the
additional important attribute of vigorously combat-
ing scaling and other types of fouling of the heat trans-
fer surface without the use of chemical additives. They
do so because the bed particles incessantly scour the
surface Kim and Lee (1997), act as alternative deposi-
tion sites for whatever precipitation does occur, and
scour each other thoroughly so that, for example, all
the precipitated calcium sulfate from saline water
leaves with the exit liquor and can be ltered out
(Hatch et al., 1966; Meijer et al., 1980). Although
tests have been performed with the liquid-uidized
bed on the unbaed shell side of a shell-and-tube
heat exchanger, both with the tubes horizontal and
with the tubes vertical (Cole and Allen, 1978), prefer-
ence has in most practical cases been given to locating
the uidized bed inside the tubes, which must therefore
be vertically oriented with a single upward pass
(Klaren and Bailie, 1988) and with proper care taken
to design for even distribution of both the liquid and
the uidized bed particles among the tubes
(Rautenbach and Kollbach, 1986). The main required
modications to a conventional shell-and-tube exchan-
ger are a larger inlet chamber with a distribution sys-
tem and a larger outlet chamber to act as freeboard for
separation of the liquid from the top of the uidized
bed, and these can sometimes be retrotted to a con-
ventional exchanger (Kollbach et al., 1987; Klaren and
Sullivan, 1999). The particles can be kept in a station-
ary uidized condition or they can be circulated by
means of an internal downcomer (Klaren and Bailie,
1988) or externally (Klaren and Sullivan, 1999). The
Figure 13 Schematic of a uidized bed electrowinning cell.
(From Van der Heiden et al., 1978.)
development of a multistage ash/uidized bed
exchanger (MSF/FBE), in which the uidized brine
feed acts to condense the stagewise ashed steam,
resulted in lower heat consumption, greater exibility
with respect to brine loads and temperatures, and
reduction in ash chamber volume by a factor of 6
relative to a conventional MSF exchanger (Veenman,
1976, 1977). In general, as summarized by Kollbach et
al. (1987) and others, FBHX units will usually main-
tain either totally clean surfaces or reduce fouling suf-
ciently that operation can be continued without
cleaning (Mu ller-Steinhagen et al., 1994); and wall-
to-bed heat transfer coecients are enhanced by fac-
tors of 2 to 7 at supercial velocities below 0.5 m/s,
maximum enhancement commonly occurring at e -
0.7 (Klaren and Halberg, 1980; Rautenbach and
Kollbach, 1986). FBHX technology has made inroads
into seawater desalination, geothermal energy utiliza-
tion, wastewater evaporation, and pulp and paper
production, and it has been proposed for lube oil
dewaxing (Kollbach et al., 1987; Klaren and Bailie,
1989; Klaren and Sullivan, 1999).
Thermal energy storage by encapsulating phase
change material into hollow spheres to be thermally
cycled by water uidization has been demonstrated
experimentally (Sozen et al., 1988). Specically, 96%
Glaubers salt (Na
2
SO
4
10H
2
O), which undergoes an
endothermic phase change to anhydrous Na
2
SO
4
(plus H
2
O) at 32.48C, and can be reversibly regener-
ated exothermically at the same temperature, was
injected along with 4% borax (a nucleation catalyst)
into thousands of 25 mm o.d. hollow polypropylene
spheres. These spheres were then cyclically uidized
at U/U
mf
= 1.22.6 in a 0.34 m i.d. column by hot
(inlet temperature - 398C) and cold (inlet temperature
-158C) water, each for intervals exceeding 1 hour.
Good heat transfer resulted because of the large cap-
sulewater surface area engaged, but more impor-
tantly, an unchanging heat storage eciency of about
60% was obtained, even after 96 cycles. Although heat
recovery eciencies up to 83% for stoichiometric
Glaubers salt were subsequently obtained using the
same capsules in a rotating drum (Sozen et al., 1988),
the greater simplicity, higher heat transfer surface area
per unit volume, and lower costs associated with
uidization continue to endow this technique with
some advantages.
Fluidized bed bioreactors have received considerable
attention during the past three decades, and the litera-
ture on this subject has mushroomed. Much of the
eort in this respect has been devoted to wastewater
treatment, which involves biodegradation of waste che-
micals (e.g., organics, ammonia, nitrates), but there is
now also a considerable literature on uidized bed fer-
mentation, the object of which is biosynthesis of useful
products (e.g., alcohol). A broad collection of papers
with useful discussions on various aspects of uidized
bed biological treatment of potable water and waste-
water has been edited by Cooper and Atkinson (1981).
An extensive review of both aerobic wastewater treat-
ment and fermentation in uidized beds has been pro-
vided by Fan (1989). There is an excellent update
mainly on wastewater treatment by Wright and
Raper (1996) and a signicant textbook entry by
Grady Jr et al. (1999). Anaerobic bioreactors, though
their uid feed is a liquid, nevertheless commonly pro-
duce in situ a gaseous product (Parkin and Speece,
1984), so that, like aerobic reactors in which air enters
with the liquid feed, they are actually three-phase
systems, but they can be treated as two-phase liquid-
uidized beds when the gaseous product is hydrodyna-
mically negligible. What most characterizes uidized-
bed bioreactors is their biocatalytic use of immobilized
enzymes or microbial cells, which are attached to the
surface of biologically inert nonporous particles, or
entrapped within the matrices of porous particles or
gels, or encapsulated within a semipermeable barrier
such as a membrane, or self-aggregated ocs (Karel
et al., 1985). The rst of these four immobilization
methods, referred to as the attached growth technique
for microbial cells, and involving growth of a biolm
(layers of cells and excreted slime) on the surface of
each support or carrier particle during the course of
the uidization, is the one most used in uidized-bed
wastewater treatment. The combination of inert carrier
particle (or biomass support particle) and attached
biolm is usually referred to as a bioparticle. The
carrier particles can be inorganic, e.g., sand, glass,
alumina, and many others, or organic, e.g., activated
carbon, coal, polyethylene, and many others; they are
often spherical or cylindrical and are usually though
not always moderately narrow cuts of sizes which for
dierent applications have varied from 0.1 to 6 mm.
For wastewater treatment, the most commonly used
biocarriers are sand and activated carbon, the latter
of which can simultaneously act as an adsorbent for
organic contaminants (Andrews and Tien, 1981). The
growth of biolm on the carrier particles changes their
size, eective density, and surface propertiesand
thereby their uidization characteristics, such as U
mf
,
U
0
, and n. For spherical carrier particles, some inves-
tigators nd that the values of these parameters given
by the equations in Sec. 4 apply reasonably well to the
resulting bioparticles, if the total equivolume sphere
diameter and weighted density of the latter at any
given time are used in the equations, while others
nd that empirical modications must be made to
the accepted equations for their prediction assuming
rigid spheres. Since the biocoated particles may be
neither smooth nor rigid, their drag coecients are
then higher than for smooth rigid spheres, so that
U
mf
and U
0
are correspondingly lower, i.e., bed expan-
sion for a given value of U is higher; and the measured
values of n are considerably higher (Thomas and
Yates, 1985). Empirical determination of U
mf
, U
0
(or
U
e
), and n as a function of bioparticle size is recom-
mended in the event that knowledge of bed expansion
characteristics is crucial to the bioreactor design. In the
case of bioocs or of particle matrices that retain some
of their porosity after cell entrapment (Atkinson et al.,
1979), it has been shown theoretically for permeable
spheres in the Stokes regime (Re
0
< 0.2; Neale et al.,
1973) and experimentally both in the Stokes regime
(Matsumoto and Suganuma, 1977) and at higher
Reynolds numbers (Masliyah and Polikar, 1980;
Webb et al., 1983) that the drag on such particles is
lower than for impermeable spheres of the same size
and bulk density, U
0
correspondingly higher, and the
bed expansion for a given supercial velocity corre-
spondingly lower. Fluidized bed bioreactors, schema-
tized in Fig. 14, come in many congurations,
including conventional upward uidization with con-
stant cross section, inverse (e.g., Nikolov and
Karamanev, 1987), tapered (e.g., Scott et al., 1978;
Allen et al., 1979), zig-zag (Nakamura et al., 1979),
baed (Parkin and Speece, 1984), with internal circu-
lation (e.g., via draft tube), and with external circula-
tion (Zhu et al., 1999). Wright and Raper (1996) list
twelve advantages of uidized bed attached growth
bioreactors as compared to more traditional suspended
growth bioreactors in which the microorganisms are
not anchored or immobilized. Of greater relevance
are the advantages that liquid-uidized beds have
over other attached growth bioreactors (e.g., xed
beds). These include (Allen et al., 1979; Grady Jr et
al., 2000) freedom from plugging, easy passage of inso-
luble foreign material or unwanted microorganisms,
lower pressure drop, superior mass and heat transfer
characteristics, better control of biolm thickness,
larger surface areas for biolm development, and
easy circulation or removal of bioparticles for excess
biomass separation (Fig. 14). Separation of the excess
biomass is commonly eected outside the reactor
either hydrodynamically or mechanically (Cooper et
al., 1981), or inside the reactor by scouring with air
fed exclusively to the draft tube (Nikolov and
Karamamev, 1987; Wright and Raper, 1996). The
overgrowth of biolms can lead to upward elutriation
of bioparticles from a conventional uidized bed or
downward elutriation from an inverse uidized bed,
Figure 14 Schematic diagram of a uidized bed bioreactor. (After Grady Jr et al., 1999.)
a problem that is sometimes dealt with by the use of a
retaining grid at the top of the conventional bed, giving
rise to semiuidization and clogging (Fan, 1989), but
is more adeptly handled by having some type of
expanded freeboard section at the top of the conven-
tionally operated column (Wright and Raper, 1996) or
at the bottom of the inverse bed (Nikolov and
Karamanev, 1987). The elutriation problem is much
attenuated or even eliminated by use of a tapered
uidized bed (Scott et al., 1978).
ACKNOWLEDGMENTS
Thanks are due to John Grace, Arturo Macchi, Dusko
Posarac, and Yunbi Zhang for help in locating refer-
ences, and to the Natural Sciences and Engineering
Research Council of Canada for continuing research
support.
NOMENCLATURE
a = exponent in Eq. (44), dimensionless
a. b. c = exponents in Eq. (167), dimensionless
a
i
= constant for particle species i in Eq. (89),
m s
1
a
/
= coecient in Eq. (81), m
(2m3),m
s
1
a
//
= constant in Eq. (176), dimensionless
A = cross-sectional area of column, m
2
A
c
= portion of submerged surface not aected
by particle contact, m
2
A
p
= portion of submerged surface aected by
particle contact, m
2
Ar = Archimedes number = d
3
p
(r
p
r)rg,m
2
,
dimensionless
b = coecient in Eq. (108), m
(m3),m
kg
1,m
s
1
b
i
= constant for particle species i in Eq. (89),
s
1
b
/
= exponent in Eq. (168), dimensionless
b
//
= exponent in Eq. (176), dimensionless
B = buoyant force on single particle in swarm
of particles, N
B
1
= parameter dened by Eq. (29b),
dimensionless
B
2
= parameter dened by Eq. (29c),
dimensionless
B
/
= surface mass ux/mainstream mass ux,
dimensionless
c = volumetric particle concentration of
monodisperse solids, m
3
m
3
c
i
= local or overall volumetric concentration
in liquid of particle species i, m
3
m
3
c
i0
= local volumetric concentration in liquid of
particle species i at z = 0, m
3
m
3
c
m
= coecient in Eq. (156), dimensionless
c
p
= specic heat capacity of liquid, J m
1
K
1
c
pp
= specic heat capacity of solid particles,
J m K
1
c
t
= local or overall volumetric concentration
in liquid of particle species i, m
3
m
3
c
+
= saturation concentration in Fig. 12, wt%
c
+
c = bulk liquid undersaturation for crystal
dissolution points in Fig. 12, wt%
c c
+
= bulk liquid supersaturation for crystal
growth points in Fig. 12, wt%
c
/
2
= volumetric concentration of smaller
particles in pseudouid through which
larger particles move, m
3
m
3
C
A
= bulk concentration of solute A in liquid
solution at given bed level, kg m
3
C
AS
= concentration of solute A in liquid
solution at particle surfaces, kg m
3
C
A1
= value of C
A
at z = 0, kg m
3
C
A2
= value of C
A
at z = L, kg m
3
(C
A
)
1.m.
= logarithmic mean of C
AS
C
A1
and C
AS
C
A2
, kg m
3
C
D
= drag coecient of particle in bed uidized
at supercial velocity U, dimensionless
C
DS
= drag coecient of isolated particle at
velocity U, dimensionless
C
D0
= drag coecient of isolated particle at
velocity U
0
, dimensionless
C
1
, C
2
= constants in Eq. (17), dimensionless
d = diameter of spherical particle, mm or m
d
avg
= average diameter of spherical particles as
given by Eq. (136), mm or m
d
eff
= eective particle diameter = d
p
K
1,3
, mm
or m
d
i
= spherical particle diameter of species i
particles, mm or m
d
p
= diameter of sphere having same volume as
particle, mm or m
d
pi
= equivolume sphere diameter of species i
particles, mm or m
dd
p
= local mass average or mass median
particle diameter, mm or m
d
s
= diameter of sphere having same surface
area as particle, mm or m
d
sv
= diameter of sphere having same surface-to-
volume ratio as particle, mm or m
D
a
= axial dispersion coecient of liquid,
cm s
1
or m s
1
D
A
= molecular diusivity of solute A in liquid,
cm
2
s
1
or m
2
s
1
D
c
= column diameter, m
D
h
= hydraulic diameter = 4 cross-sectional
area/wetted perimeter, m
D
i
= axial dispersion coecient of particle
species i with respect to particle species j,
cm
2
s
1
or m
2
s
1
D
ia
= axial self-dispersivity of particle species i,
cm
2
s
1
or m
2
s
1
D
ir
= radial self-dispersivity of particle species i,
cm
2
s
1
or m
2
s
1
D
L
= axial dispersivity of larger particles with
respect to smaller particles, cm
2
s
1
or
m
2
s
1
D
r
= radial dispersion coecient of liquid,
cm
2
s
1
or m
2
s
1
D
S
= axial dispersivity of smaller particles with
respect to larger particles, cm
2
s
1
or
m
2
s
1
e = base of natural logarithms - 2.718,
dimensionless
e
d
= rate of energy dissipation per unit mass of
liquid, m
2
s
3
f = collision frequency of particles, s
1
f (d
p
) = function of particle diameter that yields
voidage, dimensionless
f (e) = function of voidage dened by Eq. (49),
dimensionless
f
/
(e) = function of voidage that yields slip
velocity, m s
1
F = fractional free area of perforated plate
distributor, dimensionless
F
D
= drag force on single particle in particle
swarm, N
F
DS
= drag force on isolated spherical particle at
velocity U, N
F
D0
= drag force on isolated spherical particle at
velocity U
0
, N
g = acceleration of gravity, m s
2
Ga = Galileo number = d
3
p
r
2
g,m
2
, dimensionless
h = heat transfer coecient between
submerged surface and liquid in uidized
bed, W m
2
K
1
h
max
= maximum value of h with respect to U or
e, W m
2
K
1
h
p
= heat transfer coecient between particles
and liquid in uidized bed, W m
2
K
1
H = distance between pressure taps in Eqs. (9)
and (10), or thickness of transition zone in
Eqs. (89)(91), m
j
m
= ChiltonColburn mass-transfer
factor = k
m
,USc
2,3
, dimensionless
k = wall-eect factor U
e
/U
0
, dimensionless
k
i
= wall-eect factor for particle species i
alone, dimensionless
k
m
= mass transfer coecient between
bed, m s
1
k
p
= mass transfer coecient between particles
and liquid in uidized bed, m s
1
K = volume of particle and immobilized
liquid/volume of particle, dimensionless
K
/
K
/
0
L = uidized bed depth, m
L
mf
= uidized bed depth at minimum
uidization, m
m = regime-dependent exponent in Eqs. (81),
(108), and (156)
m = 2 slope of log C
D0
vs. log Re
0
at Re
0
of
monodisperse solidsliquid system under
investigation, dimensionless
m
A
= mass transfer rate of solute A between
particles and liquid, kg s
1
m
z
= mass of particles between distributor and
height z, kg
M = total mass of particles, kg
M
i
= mass of particle species i, kg
n = RichardsonZaki expansion index dened
by Eq. (32), dimensionless
n
i
= RichardsonZaki expansion index for
particle species i alone, dimensionless
n
p
= number of particles in contact with heat
transfer surface at any instant
n
/
= expansion index exceeding n for e > e
c
,
dimensionless
n
//
= expansion index less than n for e > e
c
,
dimensionless
N = total number of dierent particle species
N
p
= total number of uidized particles
Nu = Nusselt number for heat transfer between
bed = hd
p
,z, dimensionless
Nu
p
= Nusselt number for heat transfer between
particles and liquid in uidized
bed = h
p
d
p
,z, dimensionless
p = pressure drop across bed due to liquid
motion, Pa
p
d
= pressure drop across distributor, Pa
p
f
= frictional pressure drop across bed, Pa
dp
f
,dz = frictional pressure gradient, Pa m
1
P
1
= total pressure immediately above
distributor, Pa
P
2
= total pressure at plane above bed, Pa
P = total pressure drop P
1
P
2
, Pa
P,z = total pressure gradient, Pa m
1
Pe
a
= axial Peclet number of liquid = d
p
U,eD
a
,
dimensionless
Pe
i
= Peclet number of single species i particle in
bed of species j = LU
pi0
,D
i
, dimensionless
Pe
L
= Peclet number of single larger particle in
bed of smaller particles = LU
pL0
,D
L
,
dimensionless
Pe
r
= radial Peclet number of liquid = d
p
U,eD
r
Pe
S
= Peclet number of single smaller particle in
bed of larger particles = LU
pS0
,D
S
,
dimensionless
Pr = Prandtl number = c
p
m,z, dimensionless
Q = volumetric ow rate of liquid, m
3
s
1
R
2
= coecient of determination, dimensionless
Re = particle Reynolds number = d
p
Ur,m,
dimensionless
Re
mf
= particle Reynolds number at minimum
uidization = d
p
U
mf
r,m, dimensionless
Re
0
= free-settling terminal particle Reynolds
number = d
p
U
0
r,m, dimensionless
S = total surface area of particles, m
2
S
v
= surface area of particles/volume of bed,
m
2
m
3
Sc = Schmidt number = m/rD
A
, dimensionless
Sh = Sherwood number for mass transfer
between submerged surface and liquid in
uidized bed = k
m
d
p
,D
A
, dimensionless
Sh
p
= Sherwood number for mass transfer
between particles and liquid in uidized
bed = k
p
d
p
,D
A
, dimensionless
t = time interval, s
T = bulk temperature, K
T
s
= particle surface temperature, K
U = supercial liquid velocity, m s
1
U
amf
= apparent minimum uidization velocity of
multidisperse solids, m s
1
U
A
. U
B
.
U
C
. U
D
= values of U shown in Figure 8, m s
1
U
bf
= beginning uidization velocity of multi-
disperse solids, m s
1
U
e
= value of U when linear plot of log U vs.
log e at e - e
c
is extrapolated to e = 1,
m s
1
U
ei
= value of U
e
for particle species i alone,
m s
1
U
/
e
= value of U when linear plot of log U vs.
log e at e > e
c
with slope n
//
is
extrapolated to e = 1, m s
1
U
E
= dynamic velocity of elastic waves, m s
1
U
i
= supercial velocity required to uidize
monodisperse particle species i to same
voidage as exists in multispecies system,
m s
1
U
mf
= minimum uidization velocity (supercial),
m s
1
U
mfd
= maximum velocity of full deuidization of
tapered bed, m s
1
U
mff
= minimum velocity of full uidization of
tapered bed, m s
1
U
mpd
= maximum velocity of partial deuidization
of tapered bed, m s
1
U
mpf
= minimum velocity of partial uidization of
tapered bed, m s
1
U
p
= supercial velocity of solids, m s
1
U
phS
= segregation velocity of denser smaller
particles through less dense larger
particles, m s
1
U
pi
= segregation velocity of species i particles
through species j, m s
1
U
p.i
= supercial velocity of species i particles,
m s
1
U
pi0
= segregation velocity of single species i
particle through swarm of species j, m s
1
U
pL0
= segregation velocity of single larger
particle through swarm of smaller
particles, m s
1
U
pS0
= segregation velocity of single smaller
particle through swarm of larger particles,
m s
1
U
r
= relative or slip velocity between liquid and
solids, m s
1
U
ri
= relative velocity between liquid and
particle species i, m s
1
U
tf
= minimum velocity for total uidization of
multidisperse solids, m s
1
U
0
= terminal free settling velocity of particles,
m s
1
U
/
0
= terminal settling velocity of single particle
along axis of nite diameter cylindrical
column, m s
1
U
0i
= terminal free settling velocity of species i
particles, m s
1
U
/
01
= terminal free settling velocity of fastest
moving particles of a ternary particle
mixture according to model of Selim et al.
(1983a,b), m s
1
U
/
02
= terminal free settling velocity of
intermediate velocity particles of a ternary
particle mixture according to model of
Selim et al. (1983a,b), m s
1
U
//
01
= terminal free settling velocity of faster
moving particles of a binary particle
mixture according to pseudouid model,
m s
1
U
e
= kinematic velocity of continuity waves,
m s
1
v
i
= volume fraction of species i particles in
solids mixture, dimensionless
V = volume of representative particle in
uidized bed, m
3
V
i
= total volume of species i particles in
uidized bed, m
3
V
p
= constant rate particle sedimentation
velocity in downward direction, m s
1
W = width of two-dimensional bed, m
x
L
= c
L
/(1 e
L
), dimensionless
x
S
= c
S
/(1 e
S
), dimensionless
z = vertical distance above distributor, or
vertical distance above bottom of binary
transition zone, m
zz
L
= value of z when x
L
= 0.5, m
zz
S
= value of z when x
S
= 0.5, m
Z = z/L, dimensionless
ZZ
L
= zz
L
,L, dimensionless
ZZ
S
= zz
S
,L, dimensionless
Greek Letters
a = exponent in Eq. (62), dimensionless
a
/
,a
/
mf
= parameter given by Eq. (29a),
dimensionless
b = exponent in Eq. (49), dimensionless
g = sizing density
ratio = (r
BL
r
BS
),(r
p
r),
dimensionless
g
h
= sorting density
ratio = (r
Bh
r
B1
),(r
ph
r),
dimensionless
e = local or overall fractional void volume,
i.e., voidage, dimensionless
e
b
= settled bed voidage after sedimentation,
dimensionless
e
b1
= minimum bubbling voidage, dimensionless
e
b2
= maximum bubbling voidage, dimensionless
e
c
= critical bed voidage at which slope of
log U vs. log e changes, dimensionless
e
ei
= eective voidage associated with particle
species i, dimensionless
e
i
= voidage when particle species i is uidized
alone at same supercial liquid velocity as
for multispecies system, dimensionless
e
max
= voidage at which various hydrodynamic
and hydrodynamically related phenomena
maximize, dimensionless
e
mf
= voidage at minimum uidization,
dimensionless
y = included angle of conical or tapered bed,
degrees
k = permeability of porous medium, m
2
l = thermal conductivity of liquid, W m
1
K
1
l
B
= parameter dened by Eq. (29d),
dimensionless
l
eddy
= eddy conductivity of heat, W m
1
K
1
l
p
= thermal conductivity of particles,
W m
1
K
1
m = liquid viscosity, Pa s
m
2
= pseudouid viscosity for determining
motion of larger particles through
suspension of smaller particles, Pa s
r = liquid density, kg m
3
r
B
= bulk density of liquid-uidized bed,
kg m
3
r
BA
, r
BB
,
r
BD
= bulk densities, (r
B
)
max
, of bottom layer at
points A, B, and D, respectively, in Fig. 8,
kg m
3
(r
B
)
max
= maximum bulk density of binary particle
mixture over range of possible bed
inversion velocities, i.e., bulk density of
bottom layer, kg m
3
r
e
= eective particle
density = [r
p
r(K1)],K, kg m
3
r
p
= particle density, kg m
3
rr
p
= mean density of heterogeneous particle,
kg m
3
r
pi
= density of species i particle, kg m
3
r
2
= pseudouid density of binary particle
system including only the slower moving
particles, kg m
3
r
2,3
= pseudouid density of ternary particle
system excluding the fastest moving
particles, kg m
3
r
3
= pseudouid density of ternary particle
system including only the slowest moving
particles, kg m
3
= summation
f = particle sphericity, dimensionless
f
A
= fraction of total particle surface area that
is mass-transfer active, dimensionless
Subscripts
h = heavier particles of monosize binary
mixture
hS = heavier (i.e., denser) and smaller particles
of binary mixture
i = particle species i, which is representative of
all other particle species
j = particle species j
l = lighter particles of monosize binary
mixture
lL = lighter (i.e., less dense) and larger particles
of binary mixture
L = larger particles of xed density binary
mixture
mf = minimum uidization
S = smaller particles of xed density binary
mixture
1. 2. 3 = particle species 1, 2, and 3, respectively
REFERENCES
Adler IL, Happel J. The uidization of smooth spheres in
liquid media. Chem Eng Prog Symp Ser 58(38):98105,
1962.
Agarwal GP, Hudson JL, Jackson R. Fluid mechanical
description of uidized beds. Experimental investigation
of convective instabilities in bounded beds. Ind Eng Chem
Fundam 19:5966, 1980.
Agricola G. De Re Metallica. Translated from the rst Latin
edition of 1556 by H. C. Hoover and L. H. Hoover. New
York: Dover, 1950, pp. 310311.
Al-Dibouni MR, Garside J. Particle mixing and classication
in liquid uidised beds. Trans IChemE 57:94103, 1979.
Allen BR, Coughlin RW, Charles M. Fluidized-bed enzyme
reactors. Ann NY Acad Sci 326:105117, 1979.
Al-Naafa MA, Selim MS. Sedimentation of polydisperse
concentrated suspensions. Can J Chem Eng 67:253264,
1989.
Al-Naafa MA, Selim SM. Sedimentation of monodisperse
and bidisperse hard-sphere colloidal suspensions. AIChE
J 38:16181630, 1992.
Amitharajah A. Optimum backwashing of sand lters. J
Environ Engng Div ASCE 104:917932, 1978.
Anderson TB, Jackson R. Fluid mechanical description of
uidized beds: stability of the state of uniform uidiza-
tion. Ind Eng Chem Fundam 7:1221, 1968.
Anderson TB, Jackson R. A uid mechanical description of
uidized beds: comparison of theory and experiment. Ind
Eng Chem Fundam 8:137144, 1969.
Andrews GF, Tien C. Bacterial lm growth in [sic] adsorbent
surfaces. AIChE J 27:396403, 1981.
Arters DC, Fan L-S. Solidliquid mass transfer in a gas-
liquid-solid uidized bed. Chem Eng Sci 41:107115,
1986.
Asif M. Modeling of multisolid liquid uidized bed. Chem
Eng Technol 10:485490, 1997.
Asif M. Generalized Richardson-Zaki correlation for liquid
uidization of binary solids. Chem Eng Technol 21:7782,
1998.
Asif M, Petersen JN. Particle dispersion in a binary solid-
liquid uidized bed. AIChE J 39:14651471, 1993.
Asif M, Kalogerakis N, Behie LA. Distributor eects in
liquid uidized beds of low-density particles. AIChE J
37:18251832, 1991.
Asif M, Kalogerakis N, Behie LA. Hydrodynamics of liquid
uidized beds including the distributor region. Chem Eng
Sci 47:41554166, 1992.
Astarita G. Forces acting on particles in a uidised bed.
Chem Eng Sci 48:34383440, 1993.
Atkinson B, Black GM, Lewis PJS, Pinches A. Biological
particles of given size, shape, and density for use in
biological reactors. Biotechnol Bioeng 21:193200, 1979.
Babu SP, Shah B, Talwalkar A. Fluidization correlations for
coal gasication materialsminimum uidization velo-
city and uidized bed expansion ratio. AIChE Symp Ser
74(176):176186, 1978.
Backhurst JR, Coulson JM, Goodridge F, Plimley RE,
Fleischmann M. A preliminary investigation of uidized
bed electrodes. J Electrochem Soc 116:16001607, 1969.
Bamforth AW. Industrial Crystallization. London: Leonard
Hill, 1965, pp 4955.
Barnea E, Mednick RL. Correlation for minimum uid-
isation velocity. Trans Instn Chem Engrs 53:278281,
1975.
Barnea F, Mizrahi J. A generalized approach to the uid
dynamics of particulate systems. Part I. General correla-
tion for uidization and sedimentation in solid multipar-
ticle systems. Chem Eng J 5:171189, 1973.
Bascoul A, Delmas H, Couderc JP. Caracte ristiques hydro-
dynamiques de la uidisation liquidesolide: inuence du
distributeur. Chem Eng J 37:1124, 1988.
Batchelor GK. Sedimentation in a dilute dispersion of
spheres. J Fluid Mech 52(2):245268, 1972.
Batchelor GK. Sedimentation in a dilute polydisperse system
of interacting spheres. Part 1. General theory. J Fluid
Mech 119:379408, 1982.
Batchelor GK. A new theory of the instability of a uniform
uidized bed. J Fluid Mech 193:75110, 1988.
Batchelor GK, Wen CS. Sedimentation in a dilute polydis-
perse system of interacting spheres. Part 2. Numerical
results. J Fluid Mech 124:495528, 1982.
Beek WI. Mass transfer in uidized beds. In: Davidson JF,
Harrison D, eds. Fluidization. London: Academic Press,
1971, pp 431470.
Ben a J, Ilavsky J, Kossaczky E, Neuz il L. Changes of the
ow character in a uidized bed. Collection Czechoslov
Chem Commun 28:293309, 1963.
Bin AK. Minimum uidization velocity at elevated tempera-
ture and pressure. Can J Chem Eng 64:854857, 1986.
Bird RB, Stewart WE, Lightfoot EH. Transport Phenomena.
New York: John Wiley, 1960, p 663.
Bordet J, Coeuret F, Le Go P, Vergues F. E
tude par con-

ductance e lectrique des uctuations de porosite locale
dans les lits uidise s liquidesolide. Powder Technology
6:253261, 1972.
Bourgeois P, Grenier P. The ratio of terminal velocity to
minimum uidizing velocity for spherical particles. Can
J Chem Eng 46:325334, 1968.
Boyanov BS, Donaldson JD, Grimes SM. Removal of cop-
per and cadmium from metallurgical leach solutions by
uidised bed electrolysis. J Chem Technol Biotech
41:317328, 1988.
Bransom SH. Factors in the design of continuous crystal-
lizers. Brit Chem Eng 5:838844, 1960.
Bremford DJ, Mu ller-Steinhagen H, Duy GG. Heat trans-
fer to black liquor in a uidised bed. In: Celata GP, Di
Marco P, Mariani A, eds. Proc 2d European Thermal-
Sciences and 14th UIT National Heat Transfer
Conference. Edizioni ETS, 1996, pp 15451551.
Briens CL, Del Pozo M, Chiu K, Wild G. Modeling of par-
ticle-liquid heat and mass transfer in multiphase systems
with the lm-penetration model. Chem Eng Sci
48:973979, 1993.
Briens LA, Briens CL, Margaritis A, Cooke SL, Bergougnou
MA. Characterization of channeling in multiphase sys-
tems. Application to a liquid uidized bed of angular
biobone particles. Powder Technology 91:19, 1997.
Brinkman HC. A calculation of the viscous force exerted by a
owing uid on a dense swarm of particles. Appl Sci Res
A1:2734, 1947.
Brody MA, Lumkins RJ. Performance of dual-media
lters1. Chem Eng Prog 73(4), 8386, 1977.
Bro tz W. Grundlagen der wirbelschichtverfahren. Chem
IngTechnik 24(2):6081, 1952.
Brown GG, Foust AS, Kata DL, Schneidewind R, White
RR, Wood WP, Brown GM, Brownell LE, Martin JJ,
Williams GB, Banchero JT, York JL. Unit Operations.
New York: John Wiley, 1950, pp 9194.
Brownell LE, Dombrowski HS, Dickey CA. Pressure drop
through porous media part IVnew data and revised
correlation. Chem Eng Prog 46:415422, 1950.
Bruinzeel C, Reman GH, van der Laan ETR. Eddy diusion
in particulately uidized beds: model experiments for the
design of a large-scale unit. In: Proc Symp on Interaction
Between Fluids and Particles. London: Instn Chem Engrs,
1962, pp 120126.
Buscall R, Goodwin JW, Ottewill RH, Tadros ThF.
The settling of particles through Newtonian and
non-Newtonian media. J Colloid Interface Sci 85:7886,
1982.
Cairns EJ, Prausnitz JM. Longitudinal mixing in uidization.
AIChE J 6:400405, 1960a.
Cairns EJ, Prausnitz JM. Macroscopic mixing in uidization.
AIChE J 6:554560, 1960b.
Calderbank PH. Gas absorption from bubbles. Chemical
Engineer (London) 45(211):CE209-CE253, 1967.
Carlos CR, Richardson JF. Solids movement in liquid
uidised bedsI. Particle velocity distribution. Chem
Eng Sci 23:813824, 1968a.
Carlos CR, Richardson JF. Solids movement in liquid
uidised bedsII. Measurement of axial mixing co-
ecients. Chem Eng Sci 23:825831, 1968b.
Carman PC. Fluid ow through granular beds. Trans Instn
Chem Engrs (London) 15:150166, 1937.
Casal J, Puigjaner L. Segregation and apparent minimum
uidization velocity in particulate uidization. Chem
Eng Commun 23:125136, 1983.
Chavarie C, Karamanev DG. Use of inverse uidization in
biolm reactors. Proc Internat Conf Bioreactor Fluid
Dynamics, BHRA, Cambridge, England, 1986, pp
181190.
Chen JJJ. Comments on Improved equation for the calcula-
tion of minimum uidization velocity. Ind Eng Chem
Res 26:633634, 1987.
Chen Z, Gibilaro LG, Foscolo PU. Two-dimensional
voidage waves in uidized beds. Ind Eng Chem Res
38:610620, 1999.
Chhun T, Couderc JP. High ux mass transfer in liquid
uidization. Chem Eng Sci 35:17071715, 1980.
Chianese A, Frances C, Di Berardino F, Bruno L. On the
behaviour of a liquid uidized bed of monosized sodium
perborate crystals. Chem Eng J 50:8794, 1992.
Chiba T. Bed contraction of liquid-uidised binary solid
particles at complete mixing. In: Yoshida K, Morooka
S, eds. Proc Asian Conf on Fluidized Bed and Three-
Phase Reactors. Tokyo: Tokyo University, 1988,
pp 385392.
Chilton TH, Colburn AP. Mass transfer (absorption) coe-
cientsprediction from data on heat transfer and uid
friction. Ind Eng Chem 26: 11831187, 1934.
Chitester DC, Kornosky RM, Fan L-S, Danko JP.
Eng Sci 39:253261, 1984.
Chong YS, Ratkowsky DA, Epstein N. Eect of particle
shape on hindered settling in creeping ow. Powder
Chung SF, Wen CY. Longitudinal dispersion of liquid ow-
ing through xed and uidized beds. AIChE J 14:857866,
1968.
Chyang CS, Huang WC. Characteristics of large particle
uidization. J Chin Inst Chem Engrs 19(2):8189, 1988.
Cleasby JL, Baumann AR. Backwash of granular lters used
in waste water ltration. Final Report on EPA Project R
802140, Engineering Research Institute, Iowa State
University, 1976.
Cleasby JL, Fan KS. Predicting uidization and expansion of
lter media. J Environ Engng Div ASCE 107:455471,
1981.
Cleasby JL, Lorence JL. Eectiveness of backwashing for
wastewater lters. J Environ Engng Div ASCE
104:749765, 1978.
Cleasby JL, Sejkora GD. Eect of media intermixing on dual
media ltration. J Environ Engng Div ASCE
101:503516, 1975.
Cleasby JL, Woods CF. Intermixing of dual media and
multimedia granular lters. J Am Water Works Assoc
67:197203, April 1975.
Cleasby JL, Stangl EW, Rice GA. Developments in back-
washing of granular lters. J Environ Engng Div ASCE
101:713727, 1975.
Clift R. An Occamist review of uidized bed modelling.
AIChE Symp Series 89(296):117, 1993.
Clift R, Grace JR, Weber ME. Bubbles, Drops, and
Particles. New York: Academic Press, 1978, p. 114.
Clift R, Seville JPK, Moore SC, Chavarie C. Comments on
buoyancy in uidized beds. Chem Eng Sci 42:191194,
1987.
Cole LT, Allen CA. Liquid-uidized-bed heat exchanger ow
distribution models. Proc Intersoc Energy Convers Eng
Conf 13:11291134, 1978.
Cooper PF, Atkinson B, eds. Biological Fluidised Bed
Treatment of Water and Wastewater. Chichester: Ellis
Horwood, 1981.
Cooper PF, Wheeldon DHV, Ingram-Tedd PE, Harrington
DW. Sand/biomass separation with production of a
concentrated sludge. In: Cooper PF, Atkinson B, eds.
Biological Fluidised Bed Treatment of Water and
Wastewater. Chichester: Ellis Horwood, 1981, pp 361367.
Cornish ARH. Note on minimum possible rate of heat
transfer from a sphere when other spheres are adjacent
to it. Trans Instn Chem Engrs 43:T332-T333, 1965.
Couderc J-P. Incipient uidization and particulate systems.
In: Davidson JF, Clift R, Harrison D, eds. Fluidization.
2d ed. London: Academic Press, 1985, pp 146.
Couderc JP, Gibert H, Angelino H. Transfert de matie` re par
diusion en uidisation liquide. Chem Eng Sci 27:1120,
1972.
Damronglerd S, Couderc JP, Angelino H. Mass transfer in
particulate uidisation. Trans Instn Chem Engrs
53:175180, 1975.
Darcy HPG. Les Fontaines Publiques de la Ville de Dijon.
Paris: Victor Dalmont, 1856.
Delachanal M. Apparatus and process for sorting solid par-
ticle material in a uidized liquid medium. U.S. Patent
3,075,643, Jan. 29, 1963.
Dhamarajah AH, Cleasby JL. Predicting the expansion beha-
vior of lter media. J. Am Water Works Assoc 78:6676,
December 1986.
Didwania AK, Homsy GM. Flow regimes and ow transi-
tions in liquid uidized beds. Internat J Multiphase Flow
7:563580, 1981.
Di Felice, R. Mixing in segregated, binary-solid liquid ui-
Di Felice R. The voidage function for uid-particle inter-
action systems. Int J Multiphase Flow 20:153159, 1994.
Di Felice R. Hydrodynamics of liquid uidisation. Chem
Eng Sci 50:12131245, 1995.
Di Felice R. A relationship for the wall eect on the settling
velocity of a sphere at any ow regime. Intern J
Multiphase Flow 22:527533, 1996.
Di Felice R. The applicability of the pseudo-uid model to
the settling velocity of a foreign particle in a suspension.
Chem Eng Sci 53:371375, 1998.
Di Felice R., Parodi E. Wall eects on the sedimentation
velocity of suspensions in viscous ow. AIChE J
42:927931, 1996.
Di Felice R, Gibilaro LG, Waldram SP, Foscolo SP. Mixing
and segregation in binary-solid liquid uidised bed. Chem
Eng Sci 42:639652, 1987.
Di Felice R, Gibilaro LG, Foscolo PU. On the inversion of
binary-solid liquid uidised beds. Chem Eng Sci
43:979981, 1988.
Di Felice R, Gibilaro LG, Rapagna` S, Foscolo, PU. Particle
mixing in a circulating liquid uidized bed. AIChE Symp
Series 85(270):3236, 1989.
Di Felice R, Foscolo PU, Gibilaro LG, Wallis GB, Carta R.
Expansion characteristics of tapered uidized beds.
AIChE J 37:16681672, 1991a.
Di Felice R, Foscolo PU, Gibilaro LG, Rapagna` S. The
interaction of particles with a uid-particle pseudo-uid.
Chem Eng Sci 46:18731877, 1991b.
Dixon DC. Comments on Hindered settling theories,
Dorgelo EAH, van der Meer AP, Wesselingh JA.
Measurement of the axial dispersion of particles in a
liquid uidized bed applying a random walk model.
Chem Eng Sci 40:21052111, 1985.
Dweik BM, Liu CC, Savinell RF. Hydrodynamic modelling
of the liquidsolid behaviour of the circulating particulate
bed electrode. J Appl Electrochem 26:10931102, 1996.
Eastwood, J, Matzen EJP, Young MJ, Epstein N. Random
loose porosity of packed beds. Brit Chem Eng 14:1542
1545, 1969.
Eisenklam P. Discussion. In: Drinkenburg AAH, ed. Proc
Internat Conf on Fluidization. Amsterdam: Netherlands
Univ. Press, 1967, pp 549551.
El-Kaissy MM, Homsy GM. Instability waves and the origin
of bubbles in uidized bedsPart 1: experiments. Internat
J Multiphase Flow 2:379395, 1976.
El-Temtamy SA, Epstein N. Eect of low velocity liquid
pulstions on some hydrodynamic characteristics of liquid
and gasliquid uidized beds. AIChE J 32:509512, 1986.
Emersleben O, Faxe n H. The motion of a rigid sphere along
the axis of a tube lled with viscous uid. Arkiv Mat
Astron Fysik 17(27):128, 1923.
Epstein N. Hydrodynamic particle volume and settled bed
volume. Can J Chem Eng 57:383, 1979.
Epstein N. Letter to the editor. AIChE J 38:637638, 1992.
Epstein N, LeClair BP. Liquid uidization of binary particle
mixturesII. Bed inversion. Chem Eng Sci 40:15171526,
1985.
Epstein N, Masliyah JH. Creeping ow through clusters of
spheroids and elliptical cylinders. Chem Eng J 3:169175,
1972.
Epstein N, Pruden BB. Liquid uidisation of binary particle
mixturesIII. Stratication by size and related topics.
Chem Eng Sci 54:401415, 1999.
Epstein N, LeClair BP, Pruden BB. Liquid uidization of
binary particle mixturesI. Overall bed expansion.
Chem Eng Sci 36:18031809, 1981.
Ergun S. Fluid ow through packed columns. Chem Eng
Prog 48:8994, 1952.
Evans GC, Gerald CF. Mass transfer from benzoic acid
granules to water in xed and uidized beds at low
Reynolds numbers. Chem Eng Prog 49:135140, 1953.
Fan L-S. GasLiquidSolid Fluidization Engineering.
Boston: Butterworths, 1989, pp 453588, 415420.
Fan L-S, Muroyama K, Chern S-H. Hydrodynamic charac-
teristics of inverse uidization in liquidsolid and gas
liquidsolid systems. Chem Eng J 24:143150, 1982.
Fan L-T, Yang Y-C, Wen C-Y. Mass transfer in semiui-
dized beds for solid-liquid system. AIChE J 6:482487,
1960.
Fan Z, Xuanyu Z, Lichang X. Fluidization of uniformly
sized spheres. In: Kwauk M, Kunii D, Jiansheng Z,
Hasatani M, eds. Fluidization 85. Science and
Technology Conference Papers. Second ChinaJapan
Symposium. Amsterdam: Elsevier, 1985, pp 283294.
Fidleris V, Whitmore RL. Experimental determination of the
wall eect for spheres falling axially in cylindrical vessels.
Brit J Appl Phys 12:490494, 1961.
Field CJ, Riley PJ. Tripos Part II Research Report,
Cambridge University, 1970.
Finkelstein E, Letan R, Elgin JC. Mechanics of vertical mov-
ing uidized systems with mixed particle sizes. AIChE J
17:867872, 1971.
Foscolo PU, Gibilaro LG. A fully predictive criterion for the
transition between particulate and aggregative uidiza-
tion. Chem Eng Sci 39:16671675, 1984.
Foscolo PU, Gibilaro LG. Fluid dynamic stability of ui-
dised suspensions: the particle bed model. Chem Eng Sci
42:14891500, 1987.
Foscolo PU, Gibilaro LG, Waldram SP. A unied model for
particulate expansion of uidised beds and ow in xed
porous media. Chem Eng Sci 38:12511260, 1983.
Foscolo PU, Di Felice R, Gibilaro LG. The pressure eld in
an unsteady-state uidized bed. AIChE J 35:19211926,
1989.
Fouda AE, Capes CE. Hydrodynamic particle volume and
uidized bed expansion. Can J Chem Eng 55:386391,
1977.
Fouda AE, Capes CE. Hydrodynamic particle volume and
uidized bed expansion. Can J Chem Eng 57:120121,
1979.
Frances C, Biscans B, Laguerie C. Modelling of a continuous
uidized-bed crystallizer. Chem Eng Sci 49:32693276,
1994.
Francis AW. Wall eect in falling ball method for viscosity.
Physics 4:403406, 1933.
Funamizu N, Takakuwa T. A minimal potential energy
model for predicting stratication pattern in binary and
ternary solidliquid uidized beds. Chem Eng Sci
51:341351, 1996.
Galvin KP, Pratten SJ. Application of uidization to obtain
washability data. Minerals Engineering 12:10511058,
1999.
Galvin KP, Pratten S, Nguyen-Tran-Lam G. Dierential set-
tling in a teeter bed separator. Presented at Third World
Congress on Particle Technology. Brighton, England: I
Chem E, 1998, Paper 228, pp 111.
Galvin KP, Pratten S, Nguyen-Tran-Lam G. A generalized
empirical description for particle slip velocities in liquid
Ganho R, Gibert H, Angelino H. Cine tique de ladsorption
du phenol en couche uidise e de charbon actif. Chem Eng
Sci 30:12311238, 1975.
Garnier A, Chavarie C, Andre G, Klvana D. The inverse
uidization airlift bioreactor, part I: hydrodynamic stu-
dies. Chem Eng Comm 98:3145, 1990.
Garside J, Al-Dibouni M. Behaviour of liquid uidized beds
containing a wide size distribution of solids. In: Angelino
H, Couderc JP, Gibert H, Laguerie C, eds. Fluidization
and Its Applications. Toulouse: Cepadues-E
ditions, 1974,
pp 5362.
Garside J, Al-Dibouni MR. Velocityvoidage relationships
for uidization and sedimentation in solid-liquid systems.
Garside J, Gaska G, Mullin JW. Crystal growth rate studies
with potassium sulphate in a uidized bed crystallizer. J
Cryst Growth 13/14:510516, 1972.
Gasparyan AM, Ikaryan NS. Hindered settling of particles
V. Izv Akad Nauk Arm SSR Ser Tekhn Nauk
15(4):5364, 1962.
Gaudin AM. Principles of Mineral Dressing. 1st ed. New
York: McGraw-Hill, 1939, p 205.
Gibilaro LG, Waldram SP, Foscolo PU. Authors reply to
comments by N. Epstein. Chem Eng Sci 39:18191820,
1984.
Gibilaro LG, Di Felice R, Waldram SP, Foscolo PU.
Generalized friction factor and drag coecient correla-
tions for uidparticle interactions. Chem Eng Sci
40:18171823, 1985a.
Gibilaro LG, Hossain I, Waldram SP. On the Kennedy and
Bretton model for mixing and segregation in liquid
uidized beds. Chem Eng Sci 40:23332338, 1985b.
Gibilaro LG, Di Felice R, Waldram SP, Foscolo PU. A
predictive model for the equilibrium composition and
inversion of binary-solid liquid uidised beds. Chem
Eng Sci 41:379387, 1986a.
Gibilaro LG, Hossain I, Foscolo PU. Aggregate behaviour
of liquid uidised beds. Can J Chem Eng 64:931938,
1986b.
Gibilaro LG, Di Felice R, Waldram SP, Foscolo PU.
Authors reply to Clift et al. Chem Eng Sci 42:194196,
1987.
Gibilaro LG, Di Felice R, Foscolo PU. Added mass eects in
uidized beds: application of the GeurstWallis analysis
of inertial coupling in two-phase ow. Chem Eng Sci
45:15611565, 1990.
Glasby J, Ridgway K. The crystallization of aspirin
from ethanol. J Pharm Pharmac 20(suppl.):94S103S,
1968.
Goodridge F, Wright AR. Porous ow-through and uidized
bed electrodes. In: Yeager E, Bockris JOM, Conway BE,
Sarangapani S, eds. Comprehensive Treatise of
Electrochemistry. Vol. 6. Electrodics: Transport. New
York: Plenum Press, 1983, pp 393440.
Goodridge F, Holden DI, Murray HD, Plimley RF.
Fluidized-bed electrodes. Trans Instn Chem Engrs
49:128141, 1971.
Grace JR. Fluidized bed hydrodynamics. In: Hetsroni G, ed.
Handbook of Multiphase Systems. New York: Hemi-
sphere and McGraw-Hill, 1982, p 86.
Grace JR. Agricola aground: characterization and interpre-
tation of uidization phenomena. AIChE Symp Series
88(289):116, 1992.
Grady Jr CPL, Daigger GT, Lim HC. Biological Wastewater
Treatment. New York: Marcel Dekker, 1999, pp 614615,
621622, 653654, 809840, 951, 956957, 971, 975978.
Grbavc ic ZB, Vukovic DV. Single-particle settling velocity
through liquid uidized beds. Powder Technology
66:293295, 1991.
Grbavc ic ZB, Garic RV, Hadjismalovic DzE, Jovanovic S,
Vukovic DV, Littman H, Morgan III MH. Variational
model for prediction of the uidparticle interphase
drag coecient and particulate expansion of uidized
and sedimenting beds. Powder Technology 68:199211,
1991.
Grimmet ES, Brown BP. A pulsed column for countercurrent
liquidsolids ows. Ind Eng Chem 54:2428, 1962.
Gru newald M, Schmidt-Traub H. Adsorption in multistage
uidized bed column. Chem Eng Technol 22:206209,
1999.
Hadz ismajlovic DE, Popov KI, Pavlovic MG. The visualiza-
tion of the electrochemical behaviour of metal particles in
spouted, uidized and packed beds. Powder Technology
86:145148, 1996.
Haid M. Correlations for the prediction of heat transfer to
liquidsolid uidized beds. Chemical Engineering and
Processing 36:143147, 1997.
Haid M, Martin H, Mu ller-Steinhagen H. Heat transfer to
liquidsolid uidized beds. Chemical Engineering and
Processing 33:211225, 1994.
Haider A, Levenspiel O. Drag coecient and terminal velo-
city of spherical and nonspherical particles. Powder
Ham JM, Thomas S, Guazzelli E, Homsy GM, Anselmet
M-C. An experimental study of the stability of liquid-
uidized beds. Internat J Multiphase Flow 16:171185,
1990.
Hancock RT. The teeter condition. Mining Magazine
(London) 55:9094, 1936.
Hancock RT. The law of motion of particles in a uid. Trans
Inst Min Engrs (London) 94:114121, 1937/38.
Handley D, Doraisamy A, Butcher KL, Franklin NL. A
study of the uid and particle mechanics in liquid-
uidized beds. Trans Instn Chem Engrs 44:T260-T273,
1966.
Hanratty TJ, Latinen G, Wilhelm RH. Turbulent diusion in
particulately uidized beds of particles. AIChE J 2:372
380, 1956.
Happel J. Viscous ow in multiparticle systems: slow motion
of uids relative to beds of spherical particles. AIChE J
4:197201, 1958.
Happel J, Brenner H. Viscous ow in multiparticle systems:
motion of spheres and a uid in a cylindrical tube. AIChE
J 3:506513, 1957.
Happel J, Brenner H. Low Reynolds Number
Hydrodynamics. Englewood Clis, NJ: Prentice-Hall,
1965.
Happel J, Epstein N. Viscous ow in multiparticle systems:
cubic assemblages of uniform spheres. Ind Eng Chem
46:11871194, 1954.
Hartman M, Trnka D, Havlin V. A relationship to estimate
the porosity in liquidsolid uidized beds. Chem Eng Sci
47:31623166, 1992.
Hasimoto H. On the periodic fundamental solutions of the
Stokes equations and their application to viscous ow
past a cubic array of spheres. J Fluid Mech 5:317328,
1959.
Hassett NJ. Flow patterns in particle beds. Nature 189:997
998, 1961a.
Hassett NJ. The mechanism of uidization. Brit Chem Engng
19:777780, 1961b.
Hatch LP, Weth GG, Wachtel SJ. Scale control in the high
temperature distillation of saline waters by means of
uidized-bed heat exchangers. Desalination 1:156164,
1966.
Hetzler R, Williams MC. Fluidized bed viscosity and expan-
sion correlated with glass forming liquid model. Ind Eng
Chem Fundam 8:668677, 1969.
Heywood H. Uniform and non-uniform motion of particles
in uids. In: Proc Symp on Interaction Between Fluids
and Particles. London: Instn Chem Engrs, 1962, pp 18.
Hiby JW. Particle forces in homogeneous beds. In:
Drinkenburg AAH, ed. Proc Internat Conf on
Fluidization. Amsterdam: Netherlands Univ. Press,
1967, pp 2130.
Himsley A, Farkas EJ. Applications of uidized bed ion
exchange techniques to extractive hydrometallurgy.
Presented at 27th Canadian Chemical Engineering
Conference. Calgary, Alberta, 1977, Paper 10-2, pp 112.
Hiquily N, Couderc JP, Angelino H. Analyse des conditions
de fonctionnement du distributeur a` billes. Powder
Homan RF, Lapidus L, Elgin JC. The mechanics of vertical
moving uidized systems: IV. Application to batch-
uidized systems with mixed particle sizes. AIChE J
6:321324, 1960.
Holman JP, Moore W, Wong VM. Particle-to-uid heat
transfer in water-uidized systems. Ind Eng Chem
Fundam 4:2131, 1965.
Howley MA, Glasser BJ. Hydrodynamics of a binary-part-
icle uidized-bed: segregation and mixing. In: Glicksman
LR. Fluidization and FluidParticle Systems, AIChE
1999 Annual Meeting, pp 152158.
Jackson R. The mechanics of uidized beds Part I: The
stability of the state of uniform uidisation. Trans Instn
Chem Engrs 41:1321, 1963.
Jackson R. Hydrodynamic stability of uid-particle systems.
In: Davidson JF, Clift R, Harrison D, eds. Fluidization.
2d ed. London: Academic Press, 1985, pp 4772.
Jackson R. Progress toward a mechanics of dense suspen-
sions of solid particles. AIChE Symp Series 90(301):1
30, 1994.
Jamialahmadi M, Mu ller-Steinhagen H. Hydrodynamics and
heat transfer of liquid uidized bed systems. Chem Eng
Communications 179:3579, 2000.
Jamialahmadi M, Malayeri MR, Mu ller-Steinhagen H.
Prediction of heat transfer to liquidsolid uidized beds.
Can J Chem Eng 73:444455, 1995.
Jamialahmadi M, Malayeri MR, Mu ller-Steinhagen H. A
unied correlation for the prediction of heat transfer coef-
cients in liquid/solid uidized systems. Trans ASME J
Heat Transfer 118:952959, 1996.
Jamialahmadi M, Malayeri MR, Mu ller-Steinhagen H.
Prediction of optimum operating conditions of liquid
uidized bed systems. Can J Chem Eng 75:327332,
1997.
Jean R-H, Fan L-S. On the criteria of solids layer inversion
in a liquidsolid uidized bed containing a binary mixture
of particles. Chem Eng Sci 41:28112821, 1986.
Jean R-H, Fan L-S. A uid mechanic-based model for
sedimentation and uidization at low Reynolds number.
Chem Eng Sci 44:353362, 1989.
Jean R-H, Fan L-S. On the model equations of Gibilaro and
Foscolo with corrected buoyancy force. Powder
Technology 72:201205, 1992.
Jean R-H, Fan L-S. Multiphase ow: liquid/solid uidized
bed systems. In: Johnson RW, ed. Handbook of Fluid
Dynamics. Boca Raton, FL: CRC Press, 1998,
pp 19-119-25.
Jira-Arune N, Laguerie C. Croissance des cristaux de
chlorure de potassium en couche uidise e: comparaison
avec la dissolution. Chem Eng J 18:4757, 1979.
Jolls KR, Hanratty TJ. Transition to turbulence for ow
through a dumped bed of spheres. Chem Eng Sci
21:11851190, 1966.
Joshi JB. Solid-liquid uidised beds: some design aspects.
Chem Eng Res Des 61:143161, 1983.
Jottrand R. An experimental study of the mechanism of
uidization. J Appl Chem (suppl. issue 1) 2:1726, 1952.
Jottrand R, Grunchard F. Transfert de matie` re dans les lits
uidise s liquides. In: Symp on Interaction Between Fluids
and Particles. London: Instn Chem Engrs, 1962,
pp 211216.
Juma AKA, Richardson JF. Particle segregation in liquid
solid uidised beds. Chem Eng Sci 34:137143, 1979.
Juma AKA, Richardson JF. Segregation and mixing in
liquid uidized beds. Chem. Eng. Sci. 38:955967,
1983.
Kang Y, Kim SD. Radial dispersion characteristics of two-
and three-phase uidized beds. Ind Eng Chem Process
Des Dev 25: 717722, 1986.
Kang Y, Kim SD. Solid ow transition in liquid and three-
phase uidized beds. Particulate Science and Technology
6:133144, 1988.
Kang Y, Nah JB, Min BT, Kim SD. Dispersion and
uctuation of uidized particles in a liquidsolid uidized
bed. Chem Eng Comm 97:197208, 1990.
Karamanev DG, Nikolov LN. Inuence of some physico-
chemical parameters on bacterial activity of biolm:
ferrous iron oxidation by thiobacillus ferrooxidans.
Biotechology and Bioengineering 31:295299, 1988.
Karamanev DG, Nikolov LN. Free rising spheres do not
obey Newtons law for free settling. AIChE J
38:18431846, 1992a.
Karamanev DG, Nikolov LN. Bed expansion of liquidsolid
inverse uidization. AIChE J 38:19161922, 1992b.
Karamanev DG, Nikolov LN. Application of inverse
uidization in wastewater treatment: from laboratory
to full-scale bioreactors. Environmental Progress
15(3):194196, 1996.
Karel SF, Libicki SB, Robertson CR. The immobilization of
whole cells: engineering principles. Chem Eng Sci
40:13211354, 1985.
Kato Y, Uchida K, Kago T, Morooka S. Liquid hold-up and
heat transfer coecient between bed and wall in liquid
solid and gasliquidsolid uidized beds. Powder
Technology 28:173179, 1981.
Kaye BH, Boardman RP. Cluster formation in dilute suspen-
sions. In: Proc Symp on Interaction Between Fluids and
Particles. London: Instn Chem Engrs, 1962, pp 1721.
Kehlenbeck R, Di Felice R. Empirical relationships for the
terminal settling velocity of spheres in cylindrical col-
umns. Chem Eng Technol 21:303308, 1999.
Kempling JC, Eng J. Performance of dual-media lters2.
Chem Eng Prog 73(4):8791, 1977.
Kennedy SC, Bretton RH. Axial dispersion of spheres
uidized with liquids. AIChE J 12:2430, 1966.
Khan AR, Richardson JF. Fluid-particle interactions and
ow characteristics of uidized beds and settling suspen-
sions of spherical particles. Chem Eng Commun
78:111130, 1989.
Khan AR, Richardson JF. Pressure gradient and friction
factor for sedimentation and uidisation of uniform
spheres in liquids. Chem Eng Sci 45:255265, 1990.
Kikuchi K, Sugawara T, Ohashi H. Correlation of mass
transfer coecient between particles and liquid in liquid
uidized beds, J Chem Eng Japan 16:426428, 1983.
Kikuchi K, Konno H, Kakutani S, Sugawara T, Ohashi H.
Axial dispersion of liquid in liquid uidized beds in the
low Reynolds number region. J Chem Eng Japan
17:362367, 1984.
Kim N-H, Lee Y-P. A study on the pressure loss, heat trans-
fer enhancement and fouling control in liquid uidized
bed heat exchangers. In: Panchal CB, ed. Fouling
Mitigation of Industrial Heat-Exchange Equipment.
New York: Begell House, 1997, pp 421433.
King DH, Smith JW. Wall mass transfer in liquid-uidized
beds. Can J Chem Eng 45:329333, 1967.
Klaren DG, Bailie RE. Fluid bed heat exchangers solve foul-
ing problems. Amer Papermaker, Pacic Ed., Feb 1988,
pp 3233.
Klaren DG, Bailie RE. Consider nonfouling uidized bed
exchangers. Hydrocarbon Processing 68(7):4850, 1989.
Klaren DG, Halberg N. Development of a multistage ash/
uidized bed evaporator. Chem Eng Prog 43(7):4143,
1980.
Klaren DG, Sullivan DW. Nonfouling heat exchanger
performance in severe fouling services: principle, indus-
trial applications and operating installations. In:
Preprints, Internat Conf Petroleum Phase Behavior and
Fouling. Houston: AIChE Spring Meeting, March 1999,
pp 359366.
Kmiec A. Particle distributions and dynamics of particle
movement in solidliquid uidized beds. Chem Eng J
15:112, 1978.
Kola r V. Fluidization of solid particles by liquid in conical
vessels. Collection Czechoslov Chem Commun 28:1224
1231, 1963.
Kollbach J, Dahm W, Rautenbach R. Continuous cleaning
of heat exchanger with recirculating uidized bed. Heat
Transfer Engineering 8(4):2632, 1987.
Koloini T, Farkas EJ. Fixed bed pressure drop and liquid
uidization in tapered or conical vessels. Can J Chem Eng
51:499502, 1973.
Koloini T, Zumer M. Ion exchange with irreversible reaction
in deep uidized beds. Can J Chem Eng 57:183190, 1979.
Kramers H, Westermann MD, de Groot JH, Dupont FAA.
The longitudinal dispersion of liquid in a uidized bed. In:
Proc Symp on Interaction Between Fluids and Particles.
London: Instn Chem Engrs, 1962, pp 114119.
Krishnaswamy PR, Ganapathy R, Shemilt LW. Correlating
parameters for axial dispersion in liquid uidized systems.
Can J Chem Eng 56:550553, 1978.
Kuwabara S. The forces experienced by randomly distributed
parallel circular cylinders or spheres in a viscous ow at
small Reynolds numbers. J Phys Soc Japan 14:527532,
1959.
Kwauk M. Particulate uidization: an overview. Adv Chem
Eng 17:207360, 199192.
Kwauk M. Fluidization: Idealized and Bubbleless, with
Applications. New York: Science Press and Ellis
Horwood, 1992.
Kwauk M. Comment. Chem Eng Sci 51(24): iiiiv, 1996.
Ladenburg R. On the inuence of the walls on the motion of
a sphere in a viscous uid. Annalen der Physik 23:447
458, 1907.
Laguerie C, Angelino H. Comparaison entre la dissolutien et
la croissance de cristaux dacide citrique monohydrate en
lit uidise . Chem Eng J 10:4148, 1975.
Lapidus L, Elgin JC. Mechanics of vertical-moving uidized
systems. AIChE J 3: 6368, 1957.
Latif BAJ, Richardson JF. Circulation patterns and velocity
distributions for particles in a liquid uidised bed. Chem
Eng Sci 27:19331949, 1972.
Law H-S, Masliyah JH, MacTaggart RS, Nandakumar K.
Gravity separation of bidisperse suspensions: light and
heavy particle species. Chem Eng Sci 42:15271538, 1987.
LeClair BP, Hamielec AE. Viscous ow through particle
assemblages at intermediate Reynolds numbers: steady
state solutions for ow through assemblages of spheres.
Ind Eng Chem Fundam 7:542549, 1968.
Lee J-K, Chun H-S, Shemilt LW. Wall-to liquid mass trans-
fer in liquidsolid and gasliquidsolid uidized beds. J
Chem Eng Japan 30:246252, 1997.
Le Go P, Vergnes F, Coeuret F, Bordet J. Applications of
uidized beds in electrochemistry. Ind Eng Chem
61(10):817 (1969).
Letan R, Elgin JC. Fluid mixing in particulate uidized beds.
Chem Eng J 3:136144, 1972.
Lewis EW, Bowerman EW. Fluidization of solid particles in
liquids. Chem Eng Prog 48:603610, 1952.
Lewis WK, Gilliland ER, Bauer WC. Characteristics of ui-
dized particles. Ind Eng Chem 41:11041117, 1949.
Liang W-G, Zhu J-X. A coreannulus model for the radial
ow structure in a liquidsolid circulating uidized bed
(LSCFB). Chem Eng J 68:5162, 1997.
Liang W-G, Zhu J-X, Jin Y, Yu Z-Q, Wang Z-W, Zhou J.
Radial nonuniformity of ow structure in a liquidsolid
circulating uidized bed. Chem Eng Sci 51:20012010,
1996.
Liang W, Zhang S, Zhu J-X, Jin Y, Yu Z, Wang Z. Flow
characteristics of the liquidsolid circulating uidized bed.
Limas-Ballesteros R. Fluidisation de particules non sphe ri-
ques. Transfert de matie` re entre le solide et le liquide.
The` se de Docteur Inge nieur, I.N.P. Toulouse, France,
1980.
Limas-Ballesteros R, Riba JP, Couderc JP. Expansion de
couches de particules non sphe riques uidise es par un
liquide. Entropie 106:3745, 1982a.
Limas-Ballesteros R, Riba JP, Couderc JP. Dissolution of
non-spherical particles in solidliquid uidization. Chem
Eng Sci 37:16391644, 1982b.
Lin T-J, Fan L-S. Characteristics of high pressure liquid
solid uidization. AIChE J 43:4557, 1997.
Lin CS, Denton EB, Gaskill HS, Putnam GL. Diusion-con-
trolled electrode reactions. Ind Eng Chem 43:21362143,
1951.
Lockett MJ, Al-Habbooby HM. Dierential settling by size
of two particle species in a liquid. Trans Instn Chem
Engrs 51:281292, 1973.
Lockett MJ, Al-Habbooby HM. Relative particle velocities
in two-species settling. Powder Technology 10:6771,
1974.
Lopez-Cacicedo CL. The electrolytic recovery of metals from
diluent euent streams. J Separ Proc Technol 2(1):3439,
1981.
Lucas A, Arnaldos J, Casal J, Puigjaner L. Improved equa-
tion for the calculation of minimum uidization velocity.
Martin BLA, Kolar Z, Wesselingh JA. The falling velocity of
a sphere in a swarm of dierent spheres. Trans IChemE
59:100104, 1981.
Maruyama T, Maeda H, Mizushina T. Liquid uidization in
tapered vessels. J Chem Eng Japan 17:132139, 1984.
Masliyah JH. Hindered settling in a multi-species particle
system. Chem Eng Sci 34:11661168, 1979.
Masliyah JH. Eect of stirring on a liquid uidized bed.
AOSTRA J Res 5:4960, 1989.
Masliyah JH. Electrokinetic Transport Phenomena.
Edmonton: AOSTRA, 1994, pp 165171.
Masliyah JH, Polikar M. Terminal velocity of porous
spheres. Can J Chem Eng 58:299302, 1980.
Mathur KB, Epstein N. Spouted Beds. New York: Academic
Press, 1974.
Matsumoto K, Suganuma A. Settling velocity of a permeable
model oc. Chem Eng Sci 32:445447, 1977.
McCune LK, Wilhelm RH. Mass and momentum transfer
in a solidliquid system. Ind Eng Chem 41:11241134,
1949.
Medlin J, Jackson R. Fluid mechanical description of ui-
dized beds. The eect of distributor thickness on convec-
tive instabilities. Ind Eng Chem Fundam 14:315321,
1975.
Medlin J, Wong H-W, Jackson R. Fluid mechanical descrip-
tion of uidized beds. Convective instabilities in bounded
beds. Ind Eng Chem Fundam 13:247259, 1974.
Mehta SC, Shemilt LW. Frequency response of liquid ui-
dized systems. Part II: Eect of liquid viscosity. Can J
Chem Eng 54:4351, 1976.
Meijer JAM, Van Rosmalen GM, Veenman AW, Van Dissel
CM. Scale inhibition of calcium sulfate by a uidized bed.
Desalination 34:217232, 1980.
Mertes TS, Rhodes HB. Liquidparticle behavior. Chem Eng
Prog 51:429432, 517522, 1955.
Michaels AS, Bolger JC. Settling rates and sediment volumes
of occulated kaolin suspensions. Ind Eng Chem
Fundamentals 54:2433, 1962.
Mirza S, Richardson JF. Sedimentation of suspensions of
particles of two or more sizes. Chem Eng Sci 34:447
454, 1979.
Mitra AK, Epstein N. Eect of solids wettability on liquid
uidization. Can J Chem Eng 56:520522, 1978.
Miura H, Kawase Y. Hydrodynamics and mass transfer in
three-phase uidized beds with non-Newtonian uids.
Chem Eng Sci 52:40954104, 1997.
Miura H, Kawase Y. Minimum liquid uidization velocity in
two- and three-phase uidized beds with non-Newtonian
uids. Powder Technology 97:124128, 1998.
Molerus O, Wirth K-E. Heat Transfer in Fluidized Beds. 1st
ed. London: Chapman and Hall, 1997, p 167.
Moritomi H, Iwase T, Chiba T. A comprehensive interpreta-
tion of solid layer inversion in liquid uidised beds. Chem
Eng Sci 37:17511757, 1982.
Moritomi H, Yamagishi T, Chiba T. Prediction of complete
mixing of liquid-uidized binary solid particles. Chem
Eng Sci 41:297305, 1986.
Mostou N, Chaouki J. Prediction of eective drag coe-
cient in uidized beds. Chem Eng Sci 54: 851858, 1999.
Mu ller-Steinhagen H, Jamialahmadi M, Robson B.
Mitigation of scale formation during the Bayer pro-
cessmeasurements at the Alcoa of Australia renery in
Kwinana. In: Mannweiler U, ed. Light Metals 1994.
Minerals, Metals and Materials Society, 1994, pp 121
127.
Mullin JW. Crystallization. 3d ed. Oxford: Butterworth-
Heinemann, 1993.
Mullin JW, Garside J. The crystallization of aluminium
potassium sulphate: a study in the assessment of crystal-
lizer design data. Part II: Growth in a uidized bed crys-
tallizer. Trans Instn Chem Engrs 45:T291T295, 1967.
Mullin JW, Nyvlt J. Design of classifying crystallizer. Trans
Instn Chem Engrs 48:T7T14, 1970.
Munroe HS. The English vs. the continental system of jig-
gingis close sizing advantageous? Trans Am Inst Min
Engrs 17:637659, 1888/89.
Nakamura K, Kumagai H, Yano T. Performance of a zig-
zag uidized bed as an immobilized enzyme reactor. In:
Linko P, Larinkari J, eds. Food Process Engineering, Vol
2. Enzyme Engineering in Food Processing. London:
Applied Science, 1979, pp 186191.
Nakamura M, Hamada Y, Toyama S, Fouda AE, Capes CE.
An experimental investigation of minimum uidization
velocity at elevated temperatures and pressures. Can J
Chem Eng 63: 813, 1985.
Nasr-El-Din H, Masliyah JH, Nandakumar K, Law D H-S.
Continuous gravity separation of a bidisperse suspension
in a vertical column. Chem Eng Sci 43:32253234, 1988.
Nasr-El-Din H, Masliyah JH, Nandakumar K. Continuous
gravity separation of concentrated bidisperse suspension
in a vertical column. Chem Eng Sci 45: 849857, 1990.
Neale G, Epstein N, Nader W. Creeping ow relative to
Needham LW, Lynch S. The use of suspensions as heavy
liquids. Trans Instn Chem Engrs 23:93105, 1945.
Nelson PA, Galloway RT. Particle-to-uid heat and mass
transfer in dense systems of ne particles. Chem Eng Sci
30:16, 1975.
Neuz il L. Fluidization of polydisperse materials. I.
Theoretical considerations on the study of the segregation
of a mixture of particles of equal density. Collection
Czechoslov Chem Commun 29:571578, 1964.
Neuz il L, Hrdina M. Eect of walls on velocity prole of
uid in a uidized bed. Collection Czechoslov Chem
Commun 30:30633070, 1965.
Nikolov L, Karamanev D. Experimental study of the inverse
uidized bed biolm reactor. Can J Chem Eng 65:214
217, 1987.
Nikolov LN, Karamanev, DG. The inverse uidized bed
biolm reactor: a new laboratory scale apparatus for bio-
lm research. J. Fermentation and Bioengineering
69(4):265267, 1990.
Nikov I, Delmas H. Solidliquid mass transfer in three-phase
xed and uidized beds. Chem Eng Sci 42:10891093,
1987.
Nikov I, Karamanev D. Liquidsolid mass transfer in inverse
uidized bed. AIChE J 37:781784, 1991.
Nikov I, Karamanev D. Reply to letter. AIChE J 38:638
639, 1992.
Niven RK, Khalili, N. In situ uidisation by a single internal
vertical jet. J Hydraulic Res 36(2):199228, 1998a.
Niven RK, Khalili N. In situ multiphase uidization
(upow washing) for the remediation of hydrocarbon
contaminated sands. Can Geotech J 35:938960, 1998b.
Obata E, Watanabe H, Endo N. Measurement of size and
size distribution of particles by uidization. J Chem Eng
Japan 15:2328, 1982.
Oman AO, Watson KM. Pressure drops in granular beds.
National Petroleum News 36:R-795801, Nov. 1, 1944.
Page RE. Some aspects of three-phase uidization. Ph.D.
thesis, Cambridge University, 1970, p 138.
Panier P, Riba JP, Couderc JP. E
tude de lextraction liquide

solide en couche uidise e par voie electrochimique. Chem
Eng J 20:157165, 1980.
Parkin GF, Speece RE. Anaerobic biological waste
treatment. Chem Eng Prog 80(12):5558, 1984.
Patwardhan VS, Tien C. Distribution of solid particles in
liquid uidized beds. Can J Chem Eng 62:4654, 1984.
Patwardhan VS, Tien C. Sedimentation and liquid uidiza-
tion of solid particles of dierent sizes and densities.
Chem Eng Sci 40:10511060, 1985.
Peng Y, Fan LT. Hydrodynamic characteristics of uidiza-
tion in liquidsolid tapered beds. Chem Eng Sci
52:22772290, 1997.
Perry RH, Green DW, Maloney JO. Perrys Chemical
Engineers Handbook. 6th ed. New York: McGraw-Hill,
1984, pp 19.3719.38.
Phillips VR, Epstein N. Growth of nickel sulfate in a
laboratory-scale uidized-bed crystallizer. AIChE J
20:678687, 1974.
Pirozzi D, Gianfreda L, Greco G (Jr). Development of a
circulating uidized bed fermentor: the hydrodynamic
model for the system. AIChE Symp Series
85(270):101110, 1989.
Price BG, Lapidus L, Elgin JC. Mechanics of vertical moving
uidized systems II. Application to countercurrent
operation. AIChE J 5:9397, 1959.
Pruden BB, Epstein N. Stratication by size in particulate
uidization and in hindered settling. Chem Eng Sci
19:696700, 1964.
Prudhoe J, Whitmore RL. Terminal velocity of spheres in
uidized beds. Brit Chem Eng 9:371375, 1964.
Qian D, Wang Z, Chen G. On the inversion phenomena in a
liquidsolid uidized bed. Chemical Reaction Engineering
and Technology 9:485491, 1993.
Quinn JA, Lapidus L, Elgin JC. The mechanics of moving
vertical uidized systems V. Concurrent cogravity ow.
AIChE J 7:261263, 1961.
Rahman K, Streat M. Mass transfer in liquid uidized beds
of ion exchange particles. Chem Eng Sci 36:293300,
1981.
Rapagna` S, Di Felice R, Gibilaro LG, Foscolo PU. Steady-
state expansion characteristics of monosize spheres
uidised by liquids. Chem Eng Commun 79:131140,
1989.
Rautenbach R, Kollbach J. New developments in uidized
bed heat transfer for preventing fouling. Swiss Chem
8(5):4755, 1986.
Riba JP, Couderc JP. Expansion de couches uidise es par des
liquides. Can J Chem Eng 55:118121, 1977.
Riba JP, Couderc JP. Transfer de matie` re autour dune
sphe` re immerge e dans une couche uidise e par un liquide.
Intern J Heat Mass Transfer 23:909917, 1980.
Riba JP, Routie R, Couderc JP. Conditions minimales de
mise en uidisation par un liquide. Can J Chem Eng
56:2630, 1978.
Richards RH. Close sizing before jigging. Trans Am Inst Min
Engrs 24:409486, 1893.
Richards RH, Locke CE. Textbook of Ore Dressing. 3d ed.
New York: McGraw-Hill, 1940, pp 136143, 154157.
Richardson JF. Incipient uidization and particulate sys-
tems. In: Davidson JF, Harrison D, eds. Fluidization.
London: Academic Press, 1971, pp 2564.
Richardson JF, Aatin E. Fluidisation and sedimentation of
binary mixtures of particles in a liquid. Proc 9th
International Conference on Transport and
Sedimentation of Solid Particles. Cracow, Poland:
Zeszyty Naukowe Akademii Rolniczej we Wroclawiu Nr
315, 1997, pp 487502.
Richardson JF, Meikle RA. Sedimentation and uidisation
Part III: The sedimentation of uniform ne particles and
two-component mixtures of solids. Trans Instn Chem
Engrs 39:348356, 1961a.
Richardson JF, Meikle RA. Sedimentation and uidisation
Part IV: Drag force on individual particles in an assem-
blage. Trans Instn Chem Engrs 39:357362, 1961b.
Richardson JF, Shabi FA. The determination of concentra-
tion distribution in a sedimenting suspension using radio-
active solids. Trans Instn Chem Engrs 38:3342, 1960.
Richardson JF, Zaki WN. Sedimentation and uidisation:
Part I. Trans Instn Chem Engrs 32:3553, 1954.
Richardson JF, Romani MN, Shakiri KJ. Heat transfer from
immersed surfaces in liquid uidized beds. Chem Eng Sci
31:619624, 1976.
Rigby GR, Van Blockland GP, Park WH, Capes CE.
Properties of bubbles in three-phase uidized beds as mea-
sured by an electroresistivity probe. Chem Eng Sci
25:17291741, 1970.
Rowe PN. Drag forces in a hydraulic model of a uidised
bedPart II. Trans Instn Chem Engrs 39:175180, 1961.
Rowe PN. A convenient empirical equation for estimation of
the RichardsonZaki exponent. Chem Eng Sci 42:2795
2796, 1987.
Rowe PN, Claxton KT. Heat and mass transfer from a single
sphere to uid owing through an array. Trans Instn
Chem Engrs 43:T321T331, 1965.
Rowe PN, Henwood GA. Drag forces in a hydraulic model
of a uidised bedPart I. Trans Instn Chem Engrs 39:43
54, 1961.
Saman PG. On the settling speed of free and xed sus-
ensions. Studies in Appl. Math L11(2):115127, June
1973.
Salas-Morales JC, Evans JW, Newman OMG, Adcock PA.
Spouted bed electrowinning of zinc: Part 1. Laboratory-
scale electrowinning experiments. Metall Mat Trans B
28:5968, 1997.
Saxena SC, Vogel GJ. The measurement of incipient uidisa-
tion velocities in a bed of coarse dolomite at temperature
and pressure. Trans IChemE 55:184189, 1977.
Saxton JA, Fitton JB, Vermeulen T. Cell model theory of
homogeneous uidization: density and viscosity beha-
viour. AIChE J 16:120130, 1970.
Scarlett B, Blogg MJ. The motion and distribution of
particles in a liquid-uidized bed. In: Drinkenburg
AAH, ed. Proc Internat Conf on Fluidization.
Amsterdam: Netherlands Univ. Press, 1967, pp 8289.
Schiller L, Naumann A. U
ber die grundlegenden berechnun-

gen bei der schwerkragtaufbereitung. Z Ver Deutsch Ing
77:318320, 1933.
Schmidt S, Buchs J, Born C, Biselli M. A new correlation for
the wall-to-uid mass transfer in liquidsolid uidized
beds. Chem Eng Sci 54: 829839, 1999.
Scott CD, Hancher CW, Shumate II SE. A tapered uidized
bed as a bioreactor. Enzyme Eng 3:255261, 1978.
Scott TC, Cosgrove JM, Asif M, Petersen JN.
Hydrodynamic studies of an advanced uidized-bed bio-
reactor for direct interaction with coal. Fuel
72:17011704, 1993.
Selim MS, Kothari AC, Turian RM. Sedimentation of binary
suspensions. AIChE Symp Series (Fluidization and Fluid
Particle Systems: Theories and Applications, Knowlton
TM, ed.) 79(222):103108, 1983a.
Selim MS, Kothari AC, Turian RM. Sedimentation of multi-
sized particles in concentrated suspensions. AIChE J
29:10291038, 1983b.
Selke WA, Bliss H. Continuous countercurrent ion exchange.
Chem Eng Prog 47:529533, 1951.
Shiau L-D, Cheung S-H, Liu Y-C. Modelling of a uidized-
bed crystallizer in a batch mode. Chem Eng Sci 54: 865
871, 1999.
Sibony MJ. Clarication with microsand seeding. A state of
the art. Water Res 15:12811290, 1981.
Slater MJ. A review of continuous counter-current contac-
tors for liquids and particulate solids. Brit Chem Eng
14:4146, 1969.
Slater MJ. Recent industrial-scale applications of continuous
resin ion exchange systems. J Separ Proc Tech 2(3):212,
1981.
Slater MJ. The relative sizes of xed bed and continuous
counter-current ow ion exchange equipment. Trans
IChemE 60:5458, 1982.
Slater MJ, Lucas BH. Flow patterns and mass transfer rates
in uidized bed ion exchange equipment. Can J Chem Eng
54:264268, 1976.
Slis PL, Willemse Th W, Kramers H. The response of the
level of a liquid uidized bed to a sudden change in the
uidizing velocity. Appl Sci Res A8:209218, 1959.
Smith TN. A model of settling velocity. Chem Eng Sci
53:315323, 1998.
Sozen ZZ, Grace JR, Pinder KL. Thermal energy storage by
agitated capsules of phase change material. Ind Eng Chem
Res 27:679691, 1988.
Steinour HH. Rate of sedimentation: suspensions of uni-
form-size angular particles. Ind Eng Chem 36: 840847,
1944.
Storck A, Coeuret F. Mass transfer between a owing liquid
and a wall or an immersed surface in xed and uidized
beds. Chem Eng J 20:149156, 1980.
Streat M. Recent developments in continuous ion exchange.
J Separ Proc Technol 1(3):1018, 1980.
Struve DL, Lapidus L, Elgin JC. The mechanics of
moving vertical uidized systems III. Application to
concurrent countergravity ow. Can J Chem Eng
36:141152, 1958.
Svanoe H. Krystal classifying crystallizer. Ind Eng Chem
32:636639, 1940.
Svarovsky L. SolidLiquid Separation. 3d ed. London:
Butterworth, 1990, p 130.
Syamlal M, OBrien TJ. Simulation of granular layer inversion
in liquid uidized beds. Chem Eng Sci 14:473481, 1988.
Tackie E, Bowen BD, Epstein, N. Hindered settling of
uncharged and charged submicrometer spheres. In:
Pfeer R, ed. Fourth International Conference on
Physicochemical Hydrodynamics. Annals of the New
York Academy of Sciences 104:366367, 1983.
Taggart AF. Handbook of Mineral Dressing, Vol. I. New
York: John Wiley, 1953, pp 847.
Tai CY, Chen C-Y, Wu J-F. Crystal dissolution and growth
in a lean uidized-bed crystallizer. Chem Eng Commun
56:329340, 1987.
Tai CY, Chien W-C, Chen C-Y. Crystal growth kinetics of
calcite in a dense uidized-bed crystallizer. AIChE J
45:16051614, 1999.
Tang W-T, Fan L-S. Axial liquid mixing in liquidsolid and
gasliquidsolid uidized beds containing low density
particles. Chem Eng Sci 45:543551, 1990.
Tannous K, Hemati M, Laguerie C. Characte ristiques au
minimum de uidisation et expansion des couches uidi-
se es de particules de la cate gorie D de Geldart. Powder
Thomas DG. Transport characteristics of suspensions: VIII.
A note on the viscosity of Newtonian suspensions of uni-
form spherical particles. J Colloid Sci 20:267277, 1965.
Thomas CR, Yates JG. Expansion index for biological ui-
dised beds. Chem Eng Res Des 63:6770, 1985.
Thonglimp V, Hiquily N, Laguerie C. Vitesse minimale de
uidisation et expansion des couches uidise es par un gaz.
Powder Technology 38:233253 (1984).
Toor HL, Marchello JM. Film-penetration model for heat
and mass transfer. AIChE J 4:97101, 1958.
Tournie P, Laguerie C, Couderc JP. Correlations for mass
transfer between uidized spheres and a liquid. Chem Eng
Sci 34:12471255, 1979.
Trawinski H. Eective zahigkeit und inhomogenita t von wir-
belschichten. Chemie-Ing.-Tech. 25:229238, 1953.
Treybal RE. Mass-Transfer Operations. 3d ed. New York:
McGraw-Hill, 1980.
Trupp AC. Dynamics of liquid uidised beds of spheres. I
Chem E Symp Ser 30:182189, 1968.
Turton R, Clark NN. An explicit relationship to predict
spherical particle terminal velocity. Powder Technology
53:127129, 1987.
Vanadurongwan V, Laguerie C, Couderc JP. Inuence des
proprie te s physiques sur le transfert de matie` re en uidi-
sation liquide. Chem Eng J 12:2931, 1976.
Van der Heiden G, Raats MS, Boon HF. Fluidised bed elec-
trolysis for removal or recovery of metals from dilute
solutions. Chemistry and Industry 13:465468, 1978.
Van der Meer AP, Blanchard CMRJP, Wesselingh JA.
Mixing of particles in liquid uidized beds. Chem Eng
Res Des 62:214222, 1984.
Van der Wielen LAM, Van Dam MHH, Luyben KChAM.
On the relative motion of a particle in a swarm of dierent
particles. Chem Eng Sci 51:9951008, 1996.
Van Duijn G, Rietema K. Segregation of liquid-uidized
Veenman AW. The MSF/FBE: An improved multi-stage
ash distillation process. Desalination 19:114, 1976.
Veenman AW. Heat transfer in a multi-stage ash/uidized
bed evaporator (MSF/FBE). Desalination 22:5576, 1977.
Verschoor H. Experimental data on the viscous force exerted
by a owing uid on a dense swarm of particles. Appl Sci
Res A2:155161, 1951.
Volpicelli G, Massimilla L, Zenz FA. Nonhomogeneities in
solidliquid uidization. Chem Eng Prog Symp Ser
62(67):4250, 1966.
Wakeman RJ, Stopp BW. Fluidisation and segregation of
binary particle mixtures. Powder Technology 13:261
268, 1976.
Wallis GB. One-Dimensional Two-Phase Flow. New York:
McGraw-Hill, 1969.
Wasmund B, Smith JW. Wall to uid heat transfer in liquid
uidized beds, Part 2. Can J Chem Eng 45:156165, 1967.
Webb C, Black GM, Atkinson B. Liquid uidisation of
highly porous particles. Chem Eng Res Des 61:125134,
1983.
Webster GH, Perona JJ. Liquid mixing in a tapered uidized
bed. AIChE J 34:13981402, 1988.
Webster GH, Perona JJ. The eect of taper angle on the
hydrodynamics of a tapered liquidsolid uidized bed.
AIChE Symp Ser 86(276):104112, 1990.
Wen CY, Fan L-S. Axial dispersion of non-Newtonian
liquids in uidized beds. Chem Eng Sci 28:17681772,
1973.
Wen CH, Yu YH. Mechanics of uidization. Chem Eng Prog
Symp Ser 62(62):100111, 1966.
Whitmore RL. The relationship of the viscosity to the set-
tling rate of slurries. J Inst Fuel 30:238242, 1957.
Wilhelm RH, Kwauk M. Fluidization of solid particles.
Chem Eng Prog 44(3):201218, 1948.
Wright PC, Raper JA. A review of some parameters involved
in uidized bed bioreactors. Chem Eng Technol 19:5064,
1996.
Yang J, Renken A. Intensication of mass transfer in liquid
uidized beds with inert particles. Chemical Engineering
Yu H, Shi Y. Characteristics of uidization of binary particle
mixtures. In: Kwauk M, Kunii D, Jiansheng Z, Hasatani
M, eds. Fluidization 85 Science and Technology
Conference Papers. Second China-Japan Symposium.
Amsterdam: Elsevier, 1985, pp 250261.
Yutani N, Fan LT. Mixing of randomly moving particles in
liquidsolid uidized beds. Powder Technology 42:145
152, 1985.
Yutani N, Ototake N, Too JR, Fan LT. Estimation of the
particle diusivity in a liquidsolids uidized bed based
on a stochastic mode. Chem Eng Sci 37:10791085, 1982.
Zheng Y, Zhu J-X, Wen J, Martin SA, Bassi AS, Margaritis
A. The axial hydrodynamic behavior in a liquidsolid
circulating uidized bed. Can J Chem Eng 77:284290,
1999.
Zhu J-X, Zheng Y, Karamanev DG, Bassi AS. (Gas)liquid
solid circulating uidized beds and their potential applica-
tions to bioreactor engineering. Can J Chem Eng 78: 82
94, 2000.
Manuscript received February 28, 2000.
27
GasLiquidSolid Three-Phase Fluidization
Liang-Shih Fan and Guoqiang Yang
The Ohio State University, Columbus, Ohio, U.S.A.
1 INTRODUCTION
Gasliquidsolid uidization systems have been
applied extensively in industry for physical (e.g., sand
lter cleaning and granular material drying), chemical
(e.g., hydrogen peroxide production and methanol
synthesis), petrochemical (e.g., resid hydrotreating
and hydrotreating of tar sands), and biochemical
(e.g., treatment of lactose wastewater and bioleaching
of metals from ores) processing (Shah, 1979;
LHomme, 1979; Ramachandran and Chaudhari,
1983; Fan, 1989). The interest in the application of
three-phase uidization systems has promoted contin-
ued research and development eorts in these systems.
Examples are biological operation for human viral vac-
cine production using microcarrier cultures of animal
cells (Kalogerakis and Behie, 1995), ethanol fermenta-
tion using immobilized cells in a multistage uidized
bed bioreactor (Tzeng et al., 1991), cleaning and desul-
furization of high sulfur coal by selective occulation
and bioleaching (Attia et al., 1991), hydrocarbon pro-
duction in fossil fuel processing using mild hydrocrack-
ing of shale oil (Souza et al., 1992), production of
paran waxes using the SASOL slurry bed process
(slurry bubble column) for FischerTropsch synthesis
(Inga, 1994), and production of light hydrocarbon
diesel using the Texaco T-Star process (three-phase
uidized bed) for vacuum gas oil hydrotreating/hydro-
cracking (Johns et al., 1993). Most of these processes
with considerable commercial interest are conducted
under high pressure and high temperature, for exam-
ple, methanol synthesis (at P = 5.5 MPa and T = 2
608C), resid hydrotreating (at P = 5.521 MPa and
T = 3004258C), FischerTropsch synthesis (at
P = 1.55.0 MPa and T = 2508C), and benzene hydro-
genation (at P = 5.0 MPa and T = 1808C) (Fox, 1990;
Jager and Espinoza, 1995; Saxena, 1995; Mill et al.,
1996; Peng et al., 1999). Fundamental study of ow
characteristics under high-pressure and high-tempera-
ture conditions is crucial for the development of these
industrial processes, and much progress has been made
recently regarding high-pressure and high-temperature
systems with relevance to industrial processes.
The fundamental characteristics of three-phase
uidization including bubble characteristics, hydrody-
namics, and heat and mass transfer properties along
with many industrial processes have been extensively
reported in Fans (1989) book as well as its companion
book on bubble wake dynamics (Fan and Tsuchiya,
1990). As both books are widely referenced in the
eld of three-phase uidization, this chapter is pre-
sented mainly as an update to these two books. The
chapter will cover the continued research progress
made over the past ten years on the fundamentals of
three-phase uidization. Major ndings on uidization
and bubble dynamics under ambient conditions and
the relevant literature reported earlier will be covered.
Furthermore, new research on the high-pressure and
high-temperature three-phase uidization will be high-
lighted as well as computational uid dynamics.
2 BUBBLE DYNAMICS
In gasliquidsolid uidization systems, bubble
dynamics plays a key role in dictating the transport
phenomena and ultimately aects the overall rates of
reactions. It has been recognized that the bubble wake,
when it is present, is the dominant factor governing the
system hydrodynamics. In general, consideration of
the ow associated with the bubble wake near the bub-
ble rear, whether laminar or turbulent, is essential to
characterize the complete behavior of the rising bubble
including its motion. Conversely, examining the shape,
rise velocity, and motion of a bubble can provide an
indirect understanding of the dynamics of the liquid
solid ow around the bubble. Temperature and pres-
sure are the key operating variables dictating the
physical properties of the gas and liquid phases and
hence the bubble behavior. The liquidsolid suspen-
sion through which the gas bubbles rise may be char-
acterized as a pseudohomogeneous medium under
certain operating conditions. The most marked eects
are ow instability induced by bubble wake, and the
resulting large-scale vortical motion and local solids
concentration gradient. Interactions between neigh-
boring bubbles are usually recognized in terms of
bubble coalescence and/or breakup. Direct causes of
these phenomena, however, lie in the interactions
between the bubbles and the surrounding ow.
In the following, several fundamental aspects of the
dynamics of single and multibubbles in liquidsolid
suspensions are discussed, which are of paramount
importance to the transport processes in the three-
phase uidization system. Specically, four subjects
are covered: (1) plenum bubble behavior, (2) bubble
rise characteristics, (3) bubble coalescence, and (4)
bubble breakup. The eects of pressure and tempera-
ture on these phenomena are illustrated.
2.1 Plenum Bubble Behavior
Bubble behavior in the plenum region is important for
understanding bubble characteristics in multiphase
uidization systems, particularly under low gas veloci-
ties. In this section, the behavior of bubble formation
from a single orice will be emphasized.
There are two typical mechanical arrangements for
bubble formation from a single orice, that is, the ori-
ce connected or not connected to a gas chamber, as
shown in Fig. 1. For bubble formation from a single
orice without a gas chamber, the gas ow rate
through the orice is always constant, which is referred
to as constant ow conditions. The phenomenon of
bubble formation from a single orice connected to a
gas chamber varies with gas injection conditions,
which are characterized by the dimensionless capaci-
tance number N
c
dened as 4V
c
gr
l
/pD
2
o
P (Kumar and
Kuloor, 1970; Tsuge and Hibino, 1983). When N
c
is
smaller than 1, the gas ow rate through the orice is
almost constant during the bubble formation process,
similar to the rst mechanical arrangement. When N
c
is larger than 1, the gas ow rate through the orice is
not constant, and it is dependent on the pressure dif-
Figure 1 Typical mechanical arrangements for single bubble
formation: (a) single orice without a gas chamber; (b) single
orice connected to a gas chamber.
ference between the gas chamber and the bubble. Such
bubble formation conditions are characterized as vari-
able ow conditions by Yang et al. (2000a) or as con-
stant pressure and intermediate conditions by Tsuge
and Hibino (1983).
Numerous experimental and modeling studies have
been conducted over the past decades on bubble for-
mation from a single orice or nozzle submerged in
liquids, mostly under ambient conditions (Kupferberg
and Jameson, 1969; Kumar and Kuloor, 1970; Azbel,
1981; Lin et al., 1994; Ruzicka et al., 1997). Among
various factors that aect the bubble formation, the
wettability of the orice surface is one of the important
factors that aect the initial size of the bubble formed
on the orice. Lin et al. (1994) found that initial bubble
size increases signicantly with the contact angle
between the bubble and the orice surface when the
contact angle exceeds the threshold value of 458. A
few studies were conducted at elevated pressures (La
Nauze and Harris, 1974; Idogawa et al., 1987; Tsuge et
al., 1992; Wilkinson and van Dierendonck, 1994).
These studies indicated that an increase in gas density
reduces the size of bubbles formed from the orice.
Bubble formation in liquids with the presence of
particles, as in slurry bubble columns and three-
phase uidized bed systems, is dierent from that in
pure liquids. The experimental data of Massimilla et al.
(1961) in an airwaterglass beads three-phase ui-
dized bed revealed that the bubbles formed from a
single nozzle in the uidized bed are larger than
those in water, and the initial bubble size increases
with the solids concentration. Yoo et al. (1997) inves-
tigated bubble formation in pressurized liquidsolid
suspensions. They used aqueous glycerol solution and
0.1 mm polystyrene beads as the liquid and solid
phases, respectively. The densities of the liquid and
the particles were identical, and thus the particles
were neutrally buoyant in the liquid. The results indi-
cated that initial bubble size decreases inversely with
pressure under otherwise constant conditions, that is,
gas ow rate, temperature, solids concentration, orice
diameter, and gas chamber volume. Their results also
showed that the particle eect on the initial bubble size
is insignicant. The dierence in the nding regarding
the particle eect on the initial bubble size between
Massimilla et al. (1961) and Yoo et al. (1997) is pos-
sibly due to the dierence in particle density.
Bubble formation in a hydrocarbon liquid and
liquidsolid suspension with signicant density dier-
ence between the liquid and solid phases was investi-
gated by Luo et al. (1998a) and Yang et al. (2000a)
under various gas injection conditions. A mechanistic
model was developed to predict the initial bubble size
in liquidsolid suspensions at high-pressure conditions.
The model considers various forces induced by the
particles and is an extension of the two-stage spherical
bubble formation model developed by Ramakrishnan
et al. (1969) for liquids. In the two-stage spherical
bubble formation model, bubbles are assumed to be
formed in two stages, the expansion stage and the
detachment stage. The bubble expands with its base
attached to the nozzle during the rst stage. In the
detachment stage, the bubble base moves away from
the nozzle, although the bubble remains connected
with the nozzle through the neck. The shape of the
bubble is assumed to remain spherical during the entire
bubble formation process. It is also assumed in this
model that a liquid lm always exists around the
bubble. During the expansion and detachment stages,
particles collide with the bubble and stay on the liquid
lm. The particles and the liquid surrounding the
bubble are set in motion as the bubble grows and rises.
The volume of the bubble at the end of the rst
stage and during the second stage can be described by
considering a balance of all the forces acting on the
bubble if the instantaneous gas ow rate, Q
o
, or the
instantaneous gas velocity, u
o
, through the orice, is
known. The forces induced by the liquid include the
upward forces (eective buoyancy force, F
B
, and gas
momentum force, F
M
), and the downward resistance
(liquid drag, F
D
, surface tension force, F
s
, bubble iner-
tial force, F
I.g
, and Basset force, F
BA
) as shown in
Fig. 2. It is assumed that the particles aect the bubble
formation process only through two additional down-
ward forces on the bubble, that is, the particlebubble
collision force, F
C
, and the suspension inertial force,
F
I.m
. The suspension inertial force is due to the accel-
eration of the liquid and particles surrounding the
bubble. Therefore the overall force balance on the
bubble in this model can be written as
F
B
F
M
= F
D
F
s
F
BA
F
I.g
F
C
F
I.m
(1)
The expansion stage and the detachment stage fol-
low the same force balance equation, although the
expression for the same force in the two stages may
be dierent owing to dierent bubble moving velocities
in the two stages. The expressions for all the forces
under two stages are listed in Table 1. The particle
bubble collision force is merely the rate of momentum
change of the particles colliding with the bubble sur-
face. The suspension inertial force is derived from the
suspension ow eld around an accelerating bubble,
obtained from a particle image velocimetry (PIV) sys-
tem. The nal expression of the suspension inertial
force is
F
I.m
=
d
_ _ _
r
m
u
m
dV
_ _
dt
=
d
dt
r
m
4
3
pr
3
b
_ _
u
b
_ _
(2)
The coecient in Eq. (2) is equal to 3.86 for bubbles
formed in liquidsolid suspensions based on PIV mea-
surements (Luo et al., 1998a) and to 11,16 for bubbles
formed in liquids, corresponding to the added mass in
inviscid liquids (Milne-Thomson, 1955).
For bubble formation from a single orice without a
gas chamber, the motion equation of the rising bubble
itself is sucient to predict the initial bubble size. For
the case in which the orice is connected to a gas
chamber, the gas ow rate through the orice is not
constant and depends on pressure uctuations in both
the chamber and the bubble. In order to simulate
bubble formation under such conditions, the pressure
uctuations in the gas chamber and in the bubble must
be considered to account for the time-variant orice
gas ow rate as illustrated below.
The instantaneous gas ow rate through the orice
depends on the pressure dierence in the gas chamber,
P
c
, and inside the bubble, P
b
, as well as the ow resis-
Figure 2 The balance of all the forces acting on a growing
bubble. (From Luo et al., 1998a.)
Table 1 Expressions for the Forces Involved in the Bubble Formation
Process
Forces Expansion stage Detachment stage
F
B
p
6
d
3
b
(r
l
r
g
)g
Same as expansion stage
F
M
p
4
D
2
o
r
g
u
2
o
F
D
C
D
p
4
d
2
b
_ _
r
l
u
2
b
2
C
D
=
24
Re
_ _
F
s
pD
o
scos g Same as expansion stage
F
I.g
d
dt
r
g
p
6
d
3
b
_ _
u
b
_ _
F
BA
Not applicable
3
2
d
2
b
pr
l
m
l
_
_
t
0
du,dt
t t
_ dt
F
C
p
4
D
2
o
(1 e)e
s
r
s
u
2
e
p
4
d
2
b
e
s
r
s
u
2
F
I.m
d
_ _ _
r
m
u
m
dV
_ _
dt
= z
d
dt
r
m
p
6
d
3
b
_ _
u
b
_ _
Source: Luo et al. (1998a).
tance of the orice, which can be described by the
orice equation as given in Eq. (3). The pressure in
the gas chamber is evaluated by applying the rst law
of thermodynamics and choosing the gas chamber as
an open system as given in Eq. (4) (Wilkinson and van
Dierendonck, 1994). The pressure inside the bubble is
derived from the pressure balance at the bubbleliquid
interface governed by a modied Rayleigh equation
(Pinczewski, 1981). In order to simulate bubble forma-
tion in liquidsolid suspensions, the eect of particles
on the pressure balance at the bubbleliquid interface
must be considered. The inertial term in the modied
Rayleigh equation is modied by considering the con-
tribution of the suspension inertial force, as given in
Eq. (5) for two stages (Yang et al., 2000a).
P = P
c
P
b
=
Q
o
k
o
_ _
2
(3)
dP
c
dt
=
,
V
c
P
e
Q
g
P
c
Q
o
_ _
(4)
P
b
P
0
= r
m
r
b
3
d
2
r
b
dt
2

dr
b
dt
_ _
2
_ _
2s
r
b
4m
1
r
b
1
4
r
g
Q
o
1
4
pD
2
o
_ _
2
(5a)
for the expansion stage and
P
b
P
0
= r
m
r
b
3
d
2
x
dt
2

dr
b
dt
dx
dt
_ _
2s
r
b
4m
l
r
b
dr
b
dt

1
4
r
g
Q
o
1,4pD
2
o
_ _
2
(5b)
for the detachment stage. P
0
is the hydrostatic pressure
at the bubble surface. The four terms on the right-hand
side of Eq. (5) represent the contributions of inertial,
surface tension, viscous, and gas momentum forces,
respectively. Combining Eqs. (1) and (3) through (5),
and solving these coupled ordinary dierential equa-
tions simultaneously, the dynamic behavior of bubble
formation can be obtained. If a certain bubble detach-
ment criterion is used, the initial bubble size can be
estimated.
The eect of pressure on the initial bubble size
under various bubble formation conditions is shown
in Fig. 3. The solid lines in the gures represent the
model predictions. Under constant ow conditions
(N
c
_ 1), the pressure eect on the initial bubble size
is not signicant; however, under variable ow condi-
tions (N
c
> 1), pressure has a signicant eect on the
initial bubble size. The initial bubble size decreases
signicantly with an increase in pressure when
N
c
> 1. The dierent pressure eects can be explained
based on the model simulation results. The simulated
bubble formation time and instantaneous gas ow rate
through the orice are shown in Fig. 4. Under variable
ow conditions, both the bubble formation time and
the orice gas ow rate decrease with increasing pres-
sure, resulting in a signicant reduction in the initial
bubble size. On the other hand, for bubble formation
under constant ow conditions, the gas ow rate
through the orice is almost constant during the for-
mation process, and the bubble formation time slightly
Figure 3 Initial bubble size in liquidsolids suspensions as a
function of pressure and gas velocity for bubble formation
under various conditions: (a) single orice without a gas
chamber. (From Luo et al., 1998a.) (b) Single orice con-
nected to a gas chamber. (From Yang et al., 2000a.)
changes with pressure, causing insignicant pressure
eect.
Based on the model calculation, a comparison of the
magnitudes of various forces acting on the bubble at
the end of formation process in a slurry system is
shown in Fig. 5. As shown in the gure, the suspension
inertial force is the dominant downward force and the
buoyancy force is the major upward force. At low
pressures, the eect of gas momentum on bubble for-
mation is negligible; however, at high pressures, the gas
momentum force could play an important role in pro-
viding an upward force. It is noted that the Basset
force is also important in dictating the bubble forma-
tion process, especially at low pressures due to the
large bubble size and fast bubble acceleration. When
the pressure increases, both buoyancy and suspension
inertial forces decrease signicantly owing to reduced
bubble size, and all the forces are important for deter-
mining the size of the bubble at high pressures owing
to similar magnitudes.
2.2 Bubble Rise Characteristics
The bubble rise velocity is an important parameter in
characterizing the hydrodynamics and transport phe-
nomena of bubbles in liquids and liquidsolid suspen-
sions. The rise velocity of a single gas bubble depends
on its size: for small bubbles, the rise velocity also
strongly depends on liquid properties such as surface
tension and viscosity; for large bubbles, the rise velo-
city is insensitive to liquid properties (Fan, 1989).
Under limited conditions, the rise velocities of single
bubbles in liquidsolid suspensions were found to be
similar to those in highly viscous liquids (Massimilla et
al., 1961; Darton and Harrison, 1974). Liquidsolid
suspensions can thus be characterized as Newtonian
homogeneous media, but they often exhibit non-
Newtonian or heterogeneous behavior (Tsuchiya et
al., 1997). Dierences in uidizing media, pressure,
and temperature lead to dierent bubble rise charac-
teristics.
This section focuses on the bubble rise characteris-
tics in liquids and liquidsolid suspensions at elevated
pressure and temperature. A nonwater based liquid
medium is employed to illustrate the eect of liquid
properties on the bubble rise velocity. For a bubble
rising in a liquidsolid suspension, the bubble rise velo-
city is discussed in light of both the apparent homo-
geneous (or eective) properties of the suspension and
the recently evolved numerical prediction based on a
computational model for gasliquidsolid uidization
systems. Computational modeling will be covered later
in Sec. 4.
Figure 4 Simulated (a) bubble formation time and (b) instan-
taneous gas ow rate through the orice during the bubble
formation process in the liquidsolid suspension (D
o
=1.63
mm, V
c
=650 cm
3
, e
s
=0.18). (From Yang et al., 2000a.)
Figure 5 Comparison of various forces acting on a bubble
formed in the liquidsolid suspension at dierent pressures
(D
o
= 1.63 mm, V
c
= 650 cm
3
, e
s
= 0.18, Q
g
= 2 cm
3
/s).
(From Yang et al., 2000a.)
2.2.1 Single Bubble Rise Velocity in Liquids
In most applications of three-phase uidization
systems, the liquid phase is a hydrocarbon based
medium. The physical properties of hydrocarbon
liquids vary dramatically with pressure and tempera-
ture. Lin and Fan (1997) and Lin et al. (1998) devel-
oped various in-situ techniques to measure the
physical properties of a hydrocarbon liquid (e.g.,
Paratherm NF heat transfer uid) at elevated pres-
sures and temperatures. The hydrostatic weighing
method, the falling ball technique, and the emerging
bubble technique were used to measure in-situ liquid
density, viscosity, and surface tension, respectively.
Based on their measurements, the liquid and inter-
facial properties change signicantly with pressure
and temperature. For example, at room temperature,
as the pressure increases from 0.1 to 20 MPa, the
liquid density of Paratherm NF heat transfer uid
increases by approximately 5%, the liquid viscosity
increases by 65%, and the surface tension decreases
by 25% (Lin and Fan, 1997; Lin et al., 1998).
Therefore it is important to consider these changes
in processing data, developing models, and conduct-
ing numerical simulations, particularly the variations
of liquid viscosity and surface tension.
Since the bubble rise velocity depends on liquid
properties, it is expected that pressure and temperature
would also aect the bubble rise characteristics.
Krishna et al. (1994) studied the pressure eect on
the bubble rise velocity and found that the single bub-
ble rise velocity does not depend on the gas density
over the range of 0.1 to 30 kg/m
3
. The conclusion is
limited to a narrow range of pressures. Lin et al. (1998)
measured the rise velocity of single bubbles of known
sizes in the Paratherm NF heat transfer uid at various
pressures and temperatures. The pressure ranges from
0.1 to 19.4 MPa. Figure 6 shows the relationship
between the bubble rise velocity and the bubble size
for two dierent temperatures. The bubble size in the
gure is represented by the equivalent spherical dia-
meter, d
b
. For a given bubble size, u
b
tends to decrease
with increasing pressure at both temperatures. The
eects of pressure and temperature, or more directly,
the eects of physical properties of the gas and liquid
phases on the variation of u
b
with d
b
, could be repre-
sented or predicted most generally by the Fan
Tsuchiya equation (Fan and Tsuchiya, 1990) among
three predictive equations. The other two are the mod-
ied Mendelsons wave-analogy equation (Mendelson,
1967) by Maneri (1995) and a correlation proposed by
Tomiyama et al. (1995).
The FanTsuchiya equation, generalized for high-
pressure systems, can be written in a dimensionless
form:
u
/
b
= u
b
r
l
sg
_ _
1,4
=
_
Mo
1,4
K
b
r
r
l
_ _
5,4
d
/2
b
_ _
n
2c
d
/
b
r
r
l
_ _
d
/
b
2
_ _
n,2
_
1,n
(6)
where the dimensionless bubble diameter is given by
d
/
b
= d
b
r
l
g
s
_ _
1,2
(7)
Three empirical parameters, n, c, and K
b
, in Eq. (6)
reect three specic factors governing the rate of
bubble rise. They relate to the contamination level of
the liquid phase, to the varying dynamic eects of the
surface tension, and to the viscous nature of the
surrounding medium, respectively. The suggested
values of these parameters are
Figure 6 Eect of pressure on rise velocity of single bubbles
at dierent temperatures: (a) 278C and (b) 788C. (From Lin
et al., 1998.)
n =
0.8 for contaminated liquids
1.6 for purified liquids
_
(8a)
c =
1.2 for monocomponent liquids
1.4 for multicomponent liquids
_
(8b)
K
b
= max(K
b0
Mo
0.038
. 12) (8c)
where
K
b0
=
14.7 for aqueous solutions
10.2 for organic solvents/mixtures
_
(8d)
The modied Mendelsons equation is a special form
of the FanTsuchiya equation where the viscous term,
that is, the rst term on the right side of Eq. (6), is
omitted. Just as general as the FanTsuchiya equation
for bubbles in liquids, the correlation by Tomiyama et
al. (1995) given in terms of drag coecient,
C
D
=
4
3
grd
b
r
l
u
2
b
(9)
consists of three equations under dierent system
purities:
C
D
= max
_
min
16
Re
(1 0.15 Re
0.687
).
48
Re
_ _
.
8
3
Eo
Eo 4
_
(10a)
for puried systems,
C
D
= max
_
min
24
Re
(1 0.15 Re
0.687
).
72
Re
_ _
.
8
3
Eo
Eo 4
_
(10b)
for partially contaminated systems, and
C
D
= max
24
Re
(1 0.15 Re
0.687
).
8
3
Eo
Eo 4
_ _
(10c)
for suciently contaminated systems. It is noted that
u
b
can be obtained explicitly from the FanTsuchiya
equation [Eq. (6)] for a given d
b
as well as gas and
liquid physical properties, while it can only be obtained
implicitly from Tomiyamas correlation [Eq. (10)].
Based on the comparison of three predictive equa-
tions as shown in Fig. 6, the modied Mendelson equa-
tion, which is valid only under the inviscid condition,
has limited predictive capability at the low temperature
(Fig. 6a), suggesting that the viscous force predomi-
nates in the bubble rise process. On the other hand, at
the high temperature (Fig. 6b), there is a strong agree-
ment over the bubble size range of d
b
> 2 mm including
the sharp breakpoint/peak. This indicates that the
liquid used tends to behave as a pure inviscid liquid.
Note that over the pressure range from0.1 to 19.4 MPa,
the liquid viscosity varies from 29 to 48 mPa s at 278C,
whereas it is almost constant within a range from 4.7 to
5.2 mPa s at 788C (Lin et al., 1998).
The FanTsuchiya equation applied for the given
liquid, a pure (n = 1.6), multicomponent (c = 1.4),
and organic solvent (K
b0
= 10.2), demonstrates good
overall predictive capability except for the sharp peak
existing under the high temperature (Fig. 6b). The
equation by Tomiyama et al. (1995) also has good
general applicability, especially around the peak
point occurring near d
b
= 2 mm at 788C; however, it
tends to underestimate the bubble rise velocity over the
rest of the bubble size range.
The consistent dierence in u
b
prevailing between
0.1 and 19.4 MPa for d
b
> 2 mmis due to the signicant
increase in gas density (as large as a 200-fold increase
with pressure from 0.1 to 19.4 MPa). The density eect
is accounted for in FanTsuchiya equation in terms of
r
_
r
l
or in Tomiyamas equation in terms of both
r
_
r
l
and Eo. As can be seen from the equations
and gure, the density dierence between the continu-
ous liquid phase and the dispersed gas phase plays an
important role in determining bubble rise velocity,
especially for large bubbles.
The ReEo relationship is often utilized in repre-
senting the general rise characteristics of single bubbles
in liquids (Clift et al., 1978; Bhaga and Weber, 1981).
Figure 7 shows the ReEo relationship at dierent Mo
values. The thin background lines signify the general,
quantitative trend for the rise velocity of single bubbles
in puried Newtonian liquids under ambient condi-
tions, plotted with constant intervals of log Mo. The
gure shows the general agreement in correlation pre-
dictions. The experimental results under four condi-
tions are also plotted in the gure (Lin et al., 1998).
By employing accurate values for the physical proper-
ties of the liquid phase and the gas density at given
pressures and temperatures, the experimental results
can be successfully represented over the entire Eo
range by the FanTsuchiya equation. This equation
was also proven to represent single bubble rise velocity
in various liquids under ambient conditions (Tsuchiya
et al., 1997). So it is clear that much more is known
now than before about single bubbles rising in liquids
at high pressure and high temperature.
2.2.2 Single Bubble Rise Velocity in LiquidSolid
Suspensions
Much progress has also been made regarding single
bubble rise characteristics in liquidsolid suspensions
at high pressure and high temperature. This section
covers the eect of suspension properties on single
bubble rise velocity. Figure 8 shows the eect of pres-
sure on the bubble rise velocity in a uidized bed with
Paratherm NF heat transfer uid and 0.88 mm glass
beads at two dierent temperatures (Luo et al.,
1997b). At both temperatures, the bubble rise velocity
decreases with an increase in pressure for a given solids
holdup. The extent of the reduction is as high as 50%
from 0.1 to 17.3 MPa. A more drastic reduction in u
b
,
however, arises from the addition of solid particles.
While the particle eect is small at a low solids holdup
(e
s
- 0.4), the eect is appreciable at a high solids
holdup (e
s
= 0.545), especially for high liquid viscosity
(Fig. 8a). A comparison of bubble rise velocity at
26.58C and 87.58C, for the same solids holdup, indi-
cates that the viscosity eect appears to be signicant.
The bubble rise velocity in a uidized bed contain-
ing smaller particles (0.21 mm glass beads) is shown in
Fig. 9 (Luo et al., 1997b). The eect of particle size on
the bubble rise velocity can be seen from Figs. 8 and 9.
While the extent of decrease in bubble rise velocity
with increase in pressure is comparable between large
particles (0.88 mm glass beads) and small particles
(0.21 mm glass beads), the extent of decrease in bubble
rise velocity with increasing solids holdup is much
smaller for the smaller particles. This dierence in
the sensitivity of u
b
reduction to solids holdup varia-
tion is clearly seen for the high solids holdup cases.
In the presence of solid particles, as a rst approx-
imation, it can be assumed that the particles modify
Figure 7 The rise characteristics of single bubbles in liquids
represented by the ReEo relationship (symbols represent
experimental results in Paratherm NF heat transfer uid
under various pressure and temperature conditions;
and represent predictions by the FanTsuchiya equation
and correlation of Tomiyama et al. (1995) at regular intervals
of Mo values, respectively;
___
represents predictions by the
FanTsuchiya equation at measured Mo values for compar-
ison with experimental results. (From Lin et al., 1998.)
Figure 8 Eect of pressure on bubble rise velocity in a ui-
dized bed of 0.88 mm glass beads at (a) 26.58C and (b) 87.58C
(solids holdups for , open, and lled symbols are 0, 0.384,
and 0.545, respectively). (From Luo et al., 1997b.)
only homogeneous properties of the surrounding
medium. Based on this homogeneous approach, the
FanTsuchiya equation, Eq. (6), can be extended to
liquidsolid suspensions by replacing the liquid proper-
ties, r
l
and m
l
, by the eective properties of the liquid
solid suspension, r
m
and m
m
, respectively (Tsuchiya et
al., 1997). The eective density can be estimated by
r
m
= r
l
(1 e
s
) r
s
e
s
(11)
The eective viscosity of liquidsolid suspensions is
estimated by
m
m
m
l
= exp
Ke
s
1 (e
s
_
e
sc
)
_ _
(12)
with two parameters correlated by Luo et al. (1997b):
K =
3.1 1.4 tanh [0.3(10 10
2
u
t
)]
f
(13a)
and
e
sc
= 1.3 0.1 tanh[0.5(10 10
2
u
t
)]
_ _
e
s0
(13b)
where u
t
is in units of m/s. The ranges of applicability
of Eqs. (12) and (13) are 840 - r
l
- 1000 kg,m
3
;
1 - m
l
- 47 mPa s; 19 - s - 73 mN/m; 0 - e
s
-
0.95e
s0
; 7.9 10
4
- u
t
- 0.26 m/s; 0.88 - f _ 1; and
0.56 - e
s0
- 0.61.
The FanTsuchiya equation with constant values of
m
m
estimated from Eq. (12) predicts reasonably well
the general trend of bubble rise velocity variation in
liquidsolid suspensions as shown in Figs. 8 and 9.
However, a detailed match between the experimental
results and predictions appears to be dicult to attain
by assigning a constant value of m
m
for each condition.
A more elaborate analysis is required to account for
the eect of bubble size on interactions of the bubble
with the surrounding medium (non-Newtonian
approach) or with individual particles (heterogeneous
approach).
The eect of particle wettability on the bubble rise
velocity was examined by Tsutsumi et al. (1991a,b).
Nonwettable particles have a larger contact angle
than wettable particles. That is to say that nonwettable
particles favor contact between bubbles and solids.
Tsutsumi et al. (1991a,b) observed that particle
particle aggregates and particlebubble aggregates are
formed at low gas velocities with nonwettable particles.
Consequently, the bubble rise velocity is smaller than
that for wettable particles. Further, the liquid velocity
for minimum uidization is lower and the liquid velo-
city for transition to the transport regime is higher for
wettable particles than for nonwettable particles. On
the other hand, a negligible eect of particle wettability
was observed on the rise velocity of large bubbles
(d
b
> 15 mm). For large bubbles with circular cap
shapes, the attachment of particles to the bubbles
occurs only at the bubble base and was not observed
on the bubble roof owing to the uid shear eects
caused by the fast rising bubbles.
2.3 Bubble Coalescence
The coalescence of bubbles to form larger bubbles is
another quantiable area of study in three-phase ui-
dization, which is important for determining the actual
bubble size in the system. Much work has been accom-
plished to understand how bubble coalescence occurs.
For gasliquid systems, the experimental results avail-
able in the literature indicate that an increase of pres-
sure retards the bubble coalescence (Sagert and Quinn,
Figure 9 Eect of pressure on bubble rise velocity in a ui-
dized bed of 0.21 mm glass beads at (a) 26.58C and (b) 87.58C
(solids holdups for +, open, and lled symbols are 0, 0.381,
and 0.555, respectively). (From Luo et al., 1997b.)
1977, 1978). There are three steps in the bubble coales-
cence process (Vrij, 1966; Chaudhari and Homann,
1994): (1) approach of two bubbles to form a thin
liquid lm between them; (2) thinning of the lm by
the drainage of the liquid under the inuence of gravity
and suction due to capillary forces; and (3) rupture of
the lm at a critical thickness. The second step is the
rate-controlling step in the coalescence process, and
the bubble coalescence rate can be approximated by
the lm thinning rate (Vrij, 1966). The lm thinning
velocity can be expressed as (Sagert and Quinn, 1977,
1978)
dl
dt
=
32l
3
s
3fR
2
d
m
l
d
b
(14)
where the parameter f is a measure of the surface drag
or velocity gradient at the surface due to the adsorbed
layer of the gas.
It is known that surface tension decreases and liquid
viscosity increases with increasing pressure. In addi-
tion, f increases with pressure. As seen from Eq.
(14), all of these variations contribute to the reduction
of the lm thinning velocity, and hence the bubble
coalescence rate, as pressure increases. As a result,
the time required for two bubbles to coalesce is longer
and the rate of overall bubble coalescence in the bed is
reduced at high pressures. Moreover, the frequency of
bubble collision decreases with increasing pressure. An
important mechanism for bubble collision is the bub-
ble wake eect (Fan and Tsuchiya, 1990). When the
dierences in bubble size and bubble rise velocity are
small at high pressures, the likelihood of small bubbles
being caught and trapped by the wakes of large bub-
bles decreases. Therefore bubble coalescence is sup-
pressed by the increase in pressure, due to the longer
bubble coalescence time and the smaller bubble colli-
sion frequency.
The presence of pulp, even at very low pulp consis-
tencies (0.1%) in the column leads to enhanced bubble
coalescence and hence a narrowing of the gas velocity
for the dispersed bubble regime as the pulp consistency
increases (Reese et al., 1996). Bubble coalescence inhi-
bitors such as inorganic salts (e.g., sodium chloride and
sodium phosphate dibasic) and organic compounds
(e.g., ethanol, n-pentanol, iso-amyl alcohol, and ben-
zoic acid) can be eectively applied to the liquid at
concentrations up to 200 ppm to inhibit bubble coales-
cence behavior in three-phase uidization (Briens et
al., 1999). With the addition of the bubble coalescence
inhibitor, the bed hydrodynamics at low gas velocities
are signicantly dierent from the case without the
inhibitor, and the inuence of the gas distributor
becomes marked (Nacef et al., 1995).
2.4 Bubble Breakup
Another quantiable area of bubble dynamics is bub-
ble breakup. Bubble breakup is a key phenomenon to
study because it determines the bubble size in the sys-
tem and hence aects heat and mass transfer between
phases. Bubble breakup has been investigated theore-
tically and experimentally. There are two causes for
bubble breakup: bubbleparticle collision and bubble
instability. In this section, only bubble breakup due to
bubble instability will be discussed.
It is known that the variation of bubble size with
pressure is the key for understanding pressure eects
on hydrodynamics. The upper limit of the bubble size
is set by the maximum stable bubble size, D
max
, above
which the bubble is subjected to breakup and hence is
unstable. Several mechanisms have been proposed for
the bubble breakup phenomenon, and based on these
mechanisms theories have been established to predict
the maximum bubble size in gasliquid systems. In
this section, the mechanisms of bubble breakup and
the theories to predict the maximum bubble size are
covered.
Hinze (1955) proposed that bubble breakup is
caused by the dynamic pressure and the shear stresses
on the bubble surface induced by dierent liquid ow
patterns, e.g., shear ow and turbulence. When the
maximum hydrodynamic force in the liquid is larger
than the surface tension force, the bubble disintegrates
into smaller bubbles. This mechanism can be quanti-
ed by the liquid Weber number. When the Weber
number is larger than a critical value, the bubble is
not stable and disintegrates. This theory was adopted
to predict the breakup of bubbles in gasliquid systems
(Walter and Blanch, 1986). Calculations by Lin et al.
(1998) showed that the theory underpredicts the
maximum bubble size and cannot predict the eect
of pressure on the maximum bubble size.
A maximum stable bubble size exists for bubbles
rising freely in a stagnant liquid without external stres-
ses, e.g., rapid acceleration, shear stress, and/or turbu-
lence uctuations (Grace et al., 1978). RayleighTaylor
instability has been regarded as the mechanism for
bubble breakup under such conditions. A horizontal
interface between two stationary uids is unstable to
disturbances with wavelengths exceeding a critical
value if the upper uid has a higher density than the
lower one (Bellman and Pennington, 1954):
l
c
= 2p
s
g(r
l
r
g
)
_
(15)
Grace et al. (1978) modied the RayleighTaylor
instability theory by considering the time available
for the disturbance to grow and the time required for
the disturbance to grow to an adequate amplitude.
Batchelor (1987) pointed out that the observed size
of air bubbles in water was considerably larger than
that predicted by the model of Grace et al. (1978).
Batchelor (1987) further took into account the stabiliz-
ing eect of the liquid acceleration along the bubble
surface and the nonconstant growth rate of the distur-
bance. In Batchelors model, the magnitude of the dis-
turbances is required to predict maximum bubble size;
however, the magnitude of the disturbances is not
known. The models based on the RayleighTaylor
instability predict an almost negligible pressure eect
on the maximum bubble size; in fact, Eq. (15) implies
that the bubble is more stable when the gas density is
higher.
The KelvinHelmholtz instability is similar to the
RayleighTaylor instability, except that the former
allows a relative velocity between the uids, u
r
. Using
the same concept of Grace et al. (1978), Kitscha and
Kocamustafaogullari (1989) applied the Kelvin
Helmholtz instability theory to model the breakup of
large bubbles in liquids. Wilkinson and van
Dierendonck (1990) applied the critical wavelength to
explain the maximum stable bubble size in high-
pressure bubble columns:
l
c
=
2p
s
g(r
l
r
g
)
_
r
l
r
l
r
g
r
g
u
2
r
2
sg(r
l
r
g
)
_
1
r
2
l
r
2
g
u
4
r
4(r
l
r
g
)
2
sg(r
l
r
g
)
_
(16)
Disturbances in the liquid with a wavelength larger
than the critical wavelength can break up a bubble.
Equation (16) indicates that the critical wavelength
decreases with an increase in pressure and therefore
bubbles are easier to break apart by disturbances at
higher pressures. However, the critical wavelength is
not equivalent to the maximum stable bubble size,
and Eq. (16) alone cannot account for the eect of
pressure on the maximum bubble size.
All of the models mentioned above do not account
for the internal circulation of the gas. The internal
circulation velocity is of the same order of magnitude
as the bubble rise velocity. A centrifugal force is
induced by this circulation, pointing outward toward
the bubble surface. This force can suppress the distur-
bances at the gasliquid interface and thereby stabilize
the interface. Centrifugal force explains the underesti-
mation of D
max
by the model of Grace et al. (1978). On
the other hand, the centrifugal force can also break
apart the bubble, as it increases with an increase in
bubble size. The bubble breaks up when the centrifugal
force exceeds the surface tension force, especially at
high pressures when gas density is high. Levich
(1962) assumed the centrifugal force to be equal to
the dynamic pressure induced by the gas moving at
the bubble rise velocity, that is, k
f
r
g
u
2
b
,2(k
f
- 0.5),
and proposed a simple equation to calculate the
maximum stable bubble size:
D
max
-
3.63s
u
2
b
r
2
l
r
g
3
_ (17)
Equation (17) severely underpredicts the maximum
bubble size in the airwater system, although it
shows a signicant eect of pressure on the maximum
bubble size.
An analytical criterion for the bubble breakup can
be derived by considering a single large bubble rising in
a stagnant liquid or slurry at a velocity of u
b
, without
any disturbances on the gasliquid interface. The bub-
ble is subjected to breakup when its size exceeds the
maximum stable bubble size due to the circulation-
induced centrifugal force. Schematic of the internal
circulation model for bubble breakup is shown in
Fig. 10 (Luo et al., 1999). Large bubbles normally
assume a spherical cap shape, and in the model, the
spherical cap bubble is approximated by an ellipsoidal
bubble with the same volume and the same aspect ratio
(height to width). The circulation of gas inside the
bubble can be described by Hills vortex (Hill, 1894).
To model bubble breakup, it is necessary to evaluate
the x-component of the centrifugal force, F
x
, induced
by the circulation on the entire bubble surface as
shown in Fig. 10. A rigorous theoretical derivation
from Hills vortex yields the expression for F
x
:
F
x
=
9pr
g
u
2
b
a
2
64
2
_
a
(18)
The surface tension force is the product of the surface
tension and the circumference of the bubble,
F
s
= sL = s
_
ellipse
(dr
c
)
2
(dz)
2
_
= 4saE(
1 a
2
_
)
(19)
Also, the volume equivalent bubble diameter, d
b
, is
related to a and a by
a =
d
b
8a
3
_ (20)
Note that the centrifugal force is aected signicantly
by the gas density, the aspect ratio of the bubble, the
bubble size, and the bubble rise velocity. The bubble is
not stable if F
x
is larger than F
s
, that is,
u
2
b
d
b
_
8a
4,3
E(
1 a
2
_
)
0.312
s
r
g
(21)
When the centrifugal force is larger than the surface
tension force, the bubble would be stretched in the
x-direction. During the stretching, the aspect ratio, a,
becomes smaller while d
b
and u
b
can be assumed to
remain constant. As a result, the centrifugal force
increases, the surface tension force decreases, and the
bubble stretching becomes an irreversible process.
Using the DaviesTaylor equation (Davies and
Taylor, 1950) for the bubble rise velocity, the maxi-
mum stable bubble size is expressed by
D
max
- 7.16a
2,3
E(
1 a
2
_
)
1,2
s
gr
g
_
(22)
The simplied forms of Eq. (22) are:
D
max
- 2.53
s
gr
g
_
for a = 0.21 ( ) (23a)
in liquids, and
D
max
- 3.27
s
gr
g
_
for a = 0.3 ( ) (23b)
in liquidsolid suspensions. Further, based on the
DaviesTaylor equation, the rise velocity of the max-
imum stable bubble is
u
max
=
1.6sg
r
g
_ _
1,4
(24)
The comparison of experimental maximum bubble
sizes and the predictions by various instability theories
is shown in Fig. 11. The internal circulation model can
reasonably predict the observed pressure eect on the
maximum bubble size, indicating that the internal cir-
culation model captures the intrinsic physics of bubble
breakup at high pressures. The comparison of the pre-
dictions by dierent models further indicates that bub-
ble breakup is governed by the internal circulation
mechanism at high pressures over 1.0 MPa, whereas
the RayleighTaylor instability or the Kelvin
Helmholtz instability is the dominant mechanism at
low pressure.
Figure 10 Schematic of the internal circulation model for bubble breakup: (a) internal and external ow elds; (b) circulation
velocity on surface S; (c) force balance and 3-D view of surface S and the ow pattern on S. (From Luo et al., 1999.)
3 TRANSPORT PHENOMENA
Bubble dynamics and characteristics discussed above
determine the hydrodynamic and heat and mass trans-
fer behaviors in three-phase uidization systems, which
is important for better design and operation of three-
phase uidized beds. In this section, various hydrody-
namic variables and transfer properties in three-phase
systems are discussed. Specically, areas discussed in
the hydrodynamics section are minimum uidization,
bed contraction and moving packed bed phenomenon,
ow regime transition, overall gas holdup and hydro-
dynamic similarity, and bubble size distribution and
the dominant role of larger bubbles. Later in this sec-
tion, important topics covering transport phenomena
will be discussed, which include heat and mass transfer
and phase mixing.
3.1 Hydrodynamics
Flow behavior of cocurrent upward gasliquid ow to
a bed of particles is characterized by a complex inter-
action among gas, liquid, and solid phases. A typical
ow regime map for three-phase uidized beds is
shown in Fig. 12 (Zhang et al., 1997). Seven dierent
ow regimes are identied in this map, which vary with
gas and liquid velocities. The regime map is operating
condition specic, and the ow regime map shown in
Fig. 12 is obtained in an airwater1.5 mm glass bead
three-phase uidized bed. The uidized bed is 8.26 cm
in diameter and 2 m in height. This map puts the com-
mon three-phase uidization operating conditions
represented by the dispersed bubble and coalesced
bubble ow regimes in proper perspective.
3.1.1 Minimum Fluidization
The rst topic of study for the hydrodynamics of three-
phase uidization is minimum uidization. For a given
gas velocity, the minimum liquid ow rate required to
uidize a bed of particles (U
lmf
) may generally be deter-
mined from the change in the bed dynamic pressure
drop behavior that occurs as the bed changes from a
xed bed to a uidized bed. There are considerable
variations on minimum uidization phenomena
among small/light, large/heavy, and mixed particle sys-
tems. However, U
lmf
in general can be evaluated
mechanistically by considering an intrinsic condition
for minimum uidization where the total pressure
drop over a bed of particles at the xed state is equal
to the total bed weight per unit bed cross-sectional area
as formulated by Song et al. (1989). In this evaluation,
the pressure drop in the xed bed can be described by a
ow model developed by Chern et al. (1983, 1984).
Figure 11 Comparison of maximum stable bubble size obtained experimentally and the predictions by various models. (From
Fan et al., 1999.)
The minimum uidization velocity can also be eec-
tively described based on the gas-perturbed liquid
model (Zhang et al., 1998a). The model assumes that
the solid particles in the bed are fully supported by the
liquid and there is little direct interaction between gas
and solid phases. In this model, the gas phase plays a
simple role in the bed by occupying space with particles
uidized by the liquid. The minimum uidization velo-
city along with minimum liquid velocity for particle
transport for three-phase uidized beds of relatively
large particles (d
p
> 1 mm) (Zhang et al., 1998b) can
be well predicted by this model. Naturally, the model is
more suitable to be applied to systems where bubble or
wake induced ow is insignicant. For example, bed
contraction, which is a typical phenomenon of the bub-
ble wake, cannot be described by this model. The mini-
mum uidization velocity for non-Newtonian uid
following the power-law model can be obtained by
equating the pressure drop in a xed bed to that in a
uidized bed, similar to the uidization of Newtonian
uids; the pressure drop in the xed bed can be
evaluated by the free-surface cell model for non-
Newtonian uids (Miura and Kawase, 1998). New
approaches in examining the minimum uidization
condition include articial neural network (Larachi et
al., 2000) for predicting the minimum uidization velo-
city and V-statistics (Briens et al., 1997) for identifying
minimum uidization of the bed. These approaches
would not replace the more mechanistic based and
currently used techniques in the quantication of mini-
mum uidization conditions.
3.1.2 Bed Contraction and Moving Packed Bed
Phenomenon
For three-phase uidization systems, two striking
phenomena pertaining to macroscopic hydrodynamic
behavior are bed contraction and moving packed bed
ow. These two topics will be covered in this section.
3.1.2.1 Bed Contraction. Three-phase uidized
beds using small particles display unique bed expan-
sion characteristics, that is, upon initial introduction
of the gas into the liquidsolid uidized bed, con-
traction, instead of expansion, of the bed occurs
(Massimilla et al., 1959). An increasing gas ow rate
causes further contraction up to a critical gas ow
rate beyond which the bed expands (Turner, 1964;
Oestergaard, 1964). A quantitative elucidation of the
bed contraction phenomenon was reported by Stew-
art and Davidson (1964) and El-Temtamy and Ep-
stein (1979). Basically, bed contraction is caused by
the behavior of the bubble wake, which entraps li-
quid and particles and therefore is associated with
large bubble systems. The entrainment of the liquid
and particles by the bubble wake reduces the effective
amount of liquid in the bed used to uidize the re-
maining particles. The bed contraction phenomenon
has been extensively studied under ambient uidiza-
tion conditions. Bed contraction has also been ob-
served at high pressure (Jiang et al., 1997).
On the basis of the generalized wake model of
Bhatia and Epstein (1974), a criterion for the bed con-
traction was developed (Jean and Fan, 1987). In the
Figure 12 (a) Flow regime map for an airwater1.5 mm
glass bead three-phase uidized bed (shaded portions refer to
preuidized and postuidized regimes); (b) schematic dia-
gram of corresponding ow regimes. (From Zhang et al.,
1997.)
generalized wake model, the three-phase uidized bed
is assumed to consist of three regions, that is, the gas
bubble region, the wake region, and the liquidsolid
uidized region. Bed contraction will occur when
c =
n
n 1
k
_ _
U
l
e
l
xk(U
g
_
e
g
)
n 1
1 k ( )U
l
k
n 1
U
g
e
g
U
l
e
l
_ _ _ _
xk U
l
U
l
e
l
_ __
n
n 1
_
- 0 (25)
Here, n is the RichardsonZaki index, k is the ratio of
wake size to bubble size, and x is the ratio of solids
concentration in the wake region to that in the liquid
solid uidized region.
3.1.2.2 Moving Packed Bed Phenomenon. Mov-
ing packed bed ow is characterized by the motion
of solids in piston ow in a three-phase uidized bed.
Moving packed bed ow, which usually occurs dur-
ing start-up, depends not only on the gas and liquid
velocities but also on how they are introduced into
the bed. Moving packed bed ow is caused by the
surface phenomena involving ne bubbles attached
onto particles and the subsequent formation of a ne
bubble blanket under the packed solids; a liquid ow
would move the entire bed upward. This phenomen-
on is thus associated with the small bubble system.
The moving packed bed ow phenomenon in a three-
phase uidized bed is a known, anomalous event in
the resid hydrotreating industry. It was observed in
the 1960s in the bench and pilot units during the
development and commercialization of the resid
hydrotreating process (Fan, 1999). The reactor was
typically operated at pressures between 5.5 and
21 MPa and temperatures between 300 and 4258C. In
the early 1970s, moving packed bed ow was
observed in a commercial three-phase uidized bed
reactor. The occurrence of a moving packed bed in a
three-phase uidized bed could simply be circum-
vented by utilizing a start-up procedure that involves
degassing the bed rst and then introducing liquid
ow to expand the bed prior to commencing gas
ow. Commercial operators of three-phase uidized
bed reactors have long recognized and undertaken a
proper start-up procedure of this nature since obser-
ving this anomalous event. As the small bubbles can
also be generated under ambient conditions using
surfactants in an airwater system, the moving
packed bed ow phenomenon was reported in open
literature rst by Saberian-Broudjenni et al. (1984)
and later by Bavarian and Fan (1991a,b) in small
columns with small bubbles generated in the same
manner.
3.1.3 Flow Regime Transition
Two main ow regimes are commonly identied for
three-phase uidization systems based on the bubble
ow behavior: the dispersed (or homogeneous) bubble
ow and the coalesced bubble (or churn-turbulent)
ow regimes. Knowledge of the transition from the
homogeneous bubble ow to the churn-turbulent
ow regimes is important for the design and operation
of industrial reactors. The transition velocity is dened
as the supercial gas velocity at which the transition
from homogeneous bubble ow to churn-turbulent
ow occurs. The transition velocity depends on gas
distributor design, physical properties of the phases,
operating conditions, and column size. The ow
regime transition has been studied extensively under
ambient conditions over the last three decades
(Wallis, 1969; Joshi and Lali, 1984; Shnip et al.,
1992; Tsuchiya and Nakanishi, 1992; Zahradnik et
al., 1997). Most of these studies pointed out a critical
role played by the liquid-phase turbulence during the
transition and employed phenomenological models to
predict the ow transition.
The regime transition is also inuenced by the oper-
ating pressure, and the pressure eect has been exam-
ined by many researchers in gasliquid systems (Tarmy
et al., 1984; Clark, 1990; Krishna et al., 1991, 1994;
Wilkinson et al., 1992; Hoefsloot and Krishna, 1993;
Reilly et al., 1994; Letzel et al., 1997; Lin et al., 1999).
The ow regime transition is normally identied based
on the following approaches: instability theory, analy-
sis of uctuation signals, and the drift ux model.
Based on the stability theory of Batchelor (1988) and
Lammers and Biesheuvel (1996), the inuence of pres-
sure on the stability of bubbly ows in a nitrogen
water system was identied (Letzel et al., 1997). It is
found that a higher gas density has a stabilizing eect
on the ow and that the gas fraction at the instability
point (i.e., transition point) increases with gas density,
while the gas velocity at the instability point only
slightly increases with gas density. However, the con-
clusion is limited to a narrow range of operating pres-
sures (0.1 to 1.3 MPa). Using the standard deviation of
the pressure uctuation and the drift ux model, the
ow transition from the homogeneous bubble ow
regime to the churn-turbulent ow regime in a nitro-
genParatherm NF heat transfer uid system was
investigated over a wide range of operating conditions
(e.g., pressures up to 15.2 MPa and temperatures up to
788C) (Lin et al., 1999). The eects of pressure and
temperature on the transition velocity are clearly
shown in Fig. 13a. Increasing pressure or temperature
delays the regime transition. The pressure eect on the
ow regime transition is mainly due to the variation in
bubble characteristics, such as bubble size and bubble
size distribution. Bubble size and bubble size distribu-
tion are closely associated with factors such as initial
bubble size, bubble coalescence rate, and breakup rate.
Under high pressure conditions, bubble coalescence is
suppressed and bubble breakup is enhanced. Also, the
distributor tends to generate smaller bubbles. All these
factors contribute to small bubble sizes and narrow
bubble size distributions and consequently delay the
ow regime transition at high pressures.
A correlation was also proposed to estimate the gas
holdup and gas velocity at the transition point under
high-pressure conditions (Wilkinson et al., 1992). This
predictive scheme incorporates the concept of the
bimodal bubble size distribution presented by
Krishna et al. (1991), that is, the churn-turbulent
regime is characterized by a bimodal bubble size dis-
tribution, consisting of fast rising large bubbles
(d
b
> 5 cm) and small bubbles (typically, d
b
- 5 mm).
They found that the transition velocity depends on the
physical properties of the liquid and can be estimated
by the following correlations:
e
g.tran
=
U
g.tran
u
small
= 0.5 exp(193r
0.61
g
m
0.5
l
s
0.11
) (26)
u
small
= 2.25
s
m
l
_ _
s
3
r
l
gm
4
l
_ _
0.273
r
l
r
g
_ _
0.03
(27)
where u
small
is the rise velocity of small bubbles. As
shown in Fig. 13a, the correlation can reasonably
predict the transition velocity at a given temperature
and pressure when the in-situ physical properties of the
uids are used in the correlation.
The eects of pressure, particle size, and liquid velo-
city on the regime transition velocity in a three-phase
uidized bed are shown in Fig. 13b (Luo et al., 1997a).
The transition velocity is identied by analyzing the
drift ux of gas. The drift ux of gas increases with
the gas holdup in the dispersed regime; in the coalesced
bubble regime, the rate of increase is much larger. As
the pressure increases, the transition gas velocity and
the gas holdup at the transition point increase. The
pressure eect on the regime transition is signicant
at low pressures, but the eect levels o at a pressure
around 6 MPa. The transition velocity increases with
liquid velocity and slightly increases with particle size,
similar to the regime transition behavior at ambient
conditions. The addition of ne particles to the liquid
phase promotes bubble coalescence and thus acceler-
ates the transition to the churn-turbulent regime
(Clark, 1990).
3.1.4 Overall Gas Holdup and Hydrodynamic
Similarity
Gas holdup is a key parameter to characterize the
macroscopic hydrodynamics of three-phase uidiza-
tion systems. The gas holdup depends on gas and
liquid velocities, gas distributor design, column geo-
metry (diameter and height), physical and rheological
Figure 13 The regime transition velocity (a) in a bubble
column. Open symbols are obtained by standard deviation
of pressure uctuation and drift ux model; closed symbols
are calculated by the correlation of Wilkinson et al. (1992).
(From Lin et al., 1999.) (b) In a three-phase uidized bed.
(From Luo et al., 1997a.)
properties of the gas and liquid, particle concentration,
and physical properties of the particles. The gas holdup
generally increases with gas velocity, with a larger rate
of increase in the dispersed bubble regime than in
the churn-turbulent regime. Such distributors as
perforated plates, nozzle injectors, and spargers aect
the gas holdup only at low gas velocities. The eect of
column size on gas holdup is negligible when
the column diameter is larger than 0.1 to 0.15 m
(Shah et al., 1982). The inuence of the column height
is insignicant if the height is above 1 to 3 m and the
ratio of the column height to the diameter is larger
than 5 (Kastaneck et al., 1984). Gas holdup
decreases as liquid viscosity and/or gasliquid surface
tension increase; however, the eect of liquid density
is not clear. The addition of particles into a bubble
column leads to a larger bubble size and thus a
decreased gas holdup, especially when the particle
concentration is low. The particle size eect on the
gas holdup can be ignored in the particle size range
of 44 to 254 mm. When highly viscous pseudoplastic
liquids such as xanthan solution with an eective
viscosity of 300 mPa s are used, the gas holdup
increases with decreasing liquid viscosity and particle
size when the liquid velocity is less than 6 cm/s
(Zaidi et al., 1990).
Numerous studies have been conducted to investi-
gate the eects of pressure and temperature on the gas
holdup in bubble columns (Deckwer et al., 1980;
Tarmy et al., 1984; Idogawa et al., 1986; Kojima et
al., 1991; Wilkinson et al., 1992; Reilly et al., 1994;
Daly et al., 1995; Jiang et al., 1995; Inga, 1997;
Letzel et al., 1997; Lin et al., 1998) and three-phase
uidized beds (Luo et al., 1997a). Empirical correla-
tions have been proposed for gas holdup in bubble
columns operated at elevated pressure and tempera-
ture. It is commonly accepted that elevated pressures
lead to a higher gas holdup in both bubble columns
and three-phase uidized beds except in those systems
that are operated with porous plate distributors and at
low gas velocities. The increased gas holdup is directly
related to the smaller bubble size and, to a lesser
extent, to the slower bubble rise velocity at higher pres-
sures (Luo et al., 1997b). Figure 14 shows bubbles
emerging from the three-phase uidized bed of
Paratherm NF heat transfer uid and 2.1 mm glass
beads over a wide range of operating conditions. As
shown in the gure, bubble size is drastically reduced
as pressure increases. The most fundamental reason for
the bubble size reduction can be attributed to the var-
iation in physical properties of the gas and liquid with
pressure.
The gas holdup in slurry bubble columns at high
pressures was also investigated by some researchers
(Deckwer et al., 1980; Kojima et al., 1991; Daly et
al., 1995; Inga, 1997; Luo et al., 1999). A dynamic
gas disengagement technique has been proven suitable
for the measurement of gas holdup in a slurry bubble
column under high-pressure conditions (Luo et al.,
1999). The pressure eect on the gas holdup in slurry
bubble columns is similar to that in bubble columns.
Elevated pressures lead to higher gas holdups in a
slurry bubble column. The presence of particles
reduces the gas holdup at both ambient and elevated
pressures as shown in Fig. 15. An empirical correlation
was obtained to estimate the gas holdup in high-
pressure slurry bubble columns as
e
g
1 e
g
=
2.9 U
4
g
r
g
,sg
_ _
a
r
g
,r
m
_ _
b
cosh Mo
0.054
m
_ _ _ _
4.1
(28)
where Mo
m
is the modied Morton number for the
slurry phase, g r
m
r
g
_ _
m
l
_ _
4
,r
2
m
s
3
, and
a = 0.21 Mo
0.0079
m
(29a)
b = 0.096 Mo
0.011
m
(29b)
Figure 14 Visualization of bubbles emerging from the sur-
face of a three-phase uidized bed at (a) P = 0.1 MPa; (b)
P = 3.5 MPa; (c) P = 6.8 MPa; (d) P = 17.4 MPa. (From
Fan et al., 1999.)
A correction factor x accounts for the eect of particles
on the slurry viscosity
ln = 4.6e
s
5.7e
0.58
s
sinh 0.71 exp 5.8e
s
( ) [
_
ln Mo ( )
0.22
_
1
_
(30)
The physical meaning of the dimensionless group of
U
4
g
r
g
,sg in the gas holdup correlation can be shown
by substituting the rise velocity of the maximum stable
bubble [Eq. (24)] into the group:
U
4
g
r
g
sg
o
U
g
u
max
_ _
4
(31)
Clearly, the dimensionless group represents the contri-
bution of large bubbles to the overall gas holdup,
which is the major reason why the correlation is able
to cover a wide range of experimental conditions.
For high-pressure bubble columns and slurry bub-
ble columns operated under a wide range of condi-
tions, hydrodynamic similarity requires the following
dimensionless groups to be the same: U
g
,u
max
, Mo
m
,
and r
g
,r
m
(Luo et al., 1999). To simulate the hydro-
dynamics of industrial reactors, cold models could be
used and milder pressure and temperature conditions
could be chosen, as long as the three groups are similar
to those in the industrial reactor. Safoniuk et al. (1999)
identied eight key variables that aect hydrodynamics
of three-phase uidization systems, yielding ve
independent dimensionless groups according to the
Buckingham PI theorem,
Mo =
grm
4
l
r
2
l
s
3
Eo
/
=
grd
2
p
s
Re
/
=
r
l
d
p
U
l
m
l
b
d
=
r
p
r
l
b
U
=
U
g
U
l
(32)
Gas holdup data in the freeboard region of the bed and
the extent of the bed expansion from two cold models
(one 8.3 cm ID and the other 91.4 cm ID) were used in
their analysis, which satisfactorily verify the hydro-
dynamic similarity. More studies are needed based
on these two approaches (Luo et al., 1999; Safoniuk
et al., 1999) to arrive at general similarity rules for
scale-up applications.
The hydrodynamic properties such as phase holdup
for reaction systems that actually generate gas (e.g.,
bioreaction, Buere et al., 1998) and consume gas
(e.g., methanol synthesis, Fan, 1989) are dierent
from those obtained from gas injected systems. The
gas holdup in the case of the bioreaction system was
found to be higher than that in the gas injected system
at the same gas ow rate, owing to small bubbles gen-
erated in the bioreaction system (Buere et al., 1998).
3.1.5 Bubble Size Distribution and Dominant Role
of Large Bubbles
The bubble size can be measured by photographic or
probe techniques. In multibubble systems, a mean
bubble size is usually used to describe the system.
The mean bubble size is commonly expressed through
the Sauter, or volume-surface, mean. For a group of
bubbles with measured diameters, the Sauter mean is
d
vs
=
n
i
d
3
bi
n
i
d
2
bi
(33)
where n
i
is the number of bubbles in the class i with its
volume equivalent size d
bi
.
Figure 15 Eects of (a) pressure and (b) solids concentra-
tion on the gas holdup in a slurry bubble column (nitrogen-
Paratherm NF heat transfer uid100 mm glass beads).
(From Luo et al., 1999.)
Studies have been conducted to investigate pressure
or gas density eects on the mean bubble size and
bubble size distribution in bubble columns (Idogawa
et al., 1986, 1987; Jiang et al., 1995; Soong et al., 1997;
Lin et al., 1998) as well as in three-phase uidized beds
(Jiang et al., 1992, 1997). According to these experi-
mental studies, pressure has a signicant eect on the
mean bubble diameter. The mean bubble diameter
decreases with increasing pressure; however, above a
certain pressure, the bubble size reduction is not sig-
nicant. The eect of pressure on the mean bubble size
is due to the change of bubble size distribution with
pressure. At atmospheric pressure, the bubble size dis-
tribution is broad, while under high pressure, the bub-
ble size distribution becomes narrower, as shown in
Fig. 16 (Luo et al., 1999). The bubble size is mainly
determined by three factors, that is, bubble formation
at the gas distributor, bubble coalescence, and bubble
breakup. When the pressure is increased, the bubble
size at the distributor is reduced (Luo et al., 1998a),
bubble coalescence is suppressed (Jiang et al., 1995),
and large bubbles tend to break up (Luo et al., 1999).
The combination of these three factors causes the
decrease of mean bubble size with increasing pressure.
The bubble size distribution can normally be
approximated by a log-normal distribution with its
upper limit at the maximum stable bubble size. The
contribution of bubbles of dierent sizes can be exam-
ined by analyzing the relationship between overall gas
holdup and bubble size distribution. In slurry bubble
columns, the gas holdup can be related to the super-
cial gas velocity, U
g
, and the average bubble rise
velocity, uu
b
(based on bubble volume), by a simple
equation:
U
g
= e
g
uu
b
(34)
When the distributions of bubble size and bubble rise
velocity are taken into account, uu
b
can be expressed as
uu
b
=
_
d
b.max
d
b.min
V
b
(d
b
)f (d
b
)u
b
(d
b
) dd
b
_
d
b.max
d
b.min
V
b
(d
b
)f (d
b
) dd
b
(35)
The outcome of Eq. (35) and the gas holdup
strongly depend on the existence of large bubbles,
because of their large volume and high rise velocity.
An experimental study by Lee et al. (1999) revealed
that, in the coalesced bubble regime, more than 70%
of the small bubbles are entrained by the wakes of
large bubbles and consequently have a velocity close
to large bubbles. It is clear that the large bubble beha-
vior characterizes the overall hydrodynamics due to
their large volume, their high rise velocity, and their
large associated wakes.
3.2 Heat Transfer
In the last several sections, hydrodynamic characteris-
tics of three-phase uidization systems were covered.
Starting from this section, several key transport phe-
nomena will be discussed. This section examines heat
transfer characteristics in three-phase uidization sys-
tems; mass transfer will be discussed in a later section.
Comprehensive reviews on the heat transfer beha-
vior in various gasliquid and gasliquidsolid systems
under ambient conditions are given by Kim and
Laurent (1991), and more recently by Kumar et al.
(1992, 1993a,b) and Kumar and Fan (1994). There
are two types of heat transfer coecients, that is,
heat transfer coecients between the column wall
and the bed (wall-to-bed), and heat transfer coecients
Figure 16 Bubble size distribution in a N
2
-Paratherm NF
heat transfer uid100 mm glass beads slurry system at (a)
P = 0.1 MPa and (b) P = 5.6 MPa. (From Luo et al., 1999.)
between the surface of an immersed heating object and
the bed (surface-to-bed). Values of these two types of
heat transfer coecients under ambient conditions
have been reported extensively.
3.2.1 Effect of Operating Variables
Heat transfer in a three-phase uidization system has a
strong dependence on the physical properties of the
liquid phase and a weak dependence on the gas
phase properties. The heat transfer coecient has
been found to increase with an increase in the gas
and liquid velocities, the size and density of the part-
icles, the diameter of the column, and the thermal con-
ductivity and heat capacity of the liquid. On the other
hand, the heat transfer coecient decreases with an
increase in the liquid viscosity and the diameter of
the immersed object (Kim and Laurent, 1991).
Among various factors aecting the heat transfer
behavior, the eects of phase velocities are most pro-
nounced. The heat transfer coecient initially
increases quite rapidly with gas velocity but becomes
less marked at higher gas ow rates, asymptotically
approaching a maximum value (Baker et al., 1978).
The plot of heat transfer coecient vs. liquid velocity
exhibits a maximum, similar to that observed in liquid
solid uidized beds. The bed voidage corresponding to
the maximum heat transfer rate decreases with increas-
ing particle size, but increases with an increase in the
liquid viscosity (Kang et al., 1985). A higher wall-to-
bed heat transfer rate is found in a three-phase
uidized bed compared to that in liquidsolid and
gasliquid systems except under conditions of small
particles and high gas velocities (Chiu and Ziegler,
1983).
3.2.2 Bubble Wake Effect
An understanding of heat transfer mechanisms is
important for developing a heat transfer correlation
and model, and studies have been conducted to inves-
tigate the mechanism of heat transfer in three-phase
uidization systems. Deckwer proposed a heat transfer
mechanism in bubble columns based on Higbies sur-
face renewal theory combined with Kolmogoros
theory of isotropic turbulence (Deckwer, 1980).
Several other investigators extended this concept to
three-phase uidized beds by modifying the energy dis-
sipation rate to include the increased surface renewal
due to the increased turbulence created by the solid
particles (Suh et al., 1985; Magiliotou et al., 1988;
Suh and Deckwer, 1989). Suh et al. (1985) and Suh
and Deckwer (1989) neglected the eect of particle
convective transport on heat transfer, while
Magiliotou et al. (1988) proposed that the particles
also contribute to heat transfer in both liquidsolid
and gasliquidsolid uidized beds in conjunction
with isotropic uid microeddies. Being semitheoretical
in approach, these studies do not fully account for the
inherent mechanism underlying the heat transfer in
three-phase uidized beds.
Heat transfer characteristics in three-phase uidized
beds are intimately associated with bubble motion,
bubble size, and phase holdup, which are in turn
aected by the hydrodynamic behavior of the system,
including wake ow. The eect of the bubble wake on
heat transfer can be illustrated by the measurement of
the instantaneous local variations in the heat transfer
coecient due to the passage of gas bubbles in liquid
and liquidsolid systems (Kumar et al., 1993a,b;
Kumar and Fan, 1994). Simultaneous visualization of
the ow around the heat transfer probe can also be
performed to establish the correspondence between
the local hydrodynamic behavior and the instanta-
neous heat transfer rate. Figure 17 shows a represen-
tative example of the time-dependent heat transfer
coecient in a liquidsolid uidized bed of low-density
gel beads with the injection of gas bubbles (Kumar et
al., 1993a). Associated photographs of the bubble-
wake-induced liquidsolid ow patterns in the vicinity
of the heat transfer probe are also shown in the gure.
Each photograph corresponds to a specic heat trans-
fer coecient. In the photographs the bright dots are
the gel particles and the bright vertical object is the
heat transfer probe located at the center of the column.
The injected bubble volume is 3 cm
3
.
Photograph A shows the instantaneous ow eld
where a marked increase in the heat transfer coecient
results as the bubble approaches the heat transfer sur-
face. It is also observed that the instantaneous heat
transfer coecient starts increasing well before the
bubble approaches the lower edge of the probe,
although the photograph for this case is not shown
here. This increase in heat transfer coecient is attrib-
uted to the local turbulence caused by the approaching
bubble. The bubble is a spherical cap, and the wake
structure appears to be symmetrical about the vertical
axis of bubble movement.
Photograph B shows a large vortex in the primary
wake. The vortex entrains liquid and particles from
around the wake, causing rapid surface renewal at
the probe surface, resulting in increased heat transfer
rates. In the two-dimensional plane of visualization,
the probe surface appears to lie in the vortex structure
close to the wake central axis.
Photograph C shows the ow pattern near the
maximum in the heat transfer signal. The heat trans-
fer surface experiences high shear ow due to the
primary-wake-induced upward liquid and solid ow
toward the wake central axis, causing enhancement
in the heat transfer. The velocity of gel particles
near the heat transfer surface is roughly estimated
from the streak of particles. The upward particle velo-
city is in the range of 20 cm/s, which is close to the
bubble rise velocity (24 cm/s), conrming that the par-
ticle is in the central wake region. Thus the maximum
heat transfer rate is obtained along the wake central
axis in the primary wake region. Figure 17 clearly
demonstrates the importance of bubble wake beha-
vior on the heat transfer characteristics of three-
phase uidized beds.
3.2.3 Pressure Effect
It is known that pressure has a signicant eect on the
hydrodynamics of three-phase uidized beds, and it
would also aect the heat transfer characteristics,
because heat transfer behavior strongly depends on
the hydrodynamics of the system. The eect of pres-
sure on heat transfer is mainly through the variations
in liquid properties and hydrodynamic parameters, as
summarized in Table 2. The overall eect of pressure
on heat transfer behavior depends on the outcome of
the counteracting eects of each individual factor.
The eects of gas velocity and pressure on the
surface-to-bed heat transfer coecient in a three-
phase uidized bed are shown in Fig. 18 (Luo et al.,
1997a). With an increase in pressure, the heat transfer
coecient increases, reaches a maximum at pressures
of 6 to 8 MPa, and then decreases. The following cor-
relation can be used to predict the heat transfer coe-
cient in three-phase uidized beds at high pressures:
h = h
/
e
0.45
g
0.396
U
0.45
g
0.6768
u
pt.0
_ _
(36)
where h
/
is the heat transfer coecient of a liquidsolid
uidized bed with the same solids holdup, and u
pt.0
is
the particle terminal velocity in the uidizing liquid at
ambient pressure. The units for U
g
and u
pt,0
in the
equation are in m/s. The heat transfer coecient, h
/
,
can be calculated by the correlation (Richardson et al.,
1976):
Nu
/
= 0.67 Re
0.62
Pr
0.33
e
0.38
s
1 e
s
(37)
The heat transfer behavior between an immersed
solid surface and surrounding bulk uids in a slurry
bubble column (nitrogen-Paratherm NF heat transfer
uid53 mm glass beads) at elevated pressures is
shown in Fig. 19 (Yang et al., 2000b). It is found
that pressure has a signicant eect on the heat trans-
fer characteristics in a slurry bubble column. The heat
Figure 17 Variation of instantaneous heat transfer rate with bubble passage synchronized with the visualization of ow patterns
in the vicinity of the heat transfer probe in a liquidsolid uidized bed of low-density gel beads. (From Kumar et al., 1993a.)
transfer coecient decreases appreciably with increas-
ing pressure. The variation in heat transfer coecient
with pressure is attributed to the counteracting eects
of the increased liquid viscosity, decreased bubble size,
and increased gas holdup or frequency of bubble pas-
sage over the heating surface as the pressure increases.
In slurry bubble columns, it is observed that the bubble
size reduces signicantly with an increase in pressure,
especially under low pressures, which would result in a
decrease in the heat transfer coecient. Therefore the
bubble size is the most important factor aecting the
heat transfer rate in slurry bubble columns. The addi-
tion of ne particles to the liquid phase enhances heat
transfer substantially, and the eect of temperature on
the heat transfer behavior is mainly determined by the
change in liquid viscosity.
It is noted that the pressure eects on the heat
transfer coecient are dierent between large-particle
and small-particle systems (Luo et al., 1997a; Yang
et al., 2000b). Similar observations were also found
for hydrodynamics and bubble characteristics
between large-particle and small-particle systems.
Luewisutthichat et al. (1997) photographically stud-
ied bubble characteristics in multiphase ow systems.
They found that large-particle systems (i.e., three-
phase uidized beds) exhibit appreciably dierent
Table 2 Variation of Various Parameters with Pressure and Their Effect
on Heat Transfer Coefcients in Three-Phase Fluidization Systems
Parameter
Eect of the
parameter on heat
transfer coecient
Eect of pressure
increase on
parametric value
Parametric eect
on heat transfer
coecient with
increase in pressure
m
l

r
l
(small) (small)
k
l
(small) (small)
C
pl
(small) (small)
s No direct eect No direct eect
d
b

e
g

: Increase; decrease.
Source: Luo et al. (1997a).
Figure 18 Heat transfer coecients as a function of gas
velocity at dierent pressures in a three-phase uidized bed
(nitrogen-Paratherm NF heat transfer uid2.1 mm glass
beads. (From Luo et al., 1997a.)
Figure 19 Eect of pressure and gas velocity on heat trans-
fer coecients in a slurry bubble column (nitrogen-
Paratherm NF heat transfer uid53 mm glass beads).
(From Yang et al., 2000b.)
bubble behavior from small-particle systems (i.e.,
slurry bubble columns). In small-particle systems,
the bubble ow behavior is similar to that in two-
phase systems, while large-particle systems exhibit a
broad bubble size distribution with a large Sauter
mean diameter. Therefore the eect of pressure on
heat transfer behavior, which is closely dependent on
the hydrodynamics and bubble characteristics, would
be dierent in three-phase uidized beds and slurry
bubble columns. In a three-phase uidized bed, the
variation of bubble size with pressure may become
less important in aecting the heat transfer coe-
cient, while the increase in the gas holdup or the
frequency of bubble passage over the heating surface
with pressure could be a dominant factor in deter-
mining heat transfer behavior. Therefore the heat
transfer coecient would increase with increasing
pressure, as observed by Luo et al. (1997a).
Various correlations are also developed to predict
the heat transfer coecient in slurry bubble columns.
For systems where the gas holdup is not aected
by pressure, the following correlation can be used to
estimate the heat transfer coecient (Deckwer et al.,
1980):
St
m
= 0.1 Re
m
Fr Pr
2
m
_ _
0.25
(38)
This correlation is obtained based on the surface
renewal model and Kolmogoros theory of isotropic
turbulence. In the model, the liquidsolid suspension is
considered as a homogeneous phase, and consequently
an estimation scheme of the physical properties of the
suspension from the individual phase is required.
For systems in which pressure has a signicant inu-
ence on the hydrodynamics, the eect of gas holdup on
the heat transfer coecient needs to be considered, and
the following correlation can be used to quantify the
heat transfer coecient in such systems (Yang et al.,
2000b):
St
m
= 0.037 Re
m
Fr Pr
1.87
m
_ _ e
g
1 e
g
_ _ _ _
0.22
(39)
3.2.4 Heat Transfer Model
The consecutive lm and surface renewal model origin-
ally developed by Wasan and Ahluwalia (1969) can be
used to analyze the heat transfer behavior in slurry
bubble columns (Yang et al., 2000b). For ne particles,
the liquidsolid suspension can be reasonably treated
as a pseudohomogeneous uid phase. The model
assumes that a thin uid lm (liquid or liquidsolid
mixture) with a thickness d exists surrounding the heat-
ing surface; and uid elements are forced to contact the
outer surface of the lm due to the passage of bubbles.
The uid elements contact the lm for a short time, t
c
,
and then are replaced by fresh uid elements. The heat
is transferred to the bulk uid through conduction by
the uid lm and unsteady state conduction by the
uid elements. The heat transfer coecient is then
expressed in terms of the physical properties of the
uid, the lm thickness, and the contact time between
the uid elements and the lm (Wasan and Ahluwalia,
1969):
h =
2k
l
pat
c
_
k
l
d
at
c
e
at
c,d
2
1 erf
at
c
_
d
_ _
1
_ _
(40)
Based on Eq. (40), the heat transfer coecient is a
function of lm thickness and contact time between
the uid elements and the lm. A thinner lm and
shorter contact time lead to a higher heat transfer
rate. On the basis of the border diusion layer model
(Azbel, 1981), the order of magnitude of the lm thick-
ness is estimated by (Kumar and Fan, 1994)
d =
6.14L
Re
3,4
m
Pr
1,3
m
(41)
where Re
m
is equal to r
m
Lu
b
,m
m
. Assuming that the
contact time between the uid elements and the lm is
equal to the contact time between the bubbles and the
lm, that is, the bubble motion is considered as the
driving force for the uid element replacement, the
contact time can be estimated from
t
c
=
L
u
b
(42)
where u
b
is the actual bubble rise velocity in a stream
of bubbles (Kumar and Fan, 1994). By considering the
pressure eects on the physical properties of the liquid
and bubble characteristics such as bubble size and bub-
ble rise velocity, this model can be used to analyze the
heat transfer behavior in a high-pressure system. As
shown in Fig. 19, this model can reasonably predict
the pressure eect on the heat transfer coecient in
slurry bubble columns.
3.3 Mass Transfer
Mass transfer is an important component of transport
phenomena in multiphase reactor operation. Two
topics will be discussed in this section, i.e., interfacial
area and liquid-side mass transfer coecient. For mass
transfer phenomena in gasliquid or gasliquidsolid
systems, the interfacial area and the liquid-side mass
transfer coecient are considered the most important
transport properties. The liquid-side mass transfer
coecient incorporates the eects of the liquid ow
eld surrounding rising gas bubbles. The interfacial
area reects the system bubble behavior. The most
common approach in treating gasliquid mass transfer
is to combine the mass transfer coecient and inter-
facial area terms into a single volumetric mass transfer
coecient (k
l
a) averaged over the entire column height
(Fan, 1989).
3.3.1 Interfacial Area
The interfacial area depends on bubble size and the
number of bubbles in the system, and can be expressed
by the equation
a =
6e
g
d
b
(43)
An increase in pressure increases the gas holdup and
decreases the bubble size. Therefore a signicant
increase in the interfacial area at high pressures is
expected, which would result in an overall increase of
k
l
a.
Typical results regarding the inuence of pressure
on the specic gasliquid interfacial area in a bubble
column are shown in Fig. 20 (Oyevaar et al., 1991).
The interfacial area in the gure was determined by
means of CO
2
absorption into aqueous solutions of
diethanolamine (DEA) in a column of 8.1 cm ID.
The pressure has a positive eect on the interfacial
area. The study in a large bubble column with a dia-
meter of 15.6 cm also conrmed this conclusion
(Stegeman et al., 1996). The positive inuence of the
operating pressure on the interfacial area is clearly
attributed to smaller bubble size and higher gas holdup
at high pressures. The increase of the interfacial area
with pressure is more pronounced at higher gas
velocities.
On the basis of Eq. (43) and the observation that the
Sauter mean bubble diameter is proportional to r
0.11
g
,
Wilkinson et al. (1992) proposed a procedure to esti-
mate the interfacial area in a bubble column under
high pressure as shown in Eq. (44):
a(high pressure)
a(atmospheric)
=
e
g
(high pressure)
e
g
(atmospheric)
r
g
(high pressure)
r
g
(atmospheric)
_ _
0.11 (44)
Equation (44) can be used to estimate the interfacial
area at high-pressure conditions based on the atmo-
spheric data. However, this procedure needs to be
further veried.
Studies of the interfacial area in a packed-bed col-
umn at elevated pressures indicate the insignicant
eect of pressure for three-phase systems (Molga and
Westerterp, 1997). The possible explanation for the
lack of pressure eect in three-phase uidized beds is
that the column packing controls the bubble coales-
cence and breakup process. Therefore the variation
of bubble size with pressure is relatively smaller with
packing than without. A comparison of interfacial
areas obtained in bubble columns with packing and
without packing indicates higher interfacial areas in
the packed-bed bubble column, which implies that
packing tends to break up the bubbles.
3.3.2 Liquid-Phase Mass Transfer Coefcient
The second important parameter to characterize mass
transfer behavior is the liquid-phase mass transfer
coecient, k
l
. The eect of pressure on the liquid-
phase mass transfer coecient is complicated. Since
the liquid-phase mass transfer coecient depends on
the gasliquid system and the pressure aects the phy-
sical properties of gas and liquid phases signicantly, a
change in the liquid-phase mass transfer coecient at
high pressures is also expected. The following illus-
trates expected relations between the liquid-phase
mass transfer coecient and the physical properties
of the liquid phase.
Figure 20 Pressure eect on interfacial area at dierent
gas velocities in a bubble column with perforated plate gas
distributor (nitrogenaqueous diethanolamine solution,
D
c
= 8.1 cm). (From Oyevaar et al., 1991.)
Normally, an increase in pressure results in a
decrease in surface tension and an increase in liquid
viscosity (Lin et al., 1998). The decrease of surface
tension would reduce the liquid ow over the surfaces
of rising gas bubbles, resulting in a reduction of bubble
rise velocity and a longer contact time between the
liquid and bubbles (Chang and Morsi, 1992). Since
the mass transfer coecient is inversely proportional
to the square root of the contact time (Higbie, 1935), k
l
would decrease as the pressure increases due to the
decrease in surface tension. The increase of liquid visc-
osity with pressure would also decrease the mass trans-
fer coecient, since k
l
is inversely proportional to the
liquid viscosity (Calderbank and Moo-Young, 1961).
Furthermore, an increase in pressure also reduces the
bubble size, and hence the liquid-phase mass transfer
coecient, since k
l
is known to be lower for small
bubbles (Calderbank and Moo-Young, 1961). Based
on the above analysis, the liquid-phase mass transfer
coecient k
l
would decrease with increasing pressure.
Most experimental data available in the literature are
expressed by the volumetric mass transfer coecient,
k
l
a. Since the interfacial area increases signicantly
with increasing pressure, it is expected that the varia-
tion of interfacial area with pressure will be the
predominant factor in determining k
l
a.
The following are some experimental studies to
quantify the pressure eect on the volumetric mass
transfer coecient. Wilkinson et al. (1994) studied
the inuence of pressure on the volumetric mass
transfer coecient in a bubble column with an inner
diameter of 0.158 m. The uncatalyzed oxidation of
sodium sulte was chosen as a model reaction for
determining the volumetric mass transfer coecient.
The supercial gas velocity and operating pressure
varied in the ranges of 00.15 m/s and 0.10.4 MPa,
respectively. Their experimental results showed that
both the interfacial area and the volumetric mass
transfer coecient increase with increasing pressure,
especially at high gas velocities. At low supercial gas
velocities (U
g
- 0.03 m/s), the inuence of pressure is
relatively small. Similar pressure eect on the volu-
metric mass transfer coecient was also observed by
Letzel (1997) using a dynamic pressure-step method
developed by Linek et al. (1989, 1993). Figure 21
shows the eect of pressure on the volumetric mass
transfer coecient in the nitrogen-water system
(Letzel, 1997). Based on experimental results, a pro-
cedure for the estimation of the liquid volumetric
mass transfer coecient at high pressures is proposed,
which is similar to that for predicting the interfacial
area (Wilkinson et al., 1992),
k
l
a(high pressure)
k
l
a(atmospheric)
=
e
g
(high pressure)
e
g
(atmospheric)
_ _
n
with n = 1.01.2 (45)
The limitation of available high-pressure mass
transfer studies is that the measuring technique is
only suitable for low-pressure conditions (normally
less than 1.0 MPa); thus further studies to cover wide
ranges of pressure and gas velocity are needed.
Although some work has been conducted to study
the mass transfer behavior in bubble columns at ele-
vated pressures, little is known for gasliquidsolid
systems, especially with ne particles. Muroyama et
al. (1997) studied mass transfer from an immersed
cylinder to the bed and found that a 50 to 75%
enhancement of mass transfer coecients for using
ne particles (d
p
= 0.250.5 mm) over coarse particles
(d
p
= 1.1 mm). This nding is of interest to electrode
applications of uidized beds. Using the electrochemi-
cal method of Reiss and Hanratty (1963), Nikov and
Delmas (1992) studied the local liquidsolid mass
transfer coecients of a particle xed in the bed. The
time-averaged local mass transfer coecients are
related to the shear stress distribution around the par-
ticle. They found that local liquidsolid mass transfer
coecients vary with the polar angle measured with
respect to the frontal stagnant point. The presence of
gas increases the liquidsolid mass transfer coecient
as well as the velocity gradient and the shear force. An
increase in the bubble diameter and the number of
bubbles around the xed particle results in an increase
Figure 21 Volumetric mass transfer coecient, k
1
a, as a
function of supercial gas velocity at dierent pressures in
a nitrogenwater system. (From Letzel, 1997.)
in the local mass transfer coecient as well as the
velocity gradient around the particle, particularly in
the frontal zone.
3.4 Phase Mixing
Backmixing is a ow pattern that is intermediate
between the two ideal cases of plug ow and perfect
mixing. Backmixing of individual phases in three-
phase uidization systems is an important parameter
for designing reactors and predicting reactor perfor-
mance. Generally, in three-phase uidized beds appre-
ciable backmixing occurs in the liquid and solid
phases, and relatively little backmixing exists in the
gas phase. The extents of backmixing of both the liquid
and the solid phase depend heavily on the rising
motion of the gas bubbles. In this section, liquid-
phase mixing will be emphasized, and recent develop-
ment in solids mixing will be introduced briey.
Liquid-phase mixing in three-phase uidized beds
can be described using the dispersion model. A two-
dimensional model considers both radial and axial dis-
persions. Both axial and radial dispersion coecients
are strong functions of operating conditions such as
liquid and gas velocities and properties of liquid and
solid phases. Evaluations of liquid-phase dispersion
coecients are based on a tracer injection method
and subsequent analysis of the mean and the variance
of the system response curves.
Typically, there are two ways to inject tracers,
steady tracer injection and unsteady tracer injection.
It has been veried that both methods lead to the
same results (Deckwer et al., 1974). For the steady
injection method, a tracer is injected at the exit or
some other convenient point, and the axial concentra-
tion prole is measured upward of the liquid bulk ow.
The dispersion coecients are then evaluated from this
prole. With the unsteady injection method, a variable
ow of tracer is injected, usually at the contactor inlet,
and samples are normally taken at the exit. Electrolyte,
dye, and heat are normally applied as the tracer for
both methods, and each of them yields identical dis-
persion coecients. Based on the assumptions that the
velocities and holdups of individual phases are uniform
in the radial and axial directions, and the axial and
radial dispersion coecients, E
zl
and E
rl
, are constant
throughout the uidized bed, the two-dimensional
unsteady-state dispersion model is expressed by
oC
ot
= E
zl
o
2
C
oz
2

E
rl
r
o
or
r
oC
or
_ _
U
l
e
l
oC
oz
(46)
Equation (46) can be reduced to the axial dispersion
(one-dimensional) model when the radial dispersion is
negligible in comparison with the axial dispersion.
The studies regarding the axial liquid-phase mixing
in gasliquid systems at atmospheric conditions are
extensive, especially for the airwater system (Fan,
1989). Recently, some work has been conducted to
study the liquid mixing under high-pressure condi-
tions. Houzelot et al. (1985) measured the axial disper-
sion of the liquid phase in a bubble column with a
diameter of 5 cm. Their study found an insignicant
pressure eect on the axial dispersion of the liquid
phase; their study was limited by narrow experimental
conditions, that is, very low supercial gas velocity
(<6 mm/s) and pressure (<3 atm). Under such con-
ditions, the ow is always in the homogeneous bub-
bling regime and a signicant change in liquid-phase
mixing within such narrow operating conditions is not
expected. Sangnimnuan et al. (1984) experimentally
investigated the extent of liquid-phase backmixing
under coal hydroliquefaction conditions (e.g., tempera-
ture between 164 and 3848C and pressure between 4.5
and 15 MPa) in a very small bubble column reactor
(1.9 cm in diameter). They did not describe the eect
of pressure on the axial mixing, and their study was
limited by the small scale of the reactor.
Holcombe et al. (1983) determined the liquid axial-
dispersion coecient in a 7.8 cm diameter bubble col-
umn under pressures in the range of 3.07.1 atm. The
supercial gas velocity varied up to 0.6 m/s. They used
heat as a tracer to measure the thermal dispersion coef-
cient, which was found to be comparable to the mass
dispersion coecient. The eect of pressure on thermal
dispersion coecients was negligible in the pressure
range of their study. Wilkinson et al. (1993) measured
the liquid axial dispersion coecient in a batch-type
bubble column of 0.158 m in diameter for the nitro-
genwater system at pressures between 0.1 and
1.5 MPa using an electrical conductivity cell. They
found that the liquid axial dispersion coecient actu-
ally increases with increasing pressure, especially under
high gas velocity conditions (>0.10 m/s). It is noted
that the available theories in the literature to describe
liquid mixing at atmospheric pressure cannot explain
the pressure eect observed in their study. Tarmy et al.
(1984) investigated liquid backmixing in industrial coal
liquefaction reactors using radioactive tracers. The
operating pressure in their study varied up to 17 MPa,
and they found that the measured dispersion coe-
cients at high pressures were up to 2.5 times smaller
than the values predicted by literature correlations,
which were developed based on the experimental data
under ambient conditions. Studies of gasliquid dis-
persion behavior in coal liquefaction reactors using a
neutron absorption tracer technique indicated a similar
trend of the pressure eect (Onozaki et al., 2000a,b).
The axial dispersion coecients of the liquid phase
under coal liquefaction conditions (P = 16.8
18.8 MPa) are smaller than those estimated from litera-
ture correlations. These observations imply a decreas-
ing trend for the pressure eect on liquid mixing.
It has been shown that the steady-state thermal dis-
persion method is suitable for measuring the axial dis-
persion coecients of the liquid phase at high
pressures in bubble columns, particularly for hydrocar-
bon liquids in which the electrical conductivity tech-
nique is not applicable (Yang and Fan, 2002). In this
technique, heat is introduced close to the outlet of the
liquid phase and the upstream temperature prole in
the liquid is measured. The dispersion coecient is
then determined from the axial temperature prole
based on the one-dimensional dispersion model. The
eects of pressure and gas velocity on the axial disper-
sion coecients in the 5.08 cm and 10.16 cm columns
for the nitrogen-Paratherm heat transfer uid system
are shown in Fig. 22 (Yang and Fan, 2002). The axial
dispersion coecient decreases signicantly when the
pressure is increased from ambient to elevated pres-
sures for both columns, indicating distinct ow beha-
viors for ambient pressure and elevated pressure.
When the pressure is further increased, the decrease
rate of the axial dispersion coecient becomes smaller.
The pressure eect is more pronounced at higher gas
velocities and in larger columns. A decrease in the
liquid mixing at high pressures was also observed by
Tarmy et al. (1984) and Onozaki et al. (2000a,b). Other
researchers found that an insignicant eect of pres-
sure on liquid mixing in small columns (diameter nor-
mally less than 10.0 cm) is possibly due to the narrow
operating conditions in their studies, that is, either low
gas velocities or a narrow pressure range (Houzelot et
al., 1985; Holcombe et al., 1983). As shown in Fig. 22a,
in small columns, the variation of the axial dispersion
coecient with pressure is not pronounced at low gas
velocities (i.e., in the homogeneous bubbling ow
regime) and low pressures.
The eects of gas velocity and pressure on liquid
mixing can be explained based on the mixing mechan-
isms of gross liquid circulation and local turbulent uc-
tuations. The available theories in the literature
describing liquid mixing under atmospheric pressure
are based on liquid turbulence induced by rising bubbles
(Baird and Rice, 1975), large-scale liquid internal circu-
lation (Joshi, 1980), or a combination of these two
mechanisms (Degaleesan et al., 1997). Based on the
internal circulation model, Joshi (1980) proposed the
following equation to predict the average liquid circula-
tion velocity in bubble columns:
V
c
= 1.31 gD
c
U
g
e
g
1 e
g
U
l
e
g
U
bo
_ _ _ _
1,3
(47)
where U
bo
is the terminal bubble rising velocity. The
second term on the right-hand side of Eq. (47) is nor-
mally negligible compared to the other two terms at
low liquid velocity. It can be seen that both the gas
holdup and the bubble rising velocity aect the liquid
circulation velocity, which can be treated as a measure
of the extent of liquid circulation eect on the liquid
mixing. It is known that the gas holdup increases and
the bubble size and rising velocity decrease with in-
Figure 22 Pressure eect on liquid mixing in dierent sizes
of bubble columns: (a) D
c
= 5.08 cm and (b) D
c
= 10.16 cm.
(From Yang and Fan, 2002.)
creasing pressure. The counteracting eects of gas hold-
up and bubble rise velocity on the liquid circulation
velocity may result in unchanged or slightly changed
liquid circulation velocity at elevated pressures, as
shown in Eq. (47). On the other hand, when the system
pressure increases, bubbles become smaller, and the
liquid entrainment by the bubble wake and the turbu-
lence induced by the motion of bubbles are reduced.
Therefore liquid mixing is reduced at elevated pres-
sures. The combination of variations of liquid-phase
turbulent uctuations and internal liquid circulation
gives rise to the overall eect of pressure on the liquid
mixing.
Recent study of ow elds and Reynolds stresses in
high-pressure bubble columns using laser Doppler
velocimetry (LDV) also conrmed that the bubble-
induced turbulence of the liquid phase is depressed at
higher pressures (Lee et al., 2001). Figure 23 shows the
eect of pressure on the proles of axial liquid velocity
and Reynolds normal stress measured by the LDV
technique. The ensemble-averaged velocity shows the
gross circulation pattern, and as the pressure increases,
the magnitudes of the averaged velocity of the liquid
phase at the center and in the wall region decrease. The
Reynolds stresses decrease signicantly with increasing
pressure, indicating the smaller extent of liquid-phase
uctuations at high pressures. The uctuations of the
liquid phase are mainly caused by the motion of gas
bubbles, and the uctuations are damped out as the
pressure increases because of a narrower bubble size
distribution with a smaller mean bubble size.
Recently, research in mixing behaviors of both the
liquid and the solid phase in three-phase uidized beds
under ambient conditions has continued. Chen et al.
(1995) and Zheng et al. (1995) studied the axial varia-
tions of the gas holdup, axial liquid mixing, and gas
liquid mass transfer. They observed a signicant
increase of the axial dispersion coecients of the liquid
phase and a decrease of the gasliquid mass transfer
coecients with an increase of the axial distance under
certain operating conditions, particularly under high
liquid ow rates. They attributed the signicant axial
variation in part to the coexistence of the dispersed bub-
ble and coalesced bubble regimes along the axial direc-
tion. Kim et al. (1992) proposed empirical correlations
to account for the axial dispersion coecient of the
liquid phase over a wide range of operating conditions:
Pe
z
=
d
p
U
l
E
zl
_ _
= 11.96
U
l
U
l
U
g
_ _
1.03
d
p
D
c
_ _
1.66
(48a)
for the bubble coalescing regime with a range of vari-
ables of 0.004 _ (d
p
,D
c
) _ 0.024 and 0.250 _ [U
l
,
(U
l
U
g
)] _ 0.857, and
Pe
z
=
d
p
U
l
E
zl
_ _
= 20.72
U
l
U
l
U
g
_ _
1.03
d
p
D
c
_ _
1.66
(48b)
Figure 23 Pressure eects on the proles of (a) axial liquid
velocity; (b) Reynolds axial normal stress in a bubble column
(D
c
= 5.08 cm, nitrogenwater). (From Lee et al., 2001.)
for the bubble disintegrating regime with a range of
variables of 0.004 _ (d
p
,D
c
) _ 0.012 and 0.143 _ [U
l
/
(U
l
+U
g
)] _ 0.857.
The degree of solids mixing is a crucial component
for such application as bioreactors. For this type of
application, the uniformity of the particle properties
is important for determining the overall rate of reac-
tion, and this uniformity of solids is strongly depen-
dent upon the degree of solids mixing. For large
bubble systems, the mechanism for solids mixing is
primarily through the bubble wake. For small bubble
systems or any bubble system with light particles, par-
ticle drift eects or random motion of the particles can
cause solids mixing. Particle motion can be induced by
velocity gradients caused by passing gas bubbles or
from local nonuniformities in the ow eld. A detailed
introduction to solids mixing behavior was given in
Fans 1989 book; this section is an update of recent
studies on solids mixing.
The turbulence of the solid phase in three-phase
uidization measured using radioactive particle track-
ing (RPT) was found to be anisotropic (Cassanello et
al., 1995). The same technique was also used to exam-
ine the mixing behavior of a binary mixture of solid
particles (Cassanello et al., 1996; Kiared et al., 1997).
The results indicated that there are two separate
regions containing solid particles: the upward ow in
the core region established by bubble wakes (wake
phase) and the downward ow in the wall region
(emulsion phase) with cross-ow between two regions,
which could form the basis for a macromixing model
to describe the solids residence time distribution and
solids mixing in the three-phase uidized bed. In a
more phenomenological approach, axial mixing and
segregation of solids particles in a slurry bubble
column or a three-phase uidized bed are com-
monly described by a sedimentationdispersion
model (Matsumoto et al., 1997; Hidaka et al., 1998).
There are various equations employed for the sedimen-
tationdispersion model and a general form can be
given as
oC
s
ot
=
o
oz
E
oC
s
oz
(V U)C
s
_ _
(49)
where C
s
is the solids concentration in a solidliquid
mixture, E is the solid phase axial dispersion coe-
cient, and V and U are liquid and solids velocities,
respectively.
A theoretical analysis of the sedimentationdisper-
sion model by Jean et al. (1989) indicated that in a
common expression of the sedimentationdispersion
model with U
l
(dened as the cross-sectional averaged
linear liquid velocity) as V and u
t
(dened as solids
settling velocity) as U, under the coalesced bubble
ow regime and low liquid velocity or slow slurry
ow conditions, E and u
t
do not have their alleged
physically compatible denitions and should be
regarded as purely empirical parameters and as bearing
no physical signicance. Under the dispersed bubble
ow regime and high liquid velocity or fast slurry
ow conditions, the physically compatible denitions
for E and u
t
would hold for this model (Jean et al.,
1989). The general prediction for u
t
is available in the
literature (e.g., Di Felice, 1995). The surface property
of particles (e.g., biolm-coated particles compared to
regular spherical particles with the same density and
size) may aect the drag coecient of the particle and
hence the value of the terminal settling velocity. Thus
sedimentation and dispersion behavior of these two
particles would be dierent in a three-phase uidized
bed.
4 COMPUTATIONAL FLUID DYNAMICS
The earlier studies mainly focused on experimentally
examining the macroscopic uid dynamic behavior of
three-phase uidized beds, and developing empirical
correlations. With increasing computer power, the
employment of the computational uid dynamic
approach has gained considerable attention. Over the
past decade, signicant advances have been made in
numerical modeling of two-phase ows such as
gassolid and gasliquid ows. For example, the two-
uid model (e.g., Gidaspow, 1994) and the discrete par-
ticle model (e.g., Tsuji et al., 1993) were developed to
simulate collision-dominant gassolid ows. The direct
numerical simulation technique based on the front-
capturing method (e.g., Boris, 1989), the tracking
method, which includes the volume-tracking (e.g.,
Harlow and Welch, 1965) and the front-tracking (e.g.,
Unverdi and Tryggvason, 1992) methods, and the lat-
tice Boltzmann method (e.g., Sankaranarayanan et al.,
2002) were developed to capture the free surfaces of
gasliquid ows.
An understanding of three-phase ows is still lim-
ited because of complicated phenomena underlying
interactions between phases, such as the particlebub-
ble interaction and the liquid interstitial eect during
particleparticle collision. Recently, several computa-
tional uid dynamics models were reported to simulate
three-phase uidization behavior (Gidaspow et al.,
1994; Grevskott et al., 1996; Mitra-Majumdar et al.,
1997). These models are based on the multiuid
approach in which both the gas and the solid phase are
treated as pseudocontinuous phases. Alternatively, a
second-order moment three-phase turbulence model
can be used. This approach was demonstrated satisfac-
torily in the simulation of the gasliquid ow in a bub-
ble column (Zhou et al., 2002) and is readily extendable
to the gasliquidsolid ow simulation. There is, how-
ever, a practical need for a discrete simulation that
illustrates the inherent discrete ow characteristics of
individual phases such as the bubble wake structure.
The discrete phase simulation method has demon-
strated its potential in simulating three-phase ows (Li
et al., 1999; Zhang et al., 2000). In this approach, the
Eulerian volume-averaged method, the Lagrangian
discrete particle method (DPM), and the volume-of-
uid (VOF) volume-tracking method can be employed
to describe the motion of liquid, solid particles,
and gas bubbles, respectively. A bubble induced force
(BIF) model, a continuum surface force (CSF) model,
and Newtons third law are applied to account for the
couplings of particlebubble, bubbleliquid, and
particleliquid interactions, respectively. A close dis-
tance interaction (CDI) model is included in the
particleparticle collision analysis, which considers
the liquid interstitial eect between colliding particles
(Zhang et al., 1999).
In the following, the computational models for each
individual phase will be discussed. Relevant underlying
equations will be given, and the assumptions of each
model will be explained. Finally, simulated results of
several important phenomena in three-phase ows will
be demonstrated.
4.1 Liquid-Phase Model
The governing equations for the continuous phase of
multiphase ows can be derived from the Navier
Stokes equations for single-phase ows. Considering
the existence of dispersed particles, a volume-averaging
technique is used to develop a set of partial dierential
equations to describe the mass and momentum conser-
vation of the liquid phase. The continuity equation for
the liquid phase can be given as
oe
l
ot
V (e
l
v) = 0 (50)
The momentum equation for the liquid phase is
r
l
o(e
l
v)
ot
r
l
V (e
l
vv) = Vp Vt r
l
g f
b
(51)
where v is the liquid velocity vector; e
l
is the liquid
holdup; r
l
is the liquid density; p is the scalar pressure;
t is the viscous stress tensor; g is gravitational accel-
eration; f
b
is the total volumetric body force acting on
the liquid phase excluding the gravitational force, that
is, the volumetric particleuid interaction force (f
pf
)
plus the volumetric bubbleuid interaction force (f
bf
).
Based on Newtons third law of motion, the force act-
ing on a particle from the liquid phase, F
fp
, yields a
reaction force on the liquid. Therefore the momentum
transfer from particles to the liquidgas phase is taken
into account in Eq. (51) by adding the volumetric
particleuid interaction force, f
pf
, given below to
the body force term,
f
pf
=
F
k
fp
V
k
ij
x
k
p
c
ij
(52)
where the subscript ij denes the location of a compu-
tational cell; and V are the domain and volume of
this cell, respectively; x
k
p
is the location vector of par-
ticle k; F
fp
is the uidparticle interaction force acting
on any individual particles, which includes the drag
(F
D
), buoyancy (F
B
), added mass (F
AM
), and Basset
force (F
BA
):
F
fp
= F
D
F
B
F
AM
F
BA
(53)
It is noted that the buoyancy force is also included in
Eq. (53) as one of the uidparticle interaction forces.
Due to the small particle size, the Saman and Magnus
forces are ignored in Eq. (53).
The bubbleuid interaction force, f
bf
, is obtained
by using a continuum surface force (CSF) model
(Brackbill et al., 1992)
f
bf
= d(x. t)sk(x. t)Va(x. t) (54)
where the plus sign is used for the liquid phase and the
minus sign for the gas phase; d(x,t) is the Delta func-
tion, which equals 1 at the gasliquid interface and 0
elsewhere; s is the surface tension of the interface;
k(x,t) is the curvature of the free surface; a(x, t) is
the volume fraction of the uid.
The Newtonian viscous stress tensor is used in Eq.
(51) and given as
t = 2m
l
S = m
l
[(Vv) (Vv)
T
] (55)
where S is the rate-of-strain tensor and m
l
is the liquid
viscosity.
The liquid properties on a particle are obtained by
an area-weighted averaging method based on the prop-
erties at the four grid points of the computational cell
containing the particle. The cell-averaged liquid
holdup obtained as the volume fraction of the liquid
in the computational cell is only used for solving the
volume-averaged equations of the liquid phase. When
evaluating the particle drag coecient in the liquid
solid medium, the liquid holdup is based on particle-
centered averaging over a prescribed area.
4.2 Gas-Phase Model
The simulation of the ow inside the gas bubble is
important for predicting bubble behaviors under
high-pressure conditions, since the eect of gas density
on the ow behavior is signicant. The model equa-
tions for the gas phase are straightforward, as the ow
inside the gas bubble is governed by single-phase
NavierStokes equations. However, modeling the bub-
blesuspension interface is not a straightforward task.
The diculty of numerical simulation is caused by the
discontinuous jump of properties across the interface
between the gas bubble and the liquidsolid suspen-
sion. To circumvent this problem, a continuous transi-
tion method (CTM) can be employed. In this method,
the discontinuous characteristics are replaced by a
smooth variation of the properties (e.g., density and
viscosity) from one phase to another within the nite
interface thickness. The continuous transition method
can overcome the problem of numerical divergence
while simulating the ow eld at both sides of the
interface where the physical properties of the uids
strongly dier.
The scalar function of uid volume fraction, a(x, t),
solved by the VOF method (Hirt and Nichols, 1981) is
used to construct this continuous transition function.
A uid property at the interface is then given by
Q = Q
+
m
a x. t ( ) Q
+
g
1 a x. t ( ) [ ] (56)
where Q represents a property of the uid, and Q
+
m
and
Q
+
g
represent the properties of the liquidsolid suspen-
sion and the gas bubble, respectively. By denition,
a(x, t) = 1 in the liquid or liquidsolid mixture,
0 < a(x, t) < 1 at the free surface, and a(x, t) = 0
in the gas bubble. Therefore Q is replaced by Q
+
m
when a(x, t) equals 1 in the liquidsolid suspension
or Q
+
g
when a(x, t) equals 0 inside the gas bubble.
The advection equation for a(x, t) is
oa
ot
(v V)a = 0 (57)
On the gasliquid free surfaces, the stress boundary
condition follows the Laplace equation as
p
s
= p p
v
= sk (58)
where the surface pressure, p
s
, is the surface-tension-
induced pressure jump across a uid interface. The
continuum surface force (CSF) model converts the
surface force into a volume force within free surfaces,
as shown in Eq. (54).
4.3 Discrete Particle Method
The motion of a particle in a ow eld can be described
in Lagrangian coordinates with its origin attached to
the center of the moving particle. The motion of a
single particle can be described by its acceleration
and rotation in a nonuniform ow eld. The particle
accelerating in the liquid is governed by Newtons
second law of motion as
m
p
dv
p
dt
= F
total
(59)
The total force acting on a particle is composed of all
applicable forces, including drag (F
D
), added mass
(F
AM
), gravity (F
G
), buoyancy (F
B
), Basset force
(F
BA
), and other forces (F
i
):
F
total
= F
D
F
AM
F
G
F
B
F
BA
i
F
i
(60)
The drag force acting on a suspended particle is
proportional to the relative velocity between the
phases:
F
D
=
1
2
e
2
l
C
/
D
r
l
A v v
p
(v v
p
) (61)
where A is the exposed frontal area of the particle in
the direction of the incoming ow, and C
/
D
is the eec-
tive drag coecient, which is a function of the particle
Reynolds number, Re
p
. For isolated rigid spherical
particles the drag coecient C
D
can be estimated by
the following equations (Rowe and Henwood, 1961)
C
D
=
24
Re
p
(1 0.15Re
0.687
p
) Re
p
- 1000
0.44 Re
p
_ 1000
(62)
_
_
where Re
p
= r
l
e
l
d
p
[n n
p
[,m
l
. In the liquidsolid or
gassolid suspensions, the drag force depends strongly
on the local phase holdup in the vicinity of the particle
under consideration. The eective drag coecient can
be obtained by the product of the drag coecient for
an isolated particle and a correction factor as given by
(Wen and Yu, 1966)
C
/
D
= C
D
e
4.7
l
(63)
The added mass force accounts for the resistance of
the uid mass that is moving at the same acceleration
as the particle. For a spherical particle, the volume of
the added mass is equal to one-half of the particle
volume, so that
F
AM
=
1
2
r
l
V
p
d
dt
(v v
p
) (64)
The Basset force accounts for the eect of past
acceleration. The original formulation of the Basset
force is derived from the creeping ow condition.
For a particle moving in a liquid with a nite
Reynolds number, the modied Basset force is given
as (Mei and Adrian, 1992)
F
BA
= 3pm
l
d
p
_
t
0
K(t t)
d(v v
p
)
dt
dt (65)
K(t t) in Eq. (65) is given as
K(t t) =
_
p(t t)n
l
r
2
p
_ _
1,4
1
2
p
(U v
p
v)
3
r
p
n
l
f
3
H
(Re)
(t t)
2
_ _
1,2
_
2
f
H
(Re) = 0.75 0.105Re.
Re =
Ud
p
n
l
(66)
where n
l
is the kinematic viscosity of the uid and U is
the mean stream velocity.
The sum of the gravity and buoyancy forces has the
form
F
G
F
B
= r
p
V
p
g r
l
V
p
g = r
p
r
l
_ _
V
p
g (67)
An inertial term, r
l
V
p
dv,dt, is added to the buoyancy
force, which is important when the uid is accelerated.
When particles approach the gasliquid interface,
which is identied as 0 - a(x. t) - 1, the surface ten-
sion force acts on the particles through the liquid lm.
Since the size of computational cell is larger than the
thickness of the gasliquid interface lm, a bubble-
induced force model (BIF) is applied to the particle:
F
bp
= V
p
sk(x. t)Va(x. t) (68)
If the particle overcomes this bubble-induced force,
the particle will penetrate the bubble surface. The
penetrating particle breaks the bubble surface momen-
tarily upon contact. If the penetrating particle is small,
the bubble may recover its original shape upon particle
penetration (Chen and Fan, 1989a,b). However, if
there are several particles colliding with the bubble
surface simultaneously, the resulting force may cause
the bubble to break.
The general scheme of a stepwise molecular
dynamic (MD) simulation (Allen and Tildesley,
1987), based on a predictorcorrector algorithm, is
used to compute the particle motion. When particles
are moving close to each other, the close-range
particleparticle interaction including collision takes
place as illustrated in the following.
4.4 ParticleParticle Collision Dynamics
In this section, particleparticle collision dynamics
will be discussed. A hard sphere approach is used
for the particleparticle collision analysis. In this
approach, it is assumed that collisions between sphe-
rical particles are binary and quasi-instantaneous, and
further, that there is a sequence of collisions during
each time step. The equations, which are similar to
the stepwise molecular dynamic simulation, are used
to locate the minimum ight time of particles before
any collision.
4.4.1 Liquid Shear Effect
The liquid shear eect on particle motion is important
in liquidsolid systems. Particularly, the liquid shear
eect between particles becomes signicant when two
particles move close to each other in liquidsolid
systems, especially when the distance between two
particles is less than 0.1d
p
. Thus the close-distance
interaction (CDI) model is used to determine the par-
ticle contact velocity just before collision; this model
considers the strong damping eect of the liquid lm
before particle contact. The particle normal contact
velocity can be described by (Zhang et al., 1999)
1
1
2
r
l
r
p
3
16
r
l
r
p
r
3
p
h
3
_ _
dv
p
dh
=
9
2
m
l
f f
r
2
p
r
p
(v
p
v)
v
p
9
32
r
l
r
p
r
4
p
h
4
(v
p
v) v
p
v
v
p
1
r
l
r
p
_ _
g
v
p
9m
l
_
t
0
K(t t)[d(v
p
v),dt]dt
2r
2
p
r
p
v
p
(69)
where h is the distance from the center of the
approaching particle to the midpoint between the
two particles; r
p
is the radius of the particle; and f
(Schiller and Naumann, 1933) and f (Zhang et al.,
1999) are the correction functions and can be expressed
as
f = exp
Re
p
1.7
_ _
0.44
r
p
r
l
_ _
0.19
_
r
p
h
_
Re
0.47
p
_ _
f = 1 0.15Re
0.687
p
(70)
Using the RungeKutta method, Eq. (69) can be
solved to determine the particle normal contact velo-
city just before the collision.
4.4.2 Particle Collision Analysis
When two particles are in contact, a collision analysis
can be conducted to obtain the velocities of the par-
ticles after the collision. To simplify the analysis, it can
be assumed that the tangential traction and the result-
ing displacements have no eect on the normal colli-
sion. For the collision between particles a and b, the
normal components after collision can be obtained by
solving the equations for the restitution coecient and
the conservation of momentum,
U
N
/
a
U
N
/
b
U
b
N
U
N
a
= e
m
a
U
N
a
m
b
U
N
b
= m
a
U
N
/
a
m
b
U
N
/
b
(71)
where U
N
is the normal velocity of the particle (a or b)
at the contact point before or after (with superscript
/
)
the collision.
From Mindlins contact theory, there are three
kinds of frictional contact during the collision: sliding
contact, nonsliding or sticking contact, and torsion of
elastic particles in contact (Mindlin, 1949). By neglect-
ing the eect of particle torsion during the collision,
the simplied Mindlins contact theory is applied to
obtain the tangential components after the collision.
If the incident angle, dened as the ratio of the par-
ticleparticle relative velocity in the tangential
direction to velocity in the normal direction, is less
than the critical angle (a
cr
= tan
1
(2 f
k
), where f
k
is
the friction coecient), sticking collision occurs:
U
T
/
a
= U
T
/
b
(72)
Otherwise, sliding collision occurs, in which
(U
T
a
U
T
b
) (U
T
/
a
U
T
/
b
) = 2f
k
(U
N
a
U
N
b
) (73)
where U
T
is the tangential particle velocity (a or b) at
the contact point (Fan and Zhu, 1998).
The conservation of momentum is given as
m
a
U
T
a
m
b
U
T
b
= m
a
U
T
/
a
m
b
U
T
/
b
(74)
The tangential velocities after the collision can be
obtained by solving Eq. (72) or (73) together with
Eq. (74).
The collision also induces a change in particle
rotation. The angular velocities after the collision are
determined by
I
a
(o
/
a
o
a
) = m
a
(U
T
/
a
U
T
a
) r
a
I
b
(o
/
b
o
b
) = m
b
(U
T
/
b
U
T
b
) r
b
(75)
where o is the angular velocity of the particle (a or b)
and I is the moment of inertia dened by I =
(2,5)m
p
r
2
p
.
The tangential velocities of the particle center are
given as
U
T
/
ac
= U
T
/
a
o
/
a
r
a
U
T
/
bc
= U
T
/
b
o
/
b
r
b
(76)
Particle collision is important in simulating bubble
behavior in liquidsolid suspensions, especially for
high solids holdup conditions. A simulation without
considering particle collision leads to inappropriate
nonuniformity of particle distribution in the ow
eld and hence false ow eld information.
Numerical instability may also occur as a result of
inappropriate particle accumulation in a computa-
tional cell. In the collision model discussed above,
only the binary-particle collision mechanism is consid-
ered, which limits the model to low solids holdup con-
ditions (less than 3040% by volume). For higher
solids holdup cases, the multiparticle collision mechan-
ism needs to be considered.
4.5 Simulation Examples
The discrete phase simulation method described in
Secs. 4.1 through 4.4 is capable of predicting the ow
behavior in gasliquidsolid three-phase ows. In this
section, several simulation examples are given to
demonstrate the capability of the computational
model. First, the behavior of a bubble rising in a
liquidsolid suspension at ambient pressure is simu-
lated and compared to experimental observations.
Then the eect of pressure on the bubble rise behavior
is discussed, along with the bubbleparticle inter-
action. Finally, a more complicated case, that is, multi-
bubble formation dynamics with bubblebubble
interactions, is illustrated.
The behavior of a bubble rising in a liquid or liquid
solid suspension can be simulated by the computa-
tional model. Simulated results and experimental
observations of a single bubble rising in a liquid
solid suspension under ambient conditions are shown
in Fig. 24 (Fan et al., 1999). The simulation domain is
30 80 mm
2
and the computational grid size is
0.15 0.16 mm
2
. One thousand glass beads with a
density of 2,500 kg/m
3
and a diameter of 1.0 mm are
modeled as the solid phase. An aqueous glycerine
solution (80 wt%) with r
l
= 1,206 kg/m
3
, m
l
= 52.9
mPa s, and s = 62.9 mN/m is modeled as the liquid
phase. A circular bubble with a diameter of 10 mm is
initially imposed in the computational domain with its
center 15 mm above the bottom. Initially, the particles
are randomly positioned in a 30 240 mm
2
area.
Then the simulation is performed for particles settling
at a liquid velocity of 5 mm/s. At this stage, the bubble
is treated as an obstacle and xed in the original place.
An equilibrium bed height is reached at 80 mm, which
gives a three-dimensional equivalent solids holdup of
0.44. After the bed reaches its equilibrium height, the
simulation is restarted with bubble tracking and
particle movement. The time step of simulation for
the liquid and solid phases is 5 ms.
The experimental results in Fig. 24 were obtained in
a two-dimensional column with a thickness of 7 mm.
The solids holdup, the liquid velocity, and the liquid
and solids properties are the same as the simulation
conditions. As shown in the gure, the simulated
bubble rise velocity and bubble shape generally agree
well with experimental results, indicating the validity
of the computational model.
The pressure eect on the single bubble rise beha-
vior can be clearly seen from the model simulations.
The simulated behavior of a single bubble rising in a
liquidsolid suspension at elevated pressures
(P = 17.3 MPa) is shown in Fig. 25 (Zhang et al.,
2000). The size of the simulation domain is the same
as in the previous case. Paratherm NF heat transfer
uid, nitrogen, and glass beads with a diameter of
0.88 mm are used as liquid, gas, and solid phases,
respectively. The numerical simulation indicates that
the bubble rise velocity decreases with an increase in
pressure, and it is in good agreement with the experi-
mental data and the prediction by the FanTsuchiya
equation. Elevated pressure also causes the bubble
shape to become atter owing to the variation of prop-
erties inside the bubble.
The simulation also provides some information
about the bubbleparticle interaction as shown in
Fig. 25. As the bubble rises in the liquidsolid uidized
bed, an interaction between the bubble and particles
takes place. In the simulation model, the bubble-par-
ticle interaction is accounted for by adding a surface-
tension-induced force to the particle motion equation.
This force is also added to the source term of the liquid
momentum equation for the liquid elements in the
interfacial area to account for the particle eect on
the interface. The particle movement is determined
from the resulting total force acting on the particle.
From the simulation results, it is seen that most par-
ticles contacting the bubble do not penetrate the bub-
ble; only one or two particles penetrate. Instead, they
pass around the bubble surface. When the particles
penetrate the bubble, they fall through quickly to the
bubble base because of the low viscosity and density of
the gas phase.
The last example demonstrated in this section is the
simulation of multibubble formation dynamics (Li et
al., 2001). Multibubble formation behavior is more
complicated, and even dicult to study experimen-
tally. However, the simulation can provide more infor-
mation about the dynamics of multibubble formation,
particularly for cases with interaction between bubbles
formed from dierent orices. The simulation of two-
bubble formation at high pressures (P = 6.6 MPa) in
Figure 24 Simulated and experimental results of a bubble
rising in a liquidsolid uidized bed under ambient condi-
tions (nitrogen80% glycerine solution1.0 mm glass beads,
d
b
= 10.0 mm, e
s
= 0.44). (From Fan et al., 1999.)
Paratherm liquids is shown in Fig. 26. In the simula-
tions, an 80 mm (width) 50 mm (height) domain is
used with two orices at the bottom, an exit at the top,
and two sidewalls as boundary conditions. The liquid
initially lls the domain to a certain height. At the
beginning of the simulation, gas is injected into the
liquid through two orices. ParathermNF heat transfer
uid and nitrogen are used as liquid and gas phases,
respectively. The gas inlet velocity is 14.0 cm/s. The
diameter of each orice is 6 mm. To study the interac-
tion between bubbles formed from dierent orices, the
orice separation distance is set as 20 mm. As can be
seen from the gure, due to the closeness of the orices,
the uctuations of ow eld induced by the bubble
forming from one orice have an eect on the bubble
forming from the other orice. This interactive eect
yields varied bubble shapes and an alternate detach-
ment pattern of bubbles from two orices. Initially
the bubble from each orice rises rectilinearly (Fig.
26a); after detachment, however, the bubbles tend to
move to the centerline, rise in zigzag fashion, and break
up at the free surface (Figs. 26df). Based on simulation
results, it is found that owing to the closeness of the
orices, two forming bubbles induce a high liquid velo-
city in the area near the centerline. The high velocity
thus results in a lower pressure in this area and draws
bubbles toward the centerline. In other words, bubble
and bubble wake interaction leads to a zigzag bubble
rising path, and induces a complicated wake ow eld,
which, in turn, aects the bubble formation behavior.
From the above examples, the computational uid
dynamics simulation is able to capture the dynamic
behavior of each individual phase in three-phase
ows, and it also demonstrates its great potential in
other elds.
5 SUMMARY
Gasliquidsolid three-phase uidized bed systems
have been widely used in many industrial applications.
Successful application of three-phase uidization sys-
tems lies in a comprehensive understanding of bubble
dynamics, hydrodynamics, and transport phenomena,
particularly under high pressure and temperature con-
ditions. High-pressure and high-temperature operation
of three-phase uidized beds is commonly encountered
in most industrial applications of commercial interest.
The ow characteristics of reactors at high pressure
and temperature are distinct from ambient conditions;
for example, elevated pressure leads to higher gas
holdup and smaller bubble size in the system, and
thus dramatically aects transport phenomena includ-
ing heat and mass transfer, and phase mixing. The
eects of pressure and temperature on uid dynamics
and transport properties are mainly in turn due to
Figure 25 Simulation of a single bubble rising in a liquidsolid suspension at high pressure (nitrogen-Paratherm NF heat
transfer uid0.88 mm glass beads, P = 17.3 MPa, e
s
= 0.384, d
b
= 7.5 mm). (From Zhang et al., 2000.)
variations in bubble characteristics and changes in the
physical properties of uid phases. General ow char-
acteristics in a three-phase uidized bed are described
in this chapter with specic attention given to high-
pressure phenomena.
To date, the design of three-phase uidized beds still
relies heavily on experimental observations, empirical
correlations, and engineering models. With increasing
computer power, the employment of the computa-
tional uid dynamics approach has gained consider-
Figure 26 Simulation of two-bubble formation in Paratherm NF heat transfer uids with bubblebubble interactions
(P = 6.6 MPa, D
o
= 6 mm, L
or
= 20 mm). (From Li et al., 2001.)
able attention in recent years. The computational
approach oers viable and attractive options that com-
plement the traditional experimental approach. In this
chapter, a discrete phase simulation method is demon-
strated in the simulation of three-phase ows. This
method has proven to be eective in simulating bubble
behavior and uid dynamics in three-phase uidization
systems. However, many challenges still remain in
computation, such as how to incorporate kinetics
and transport properties into uid dynamics calcula-
tions, how eectively to simulate the full ow eld in
large columns, and how to employ properly the closure
relationships into turbulence calculations. Further-
more, more experimental studies on the liquid turbu-
lence induced by rising bubbles are required for
improving modeling and exploring the mechanisms
of various transport phenomena.
NOMENCLATURE
A = cross-sectional area of particle
a = half-width of bubble; interfacial area
b = bubble breadth
C = tracer concentration
C
D
= drag coecient
C
/
D
= modied drag coecient
C
pl
= liquid heat capacity
C
s
= solids concentration in a liquidsolid
mixture
c = parameter in FanTsuchiya equation
reecting surface tension eect; half-
height of ellipsoidal bubble
D
c
= column diameter
D
max
= maximum stable bubble size
D
o
= orice diameter
d
b
= volume equivalent bubble diameter
d
/
b
= dimensionless bubble diameter
d
p
= particle diameter
d
vs
= Sauter mean bubble diameter
E = solid phase axial dispersion coecient
E(
1 a
2
_
) = complete second kind elliptic integral
Eo = Eo tvo s number based on bubble
diameter
Eo
/
= Eo tvo s number based on particle
diameter
E
rl
= radial liquid dispersion coecient
E
zl
= axial liquid dispersion coecient
e = restitution coecient
F
AM
= added mass force
F
B
= eective buoyancy force; buoyancy force
F
BA
= Basset force
F
bp
= bubbleparticle interaction force
F
C
= particlebubble collision force
F
D
= liquid drag force
F
fp
= uidparticle interaction force
F
G
= gravity force
F
i
= other forces
F
I,g
= bubble inertial force
F
I,m
= liquidsolid suspension inertial force
F
M
= gas momentum force
F
total
= total force
F
x
= x-component of centrifugal force
induced by gas circulation inside bubble
F
s
= surface tension force
Fr = Froude number
f = correction factor for particle collision
f
b
= volumetric body force
f
bf
= volumetric bubbleuid interaction force
f
k
= friction coecient
f
pf
= volumetric particleuid interaction
force
f d
b
( ) = probability density function of bubble
size
H = column height
h = level of liquid or liquidsolid suspension
in the column; time-averaged heat
transfer coecient; separation distance
from the center of the approaching
particle to the symmetric plane of two
colliding particles
h
/
= heat transfer coecient in liquidsolid
uidized beds
h
i
= instantaneous heat transfer coecient
I = moment of inertia
K = proportionality constant for calculating
the eective viscosity of liquidsolid
suspensions
K
b
= parameter in FanTsuchiya equation
reecting viscous nature of surrounding
medium
K
b0
= proportionality constant in Fan
Tsuchiya equation
k = ratio of wake size (or volume) to bubble
size (or volume) in the generalized wake
model
k
l
= liquid thermal conductivity; liquidphase
mass transfer coecient
k
o
= orice constant
L = circumference of the ellipsoidal bubble;
length of the heat transfer probe
L
or
= distance between two orices
l = thickness of the liquid lm between two
coalescing bubbles
Mo = Morton number based on liquid
properties
Mo
m
= modied Morton number based on
slurry properties
m
p
= particle mass
N
c
= dimensionless capacitance number
Nu
/
= Nusselt number in liquidsolid uidized
beds
n = parameter in FanTsuchiya equation
reecting system purity; Richardson
Zaki index in the generalized wake
model
n
i
= number of bubbles
P = system pressure
P
b
= pressure in the bubble
P
c
= pressure in the gas chamber
P
e
= pressure at the gas inlet to the chamber
Pe
z
= axial liquid Peclet number
P
0
= hydrostatic pressure at the bubble
surface
Pr = Prandtl number based on liquid
properties
Pr
m
= Prandtl number based on slurry
properties
p = scalar pressure
p
s
= surface pressure
p
v
= vapor pressure
Q = property of uid
Q
g
= volumetric gas ow rate into the gas
chamber
Q
+
g
= property of gas
Q
+
m
= property of liquidsolid suspension
Q
0
= volumetric gas ow rate through the
orice
R = column radius
R
d
= radius of a contacting circle between two
bubbles
Re = bubble Reynolds number based on
liquid properties; particle Reynolds
number based on mean stream velocity
Re
/
= Reynolds number based on particle
diameter
Re
m
= Reynolds number based on slurry
properties
Re
p
= particle Reynolds number
r = r-axis in a cylindrical coordinate system
r
b
= radius of bubble
r
c
= radius in a cylindrical coordinate system
r
o
= radius of orice
r
p
= radius of particle
S = rate-of-strain tensor
St
m
= Stanton number based on slurry
properties
T = temperature
T = matrix transform
t = time
t
c
= contact time between liquid element and
lm
U = solids velocity in the sedimentation
dispersion model; mean stream velocity
U
N
= normal velocity of particle
U
T
= tangential velocity of particle
U
b
= bubble rise velocity relative to the liquid
phase
U
bo
= terminal bubble rising velocity
U
g
= supercial gas velocity
U
g,tran
= regime transition gas velocity
U
l
= supercial liquid velocity
U
lmf
= liquid minimum uidization velocity
u = rise velocity of bubble base
u
b
= bubble rise velocity relative to the liquid
phase; bubble rise velocity in a stream of
bubbles
u
/
b
= dimensionless bubble rise velocity
uu
b
= average bubble rise velocity
u
e
= bubble expansion velocity
u
m
= suspension velocity
u
max
= rise velocity of maximum stable bubble
u
o
= supercial gas velocity through the
orice
u
pt,0
= particle terminal velocity in the uidizing
liquid at the ambient pressure
u
r
= relative velocity between liquid and gas
bubble
u
small
= small bubble rise velocity
u
t
= particle terminal velocity in liquid
u
x
= x-component of the circulation velocity
of gas inside a bubble
V = liquid velocity in the sedimentation
dispersion model
V
b
= bubble volume
V
c
= volume of gas chamber; average liquid
circulation velocity
V
p
= particle volume
v = liquid velocity vector
v
p
= particle velocity vector
x = distance between bubble center and
orice plate; ratio of solids
concentration in the bubble wake region
to that in the liquidsolid uidized
region in the generalized wake model
x
k
p
= location vector of particle k
y = lateral displacement from the axis of
symmetry of a bubble
z = z-axis in a cylindrical coordinate system
v) = ensemble-average axial liquid velocity
v
/
v
/
) = Reynolds axial normal stress
Greek Letters
a = aspect ratio of bubble; thermal
diusivity; volume fraction of uid
a
cr
= critical angle for sticking collision
b
d
= ratio of particle density to liquid density
b
U
= ratio of supercial gas velocity to liquid
velocity
d = thickness of uid lm surrounding the
heating surface; delta function
e
g
= gas holdup
e
g,tran
= gas holdup at the regime transition point
e
l
= liquid holdup
e
s
= solids holdup
e
sc
= critical solids holdup
e
s0
= solids holdup at incipient uidization
f = particle sphericity; parameter reecting
the surface drag in the equation
calculating lm thinning velocity;
correction factor; azimuthal angle in
spherical or cylindrical coordinates
g = contact angle between bubble and orice
surface; heat capacity ratio
k = free surface curvature
l
c
= critical wavelength for bubble breakup
m
g
= gas viscosity
m
l
= liquid viscosity
m
m
= eective viscosity of liquidsolid
suspension
n
l
= liquid kinematic viscosity
r
g
= gas density
r
l
= liquid density
r
m
= density of liquidsolid suspension
r
p
= particle density
r
s
= solids density
s = surface tension
t = viscous stress tensor
= computational domain
o = angular velocity of particle
x = correction factor reecting particle
eects on the slurry viscosity in the gas
holdup correlation
z = coecient of suspension inertial force
V = volume of a computational cell
r = density dierence between liquid and gas
phases
REFERENCES
Allen MP, Tildesley DJ. Computer Simulation of Liquids.
Oxford: Clarendon Press, 1987.
Attia YA, Elzeky M, Bavarian F, Fan LS. Cleaning and
desulfurization of high sulfur coal by selective occulation
and bioleaching in a draft tube uidized bed reactor. In:
Dugan P, Quigly D, Attia YA, eds. Processing and
Utilization of High Sulfur Coals IV. Elsevier, 1991,
pp 181201.
Azbel D. Two-Phase Flows in Chemical Engineering.
Cambridge, UK: Cambridge University Press, 1981.
Baird M, Rice R. Axial dispersion in large scale unbaed
columns. Chem Eng J 9:171174, 1975.
Baker CGJ, Armstrong ER, Bergougnou MA. Heat transfer
in three-phase uidized beds. Powder Technol
21:195204, 1978.
Batchelor GK. The stability of a large gas bubble rising
through liquid. J Fluid Mech 184:399422, 1987.
Batchelor GK. A new theory of the instability of a uniform
uidized bed. J Fluid Mech 193:75110, 1988.
Bavarian F, Fan LS. Mechanisms of hydraulic transport of a
packed bed at the start-up of a three-phase uidized bed.
Chem Eng Sci 46:30813087, 1991a.
Bavarian F, Fan LS. Hydraulic transport of a packed bed
during the start-up of a three-phase uidized bed with
large gas holdups. Ind Eng Chem Res 30:408414, 1991b.
Bellman R, Pennington RH. Eect of surface tension and
viscosity on Taylor instability. Quarterly of Applied
Math 12:151162, 1954.
Bhaga D, Weber ME. Bubbles in viscous liquids: shapes,
wakes and velocities. J Fluid Mech 105:6185, 1981.
Bhatia VK, Epstein N. Three phase uidization: a general
wake model. In: Angelino H, Couderc JP, Gibert H,
Laguerie C, eds. Fluidization and Its Applications.
Toulose: Cepadues-editions, 1974, pp 372392.
Boris JP. New directions in computational uid dynamics.
Annu Rev Fluid Mech 21:345385, 1989.
Brackbill JU, Kothe DB, Zemach C. A continuum method
for modeling surface tension. J Comput Phys
100:335354, 1992.
Briens LA, Briens CL, Margaritis A, Hay J. Minimum liquid
uidization velocity in gasliquidsolid uidized beds of
low-density particles. Chem Eng Sci 52:42314238, 1997.
Briens CL, Ibrahim YAA, Margaritis A, Bergougnou MA.
Eect of coalescence inhibitors on the performance of
three-phase inverse uidized-bed columns. Chem Eng
Sci 54:49754980, 1999.
Buere P, Fonade C, Moletta R. Liquid mixing and phase
hold-ups in gas producing uidized bed bioreactors.
Chem Eng Sci 53:617627, 1998.
Calderbank PH, Moo-Young MB. Continuous phase heat
and mass transfer properties of dispersions. Chem Eng
Sci 16:3954, 1961.
Cassanello M, Larachi F, Marie MN, Guy C, Chaouki J.
Experimental characterization of the solid phase chaotic
dynamics in three-phase uidization. Ind Eng Chem Res
34:29712980, 1995.
Cassanello M, Larachi F, Guy C, Chaouki J. Solids mixing
in gasliquidsolid uidized beds: experiments and mod-
elling. Chem Eng Sci 51:20112020, 1996.
Chang MY, Morsi BI. Mass transfer in a three-phase reactor
operating at elevated pressures and temperatures. Chem
Eng Sci 47:17791790, 1992.
Chaudhari RV, Homann H. Coalescence of gas bubbles in
liquids. Reviews in Chemical Engineering 10:131190,
1994.
Chen YM, Fan LS. Bubble breakage mechanisms due to
collision with a particle in a liquid medium. Chem Eng
Sci 44:117132, 1989a.
Chen YM, Fan LS. Bubble breakage due to particle collision
in a liquid medium: particle wettability eects. Chem Eng
Sci 44:27622767, 1989b.
Chen Z, Zheng C, Feng Y, Hofmann H. Distributions of
ow regimes and phase hold-ups in three-phase uidized
beds. Chem Eng Sci 50:21532159, 1995.
Chern SH, Muroyama K, Fan LS. Hydrodynamics of
constrained inverse uidization and semiuidization in a
gasliquidsolid system. Chem Eng Sci 38:11671174,
1983.
Chern SH, Fan LS, Muroyama K. Hydrodynamics of cocur-
rent gasliquidsolid semiuidization with a liquid as the
continuous phase. AIChE J 30:288294, 1984.
Chiu TM, Ziegler EN. Heat transfer in three-phase uidized
beds. AIChE J 29:677685, 1983.
Clark KN. The eect of high pressure and temperature on
phase distributions in a bubble column. Chem Eng Sci
45:23012307, 1990.
Clift R, Grace JR, Weber ME. Bubbles Drops, and Particles.
New York: Academic Press, 1978.
Daly JG, Patel SA, Bukur DB. Measurement of gas holdups
in a three-phase bubble column by gamma-ray densitome-
try. In: Large JF, Laguerie C, eds. Fluidization VIII. New
Darton RC, Harrison D. The rise of single gas bubbles in
liquid uidised beds. Trans Instn Chem Engrs 52:301306,
1974.
Davies RM, Taylor G. The mechanics of large bubbles rising
through extended liquids and through liquids in tubes.
Proc Roy Soc London A200:375390, 1950.
Deckwer WD. On the mechanism of heat transfer in bubble
column reactors. Chem Eng Sci 35:13411346, 1980.
Deckwer WD, Burckhart R, Zoll G. Mixing and mass trans-
fer in tall bubble columns. Chem Eng Sci 29:21772188,
1974.
Deckwer WD, Louisi Y, Zaidi A, Ralek M. Hydrodynamic
properties of the FischerTropsch slurry process. Ind Eng
Chem Process Des Dev 19:699708, 1980.
Degaleesan S, Dudukovic MP, Toseland BA, Bhatt BL. A
two-compartment convective-diusion model for slurry
bubble column reactors. Ind Eng Chem Res
36:46704680, 1997.
Di Felice R. Hydrodynamics of liquid uidization. Chem
Eng Sci 50:12131245, 1995.
El-Temtamy SA, Epstein N. Contraction or expansion of
three-phase uidized beds containing ne/light solids.
Can J Chem Eng 57:520522, 1979.
Fan LS, GasLiquidSolid Fluidization Engineering.
Stoneham, MA: Butterworth, 1989.
Fan LS. Moving packed bed phenomenon in three-phase
uidization. Powder Technology 103:300301, 1999.
Fan LS, Tsuchiya K. Bubble Wake Dynamics in Liquids and
LiquidSolid Suspensions. Stoneham, MA: Butterworth-
Heinemann, 1990.
Fan LS, Zhu C. Principles of GasSolid Flows. New York:
Cambridge Press, 1998.
Fan LS, Yang GQ, Lee DJ, Tsuchiya K, Luo X. Some
aspects of high-pressure phenomena of bubbles in liquids
and liquidsolid suspensions. Chem Eng Sci
54:46814709, 1999.
Fox JM. FischerTropsch reactor selection. Catal Lett
7:281292, 1990.
Gidaspow D. Multiphase Flow and Fluidization. New York:
Academic Press, 1994.
Gidaspow D, Bahary M, Jayaswal UK. Hydrodynamic mod-
els for gasliquidsolid uidization. Numerical Methods
in Multiphase Flows, FED 185, ASME, New York, 1994,
pp 117129.
Grace JR, Wairegi T, Brophy J. Break-up of drops and
bubbles in stagnant media. Can J Chem Eng 56:38, 1978.
Grevskott S, Sannas BH, Dudukovic MP, Hjarbo KW,
Svendsen HF. Liquid circulation, bubble size distribu-
tions, and solids movement in two- and three-phase
bubble columns. Chem Eng Sci 51:17031713, 1996.
Harlow FH, Welch JE. Numerical calculation of time-depen-
dent viscous incompressible ow of uid with free surface.
Phys Fluids 8:21822193, 1965.
Hidaka N, Kakoi K, Matsumoto T. Flow behavior of a gas
liquidsolid uidized bed in a self-circulating operation
through an external conduit. Ind Eng Chem Res
37:240246, 1998.
Higbie R. The rate of absorption of pure gas into a still liquid
during short periods of exposure. Trans Am Inst Chem
Eng 31:365375, 1935.
Hill MJM. On a spherical vortex. Phil Trans Roy Soc
London 185:213223, 1894.
Hinze JO. Fundamentals of the hydrodynamic mechanism of
splitting in dispersion processes. AIChE J 1:289295,
1955.
Hirt CW, Nichols BD. Volume of uid (VOF) method for
the dynamics of free boundaries. J Comput Phys
39:201225, 1981.
Hoefsloot HCJ, Krishna R. Inuence of gas density on the
stability of homogeneous ow in bubble columns. Ind
Eng Chem Res 32:747750, 1993.
Holcombe NT, Smith DN, Knickle HN, ODowd W.
Thermal dispersion and heat transfer in nonisothermal
bubble columns. Chem Eng Commun 21:135150, 1983.
Houzelot JL, Thiebaut MF, Charpentier JC, Schiber J.
Contribution to the hydrodynamic study of bubble col-
umns. Int Chem Eng 25:645650, 1985.
Idogawa K, Ikeda K, Fukuda T, Morooka S. Behavior of
bubbles of the air-water system in a column under high
pressure. Int Chem Eng 26:468474, 1986.
Idogawa K, Ikeda K, Fukuda T, Morooka S. Formation and
ow of gas bubbles in a pressurized bubble column with a
single orice or nozzle gas distributor. Chem Eng Comm
59:201210, 1987.
Inga JR. Personal communication, 1994.
Inga JR. Scaleup and scaledown of slurry reactors: a new
methodology. PhD diss., University of Pittsburgh, PA,
1997.
Jager B, Espinoza R. Advances in low temperature Fischer
Tropsch synthesis. Catal Today 23:1728, 1995.
Jean RH, Fan LS. Letter to the editor: Bed contraction
criterion for three phase uidization. Can J Chem Eng
65:351352, 1987.
Jean RH, Tang WT, Fan LS. The sedimentationdispersion
model for slurry bubble columns. AIChE J 35:662665,
1989.
Jiang P, Arters D, Fan LS. Pressure eects on the hydrody-
namic behavior of gasliquidsolid uidized beds. Ind
Eng Chem Res 31:23222327, 1992.
Jiang P, Lin TJ, Luo X, Fan LS. Visualization of high
pressure (21 MPa) bubble column: bubble characteristics.
Chem Eng Res Des 73:269274, 1995.
Jiang P, Luo X, Lin TJ, Fan LS. High temperature and high
pressure three-phase uidizationbed expansion phenom-
ena. Powder Technology 90:103113, 1997.
Johns WF, Clausen GA, Nongbri G, Kaufman H. Texaco
T-star process for ebullated bed hydrotreating/hydro-
cracking. National Petroleum Reners Association
(NPRA) Conference, San Antonio, TX, 1993.
Joshi JB. Axial mixing in multiphase contactorsa unied
correlation. Trans Instn Chem Engrs 58:155165, 1980.
Joshi JB, Lali AM. Velocityholdup relationship in multi-
phase contactorsa unied approach. In: Doraiswamy
LK, Mashelkar RA, eds. Frontiers in Chemical Reaction
Engineering. New York: Wiley Eastern, 1984, pp 314329.
Kalogerakis N, Behie LA. Oxygenation capabilities of new
generation three phase two region bioreactors for micro-
carrier cultures of animal cells. In: Large JF, Laguerie C,
eds. Fluidization VIII. New York: Engineering
Kang Y, Suh IS, Kim SD. Heat transfer characteristics of
three phase uidized beds. Chem Eng Commun 34:113,
1985.
Kastaneck F, Zahradnik J, Kratochvil J, Cermak J.
Modeling of large-scale bubble column reactors for non-
ideal gasliquid systems. In: Doraiswamy LK, Mashelkar
RA, eds. Frontiers in Chemical Reaction Engineering.
New York: Wiley Eastern, 1984, pp 330344.
Kiared K, Larachi F, Guy C, Chaouki J. Trajectory length
and residence-time distributions of the solids in three-
phase uidized beds. Chem Eng Sci 52:39313939, 1997.
Kim SD, Laurent A. The state of knowledge on heat transfer
in three-phase uidized beds. Int Chem Eng 31:284302,
1991.
Kim SD, Kim HS, Han JH. Axial dispersion characteristics
in three-phase uidized beds. Chem Eng Sci
47:34193426, 1992.
Kitscha J, Kocamustafaogullari G. Breakup criteria for uid
particles. Int J Multiphase Flow 15:573588, 1989.
Kojima H, Okumura B, Nakamura A. Eect of pressure on
gas holdup in a bubble column and a slurry bubble
column. J Chem Eng Japan 24:115117, 1991.
Krishna R, Wilkinson PM, van Dierendonck LL. A model
for gas holdup in bubble columns incorporating the
inuence of gas density on ow regime transitions.
Chem Eng Sci 46:24912496, 1991.
Krishna R, de Swart JWA, Hennephof DE, Ellenberger J,
Hoefsloot CJ. Inuence of increased gas density on
hydrodynamics of bubble column reactors. AIChE J
40:112119, 1994.
Kumar S, Fan LS. Heat-transfer characteristics in viscous
gasliquid and gasliquidsolid systems. AIChE J
40:745755, 1994.
Kumar R, Kuloor NR. The formation of bubbles and drops.
Advances in Chemical Engineering 8:255368, 1970.
Kumar S, Kusakabe K, Raghunathan K, Fan LS.
Mechanism of heat transfer in bubbly liquid and liquid
solid systems: single bubble injection. AIChE J 38:733
741, 1992.
Kumar S, Kusakabe K, Fan LS. Heat transfer in three-phase
uidized beds containing low-density particles. Chem Eng
Sci 48:24072418, 1993a.
Kumar S, Kusakabe K, Fan LS. Heat transfer in three-phase
uidization and bubble-columns with high gas holdups.
AIChE J 39:13991405, 1993b.
Kupferberg A, Jameson GJ. Bubble formation at a sub-
merged orice above a gas chamber of nite volume.
Trans Instn Chem Engrs 47:T241250, 1969.
Lammers JH, Biesheuvel A. Concentration waves and the
instability of bubbly ows. J Fluid Mech 328:6793, 1996.
La Nauze RD, Harris IJ. Gas bubble formation at elevated
system pressures. Trans Instn Chem Engrs 52:337348,
1974.
Larachi F, Iliuta I, Rival O, Grandjean BPA. Prediction of
minimum uidization velocity in three-phase uidized-bed
reactors. Ind Eng Chem Res 39:563572, 2000.
Lee DJ, Luo X, Fan LS. Gas disengagement technique in a
slurry bubble column operated in the coalesced bubble
regime. Chem Eng Sci 54:22272236, 1999.
Lee DJ, McLain BK, Cui Z, Fan LS. Pressure eect on the
ow elds and the Reynolds stresses in a bubble column.
Ind Eng Chem Res 40:14421447, 2001.
Letzel HM. Hydrodynamics and mass transfer in bubble
columns at elevated pressures. PhD diss., Delft
University of Technology, The Netherlands, 1997.
Letzel HM, Schouten JC, van den Bleek CM, Krishna R.
Inuence of elevated pressure on the stability of bubbly
ows. Chem Eng Sci 52:37333739, 1997.
Levich VG. Physiochemical Hydrodynamics. Englewood
Clis, NJ: Prentice Hall, 1962.
LHomme GA. Chemical Engineering of GasLiquidSolid
Catalyst Reactions. Liege: CEBEDOC, 1979.
Li Y, Zhang J, Fan LS. Numerical simulation of gasliquid
solid uidization systems using a combined CFD-VOF-
DPM method: bubble wake behavior. Chem Eng Sci
54:51015107, 1999.
Li Y, Yang GQ, Zhang JP, Fan LS. Numerical studies of
bubble formation dynamics in gasliquidsolid uidiza-
tion at high pressures. Powder Technology 116:246260,
2001.
Lin TJ, Fan LS. Characteristics of high-pressure liquidsolid
uidization. AIChE J 43:4557, 1997.
Lin JN, Banerji SK, Yasuda H. Role of interfacial tension in
the formation and the detachment of air bubbles: a single
hole on a horizontal plane immersed in water. Langmuir
10:936942, 1994.
Lin TJ, Tsuchiya K, Fan LS. Bubble ow characteristics in
bubble columns at elevated pressure and temperature.
AIChE J 44:545560, 1998.
Lin TJ, Tsuchiya K, Fan LS. On the measurements of regime
transition in high pressure bubble columns. Can J Chem
Eng 77:370374, 1999.
Linek V, Benes P, Vacek V. Dynamic pressure method for k
l
a
measurement in large-scale bioreactors. Biotechnol
Bioeng 33:14061412, 1989.
Linek V, Benes P, Sinkule J, Moucha T. Non ideal pressure
step method for k
1
a measurement. Chem Eng Sci
48:15931599, 1993.
Luewisutthichat W, Tsutsumi A, Yoshida K. Bubble charac-
teristics in multi-phase ow systems: bubble sizes and size
distributions. J Chem Eng Japan 30:461466, 1997.
Luo X, Jiang P, Fan LS. High pressure three-phase uidiza-
tion: hydrodynamics and heat transfer. AIChE J 43:2432
2445, 1997a.
Luo X, Zhang J, Tsuchiya K, Fan LS. On the rise velocity of
bubbles in liquidsolid suspensions at elevated pressure
and temperature. Chem Eng Sci 52:36933699, 1997b.
Luo X, Yang GQ, Lee DJ, Fan LS. Single bubble formation
in high pressure liquidsolid suspensions. Powder
Technology 100:103112, 1998a.
Luo X, Lee DJ, Lau R, Yang GQ, Fan LS. Maximum stable
bubble size and gas holdup in high-pressure slurry bubble
columns. AIChE J 45:665680, 1999.
Magiliotou M, Chen YM, Fan LS. Bed immersed object heat
transfer in a three-phase uidized bed. AIChE J
34:10431047, 1988.
Maneri CC. New look at wave analogy for prediction of
bubble terminal velocities. AIChE J 41:481487, 1995.
Massimilla L, Majuri N, Signorini P. Gas absorption in
solidliquid uidized systems. Ricerca Sci 29:19341940,
1959.
Massimilla L, Solimando A, Squillace E. Gas dispersion in
solidliquid uidized beds. British Chem Eng 6:232239,
1961.
Matsumoto T, Hidaka N, Takebayasi Y, Morooka S. Axial
mixing and segregation in a gasliquidsolid three-phase
uidized bed of solid particles of dierent sizes and den-
sities. Chem Eng Sci 52:39613970, 1997.
Mei R, Adrian RJ. Flow past a sphere with an oscillation in
the free-stream and unsteady drag at nite Reynolds num-
ber. J Fluid Mech 237:323341, 1992.
Mendelson HD. The prediction of bubble terminal velocities
from wave theory. AIChE J 13:250253, 1967.
Mill PL, Turner JR, Ramachandran PA, Dudukovic MP.
The FischerTropsch synthesis in slurry bubble column
reactors: analysis of reactor performance using the axial
dispersion model. In: Nigam KDP, Schumpe A, eds.
Three-Phase Sparged Reactors. Gordon and Breach,
1996, pp 339386, 679739.
Milne-Thomson LM. Theoretical Hydrodynamics. 3d ed.
London: Macmillan, 1955.
Mindlin RD. Compliance of elastic bodies in contact. J Appl
Mech 16:259268, 1949.
Mitra-Majumdar D, Farouk B, Shah YT. Hydrodynamic
modeling of three-phase ows through a vertical column.
Chem Eng Sci 52:44854497, 1997.
Miura H, Kawase Y. Minimum liquid uidization velocity in
two- and three-phase uidized beds with non-Newtonian
uids. Powder Technology 97:124128, 1998.
Molga EJ, Westerterp KR. Gasliquid interfacial area and
holdup in a cocurrent upow packed bed bubble column
reactor at elevated pressure. Ind Eng Chem Res
36:622631, 1997.
Muroyama K, Yoshikawa T, Takakura S, Yamanaka Y.
Mass transfer from an immersed cylinder in three-phase
systems with ne suspended particles. Chem Eng Sci
52:38613868, 1997.
Nacef S, Poncin S, Wild G. Onset of uidization and ow
regimes in gasliquidsolid uidizationinuence of the
coalescence behavior of the liquid. In: Large JF, Laguerie
C, eds. Fluidization VIII. New York: Engineering
Nikov I, Delmas H. Mechanism of liquidsolid mass transfer
and shear stress in three-phase uidized beds. Chem Eng
Sci 47:673681, 1992.
Oestergaard K. Fluidization. Fluidisation Papers Joint
Symp, London, 1964, pp 5766.
Onozaki M, Namiki Y, Sakai N, Kobayashi M, Nakayama
Y, Yamada T, Morooka S. Dynamic simulation of gas
liquid dispersion behavior in coal liquefaction reactors.
Chem Eng Sci 55:50995113, 2000a.
Onozaki M, Namiki Y, Ishibashi H, Takagi T, Kobayashi
M, Morooka S. Steady-state thermal behavior of coal
liquefaction reactors based on NEDOL process. Energy
Fuels 14:355363, 2000b.
Oyevaar MH, Bos R, Westerterp KR. Interfacial areas and
gas hold-ups in gasliquid contactors at elevated-
pressures from 0.1 to 8.0 MPa. Chem Eng Sci
46:12171231, 1991.
Peng XD, Toseland BA, Tijm PJA. Kinetic understanding of
chemical synergy under LPDME conditionsonce-
through applications. Chem Eng Sci 54:27872792, 1999.
Pinczewski WV. The formation and growth of bubbles at a
submerged orice. Chem Eng Sci 36:405411, 1981.
Ramachandran PA, Chaudhari RV. Three-Phase Catalytic
Reactors. Gordon and Breach Science, 1983.
Ramakrishnan S, Kumar R, Kuloor NR. Studies in bubble
formationI: Bubble formation under constant ow con-
ditions. Chem Eng Sci 24:731747, 1969.
Reese J, Jiang P, Fan LS. Bubble characteristics in three-
phase systems used for pulp and paper processing.
Chem Eng Sci 51:25012510, 1996.
Reilly IG, Scott DS, de Bruijn TJW, MacIntyre D. The role
of gas phase momentum in determining gas holdup and
hydrodynamic ow regimes in bubble column operations.
Can J Chem Eng 72:312, 1994.
Reiss LP, Hanratty TJ. An experimental study of the
unsteady nature of the viscous sublayer. AIChE J
9:154160, 1963.
Richardson JF, Roman MN, Shakiri KJ. Heat transfer from
immersed surfaces in liquid uidized beds. Chem Eng Sci
31:619624, 1976.
Rowe PN, Henwood GA. Drag forces in a hydraulic model
of a uidized bedPart I. Trans Instn Chem Engrs
39:4354, 1961.
Ruzicka MC, Drahos J, Zahradnik J, Thomas NH.
Intermittent transition from bubbling to jetting regime
in gasliquid two phase ows. Int J Multiphase Flow
23:671682, 1997.
Saberian-Broudjenni M, Wild G, Charpentier JC, Fortin Y,
Euzen JP, Patoux R. Contribution to the hydrodynamic
study of uidized gasliquidsolid reactors. Entropie
120:3044, 1984.
Safoniuk, M, Grace JR, Hackman L, McKnight CA. Use of
dimensional similitude for scale-up of hydrodynamics in
three-phase uidized beds. Chem Eng Sci 54:49614966,
1999.
Sagert NH, Quinn MJ. Inuence of high-pressure gases on
the stability of thin aqueous lms. J Colloid Interface Sci
61:279286, 1977.
Sagert NH, Quinn MJ. Surface viscosities at high pressure
gasliquid interfaces. J Colloid Interface Sci 65:415422,
1978.
Sangnimnuan A, Prasad GN, Agnew JB. Gas hold-up and
backmixing in a bubble-column reactor under coal-hydro-
liquefaction conditions. Chem Eng Commun 25:193212,
1984.
Sankaranarayanan K, Shan X, Kevrekidis IG, Sundaresan
S. Analysis of drag and virtual mass forces in bubble
suspensions using an implicit formulation of the Lattice
Boltzmann method, J. Fluid Mech., in press, 2002.
Saxena SC. Bubble column reactors and FischerTropsch
synthesis. Catal Rev Sci Eng 37:227309, 1995.
Schiller L, Naumann A. U
ber die grundlegenden berechnun-

gen bei der schwerkraftaufbereitung. Ver Deut Ing
77:318324, 1993.
Shah YT. GasLiquidSolid Reactor Design. McGraw-Hill,
1979.
Shah YT, Kelkar BG, Godbole SP, Deckwer WD. Design
parameter estimations for bubble column reactors.
AIChE J 28:353379, 1982.
Shnip AI, Kolhatkar RV, Swamy D, Joshi JB. Criteria for
the transition from the homogeneous to the heteroge-
neous regime in two-dimensional bubble column reac-
tors. Int J Multiphase Flow 18:705726, 1992.
Song GH, Bavarian F, Fan LS, Buttke RD, Peck LB.
Hydrodynamics of three-phase uidized bed containing
cylindrical hydrotreating catalysts. Can J Chem Eng
67:265275, 1989.
Soong Y, Harke FW, Gamwo IK, Schehl RR, Zarochak
MF. Hydrodynamic study in a slurry-bubble-column
reactor. Catalysis Today 35:427434, 1997.
Souza GLM, Afonso JC, Schmal M, Cardoso JN. Mild
hydrocracking of an unstable feedstock in a three-phase
uidized-bed reactor: behavior of the process and of the
chemical compounds. Ind Eng Chem Res 31:21272133,
1992.
Stegeman D, Knop PA, Wijnands AJG, Westerterp KR.
Interfacial area and gas holdup in a bubble column reac-
tor at elevated pressures. Ind Eng Chem Res 35:3842
3847, 1996.
Stewart PSB, Davidson JF. Three-phase uidization: water,
particles, and air. Chem Eng Sci 19:319321, 1964.
Suh IS, Deckwer WD. Unied correlation of heat transfer
coecients in three-phase uidized beds. Chem Eng Sci
44:14551458, 1989.
Suh IS, Jin GT, Kim SD. Heat transfer coecients in three
phase uidized beds. Int J Multiphase Flow 11:255259,
1985.
Tarmy B, Chang M, Coulaloglou C, Ponzi P. Hydrodynamic
characteristics of three-phase reactors. Chemical Engineer
407:1823, 1984.
Tomiyama A, Kataoka I, Sakaguchi T. Drag coecients of
bubbles. 1. Drag coecients of a single bubble in a stag-
nant liquid. Nippon Kikai Gakkai Ronbunshu B-hen
61:23572364, 1995.
Tsuchiya K, Nakanishi O. Gas holdup behavior in a tall
bubble column with perforated plate distributors. Chem
Eng Sci 47:33473354, 1992.
Tsuchiya K, Furumoto A, Fan LS, Zhang J. Suspension
viscosity and bubble velocity in liquidsolid uidized
beds. Chem Eng Sci 52:30533066, 1997.
Tsuge H, Hibino S. Bubble formation from an orice sub-
merged in liquids. Chem Eng Comm 22:6379, 1983.
Tsuge H, Nakajima Y, Terasaka K. Behavior of bubbles
formed from a submerged orice under high system pres-
sure. Chem Eng Sci 47:32733280, 1992.
Tsuji Y, Kawaguchi T, Tanaka T. Discrete particle simula-
tion of two-dimensional uidized bed. Powder Technol
77:7987, 1993.
Tsutsumi A, Nieh JY, Fan LS. Particle wettability eects on
bubble wake dynamics in gasliquidsolid uidization.
Chem Eng Sci 46:23812384, 1991a.
Tsutsumi A, Dastidar AG, Fan LS. Characteristics of gas
liquidsolid uidization with nonwettable particles.
AIChE J 37:951952, 1991b.
Turner R. Fluidization in the petroleum industry.
Fluidisation Papers Joint Symp, London, 1964, pp 4756.
Tzeng JW, Fan LS, Gan YR, Hu TT. Communication to the
editor: Ethanol fermentation using immobilized cells in
multistage uidized bed bioreactor. Biotech Bioeng
38:12531258, 1991.
Unverdi SO, Tryggvason G. A front-tracking method for
viscous, incompressible, multi-uid ows. J Comput
Phys 100:2537, 1992.
Vrij A. Possible mechanism for the spontaneous rupture of
thin, free liquid lms. Discuss Faraday Soc 42:2333,
1966.
Wallis GB. One-Dimensional Two-Phase Flow. New York:
McGraw-Hill, 1969.
Walter JF, Blanch HW. Bubble break-up in gasliquid
bioreactors: break-up in turbulent ows. Chem Eng J
32:B7B17, 1986.
Wasan DT, Ahluwalia MS. Consecutive lm and surface
renewal mechanism for heat and mass transfer from a
wall. Chem Eng Sci 24:15351542, 1969.
Wen CY, Yu YH. Mechanics of uidization. Chem Eng
Progr Symp Ser 62:100111, 1966.
Wilkinson PM, van Dierendonck LL. Pressure and gas den-
sity eects on bubble break-up and gas hold-up in bubble
columns. Chem Eng Sci 45:23092315, 1990.
Wilkinson PM, van Dierendonck LL. A theoretical model
for the inuence of gas properties and pressure on
single-bubble formation at an orice. Chem Eng Sci
49:14291438, 1994.
Wilkinson PM, Spek AP, van Dierendonck LL. Design para-
meters estimation for scale-up of high-pressure bubble
columns. AIChE J 38:544554, 1992.
Wilkinson PM, Haringa H, Stokman FPA. Liquid mixing in
a bubble column under pressure. Chem Eng Sci
48:17851191, 1993.
Wilkinson PM, Haringa H, van Dierendonck LL. Mass
transfer and bubble size in a bubble column under pres-
sure. Chem Eng Sci 49:14171427, 1994.
Yang GQ, Fan LS. Axial liquid-phase mixing in high-
pressure bubble columns. AIChE J, In press, 2002.
Yang GQ, Luo X, Lau R, Fan LS. Bubble formation in high-
pressure liquidsolid suspensions with plenum pressure
uctuation. AIChE J 46:21622174, 2000a.
Yang GQ, Luo X, Lau R, Fan LS. Heat-transfer character-
istics in slurry bubble columns at elevated pressures and
temperatures. Ind Eng Chem Res 39:25682577, 2000b.
Yoo DH, Tsuge H, Terasaka K, Mizutani K. Behavior of
bubble formation in suspended solution for an elevated
pressure system. Chem Eng Sci 52:37013707, 1997.
Zahradnik J, Fialova M, Ruzicka M, Drahos J, Kastanek F,
Thomas NH. Duality of the gasliquid ow regimes in
bubble column reactors. Chem Eng Sci 52:38113826,
1997.
Zaidi A, Deckwer WD, Mrani A, Benchekchou B.
Hydrodynamics and heat transfer in three-phase uidized
beds with highly viscous pseudoplastic solutions. Chem
Eng Sci 45:22352238, 1990.
Zhang JP, Grace JR, Epstein N, Lim KS. Flow regime iden-
tication in gasliquid ow and three-phase uidized
beds. Chem Eng Sci 52:39793992, 1997.
Zhang JP, Epstein N, Grace JR. Minimum uidization velo-
cities for gasliquidsolid three-phase systems. Powder
Technology 100:113118, 1998a.
Zhang JP, Epstein N, Grace JR. Minimum liquid velocity for
particle transport from gasliquid uidized beds. In: Fan
LS, Knowlton TM, eds. Fluidization IX. New York:
Engineering Foundation, 1998b, pp 645652.
Zhang J, Fan LS, Zhu C, Pfeer R, Qi D. Dynamic behavior
of collision of elastic spheres in viscous uids. Powder
Technology 106:98109, 1999.
Zhang J, Li Y, Fan LS. Numerical studies of bubble and
particle dynamics in a three-phase uidized bed at ele-
vated pressures. Powder Technology 112:4656, 2000.
Zheng C, Chen Z, Feng Y, Hofmann H. Mass transfer in
dierent ow regimes of three-phase uidized beds. Chem
Eng Sci 50:15711578, 1995.
Zhou LX, Yang M, Lian CY, Fan LS, Lee DJ. On the
second-order moment turbulence model for simulating a
bubble column. Chem Eng Sci 57:32693281, 2002.
28
LiquidSolids Separation
Shiao-Hung Chiang, Daxin He, and Yuru Feng
1 INTRODUCTION
Liquidsolids separation represents a group of unit
operations widely used by chemical, mineral, paper,
food, biotechnology, water treatment, waste remedia-
tion and other activities. Technically, it involves the
separation, removal, and collection of solid particulate
matter existing in a dispersed or colloidal state in a
liquid suspension. The separation is most often per-
formed by employing mechanical forces causing uid
ow through a complex media structure. Thus the
design of a liquidsolids separation process requires a
combined knowledge of uid mechanics, particle
mechanics, and material properties of the media.
Generally, the separation of solid particulates from
liquid suspensions consists of one or more process
steps: pretreatment, solids concentration, solids
separation, and posttreatment. To implement these
steps four major unit operations are commonly used:
1. Filtration (cake ltration, deep-bed ltration,
and membrane ltration)
2. Sedimentation (thickening and clarication)
3. Centrifugation (centrifugal ltration and sedi-
mentation)
4. Hydrocyclones (classication and clarication)
These unit operations are integral parts of many indus-
trial processes mainly to serve the purposes of recover-
ing
1. Valuable solids (the liquid being discarded)
2. The liquid (the solids being discarded)
3. Both the solids and the liquid
4. Neither (e.g., to prevent water pollution)
In all cases, the liquidsolids separation plays a crucial
role in producing materials to meet the desired product
quality and in maintaining environmental protection.
1.1 LiquidSolids Systems
There are three major classes of liquidsolids systems:
aqueous systems, nonaqueous systems, and biological
systems. The following provides a brief description of
these systems from the standpoint of liquidsolids
separation.
1.1.1 Aqueous Systems
Water is the most commonly used liquid in industry
and in all human activities. Thus most liquidsolids
separation equipment is specically designed for treat-
ing aqueous systems. In water suspensions, the sizes of
solid particles are usually very small (in the micron and
submicron size range) and surface-active forces often
play an important role. Representative examples are
the mineral slurries, lyophilic colloids, slimes, and
wastewater mixtures. The treatment of aqueous sys-
tems involves four stages: occulation (clarication),
sedimentation (settling), consolidation (compression
or compaction), and phase separation (ltration or
centrifugation).
1.1.2 Nonaqueous Systems
A major problem in the treatment of nonaqueous sys-
tems is the removal of colloidal particles to produce
acceptable liquid products (e.g., fuel oil). Such a situa-
tion is encountered in hydrocarbon production from
tar sands and oil shale. The particles (such as oxides,
silicates, and clay mineral) suspended in the hydro-
carbon liquids originate from a rock matrix. Particle
separation problems occur in the solvent extraction of
bitumen with nonaqueous media such as toluene.
Electrostatic forces (bonding forces) play a predomi-
nant role in the physical state of these nonaqueous
systems. In many instances, by addition of anti-
solvents, and selecting the proper temperature and
agitation, these systems can be altered to improve
solid separation. Separation of carbon black particles
suspended in tetralin using Aerosol OT as a surfactant
and by ltration through a bed of sand (deep-bed l-
tration) is another example of liquidsolids separation
in nonaqueous systems.
1.1.3 Biological Systems
Biological treatment is often employed to decompose
organic substances in wastewater to remove biological
oxygen demand (BOD) or chemical oxygen demand
(COD) components using microorganisms under aero-
bic or anaerobic conditions. As a result of these biolo-
gical reactions, cells (or biomass) are produced as an
aqueous sludge that needs to be separated from the
euent stream. In a microbial fermentation process,
fermentation broth contains a complex aqueous
mixture of cells, soluble extracellular products, intra-
cellular products, and unconverted substrate or uncon-
vertible components. In order to produce the desired
product, a biological system always involves both a
bioreactor and a separation/product recovery section.
The product recovery section usually consists of solids
removal (ltration), primary isolation, purication,
and nal isolation (crystallization) steps.
1.2 Industrial Applications
As mentioned earlier, liquidsolids separation technol-
ogy is basic to many manufacturing industries (chemi-
cal, mineral, food, beverages, etc.) as well as to
pollution abatement and environmental control. It is
dicult to nd any important engineering enterprises
in which liquidsolids separation does not play an
important part. Major industrial and commercial
applications of four key unit operations for liquid
solids separation are summarized here:
Unit Operation Applications
Filtration Suspended, precipitated, and oversize
particle removal; process water, wash
water, waste oil, boiler feed water
cleanup; mineral dewatering; and recov-
ery of valuable products.
Sedimentation Portable water clarication; municipal
treatment; storage pond for toxic waste
and mineral processing waste; and pre-
treatment step for feed to lters and
centrifuges.
Centrifugation Wastewater and sludge thickening;
metalworking coolants cleaning; puri-
cation of marine fuels; dehydration of
tar; and clarication of beer, wine, fruit
juice, and varnishes.
Hydrocyclones Mineral processing and coal cleaning;
cooling oils cleaning and clarication;
and industrial wastewater treatment.
These unit operations will be discussed separately in
the following sections.
2 FILTRATION
The ltration operation involves the separation,
removal, and collection of a discrete phase of matter
existing in suspension. The undissolved solid particles
are separated from the liquid suspension by means of a
porous medium (i.e., lter medium). Filtration leads to
the formation of a cake containing a relatively low
proportion of residual ltrate. Depending upon the
mechanism for arrest and accumulation of particles,
the ltration operation can generally be classied
into three types: cake ltration, deep-bed ltration,
and membrane ltration (see Fig. 1).
2.1 Cake Filtration
Most of the liquid ltration operations follow the
mechanism of cake ltration. As the ltration pro-
ceeds, the particles retained on the lter medium will
form a growing cake with porous structure. The small
particles that are able to pass through the pores initi-
ally will be trapped at a greater depth as they traverse
through this porous cake. This cake becomes the true
lter medium and hence plays a very important part in
the entire ltration operation. The mechanism of ow
within the cake and the external conditions imposed on
the cake are the basis for modeling a ltration process.
Under a pressure gradient, if the particle packing
arrangement in the cake can sustain the drag force
without deformation, the cake is regarded as incom-
pressible. However, stresses developed in the parti-
culate structure usually lead to deformation and
compression with substantial changes in the porosity
and permeability as a load (uid drag) is applied. This
kind of lter cake is known as a compressible cake.
2.1.1 Cake Filtration EquationTwo Resistance
Model
The structure of the deposited cake resembles the
structure of a packed bed. Based on the principles
for ow through a packed bed with the consideration
of resistances oered by both the lter medium and the
lter cake, an equation of cake ltration can be
obtained (Akers and Ward, 1977; Cheremisino and
Azbel, 1989; Tiller et al., 1987a):
dV
dt

AP
macV=A R
m
1
where V is the ltrate volume, t the ltration time, m
the uid viscosity, c the mass of solid per unit volume
of ltrate, A the cross-sectional area of lter, P the
pressure drop across the lter, and R
m
the resistance of
lter medium. The specic cake resistance, a, can be
expressed as
a
1
kr
s
1 e
2
where k is the permeability, r
s
the density of the solid
particles, and e the porosity of the lter cake, dened
as the ratio of the volume of voids in the cake to the
total volume of lter cake.
Equation (1), known as the two-resistance ltration
model, is a simple expression for describing the ltra-
tion of incompressible cakes, the specic cake resis-
tance, a, can be regarded as a constant. In the case
of compressible cakes, the eect of variation in cake
porosity on specic cake resistance must be considered
(Tiller and Shirato, 1964). Filtration tests should be
performed under dierent pressure drops to establish
an empirical relation between the specic cake resis-
tance, a, and the pressure drop across the lter cake,
P
c
(McCabe et al., 1993):
a a
0
P
c
n
3
where a
0
and n are empirical constants. The constant n
represents the compressibility coecient of the lter
cake. For an incompressible cake, n equals zero. The
value of n ranges from 0.25 for moderately compress-
Figure 1 Mechanisms of ltration: (a) cake ltration, (b) deep-bed ltration, (c) cross-ow membrane ltration. (From Chiang
and He, 1995.)
ible to 1.0 for supercompressible cakes (Tiller et al.,
1987b).
2.1.2 Constant Pressure Filtration
For constant pressure ltration, Eq. (1) can be re-
arranged to give
dt
dV

mars
P1 msA
2
V
mR
m
AP
4a
where
rs
1 ms
c 4b
In Eq. (4b), s represents the slurry concentration, r the
ltrate density, and m the mass ratio of wet cake to dry
cake. For incompressible cakes (a constant, Eq.
(4a) may be simplied to
dt
dV
K
0
V R
0
5
where K
0
and R
0
are constants.
2.1.3 Constant Rate Filtration
Many industrial ltration processes can be approxi-
mated as constant rate ltration. In this case, the linear
velocity of ltrate, u V=At, remains constant. The
relation between the overall pressure drop, P, and
the ltration time, t, can be expressed as (McCabe et
al., 1993)
P P
m
1n
a
0
mc
V
At

2
t K
r
t 6
where P
m
is the pressure drop across the lter
medium and K
r
a
0
mcV=At
2
is a constant.
In industrial ltration operation carried out under
variable pressure and variable rate conditions, the
method of Tiller must be employed to integrate Eq.
(1) for the general case (Tiller, 1958).
2.2 Deep-Bed Filtration
Deep-bed ltration (or depth ltration) is known by
various terms like blocking ltration, surface ltration,
and clarication. Deep-bed ltration (Rajagopalan
and Tien, 1979; Stamatakis and Tien, 1993; Tien and
Payatakes, 1979) is normally preferred in treating large
quantities of liquids containing low solid concentration
(less than 500 mg/liter) with particles size less than
30 mm. In this operation, the particles to be removed
are often substantially smaller than the pores of the
lter medium and will penetrate a considerable depth
before being captured. Silica sand, anthracite coal,
active carbon, and bers are most commonly employed
as lter media.
A rational design of a deep-bed ltration process is
based on the rate of clarication (removal of sus-
pended particulate) and the pressure drop (due to
medium clogging) required for maintaining a given
throughput. The ow of a suspension through a
deep-bed of grains (lter media) results in the penetra-
tion of the particles into the lter medium where
deposition takes place on the grain surfaces at various
depths.
2.2.1 Basic Equations for Deep-Bed Filtration
The fundamental equations describing the particle
retention behavior in a deep-bed ltration are the con-
tinuity equation, the rate equation, and the expression
for pressure drop. The removal rate of suspended
solids as a function of solid concentration is written as
@C
@L
lC 7
Where C is the solid concentration in the suspension, L
the distance from the top of the bed to the section
under consideration, and l the lter coecient or
impediment modulus. The variation of l depends on
the extent of particle retention and parameters that
determine the mode of deposit morphology. A pro-
posed expression for l can be written in a general
functional form:
l l
0
Fa; s 8
where l
0
is the lter coecient of the clean lter bed, a
a parameter vector, s the specic deposit (the volume
of deposited matter per unit volume of lter bed), and
Fa; s is a function representing the eect of particle
deposition in the lter bed on the lter coecient, l. A
list of useful expressions for function F may be found
in the reference (Tien, 1989).
2.2.2 Pressure Drop in Deep-Bed Filtration
For a clean lter bed, the pressure drop can be calcu-
lated using the CarmanKozeny equation,
@P
@L

0
150
d
2
p
mu
1 e
0
2
e
3
0
9
where d
p
is the particle diameter in the packed bed.
This can serve as a basis for evaluating the change of
pressure drop due to lter bed clogging by using
empirical correlations in terms of volume occupied by
the deposited particles (Ives and Pienvichitr, 1965;
Ives, 1985; Tien, 1989).
2.3 Membrane Filtration
Membrane separation processes have been applied to
many industrial production systems for the purpose of
clarication, concentration, desalting, waste treatment,
or product recovery. Broadly speaking, membrane l-
tration can be classied as microltration, ultral-
tration, nanoltration, reverse osmosis, and dialysis
or electrodialysis. In this section, the discussion will
only cover microltration and ultraltration, both of
which are pressure-driven membrane processes.
2.3.1 Basic Concepts
A membrane ltration process is illustrated in Fig. 1c.
A suspension ows tangentially to the membrane sur-
face into the lter under an applied pressure. The mem-
brane functions as a lter medium. The components
that pass through the membrane constitute the per-
meate (or ltrate), while the components retained on
the membrane surface form the retentate.
The membrane retention is determined almost solely
by the size and shape of the particles (including macro-
molecules) in suspension. For particle sizes larger than
the pore size of the membrane, capture of particles is
by means of interception. For particles that are smaller
than the pore size, the predominant capture mechan-
ism is inertial impaction. As the particle size and/or
inertial mass decreases, diusion is the primary capture
mechanism. In addition, increased viscosity of the uid
will greatly diminish the eect of inertial impaction
and diusion. In this case, interception would be the
primary collection mechanism.
In a membrane ltration process, the retentate in
suspension may build up a high concentration adjacent
to the membrane surface forming a dyamic boundary
layer (a gel layer). This concentration gradient
becomes a driving force to pull the retentate from the
boundary layer back to the bulk ow. This phenom-
enon is referred to as concentration polarization. The
accumulation of retentate at the membrane surface will
result in a hydraulic resistance that may reduce the
permeability of the membrane. This phenomenon is
called fouling. Membrane fouling is a common phe-
nomenon observed in the operation of any membrane
ltration process, which leads to a reduction in
permeate ux and selectivity.
The retention characteristics of a membrane can be
described using a retention coecient (rejection co-
ecient), which is dened as
R
C
S
C
P
C
S
10
and
R
0

C
b
C
P
C
b
11
where R is the true retention coecient and R
0
the
observed retention coecient. The variables C
S
and
C
b
are the solid concentration in the retentate at the
membrane surface and in the bulk ow, respectively,
and C
P
is the solid concentration in the permeate.
Membrane lters of early design can foul quickly
owing to the concentration polarization eect. To con-
trol this eect, the conguration of a membrane lter is
designed as a cross-ow mode (Belfort, 1986). In cross-
ow conguration, the main direction of the suspen-
sion ow is perpendicular to the permeate ow. As
permeate ow passes through the membrane, a portion
of the particles is deposited at the surface of the mem-
brane to form a cake layer. The cake thickness varies
with the operating time and so does the rate of perme-
ation. Thus the time-dependent behavior represents
a major factor of the cross-ow membrane ltration.
2.3.2 Microltration
Microltration (MF) is a membrane ltration in which
the lter medium is a porous membrane with pore sizes
in the range of 0.0210 mm. It can be utilized to sepa-
rate materials such as clay, bacteria, and colloid par-
ticles. The membrane structures have been produced
from the cellulose ester, cellulose nitrate materials,
and a variety of polymers. A pressure of about 15
atm is applied to the inlet side of suspension ow dur-
ing the operation. The separation is based on a sieve
mechanism. The driving force for ltration is the dif-
ference between applied pressure and back pressure
(including osmotic pressure, if any). Typical congura-
tions of the cross-ow microltration process are illu-
strated in Fig. 2. The cross-ow membrane modules
are tubular (multichannel), plate-and-frame, spiral-
wound, and hollow-ber as shown in Fig. 3. The
design data for commercial membrane modules are
listed in Table 1.
The microltration membranes are known to be
highly porous. Thus the separation behavior within
these membranes is mainly based on pore size. As the
ow paths through the membrane are tortuous with
dead-end passages and variable channel dimensions,
the CarmanKozeny equation may be used. There-
fore, the ux across the microltration membrane J
can be calculated by
J
dV
Adt

e
3
K1 e
2
S
2
p
P
mL
m
12
where K is the Kozeny coecient, depending only on
the pore conguration, S
p
the surface area of a single
particle; and L
m
the eective membrane thickness. All
other parameters are as dened in the previous section.
For more detailed information regarding the per-
formance of microltration, readers are referred to
Microltration and Ultraltration by Zeman and
Zydney (1996).
2.3.3 Ultraltration
Ultraltration (UF) is a membrane ltration process
used for separating or collecting submicron-size par-
ticles (0.001 to 0.02 mm) from a suspension (Cheryan,
1986). It is usually employed to concentrate or frac-
tionate a solution containing macromolecules, colloids,
salts, or sugars. The UF membrane functions as a sieve
with the pore size of molecular dimensions.
A UF process can be operated in either batch or
continuous mode. The determination of membrane
surface area A for the UF process requires three
parameters: (1) ux, J, which is the measure of the
membrane productivity; (2) the volume of permeate,
V
p
, passed through the membrane, and (3) the volume
of retentate, V
r
, retained on the membrane surface.
The average ux, J
av
, can be estimated from the
equation
J
av
J
f
0:33J
i
J
f
13
where J
f
is the nal ux at the highest concentration
and J
i
the initial ux. The material balance gives
V
f
V
r
V
p
14
where V
f
is the volume of feed. The volume concen-
tration ratio (VCR) is dened as
VCR
V
f
V
r
15
Eqs. (13) through (15) can be used to estimate the
membrane surface area (Chiang and He, 1995),
which can be expressed as
A
V
f
V
p
J
av
16
The conguration of UF is usually designed as poly-
meric and asymmetric modules for high productivity
and resistance to plugging. Membrane modules used in
the UF process design are similar to those adopted in
MF process design (see previous section).
2.4 Filter Media
After specifying the lter type and the optimum oper-
ating conditions, the remaining issue for a lter design
is the selection of the most suitable lter medium. In a
ltration process, the fundamental role of a lter med-
ium is to separate eectively the particulates from
a owing uid to provide a suciently clean ltrate
without clogging and damaging the medium (i.e., low
Figure 2 Membrane ltration process congurations.
(From Zeman and Zydney, 1996.)
energy consumption and long operating period). Thus
proper selection of the lter medium is often the most
crucial step for assuring ecient ltration operation.
The major considerations include the permeability of
the medium relative to a pure liquid, its retention
capacity relative to particulates of known size, and
the medium pore size distribution.
The resistance (or permeability) of a lter medium
directly aects the capital and operating costs. Most
manufacturers also employ the permeability of lter
media as a measure of particle retention, which is
related to its pore size and porosity. A high permeabil-
ity of the media is, therefore, used as an indication of
high porosity and in turn a low particle retentivity. The
ideal medium would provide the maximum open (free)
area for ow while it meets the required retentivity. A
comparison of the free area for commonly used lter
media is presented in Table 2.
An alternative classication of lter media is based
on the minimum size of the trapped particles (see Table
3). Obviously, this classication provides only a gen-
eral guideline to the types of media available. In fact,
many process parameters (such as the particulate
concentration, etc.) aect the retention behavior of
Figure 3 Membrane modules for microltration and ultraltration. (a) Tubular membrane module. (From Zeman and Zydney,
1996.) (b) Cassette membrane assembly. (From Perry et al., 1997.) (c) Spiral wound membrane module. (From Purchase, 1996.)
(d) Hollow ber membrane module. (From Zeman and Zydney, 1996.)
Table 1 Design Data for Commercial Membrane Modules
Congur-
ation
Company
Location Product Code
a
Membrane
type
U
m
(ms
1
)
Height or
i.d. (mm)
L
(m)
Membrane
area (m
2
)
V
W
(ms
1
)
Re
c
10
3
Re
w
d
10
3
Ultraltration
Flat plate Dorr-Oliver USA Ioplate US1 Polysulphone 5.012.2 2.54 0.37 0.067 1:894:25 10
4
21.84 48.0108.0
US2 Polyvinylidene uoride 5.012.2 2.54 0.37 0.067 1.801:42 10
4
21.84 30.036.0
US3 VC-AN copolymer 5.012.2 2.54 0.37 0.067 0.951:18 10
4
21.84 24.030.0
Flat plate DDS
Denmark
Lab Unit 35 US4 Polysulphone 0.681.36
b
5.90
b
0.22 0.150 0.696:94 10
5
6.02 4.041.0
Lab Unit 20 US5 Polysulphone 0.100.20
b
5.19
b
0.076 0.018 0.168:10 10
5
0.78 6.042.0
Tube Abcor 24.0
USA
HFA UT1 Cellulose acetate 7.60
b
2.54 3.05 0.204 9:43 10
5
19.30
HFM UT2 Polyvinylidene uoride 11.3622.7
b
2.54 3.05 0.204 1.412:82 10
4
43.26 36.072.0
Tube Romicon-
Amicon
USA
PM UT3 Polysulphone 3.326.48 1.10 0.635 0.022 0.1161:62 10
4
5.39 1.318.0
XM UT4 Modacryl polymer 2.846.02 1.10 0.635 0.0021 1.391:74 10
4
4.87 15.019.0
Tube Berghof
Germany
BM UT5 Polyamidimide 0.0110.266 0.60 0.30 5:65 10
4
0.1166:94 10
5
0.083 0.074.20
Microltration
Tube Enka/
Membrana
Germany
Dyna-Sep
Sampler
MT1 Polypropylene 4.217.97 5.50 1.83 0.032 0.3651:62 10
4
33.44 20.089.0
Microdyne MT2 Polypropylene 1.03.0 1.80 0.56 1.2 0.5562:78 10
4
3.60 10.050.0
a
US ultraltration slit, UT ultraltration tube, MT microltration tube.
b
Estimate
c
v 10
6
m
2
=s for water at 258C
d
Re
w
hV
w
=v or dV
w
=v
Source: Belfort G. In: Muralidhara HS, ed. Advances in SolidLiquid Separation. Columbus: Battelle Press, 1986, pp 182183, Table 2.
particulates within the lter media. The detailed infor-
mation regarding the lter media properties and their
selection can be found in the Handbook of Filter
Media (Purchas, 1996) and the Filters and Filtration
Handbook (Dickenson, 1994).
A good lter medium should have the following
features:
1. A wide size distribution of particles from the
slurry, producing a clean ltrate
2. An economic ltration time, i.e., minimum
ltrate ow resistance
3. Easy discharge of the lter cake from the
medium
4. Sucient strength to withstand ltering pres-
sure and mechanical wear
5. Avoidance wedging of particles into its pores
and an adequate medium lifetime
6. Low cost
2.5 Filter Aids
In order to improve the ltration characteristics of
hard-to-lter suspensions, such as those with slow
ltration rate, rapid medium binding, or unsatisfac-
tory clarity, addition of another particulate solid
material is one of the best alternative measures.
Such solid material is termed a lter aid. Filter aids
added to the suspensions build up a porous, perme-
able, and rigid lattice structure for retaining solid
particles and allowing the liquid to pass through.
As a general rule, a suitable lter aid should be cap-
able of creating a thin layer structure with a maxi-
mum pore size over the mediums external surface
and producing a prespecied ltrate clarity at an opti-
mum ltration rate.
Filter aids may be applied to the ltration operation
in two ways. The rst way is to precoat the lter med-
ium using a precoat lter aid. The precoat is to behave
as the actual lter medium. The function of such an
application is to prevent the lter medium from clog-
ging or fouling as well as to facilitate the removal of
the formed cake at the end of ltration. The second
way is to pretreat the suspension using a lter aid
powder with a coarser size distribution prior to the
ltration process. Such material is called body aid or
admix. The functions of body aid are to increase the
porosity of lter cake and to decrease its compressibil-
ity, resulting in a decrease in the cake resistance and in
turn an increase in the ltration rate.
2.5.1 Requirements for Filter Aid Selection
Filter aid selection should be based on laboratory
tests. The requirements for preliminary evaluation
of the selected lter aids may be summarized as
follows:
1. Form a thin and rigid lattice layer with high
porosity
2. Have low specic surface or coarse size
3. Have a narrow fractional size distribution by
removing the ner size fractions
4. Create a rapid particle bridging and settling or
a uniform lter aid layer
5. Be chemically inert and able to prevent
medium cracking and clogging
These requirements are all found in the two most com-
mon lter aids. The rst one is the diatomaceous silica
type lter aid (also called diatomite, kieselguhr or dia-
tomaceous earth), which contains 90% or almost pure
silica and particle size mostly smaller than 50 mm. Its
bulk density ranges from 128 to 320 kg/m
3
. Calcinated
diatomaceous additives display their high retention
ability with relatively low hydraulic resistance. the rela-
tive permeability of the calcinated diatomite increases
up to 320 times that of natural diatomite. The disad-
vantage of diatomite is that it may foul ltering liquids
by dissolved salts and colloidal clays. The second most
common lter aid is expended perlite, which is a
glasslike volcanic rock. The porosity of this lter
aid is in the range of 0.850.9. Its bulk density ranges
Table 2 Typical Porosity of Filter Media
Filter media % free area
Wedge wire screen 540
Woven wire
twill weave 1520
square 2550
Perforated metal sheet 3040
Porous plastics (molded powder) 45
Sintered metal powder 2555
Crude kieselguhr 5060
Membranes 80
Paper 6095
Sintered metal bers 7085
Rened lter aids (diatomite, perlite) 8090
Plastic, ceramic foam 93
Source: Purchas D. Handbook of Filter Media. 1st ed.
UK: Elsevier Science, 1996, p 31.
from 4896 kg/m
3
. The key advantage of perlite over
diatomite is its relative purity.
Cellulosic ber (ground wool pulp) is applied to
cover metallic cloths. This lter aid forms a much
more compressible cake with good permeability but
displays a smaller particle retentivity than diatomite
or perlite. The cost of cellulose is higher than that of
diatomite or perlite. Thus this and other lter aids are
only applied to special cases (such as precoat stabiliza-
tion or chemical resistance).
2.5.2 Filtration with Filter Aids
In general, lter aid ltration should be used only for
systems that meet two key requirements. First, the
desired product is the ltrate, not the cake. Second,
the lter aid is acceptable in the lter cake or the lter
cake can be easily repulped and reltered to remove the
lter aid. In addition, the particle-settling rate should
be less than 0.012 m/min and the particle concentration
lower than 0.1 wt% (<0.3% for rotary drum precoat).
Table 3 Generalized Summary of Filter Media Based on Rigidity
Main type Subdivisions
Smallest particle
retained, microns*
Solid fabrications (a) at wedge-wire screens 100
(b) wire-wound tubes 10
(c) stacks of rings 5
Metal sheets (a) perforated 20
(b) woven wire 5
Rigid porous media (a) plastics 10
(b) sintered metals 5
(c) ceramics & stoneware 1
(d) carbon 1
Plastic sheets (a) woven monolaments 10
(b) porous sheets 10
(c) membranes <0.1
Woven fabrics (a) mono- or multilaments 10
(b) staple ber yarns 5
Nonwoven media (a) polymeric nonwovens
(melt blow, spun bonded, etc)
10
(b) felts & needle felts 10
(c) paper media
cellulose
glass
5
2
(d) lter sheets 0.5
Cartridges (a) yarn wound 5
(b) bonded beds 5
(c) sheet fabrications 3
Loose media (a) bers 1
(b) powders <0.1
Membranes (a) ceramic 0.2
(b) metal 0.2
(c) polymeric <0.1
*Very rough indication.
Source: Purchas D. Handbook of Filter Media. 1st ed. UK: Elsevier Science, 1996, p 4.
Pressure or vacuum cake lters can be operated with
both precoat and body aids. Most lter aids are used
on a one-time basis, although tests reusing precoat
lter aid have been demonstrated in a few pressure
lters (Schweitzer, 1997).
2.6 Filtration Equipment
Filtration equipment selection is often complex. First,
the suspension properties and process conditions can
change tremendously. Secondly, a wide variety of l-
ters are commercially available. In practice, projected
results of the selection are worked out most reliably
from actual or pilot plant database or analogous appli-
cation proles. The basic ltration equations can serve
as a useful guide for performing the comparison
among the selected lters and evaluating their accept-
ability. In specifying and selecting ltration equipment,
attention should be placed on options that would oer
a low cake resistance. This resistance is directly related
to the lter capacity. The cost proles are also impor-
tant information for making the nal decision.
Therefore a proper selection of a lter should consider
a number of factors including slurry properties, cake
characteristics, anticipated capacity, process operation
requirements, and cost estimate (Fitch, 1977; Mayer,
1988; Rushton et al., 2000).
2.6.1 Cake Filtration Equipment
Many dierent types of equipment are being marketed
for cake ltration. Only the three most commonly used
ones are discussed here.
Rotating Drum Filters. Rotating drum lters
most frequently work under vacuum conditions, such
as rotary vacuum drum lters with external ltering
surfaces (see Fig. 4). The design conguration con-
sists of a cloth-covered hollow drum with a slotted
face, the outer circumference containing a shallow
tray-shaped compartment. The drum surface is par-
tially submerged in the feed supension. The lter
drum is divided into several operating zones: ltra-
tion, rst dewatering, cake washing, second dewater-
ing, cake removal, and cloth cleaning. The drum
rotates at 10 to 60 revolutions per hour by a variable
speed motor. Each zone of the drum is connected to
a collection port on the automatic valve. In the
course of one revolution, the drum area passes
through these zones in succession by means of a con-
trol head device. In the ltration zone, the lter cake
builds up on the drum outer surface under vacuum
conditions (approximately 400 to 160 torr). After
cake dewatering and washing, compressed air and a
scraper are used for cake removal and clean-blowing
the lter cloth. Fouling by small particles is a fre-
quent problem in the cases of suspended particles
with wide size ranges. Rotary drum lters also have a
design version used as a pressure lter. Unlike the
vacuum drum lter, the pressure drop required for
cake formation is controlled between the lter pres-
sure vessel and the ltrate separator. The lter cake
can be washed and discharged by the same method
as described previously. Pressure drum lters are par-
ticularly suitable for processing of foodstuffs, antibio-
tics, dyestuffs and solvent, and water treatment. The
typical applications and performance guide for rotary
drum vacuum lters can be found in Dickenson
(1994).
Rotary Disc Filters. Rotary disc lters are an-
other kind of rotating lters, as shown in Fig. 5
(Svarovsky, 1990). This type of lter provides a much
larger lter area per unit of oor area at lower cost
than those of the rotary drum lter. Their applica-
tions are in coal preparation, ore dressing, and pulp
or paper processing. The rotary disc lter is con-
structed by a number of discs (up to 12 or more)
mounted on a horizontal hollow shaft. Each disc is
equipped with interchangeable elements and has an
individual slurry compartment. Submergence up to
50% of the ltering surface can be attained by a level
control. The cake formed on the emergent sector of
the disc is treated and removed by washing and
scraper before reentering the trough. The lter area
or ltering capacity can be adjusted by the change in
the number of discs. Disc lters are available in lter-
ing areas from 0.5 to 300 m
2
. The disc lter is mostly
used as a dewatering or thickening device. Owing to
the vertical ltering surface, cake washing in the disc
lter is not as efcient as in a drum lter. The rotary
disc lter is particularly benecial in cases of limited
space and where cakes do not require washing.
Horizontal Filters. To avoid the poor cake pick-
up and washing inherent in the design of rotary drum
and vertical disc lters, one such design, the Nutsch
lter, employs a at ltering plate covered with lter-
ing cloth. This type of lter basically consists of a
large false-bottom tank with a loose lter medium.
The Nutsch lter takes advantage of the effect of
gravity in cake formation and is able to provide uni-
form washing.
The horizontal belt lter consists of a series of
Nutsch lters as a long chain moving along a closed
path with the belt speed in the range of 330 m/min. A
Figure 4 Rotary drum vacuum lter. (a) Cut-away view. (From Dickenson, 1994.) (b) Filtration cycle. (From Rushton, 2000.)
simple belt vacuum lter is illustrated in Fig. 6. Filtrate
permeates through the lter medium and is directed to
centrally situated drainage pipes. In this lter, the sus-
pension feed, the wash liquor feed, the cake discharge,
and cloth cleaning are continuous. The cake thickness
generally ranges from 1 to 25 mm. Cakes up to 100
150 mm thick are possible with some fast-draining
materials. The chamber and collector are divided into
sections from which ltrate and washing liquid can be
discharged. Ecient cake discharge can be accom-
plished by separating the belt from the lter cloth
and directing the latter over a set of discharge rollers.
The simplicity in design is an advantage of belt lters.
In addition, the countercurrent cake washing and
removal of thin cake can be easily achieved in such
lters. The shortcomings of this type of lter are
large area requirements, inecient use of the available
lter area, and ineective washing at the belt edges.
Other types of lters, such as the rotating table
lter, the candle lter, lter presses, etc., are described
in detail in the recent ltration literature (Dickenson,
1994; Schweitzer, 1997; Svarovsky, 1990).
2.6.2 Deep-Bed Filtration Units
Commercial deep-bed lters consist of a cylindrical or
rectangular packed bed through which the suspension
to be ltered is passed. The common types of deep-bed
lter include the slow sand lter, the rapid lter, and
direct ltration with a occulated mixture. Typical
deep-bed lters are 0.50.3 m in height and 1 m in dia-
meter. Smaller packing material (lter media) provide
a greater surface area and result in a more eective
capture of suspended solids, but the bed pressure
drop and clogging tendency also increase. Often, the
design of the deep-bed lter is to employ mixed size
media packed in multilayers, as shown in Fig. 7
(Cheremisino, 1998).
In water ltration, the dual media lters are usually
designed using coarse anthracite coal on the top of ne
silica sand. The coarse anthracite layer serves to prevent
the formation of surface deposits on the sand bed,
resulting in the formation of a compressible cake
along the bed depth. Thus the depth removal of parti-
culates throughout the bed would be the key feature.
Figure 5 Rotary vacuum disc lter (Eimco). (From Osborne, 1990.)
This arrangement leads to longer operating cycles, par-
ticularly in low-pressure systems. For potable or waste-
water ltration, a minimum top eective size of 1.1 to
1.7 mm for the coarse medium (anthracite) is recom-
mended. It has been demonstrated that the use of coar-
ser media can reduce the ltration costs, because such a
lter can be run at a higher ow rate with a lower pres-
sure drop. Obviously, the use of coarser particles
requires deeper bed media to meet the demand of
equivalent media surface area for ltrate quality. In
some cases, a thin layer of dense solids, such as alumina
or gravel, has been used as a third layer situated beneath
the sand in the modied design of a dual layer deep-bed
lter. However, modern lter design still mostly stems
from the design version used in the 1980s. It uses coarse
media at the bottomand ner sand media at the top (see
Fig. 7ac). It was reported that good quality water with-
out chemical pretreatment was produced at a rate of 45
million gallons per acre per day. A comparison of oper-
ating characteristics of various deep-bed lters is pre-
sented in Table 4 (Rushton et al., 2000).
NOMENCLATURE
A cross-sectional area of lter
C solid concentration in the suspension
C
b
retentate concentration in the bulk ow
C
P
permeate concentration
C
S
retentate concentration at the membrane
surface
c mass of solid per unit volume of ltrate
d
p
particle (grain) diameter
F function describing the eect of particle
deposition on deep-bed ltration coecient
J ux
J
av
average ux
J
f
nal ux at the highest concentration
J
i
initial ux
K Kozeny coecient
K
0
constant
K
r
constant
k permeability
L distance from the top of the bed to the section
under study or bed depth
L
m
eective membrane thickness
m mass ratio of wet cake to dry cake
n compressibility constant of lter cake
P overall pressure drop across the lter
P
c
pressure drop across the lter cake
P
m
pressure drop across the lter septum
R true retention coecient
R
0
observed retention coecient
R
m
resistance of lter medium (e.g., septum)
R
0
constant
S
p
surface area of a single particle
s slurry concentration
t ltration time
Figure 6 Horizontal belt lter. (From McCabe et al., 1993.)
u liquid ow rate
V ltrate volume
V
f
volume of feed
V
p
volume of permeate passed through the
membrane
V
r
volume of retentate, retained on the membrane
surface
VCR volume concentration ratio
Greek Symbols
a specic cake resistance
a
o
empirical constant dened
a vector parameter in function F, Eq. (8)
e porosity of the lter cake
e
0
porosity of clean lter bed
l deep-bed lter coecient or impediment
modulus
l
0
deep-bed lter coecient of the clean lter
bed
m uid viscosity
r ltrate density
r
s
density of the solid particles
s specic deposit in the lter bed
Figure 7 Common deep-bed lter operating conguration. (From Cheremisino, 1998.)
REFERENCES
Akers, RJ, Ward AS. Liquid ltration theory and ltration
pretreatment. In: Orr, CC, ed. Filtration: Principles and
Practices, Part I. New York: Marcel Dekker, 1977, pp
169250.
Belfor G. Fluid mechanics and cross-ow membrane ltra-
tion. In: Muralidhara HS, ed. Advances in SolidLiquid
Separation. Columbus: Battelle Press, 1986, pp 182183.
Cheremisino NP. Solid/Liquid Separation. 2d ed.
Lancaster, PA: Technomic, 1995, p 240.
Cheremisino NP. Liquid Filtration. 2d ed. Boston:
Butterworth-Heinemann, 1998, p 277.
Cheremisino NP, Azbel DS. Liquid Filtration for Process
and Pollution Control. New Jersey: SciTech Publishers,
1989, p 59.
Cheryan M, Ultraltration Handbook, Lancaster, PA:
Technomic, 1986, p 2.
Chiang SH, He DX. Section 62: Fluid/particle separation. In:
Dorf RC, ed. Engineering Handbook. Boca Raton, FL:
CRC Press, 1995, pp 644651.
Dickenson C. Filters and Filtration Handbook. 3d ed. UK:
Elsevier Science, 1994, p 214.
Fitch B. When to Use Separation Techniques other than
Filtration. In: Shoemaker W, ed. AIChE Symposium
Series, 171, Vol. 73. New York: AIChE, 1977, pp
107108.
Ives KJ. Deep bed lters. In: Mathematical Models and
Design Methods. In: Rushton A, ed. SolidLiquid
Separation. Boston: Martinus Nijho, 1985, pp 90149.
Ives KJ, Pienvichitr V. Kinetics of the ltration of dilute
suspensions. Chem Eng Sci (20):965, 1965.
Jacobs LJ, Penney WR. In: Rousseau RW, ed. Handbook of
Separation Process Technology. New York: John Wiley,
1987, p 168.
Mayer E. Solid/liquid separation-selection techniques. Fluid/
Particle Separation Journal 1(2):132139, 1988.
McCabe WL, Smith JC, Harriott P. Unit Operations of
Chemical Engineeering. 5th ed. New York: McGraw-
Hill, 1993, pp 10101027.
1997, p 2242.
Purchas D. Handbook of Filter Media. 1st ed. UK: Elsevier
Science, 1996, p 4, 31, 415.
Rajagopalan R, Tien C. The theory of deep bed ltration. In:
Wakeman RJ, ed. Progress in Filtration and Separation.
New York: Elsevier Scientic, 1979, pp 179269.
Rushton A, Ward AS, Holdich RG. SolidLiquid Filtration
and Separation Technology. 2d ed. Weinheim, Germany:
Wiley-VCH, 2000, pp 225, 453.
Table 4 Characteristics of Different Types of Deep-Bed Filters
Characteristic Continuous upow Coarse downow Multimedia
Direction of ow upwards downwards downwards
Cleaning continuous sequenced sequenced
Media size, mm sand, 12 sand, 13 sand, 0.41
anthracite, 0.82
Sphericity of media 0.8 0.80.9 sand, 0.8
anthracite, 0.7
Media depth, m 1.2 1.52 sand, 0.5
anthracite, 0.3
Vessel depth, m 3.57 35 23.5
Surface loading, m
3
=m
2
h 814 618 816
Hydraulic stability low high high
Removal eciency moderate low sand: high
anthracite: low
Back-wash cycle
Water, m
3
=m
2
d 19 30 30
Air, m
3
=m
2
1012 at 7.5 bar 120 at 1 bar 50 at 0.5 bar
Time continuous 2 h/d 1 h/d
Relative capital cost high low moderate
Relative operating cost high low low
Source: Rushton et al., 2000, p 225.
Schweitzer PA. Handbook of Separation Techniques for
Chemical Engineers. 3d ed. New York: McGraw-Hill,
1997, pp 4.3, 4.61, 4.140.
Stamatakis K, Tien C. A simple model of cross-ow ltration
based on particle adhesion. AIChEJ 39(8): 12921302,
1993.
Svarovsky L. Pressure ltration and vacuum ltration. In:
Svarovsky L, ed. SolidLiquid Separation. 3d ed.
London: Butterworths, 1990, pp 402475.
Tien C. Granular Filtration of Aerosols and Hydrosols.
Tien C, Payatakes AC. Advances in deep bed ltration.
AIChE J 25(5):737759, 1979.
Tiller FM. The role of porosity in ltration, Part 3: Variable
pressurevariable rate ltration. AIChE J 4(2):170175,
1958.
Tiller FM, Shirato M. The role of porosity in ltration, VI.
New denition of ltration resistance. AIChE J
10(1):6168, 1964.
Tiller FM, Alciatore A, Shirato M. Filtration in the chemical
process industry. In: Matteson MJ, Orr C, eds. Filtration:
Principles and Practices. 2d ed. New York: Marcel
Dekker, 1987a, pp 361474.
Tiller FM, Yeh CS, Leu W. Compressibility of particulate
structures in relation to thickening, ltration and ex-
pressionA review. Separa Sci Technol 22(2&3):1044,
1987b.
Zeman LJ, Zydney AL. Microltration and Ultraltration:
Principles and Applications. New York: Marcel Dekker,
1996, pp 332, 335, 381.
3 SEDIMENTATION
Sedimentation is the separation of suspended solid par-
ticles from a uid stream by the action of a body force
on the settling behavior of the particle. The body force
may be either gravitational or centrifugal force. This
section covers gravity sedimentation, represented by
clarication and thickening. The equipment used for
these two operations are called clariers and thick-
eners, respectively. Centrifugal sedimentation will be
discussed in the next section.
From a unit operation standpoint, clarication and
thickening are essentially based on the same design
principles, and each combines features of the other.
The key objective of a clarication operation is to
remove small quantities of suspended particulates
from the liquid stream to produce a claried euent
or overow stream. In thickening operation, the goal
is to concentrate the dilute suspensions for their
further treatment in lters or centrifuges. Key fea-
tures of these two types of operation are presented
in Fig. 8.
3.1 Sedimentation Fundamentals
The settling behavior of suspended particulates in a
gravity eld is mainly aected by the concentration
of the particulate solids and their aggregation status.
In a dilute suspension, the settling solid behaves as
individual particles, and the process is regarded as
dilute sedimentation. This operating regime is called
a particulate (or free) settling regime (see Fig. 9).
Most clarier operations fall into this regime. As the
solid concentration increases, the suspended particles
have more chances to approach each other closely and
gradually form an aggregation (or cloud) state.
Once the concentration reaches a level in which the
suspended particles settle as a mass, the corresponding
sedimentation is known as hindered or zone settling. In
this regime, the settling behavior is related more to the
solid concentration rather than to particle size. Most
thickener sizing calculations are based on this regime.
As the solid concentration further increases to a higher
level, a settled bed of sediment mass (or settled units) is
compressed by the overburden of sediment on top of
them. Liquid is expressed from the lower sediment
layers and ows upward through the sediment. Their
settling behavior is aected not only by hydrodynamic
forces but also by the depth of the settling layers. This
regime is termed a compression regime. Sedimentation
Figure 8 Pretreatment requirements and key features of
gravity sedimentation.
with the addition of occulant usually falls into this
regime. It should be noted that when the suspension
becomes concentrated, the separation behavior is more
like ltration than settling in certain cases. Under this
situation, the development of channels is quite possi-
ble. This type of microstructure formation is, however,
seldom achievable in industrial settlers due to horizon-
tal shear. A typical thickener is illustrated in Fig. 10. A
feed suspension to be claried or thickened can be
operated in any regime. When it becomes concentrated
because of sedimentation, the settling particles may
initially behave in the particulate settling regime and
subsequently in the zone settling and compression
regimes. Therefore the design of sedimentation equip-
ment must consider all three regimes.
In dilute sedimentation, solid particulates have no
tendency to aggregate with one another. In this case,
the detention time has little or no eect on the
settling behavior. The sedimentation rate is equal to
the particle settling rate, which may be evaluated
in terms of the terminal settling velocity of the par-
ticle. On the other hand, in the hindered settling
regime, clusters of particles (settling units) develop
and settle as a sediment mass. The primary feature
of this regime is that the settling rate of the suspen-
sion is a function of particle concentration. The per-
tinent equations and related constant are given in
Tables 5a and 5b for both free and hindered settling
regimes.
3.2 Thickeners
As shown in Fig. 10, there are four zones existing in a
thickener: a clarication zone, a feed zone, a transition
zone, and a compression zone. As solids thicken, a
critical concentration will be reached. This concentra-
tion would aect the passage of solids to the under-
ow, causing a buildup in the compression zone
thickness. The thickener design and operation must
prevent the solid concentration from reaching the cri-
tical value.
3.2.1 Thickener Design
Three key parameters are required for thickener
design: the thickener basin area (or unit area), the
thickener basin depth, and the torque for the rake.
These are discussed here.
Thickener Basin Area (Unit Area). The thickener
basin area is determined from the solid ux rates
at the critical concentration in the hindered settling
regime. By denition, the critical point can be ex-
perimentally determined from the solids concen-
tration just prior to the beginning of the compression
zone. For a nonocculated system, the basin area,
expressed as the unit area (m
2
/ton/d), A
0
, can be
calculated using the equation (Osborne, 1990)
A
o

1=C 1=C
u
u
i
17
where C is the test solid concentration; C
u
the under-
ow solids concentration; and u
i
the initial settling rate
at the test condition.
For a occulated suspension, the unit area of the
thickener is determined by
A
o

t
u
C
o
H
o
18
where t
u
is the settling time; C
o
the test or feed solids
concentration; and H
o
the initial height of suspension
in the test. Scale-up factors used in the thickener design
usually vary, but a 1.2 to 1.3 multiplier applied to the
unit area calculated from laboratory data is sucient.
Thickener Basin Depth. In the hindered settling
conditions, the pulp depth is unimportant in the de-
termination of thickening rate and can be omitted.
As the pulp enters into the compression regime, the
pulp depth and the agitation affect the thickening
rate. In this case, the compression zone unit volume
may be calculated from the equation (Perry et al.,
1997)
Figure 9 Characteristics of sedimentation processes in a sus-
pension. (From Perry et al., 1997.)
V
c

t
c
r
s
r
l
r
s
r
p
r
l
19
where V
c
is the unit volume of the compression zone
dened as volume per weight of solids per day, t
c
the
compression time for a required particular underow
concentration, and r
s
, r
l
, and r
p
are the average
densities of solids, liquid, and pulp, respectively.
The depth of the compression zone equals to
V
c
=A
0
. The design thickener depth is calculated as
the sum of the depth for the compression zone plus a
clear zone of approximately 2 to 3 m allowing for clar-
ication and transition from feed concentration to
compression zone concentration. However, a greater
depth may be used to attain a better overow clarity.
Torque Requirement. The torque requirement for
the thickener operation is based on the force neces-
sary to drive the rake mechanism through the thick-
ened slurry. All other mechanical parts must also be
designed for this same load. Most thickener suppliers
base the torque requirements on operating data from
experience with similar applications. The maximum
torque selection for a thickener may follow the ex-
pression written as (Perry et al., 1997)
T k
T
D
2
20
where T is the torque; D the thickener diameter; and
k
T
a constant dependent upon the application. The
value of k
T
can be estimated from the process data
shown in Table 6. A much higher value of k
T
than
the required one would lead to an increase in the
unnecessary capital cost, while too small a value of
k
T
would severely aect the capability of the unit to
handle process upsets and produce the desired under-
ow concentration. Normally, sedimentation units
should be operated at the torque capability that is
Figure 10 Operating zone in a typical thickener.
not greater than 20% of the design level to avoid shut-
down due to inevitable process upsets or overloads.
Rake speed requirements depend on the type of solids
entering the thickener. Rake speed ranges used are 3
to 8 m/min for slow-settling solids, 8 to 12 m/min for
fast settling solids, and 12 to 30 m/min for coarse
solids or crystalline materials.
The detailed description of the major mechanical
components used for the thickeners, such as tank,
feed well, drive support structure, drive and lifting
devices, rake structure, underow withdrawal, and
overow collection systems can be found in the litera-
ture (Osborne, 1990; Perry et al., 1997; Schweitzer,
1997).
3.2.2 Thickener Types and Selection
A thickener consists of several basic components: a
tank to contain the slurry, a feed well for feed supply
(to minimize the turbulence eect), a rotating rake
mechanism, an underow solids withdrawal system,
and an overow launder. Recirculation of the under-
ow back to the thickener feed line is a common prac-
tice, but care must be exercised in the design to avoid
Table 5a Equations for the Settling Rate Calculations
Operating regime Equations Note
Particulate settling
u
t

Kd
2
p
r
s
r
m
Re
p

d
p
u
t
r
m
< 0:1
particle size range of 1 to 200 mm
u
t

4r
s
rgd
p
3rC
D

0:5
Re
p
0:2
u
t
terminal settling velocity; d
p
the particle size;
r
s
the solid particle density; r the liquid density;
m the liquid viscosity; K (Kozeny constant) 0:002 for u
t
in
m/h, d
p
in micrometer, and m in cp;
g gravitational constant; C
d
drag coefficient
Hindered setting u u
t
e
n
non-occulated suspension
(Richardson and Zaki, 1954a,b)
n is a function of the particle Reynolds number (see Table 5b)
u u
t
1 k
f
f
S
4:7017:8
d
=D
occulated suspension
(Scott, 1984)
f k
f
f
s
f settling units
f
s
volumetric dry concentration
k
f
the factor allowing for the liquid closely associated with the solids
d
the mean volume-surface-length diameter relevant to sedimentation

Table 5b Exponent n as a Function of the Particle Reynolds
Number and Vessel Diameter (D)
Re
p
d
p
u
t
r=m n for small vessel n for large vessel
Re
p
< 0:2 4:65 19:5 d
p
=D 4.65
0:2 < Re
p
< 1 4:35 1:75 d
p
=D Re
0:03
p
4:35 Re
0:03
p
1 < Re
p
< 200 4:45 18 d
p
=D Re
0:1
p
4:45 Re
0:1
p
200 < Re
p
< 500 4:45 Re
0:1
p
4:45 Re
0:1
p
Re
p
> 500 2.39 2.39
Source: Rushton et al., 2000, p 109.
the overload and underow line plugging. There are
two basic types of continuous thickeners: conventional
thickeners and high-rate thickeners.
The conventional thickeners can be operated with
or without occulants. They may be divided into three
classes based on their specic drive-support congura-
tions: bridge supported, center column supported, and
traction driven. The design conguration of bridge-
supported thickener is illustrated in Fig. 11. The dia-
meter of this type of thickener can be selected up to
45 m. The key advantages of this thickener include (1)
ability to produce a denser and more consistent under-
ow concentration; (2) use of a simplied lifting; and
(3) fewer parts subject to mud accumulation. The cen-
ter-column-supported thickeners are usually designed
for large diameter units (>20 m). The mechanism is
supported by a stationary center column. The traction
thickeners are an economical conguration of the
center column support. They are mostly adaptable to
the larger tanks over 60 m in diameter. For example,
the Superthickener or Caisson thickener is a very big
center-column-supported unit over 120 m in diameter.
In this thickener, the underow is withdrawn into the
column and pumped back to the circumference by the
underow pump. Its maintenance generally is easier
than other types of thickeners, but its installed cost
may be higher.
High-rate thickeners or high-capacity thickeners are
designed to take advantage of maximizing the occula-
tion eciency. They have specially designed feed wells
and occulation systems as illustrated in Fig. 12. The
feed well design is aimed at providing a good dispersion
of the occulants in the feed and transporting the oc-
culated feed suspension into the settling zone of the
thickener without destruction of the newly formed
ocs. The increase in occulation eciency will
normally increase the solid settling rate 210 times
over that obtained from conventional thickeners.
Obviously, the required unit area will reduce by the
same factor. As a result, the capital cost will be reduced
signicantly. However, care must be exercised in adding
occulants, since overocculation will produce a gellike
structure that is hard to remove from the thickener.
Numerous combination and component design var-
iations are possible. Selection of the type of thickener
should primarily be based on installation and operat-
ing costs. The selection of the nal design is made after
an initial conceptual approach is dened. To specify a
thickening system, one should consider the following:
(1) the characteristics of feed suspension: ow rate,
solid loading, particle-size distribution, and pH value;
(2) the model experimental results for underow con-
centration and overow clarity; (3) a tank size esti-
mate; (4) the required power or torque; (5) the
capability of the selected rake; (6) the materials of con-
struction; (7) a control scheme; and (8) a cost estimate.
Table 7 lists typical design sizing criteria and operating
conditions for a number of commercial thickeners and
clariers. The information presented in the table can be
used as a guide for selecting the gravity sedimentation
units and preliminary cost estimate.
3.3 Clariers
Continuous clariers arre generally employed to treat
dilute suspensions (industrial process streams and
domestic wastewater) with a large percentage of
relatively ne (usually smaller than 10 mm) solids.
The claried liquid is the main product. Clariers typi-
cally are designed for lighter duty operation than
thickeners.
Table 6 A Guide to Torque Specication for Gravity
Sedimentation Devices
Duty classication
Light Standard Heavy Extra heavy
Unit area, m
2
/(t/d) 5 15 0.21 <0.2
Underow, wt% 5 525 2555 >55
Particle size
% 65 mesh 0 05 515 >15
% 200 mesh 100 85100 5085 <50
Solids sp. gr. 1.251.5 1.52.7 2.74.0 >4.0
k factor 1575 75150 150300 >300
Source: Schweitzer PA. Handbook of Separation Techniques for Chemical
Engineers. 3d ed. New York: McGraw-Hill, 1997, p 4-140, Table 1.
3.3.1 Clarier Design
As mentioned earlier, clariers and thickeners are very
similar in their overall designs and layouts. The design
considerations of the thickeners can also be used as a
basis for clarier design. The design conguration of
clariers may have the following features: (1) rapid and
good mixing of chemical additives with feed slurry and
adjustable recirculator speed with at least a range of
3 : 1; (2) control of the mixer speed (not exceeding 2 m/
s) and the scraper tip speed (less than 3 m/min with
speed variation of 3 : 1); (3) a discharge system for
easy automation and variation of amount discharged;
(4) devices for measuring and varying the slurry con-
centration in the contacting zone; (5) suspension blan-
ket levels at a minimum of 1.6 m below the clean water
surface. The design criteria and operating conditions
for commercial clariers are given in Table 7.
Clarier design has traditionally been based on the
principle of dilute sedimentation. The design calcula-
tions of a clarier include feed well design, sedimenta-
tion basin design (providing enough residence time for
the separation), and solid ux (G). The other design
parameters, such as the torque requirement and clari-
er area, are similar to those of the thickeners, which
have been discussed in the previous section (3.2.1).
The design congurations of the feed well should
provide: (1) a decreasing velocity gradient inside and
outside the feed well; (2) gentle mixing to promote
ideal conditions for dispersing occulant or coagulant
Figure 11 Conventional thickener with bridge-supported mechanism (Eimco). (From Schweitzer, 1997.)
Figure 12 High-capacity thickener (Eimco). (From Osborne, 1990.)
Table 7 Design Criteria and Operating Conditions for Commercial Thickeners and
Clariers
Slurry
Solid %
Unit area,
m
2
day/mg
Overow rate,
m
3
/(m
2
h) Feed Underow
Alumina-hydrate 110 2050 1.23 0.070.12
Alumina-red mud (primary) 34 1025 25
Coal refuse 0.56 2040 0.51
a
0.71.7
FGD
b
waste 312 2045 0.33
a
Magnesium hydroxide (seawater) 14 1520 310 0.50.8

Metallurgical
Copper concentrates 1450 4075 0.22
Copper tailings 1030 4565 0.41
Magnetic tailings 25 4560 0.51.5 1.22.4
Municipal waste
Primary clarier 0.020.05 0.51.5 11.7
Primary thickening sludge 13 510 8
Phosphate slimes 13 515 1.218
a
De-inking waste 0.010.05 47 11.2

Paper-mill waste 0.010.05 28 1.22.2
Water softening lime-sludge 510 2040 0.62.5
a
High-rate thickeners using required occulant dosages operate at 10 to 50% of these unit areas.
b
FGD represents ue-gas desulfurization.
Source: Perry RH, Green DW, Maloney JO. Perrys Chemical Engineers Handbook. 7th ed. New
York: McGraw-Hill, 1997, pp 1872, Table 18-7.
into the feed stream without creating too much turbu-
lence, while keeping the maximum overow clarity.
The usual operating range for the upow clarier is 2
to 5 m
3
/(h m
2
). The retention time required for the oc
separation in a sedimentation basin can be estimated
using Stokes law or Newtons law, especially for well-
dened homogeneous systems. It should be noted that
allowances must be made for deleterious eects due to
convection, air-induced surface waves, and inlet and
outlet turbulence.
If the clarier is assumed to operate close to its
capacity limit, the solids ux, G, into the clarier at a
steady state can be determined using a simple material
balance:
G
CQ Q
u
A
21
where C is the solids concentration in feed suspension,
Q the euent ow rate, Q
u
the underow rate, and A
the surface area of clarier.
3.3.2 Clarier Types and Selection
A clarier typically consists of a concentric circular
compartment for conditioning and settling, a claried
liquid overow section (euent launder), and a con-
centrated underow discharge system. Figure 13 illus-
trates the design of an upow clarier, which combines
the functions of mixing, occulation, and sedimenta-
tion in a single unit. This clarier is also known as a
solid-contact clarier. In such a clarier, the impeller is
installed underneath the draft tube as a primary mixing
device. In another version of the upow clarier design
(Eimco Process Equipment Co.), the impeller is set
above the draft tube to provide secondary mixing for
occulationfeed contact. Also, this new type of
upow clarier is equipped with a rake. The operation
of the clarier may be divided into several distinct
layers: the claried water zone, the feed zone (co-
agulant and raw water addition with primary and sec-
ondary mixing), and the compacting zone. In the feed
zone, the coagulants and/or occulants are mixed
rapidly with the raw water and the return ow under
agitation. With proper coagulation/occulation a
uniform dense oc can be produced. For operating
exibility, it is desirable to control the mixing intensity
and the sediment rake speed independently. In the
claried zone, the suspended solid concentration is
very low, and the claried stream will escape over
weirs as puried euent. In the compacting zone,
the formed ocs with the treated water pass
through the sedimentation basin, which provides a
retention time of 1 or 2 hours for the oc particles to
settle. When the oc particles are too heavy to
circulate up through the draft tube, a modied
design using external recirculation of a portion of
the thickened underow is recommended. Solid-
contact clariers are usually used for clarifying turbid
water or slurries that require adding coagulant or
occulant for the removal of bacteria, suspended
solids, or color.
Inverted cone clariers are another type of solid-
contact clariers. The mixed feed suspension is intro-
duced from the top of a cone, which has an angle
ranging from 608 to 908. These clariers are operated
without rakes or agitators. The slurry interface is
usually within 1 or 2 m of the overow level. The sus-
pended solids mixed with the chemical additives are
Figure 13 Structure of an upow clarier. (From Mukai, 1986).
kept in a uidized state. Typical operating rates are
around 5 to 10 m
3
/(m
2
h). The concentration of sus-
pended solids in the overow ranges from 10 to
40 mg/L. The operating and maintenance costs are
low compared to other types of solid-contact clariers.
These clariers are well suited to the case where the
ow rate varies only slightly.
In recent years, tilted-plate clariers (see Fig. 14)
have been widely used. This type of clarier is also
known as lamella or tube settler. It contains a multiple
plate assembly inclined at 308 to 608 from the horizon-
tal. Claried liquid and settled solids may ow counter-
currently or cocurrently along each channel. Generally,
a plate angle of 308 to 408 is a suitable inclination when
operating cocurrently, whereas an increase of up to 558
to 608 may be necessary with the countercurrent
arrangement. The suspended solids settle only a short
distance in the channel before falling into the base.
Settled solids are collected in a lower compartment
and discharged by pumping. The selection of the plate
spacing should be based on two considerations. First,
the plate spacing must be large enough to accommo-
date the opposite ows. Secondly, the channel space
arrangement should be able to limit the interference
and plugging as well as to provide enough residence
time for the solids to settle in a short distance. The
tilted-plate clariers usually use 10 to 50 plates with
spacing from 0.05 to 0.1 m. The channel lengths range
from 1 to 3 m with widths of about 1.2 m. Operating
capacities vary from 0.5 to 3 m
3
=m
2
h.
The major advantage of the tilted-plate clarier is its
increased capacity per unit of plate area compared to
the conventional clarier. There are two key shortcom-
ings existing in this type of clarier. First, it produces a
varying underow solids concentration lower than
other type of gravity clariers. Secondly, it is dicult
to clean the scale formed within the channel. The appli-
cations of tilted-plate clariers include clarication of
plating and pickling wastes, paper mill euent, and
tertiary wastewater treatment. For the clarier selec-
tion one can follow the same guidelines as for thickener
selection, as discussed in the last section.
NOMENCLATURE
A clarier surface area
A
o
unit area
C solid concentration in the feed
C
o
test or feed solid concentration
C
u
underow solids concentration
D thickener diameter
G solid ux into the clarier
H
o
initial height of suspension in the test
k
T
constant dependent upon the application of
thickener
Q euent ow rate
Q
u
underow rate
T torque
t
C
compression time for a required particular
underow concentration
Figure 14 Schematic of a tilted-plate clarier (Heil Process Equipment Co.). (From Schweitzer, 1997.)
t
u
settling time
u
i
initial settling rate or hindered settling velocity
at solids concentration C
i
V
C
unit volume of the compression zone
Greek Symbols
r
l
average density of liquid
r
p
average density of suspension or pulp
r
s
solid particle density
REFERENCES
Mukai T. Solidliquid separations of water and wastewater
treatment systems in Japan. In: Muralidhara HS, ed.
Advances in SolidLiquid Separation. Columbus:
Battelle Press, 1986, p 409.
Osborne DG. Gravity thickening. In: Svarovsky L, ed. Solid
Liquid Separation. 3d ed. London: Butterworths, 1990,
pp 132201.
1997, pp 1860 to 1872.
Richardson JF, Zaki WN. Sedimentation and uidization:
Part 1. Trans Inst Chem Engrs 32:3553, 1954a.
Richardson JF, Zaki WN. The sedimentation of a suspension
of uniform spheres under conditions of viscous ow.
Chem Eng Sci 3:6573, 1954b.
Rushton A, Ward AS, Holdich RG. SolidLiquid Filtration
and Separation Technology. 2d ed. Reinheim, Germany:
Wiley-VCH, 2000, p 109.
Schweitzer PA. Handbook of Separation Techniques for
Chemical Engineers. 3d ed. New York: McGraw-Hill,
1997, pp 4.1404.156.
Scott KJ. Hindered settling of a suspension of spheres.
Critical evaluation of equations relating settling rate to
mean particle diameter and suspension concentration.
CSIR Report CENG 497. CSIR, Pretoria, South Africa,
1984.
4 CENTRIFUGATION
Centrifugation can be viewed as an extension of the
conventional ltration and gravity sedimentation. In
this case, the centrifugal force replaces the pressure
force in ltration and the gravitational force in sedi-
mentation, respectively. However, the operating prin-
ciples remain the same. The deciding factor for
separating particles from liquid is the density dierence
between the solids and the suspending liquid.
Centrifuges are available in a wide variety of types
and sizes with a centrifugal force ranging from less
than 100G up to 10,000G, where G is the gravitational
acceleration. They can be broadly divided into two
major types: sedimentation centrifuges and lter
centrifuges.
4.1 Sedimentation Centrifuges
Centrifugal sedimentation removes solid particles from
a solidliquid suspension by employing centrifugal
force to induce setting eects. Owing to its much
higher acceleration, centrifugation can extend the
range of sedimentation to ner particles and it can
also separate emulsions, which might normally be
stable in the gravity eld. The sedimentation centri-
fuges are not usually sensitive in their solids handling
capacity to feed solid concentration because the liquid
does not have to lter through the solids or a lter
medium. They are eective for separating particles
ranging from 6 mm (1/4 in.) down to submicron sizes.
Flocculants are often used to promote agglomeration
of particles to accelerate the settling rate of very ne
materials.
4.1.1 Principles of Sedimentation Centrifuges
In sedimentation centrifuges, liquid and solid are acted
on by two forces: gravity acting downward and centri-
fugal force acting horizontally. In commercial units,
however, the centrifugal force component is normally
so large that the gravitational component may be
neglected. The separating power of the sedimentation
centrifuges is often measured by comparing the centri-
fugal force R
c
in the device with the gravity accelera-
tion G, which is referred to as the relative centrifugal
force or the centrifugal number N
c
. The centrifugal
number, N
c
, typically varies from 200 times gravity to
360,000 times gravity.
The separation happens in a sedimentation centri-
fuge when the solid particles are removed from the
uid. In order for a particle of a given size to be
removed from the uid a sucient time should be
allowed for the particle to settle and reach the wall
of the separator bowl. If it is assumed that the solid
particle moves at its terminal velocity at all times, the
smallest particle that should just be removed can be
calculated. Consider the simplest, tubular type centri-
fuge, which is shown in Fig. 15 (McCabe et al., 1993).
In a tubular centrifuge, the bowl consists of a vertical
tube with a large height-to-diameter ratio, which
rotates at a high speed about its vertical axis. The
feed point is at the bottom, and the liquid discharge
is at the top. Assuming that the incoming uid starts to
rotate with the bowl, its angular velocity will soon
become identical with that of the bowl. There is there-
fore no tangential ow in the bowl, and the uid only
rotates with the bowl and moves upward through the
bowl at a constant velocity, carrying solid particles
with it. As the solid particles move with the uid,
they are subjected to high centrifugal forces and
begin to settle at some position in the uid, say at a
distance r
A
from the vertical axis, following trajectory
similar to that shown in Fig. 16. Its settling time is
limited by the residence time of the uid in the bowl;
at the end of this time let the particle be at a distance r
B
from the axis of the rotation. If r
B
< r
2
, the solid par-
ticle leaves the centrifuge with the liquid; if r
B
r
2
, it is
deposited on the bowl wall and removed from the
liquid.
Ambler (1952) introduced the cut point concept in
the sedimentation centrifuge separation. It is dened as
the diameter of the particle that just reaches one-half
the distance between r
1
and r
2
. If a solid particle is to
be removed from the uid, it must travel the distance
r
2
r
1
=2 to the bowl wall in the available residence
time. The following equation gives the relationship
between the feed rate and the particle cut diameter
(McCabe et al., 1993):
q
c

pbo
2
rD
2
pc
18m
r
2
2
r
2
1
ln2r
2
=r
2
r
1
22
where r
1
is the radius of inner surface of the liquid, r
2
the radius of inner surface of the bowl, o the angular
velocity, r the density dierence between solids and
liquid, and q
c
the feed volumetric ow rate correspond-
ing to the cut diameter D
pc
. At this feed rate most of
the particles having diameters greater than D
pc
will be
removed from the liquid by the centrifuge, and the
particles remaining in the liquid will be smaller than
D
pc
. Most of the sedimentation centrifuges are capable
of removing solids particle sizes well into the sub-
micron range.
This analysis is an oversimplication, since the ow
pattern of the uid in the centrifuge bowl is much more
complicated than that assumed in the Fig. 16. The only
way to describe fully the separation performance of a
sedimentation centrifuge is by the grade eciency
curve. Knowledge of the grade eciency curve allows
accurate and reliable predictions of total eciency with
dierent feed solids, subject to the operating character-
istics, the state of the dispersion of solids, other vari-
ables remaining constant.
4.1.2 Major Types of Sedimentation Centrifuges
A sedimentation centrifuge consists of an imperforate
bowl into which a suspension feed is fed. The bowl
rotates at high speed. The liquid after separation is
removed through a skimming tube or over a weir
Figure 15 Schematic of a tubular centrifuge. (From
McCabe et al., 1993.)
Figure 16 Particle trajectory in sedimenting centrifuge.
(From McCabe et al., 1993.)
while the separated solids either remain in the bowl or
are intermittently (or continuously) discharged from
the bowl. Typically a sedimentation centrifuge will
have the following main components (as shown in
Fig. 15):
1. A bowl or rotor in which the centrifugal force
is applied to a solidliquid suspension
2. A means for feeding the suspension into the
rotor and a means for discharging the sepa-
rated components from the bowl either in
batch or continuous mode
3. A drive shaft, axial, and thrust bearings
4. A drive mechanism to rotate the shaft and
bowl
5. A casing to contain the separated components
6. A frame for support and alignment
Sedimentation centrifuges can be classied by several
criteria, including the centrifugal number, N
c
, the
range of throughputs, the solids concentration in sus-
pension that can be handled, the bowl design, and the
solids discharge mechanism. Figure 17 gives a classi-
cation of sedimentation centrifuges based on the design
of the bowl and of the solids discharge mechanism:
tubular, multichamber, imperforate basket, scroll
type, and disk centrifugres (Svarovsky, 1985). Due to
the design of the bowl structure, the bowl has to be
cleaned manually for both tubular and multichamber
centrifuges, so they are usually used as liquid classiers
and suitable for relatively low solids concentrations.
The disc type and the scroll type centrifuges can be
operated continuously so that they are suitable for
very high solids concentrations, sometimes up to
50% by volume. The imperforate basket type centri-
Figure 17 Classication of sedimentation centrifuges. (From Svarovsky, 1985.)
fuge is somewhere in between and suitable for moder-
ate feed solids concentrations.
4.2 Filter Centrifuges
Filter centrifuges separate solid particles and liquid
from a solidliquid suspension by employing pressure
resulting from the centrifugal action to force the liquid
through the lter medium, leaving the solid particles
behind. The density dierence between the solids and
the liquid, which governs the separation in sedimenta-
tion, is no longer a necessary prerequisite.
The common feature of all lter centrifuges is a
rotating basket equipped with a lter medium.
Solids, which form a porous cake, can be separated
from the suspension liquid in a lter centrifuge.
Slurry is fed to a rotating basket having a slotted or
perforated wall covered with a lter medium such as
canvas or metal cloth. The separation process takes
place much like the cake ltration (see Sec. 2.1)
process, except the pressure gradient is created by the
centrifugal action of the rotating basket.
4.2.1 Principles of Filter Centrifuges
The centrifugal ltration process starts with the
feed slurry to an empty centrifuge and is followed
by the lter cake deposition and the ow of clear
ltrate or wash water through the cake. The basic
theory of constant pressure ltration can be mod-
ied to apply to the lter centrifuges (McCabe et
al., 1993).
Based on the two-resistance model (see Sec. 2.1.1),
the volumetric ow rate of ltrate, q, is expressed as
q
ro
2
r
2
2
r
2
1
2mam
C
=

AA
L

AA
a
R
m
=A
2
23
where A
2
is the area of the lter medium,

AA
L
the arith-
metic mean cake area,

AA
a
the logarithmic mean cake
area, R
m
the lter medium resistance, a the specic
cake resistance, and m
c
the total mass of the solids in
the lter. The mean areas

AA
L
and

AA
a
are dened by the
equations
AA
a
r
i
r
2
pb 24
AA
L

2pbr
i
r
2
lnr
2
=r
i
25
where b is the hieght of the basket and r
i
the inner
radius of the cake.
4.2.2 Major Types of Filter Centrifuges
In general, the lter centrifuges can be classied into
two groups: batch and continuous type. Several
selected ltering centrifuges are listed in Fig. 18.
Figure 18 Classication of lter centrifugal lters. (From Alt, 1986.)
More complete information can be found in several
references (Perry et al., 1997; Svarovsky, 1990;
Zeitsch, 1990).
Batch-Type Filter Centrifuges. Batch-type lter
centrifuges are much older in conception than contin-
uous ones, but they are by no means obsolete. The
total number of batch type far exceeds that of the
continuous type. Their main advantage is high
separation eciency linked to the outstanding purity
of the separated solids and liquids. Among the
commonly used batch-type lter centrifuges are the
three-column basket centrifuges and the peeler centri-
fuges.
Basket Centrifuges The simplest and most com-
mon form of the batch-type lter centrifuge is the bas-
ket centrifuge. It consists of a cylindrical basket, or
drum, which is suspended on three columns and
because of this characteristic it is also known as the
three-column centrifuge. The basket centrifuges are
normally arranged with a vertical axis of rotation. A
schematic of this design is shown in Fig. 19. The basket
centrifuge represents the earliest centrifuge used for
liquidsolid separations. It remains extensively used
throughout the process industries worldwide.
The basket centrifuges are constructed with baskets,
as their name suggests. Generally, the mantle surface
of the basket is perforated with a large number of holes
covered on the inside by one or more coarse screens.
The latter serves as a backup for the lter medium that
can have the shape of a bag lining the perforated wall
of the basket. The feed slurry enters the rotating basket
from the top of the basket, before or after the start of
the rotation. Liquor (ltrate) drains through the lter
medium into the casing and out a discharge pipe. The
solid particles deposit against the basket wall and form
cake of 5 to 15 cm thick. The discharge of the solids is
achieved by stopping the machine and manually
removing the solid cake or replacing the bag.
Operation cycles of these machines can be varied to
achieve the desired performance. The basket provides
a good surface for washing the lter cake. Wash liquid
may be sprayed through the cake to remove the soluble
material. The cake is then spun as long as needed,
often at a higher rotating speed than those during
the loading, ltering, and washing steps, to provide
maximum dryness.
The dimensions of the baskets range from 76 to
122 cm in diameter and 46 to 76 cm deep and turn at
speeds from 600 to 1,800 rpm. These machines can
usually handle a capacity up to about 12 ft
3
of product
per cycle. They are widely employed in the ne chemi-
cal and pharmaceutical industries, where products are
produced on a batch basis. The characteristics of batch
operation allow easy changing of the operation para-
meters from batch to batch to meet production needs.
In these machines the solid cake is removed by mov-
ing a plough into the cake after the basket is slowed
down to a preset speed. The plough directs the solids
toward the center, where they fall through the bottom
openings of the basket. The entire operation can be
automated by means of timers and solenoid-operated
valves, which control the various parts of the opera-
Figure 19 Schematic of a three-column centrifuge. (From Alt, 1986.)
tion: feeding, ltering, washing, spinning, rinsing, and
unloading. Any portion of the cycle may be lengthened
or shortened as desired.
Peeler Centrifuges. The peeler centrifuge (as
shown in Fig. 20) is a variation of the basket centri-
fuge. It usually rotates at constant speed about a hor-
izontal axis instead of a vertical axis. The lter cake in
the peeler centrifuge is unloaded periodically while
turning at full speed by a heavy knife: the peeler.
During the unloading, the peeler rises and cuts solid
cake out with considerable force through a discharge
chute. In general, the peeler centrifuge is fully auto-
mated in its operation.
In the three-column basket centrifuge the basket
rotates vertically. The eect of the gravity tends to
form a nonuniform cake, its lower portion being
thicker and containing larger particles than its upper
portion. Because of the unevenly distributed solid
cake, the washing can cause variations in the purity
of the solid product. In such cases, peeler centrifuges
are preferable over three-column centrifuges.
Furthermore, the peeler centrifuges have higher
productive capacity since they do not require any non-
productive periods of deceleration and acceleration for
unloading solids. Usually they are not used for treating
feed containing solid particles ner than 150 mesh.
Hence these machines have been found particularly
attractive where the ltration and drainage periods
are relatively short. On the other hand, they are not
suitable for handling slow-draining solids, which
would give uneconomically long cycles, or sticky
solids, which do not discharge cleanly through the
chute. There is considerable breakage or degradation
of the particles during high-speed unloading by the
peeler. Also, the permeability of the residual heel
could be unfavorably aected by plugging it with frac-
tured nes. Consequently, it may require washing to
recondition the plugged residual heel.
Continuously Fed Filter Centrifuges. There are
several different types of continuously fed lter cen-
trifuges. Among them, the pusher centrifuges and the
conical screen lter centrifuges are the most common
designs.
Pusher Centrifuges. The pusher centrifuge is so
named because of the pushing mechanism employed
to transport the solids across the basket. A schematic
of a one-stage pusher centrifuge is shown in Fig. 21.
The rst pusher centrifuge was designed more than a
century ago (Alt, 1986). It consists of a rotating per-
forated rotor with a slotted wall and the circular
pusher plate reciprocating with frequencies ranging
from 20 to 100 strokes per minute. The plate attached
to the feed funnel rotates in the same direction as the
rotor. Feed enters the small end of the feed funnel from
a stationary pipe at the axis of the rotation of the
rotor. It travels toward the large end of the feed funnel.
When it spills o the feed funnel onto the rotor wall, it
moves in same direction as the wall and very nearly the
same rotation speed. The main ltration occurs within
Figure 20 Peeler centrifuge: (a) feed and separation phase, (b) solids discharge phase. (From Jacobs and Penney, 1987.)
the stroke zone of the pusher plate. A layer of solid
cake 2.5 to 7.5 cm thick is formed. This solid layer is
moved over the ltering surface by the reciprocating
pusher plate. Each stroke of the pusher plate moves the
solid layer a few inches toward the lip of the rotor; on
the return stroke a space is opened on the ltering
surface in which more solid cake can be deposited.
When the cake reaches the lip of the rotor, it is thrown
into the large casing by the centrifugal acceleration.
The liquor passes the ltering screen and leaves the
machine by ducts well separated from the solids dis-
charge.
The pusher centrifuges are normally used when the
feed can be concentrated above 60% by volume, since
the operation capacity of the pusher centrifuge is
greatly increased as the solid concentration is
increased. They are capable of handling fragile crys-
tals, which may be damaged in other types of centri-
fuges. The small pusher centrifuges can handle a few
grams per second, while the large ones have capacities
up to 55,000 kg/h. Based on the number of rotors,
pusher centrifuges are divided into two types: single
stage pusher centrifuges with only one rotor and multi-
stage pusher centrifuges with two or more rotors. The
multistage pusher centrifuge is particularly suitable for
particles forming a soft cake or having a high frictional
resistance to sliding on the lter medium. A schematic
of a two-stage pusher centrifuge is also shown in
Fig. 21. The purpose of the multistage design is to
keep the length of the rotor short.
Figure 21 Schematic of (a) single-stage pusher centrifuge, (b) two-stage pusher centrifuge. (From Perry et al., 1997.)
The capacity and performance of the pusher centri-
fuge depend on the particle size and shape of the feed:
the coarser the feed particle size, the higher the opera-
tion capacity and the lower the moisture. The thickness
of the cake is one of the most important factors aect-
ing the operation of the pusher centrifuge. The pusher
centrifuge has three signicant operating characteristics
dependent on the cake thickness: the solid discharge,
the lter rate, and the pushing force. Generally, the
thickness of the cake on the rotor is proportional to
the friction of the cake over the lter medium, the
centrifugal force, and the length of the rotor.
Conical Screen Filter Centrifuges The conical
screen centrifuge is another very commonly used con-
tinuous centrifuge. Instead of using the pusher action
to discharge the solids, it uses a helical conveyor that is
turning slightly slower or faster than the rotor to dis-
charge the solids. Such a solid discharge mechanism
can be used for both cylindrical and conical rotors.
The dierential speed of the conveyor controls the
rate at which the solids move through the drainage
zone. A schematic of a conical screen centrifuge is
shown in Fig. 22. The conical screen centrifuges may
have vertical or horizontal axes of rotation.
The conical screen centrifuges are used mainly for
processing coarser particles, for example crystalline
salts, coal, and minerals. These centrifuges are suita-
ble for relatively large throughputs, up to 320,000 kg/
h. The various types of conical screen centrifuges
operate successfully in a wide range of applications.
Among them, the sliding lter centrifuges and the
vibrating lter centrifuges are the most commonly
used. In some cases, the angle of the conical screen
can be set suciently large to permit the cake to
overcome its friction on the lter medium so that it
becomes self-discharging. Compared with the pusher
centrifuge, the conical screen centrifuge has a much
greater particle breakage.
Table 8 shows the application range of dierent
types of centrifuges. Several important feed character-
istics, including the minimum particle size, the maxi-
Figure 22 Schematic of a conical conveyor discharge cen-
trifuge. (From Zeitsch, 1990.)
Table 8 Application Range of Centrifuges
Tubular
Disc
(nozzle type)
Conveyor
bowl
A. Sedimentation type centrifuges
Minimum particle size, micron 0.1 0.25 2
Maximum particle size, micron 200 50 5,000
Allowable concentration of feed solid, % 0.1 220 260
Condition of cake Pasty, rm Fluid Firm, pasty
Typical solids handling rate, lb/h 0.15 1030,000 100100,000
Batch
vertical
Conical
basket Pusher
B. Filtering Centrifuges
Minimum particle size, micron 10 250 40
Maximum particle size, micron 1,000 10,000 5,000
Allowable concentration of feed solid, % 210 4080 1575
Condition of cake Pasty Relatively dry Relatively dry
Typical solids handling rate, lb/h 0.11.0 540 0.55.0
Source: Cheremisinoff PN. Solid/Liquids Separation. Lancaster PA: Technomic, 1995, p 240, Table 6.3.
mum particle size, the maximum solid concentration,
and the solid handling capacity are presented. This list
can serve as a useful guideline for an initial selection of
suitable centrifuges.
NOMENCLATURE
AA
L
arithmetic mean cake area
AA
a
logarithmic mean cake area
A
2
area of lter medium
b height of the basket in a lter centrifuge
D
pc
cut diameter
G gravity acceleration
m
c
total mass of the solids in the lter
N
c
centrifugal number
q
c
feed volumetric ow rate
r
1
distance from the surface of liquid to the axis
of rotation
r
2
distance from the wall of the bowl to the axis
of rotation
r
A
; r
B
distance from the particle to the axis of
rotation
r
i
inner radius of the cake
R
c
centrifugal force
R
m
lter medium resistance
Greek Symbols
a specic cake resistance
r density of liquid (ltrate)
r density dierence between solids and liquid
m liquid viscosity
o angular velocity
REFERENCES
Alt C. Centrifugal separation. In: Muralidhara HS, ed.
Advances in SolidLiquid Separation. Columbus, Ohio:
Battelle Press, 1986, pp 107139.
Ambler CM. The evaluation of centrifuge performance.
Chem Eng Progress 48:150158, 1952.
Jacobs LJ, Penney WR. In: Rousseau RW, ed. Handbook of
Separation Process Technology. New York: John Wiley,
1987, p 168.
McCabe WL, Smith JC, Harriott P. Unit Operations of
Chemical Engineering. 5th ed. New York: McGraw-Hill,
1993, pp 10111028, 10641072.
1997, pp 1864.
Svarovsky L. SolidLiquid Separation Processes and
Technology. Amsterdam: Elsevier, 1985, pp 72106.
Svarovsky L. Separation by centrifugal sedimentation. In:
Svarovsky L, ed. SolidLiquid Separation. 3d ed. UK:
Butterworths, 1990, p 264.
Zeitsch K. Centrifugal ltration. In: Svarovsky L, ed. Solid
Liquid Separation. 3d ed. UK: Butterworths, 1990, pp
476532.
5 HYDROCYCLONES
The hydrocyclone is based on the principle of centri-
fugal force causing the separation of solids from a
liquid by the dierences in density and particle size.
A typical hydrocyclone consists of a cylindrical section
and a conical section, as shown in Fig. 23. It does not
have any internal rotating parts. An external pump is
used to transport the liquid suspension to the hydro-
cyclone through a tangential inlet at high velocity,
which in turn generates the uid rotation and the
necessary centrifugal force. The outlet for the bulk of
the liquid is connected to a vortex nder located on the
axis of the cylindrical section of the vessel. The under-
ow, which carries most of the solids, leaves through
an opening (apex) at the bottom of the conical section.
The principle and basic design of the hydrocyclone
has been known for more than a century (Bretney,
1891), but it did not nd signicant application in
industry until the late 1940s. These separation devices
were rst used in mining and mineral processing, but in
recent years their applications have spread to many
other industries, including chemical manufacturing,
power generation, and environmental cleanup.
5.1 Separation Efciencies
The solidliquid separation in hydrocyclones is never
complete, because there is always liquid discharging
with the solids through the underow. The term
separation eciency used for the hydrocyclone is
usually dened for measuring the capability of the
hydrocyclone of separating the solids from the feed
into the underow. There are a number of dierent
terms for the separation eciency used in the litera-
ture. They include total eciency, reduced total
eciency, grade eciency, reduced grade eciency,
and cut size.
5.1.1 Total Efciency
Total eciency is dened as the total solids (mass or
volume) reported in the underow as a fraction of the
total solid in the feed:
E
T

UC
u
QC
f
26
where U is the underow volumetric ow rate, Q the
feed volumetric ow rate, C
u
the solid concentration in
the underow, volume, or mass fraction, and C
f
the solid concentration in the feed, volume, or mass
fraction.
Because an overall mass balance must apply, the
total eciency E
T
can be determined by measuring
any two of the three streams (feed, underow, and
overow) for total solid amount, assuming no accumu-
lation of solids in the hydrocyclone.
5.1.2 Reduced Total Efciency
There are two problems associated with total e-
ciency. First, if a hydrocyclone delivers both liquid
and solid to the underow and nothing to the overow,
an ideal total eciency of 1 will result. Second, without
any separation, a hydrocyclone, by simply splitting the
feed to an overow and an underow, will result in
certain guaranteed total eciency. In order to over-
come these weaknesses, several alternative denitions
of eciency are used. One of the denitions is the
reduced total eciency E
0
T
:
E
0
T

E
T
R
f
1 R
f
27
where R
f
is the underow-to-throughput ratio dened
as
R
f

U
Q
28
5.1.3 Grade Efciency
In place of the total eciency, a grade eciency corre-
sponding to a particular particle size is used, since
a hydrocyclone is a size-dependent separator. A
graphical representation of the relationship between
the grade eciency and the particle size is called the
grade eciency curve, as shown in Fig. 24 (Svarovsky,
1985).
The grade eciency curve can be determined by
measuring the total eciency and the particle size
distribution of any two of the three streams (feed,
underow, and overow). One of the following
equations can be used for calculating the grade
eciency:
Figure 23 Hydrocyclone: (a) schematic diagram, (b) ow pattern. (From Rushton, 2000.)
Feed and underflow: Gx E
T

dF
u
x
dF
f
x
29
Feed and overflow:
Gx 1 1 E
T

dF
o
x
dF
f
x
30
Overflow and underflow:
Gx 1
1
E
T
1

dF
o
x
dF
u
x
31
where F
f
x is the cumulative percentage of solids with
particle size x in the feed, F
o
x the cumulative percen-
tage of solids with particle size x in the overow, and
F
u
x the cumulative percentage of solids with particle
size x in the underow.
Similar to the reduced total eciency, a reduced
grade eciency G
0
x is introduced as
G
0
x
Gx R
f
1 R
f
32
5.1.4 Cut Size
A term closely related to separation eciency is the cut
size d
50
, which is dened as the particle size that has a
50% chance of being separated when it is subject to the
action of a hydrocyclone. The majority of the solid
particles ner than the cut size in the feed will report
to the overow, while the majority of those coarser will
be separated and report to the underow. The cut size
can be determined from the grade eciency curve (see
Fig. 24).
5.2 Fundamentals of Hydrocyclone Separation
A full understanding of the hydrocyclone requires a
detailed analysis of the ow pattern within its body.
A number of reviews on this subject may be found in
the literature (Bradley and Pulling, 1959; Fontein,
1951; Kelsall, 1952). Only a brief qualitative descrip-
tion will be presented in this section.
5.2.1 Fluid Flow in Hydrocyclones
The pattern of uid ow within the hydrocyclone body
is best described as a spiral within a spiral with circular
symmetry. A schematic view of the spiral ow inside a
hydrocyclone is shown in Fig. 23b. The entering uid
ows down the outer regions of the hydrocyclone
body. This combined with the rotational motion
creates the outer spiral. At the same time, because of
the wall eect, some of the downward moving uid
begins to feed across toward the center. The amount
of inward motion of uid increases as the uid
approaches the cone apex, and uid that ows in this
inward stream ultimately reverses its direction and
ows upward to the cyclone overow outlet via the
vortex nder. This reversal applies only to the vertical
component of velocity, and the spirals still rotate in the
same circular direction. In the meantime, the down-
ward ow near the wall carries solid particles to the
apex opening (bottom outlet).
In addition to the main ow pattern there exists a
secondary ow pattern, short circuit ow, at the top of
the hydrocyclone. The short circuit ow is a ow
pattern that moves across the cover of the cylindrical
section to the base of the vortex nder. It ows along
the outer wall of the vortex nder until it combines with
the uid in the overow created by the main ow
pattern. This short circuit ow pattern is due to the
wall eect of the cyclone top cover and the outer wall
of the vortex nder. The quantity of the short circuit
ow can be as much as 15% of the total feed ow.
The central air core is another important ow
pattern in the hydrocyclone. The rotation of the uid
in the hydrocyclone creates a low-pressure axial area
that results in the formation of a rotating free liquid
surface. The central air core is cylindrical in shape and
lled with air the whole way through the length of the
hydrocyclone. the central air core tends to stabilize the
vortex ow pattern within the hydrocyclone.
5.2.2 Particle Motion in Hydrocyclones
Particles (including immiscible liquid droplets) will
separate from the suspending liquid if their density is
Figure 24 Grade eciency and cut size in hydrocyclone
operation. (From Svarosvsky, 1985.)
dierent from that of the suspending liquid owing to
the centrifugal eects generated by the spiral ow in
the hydrocyclone. Usually it is assumed that particles
are heavier than that of the suspending uid and that
they move readily outward.
As discussed earlier, there are two spiral ow pat-
terns existing in the hydrocyclone. Only particles exist-
ing in the outer spiral ow will be separated by the
centrifugal force. Any particles in the inner spiral
ow will pass upward to the overow outlet. It should
be noted that there are two important stages in the
process of particle separation. One is the separation
of the solids from the main body of the ow into the
boundary layer adjacent to the inner wall of the hydro-
cyclone by centrifugal forces. The other is the removal
of the separated solids from the boundary layer by
downward uid ow (not by gravity) to the apex of
the cone and out of the hydrocyclone.
5.2.3 Velocity and Pressure Distributions
The velocity of the uid ow in a hydrocyclone can be
resolved into three components: tangential, axial, and
radial. The most useful and signicant of these three
components is the tangential velocity.
The tangential velocity of the uid in the hydro-
cyclones increases as the radius decreases, which is
expressed by the empirical relationship
VR
n
const 33
where V is the tangential velocity, R the radius, and n
the empirical exponent, usually between 0.5 and 0.9.
For the outer regions of a free vortex by denition n
will be 1. From Eq. (33) one can conclude that as the
radius R approaches zero the tangential velocity V will
approach innity. In practice, this cannot happen since
Eq. (33) only holds true until small values of the radius
are reached, when the tangential velocity starts to fall
with a further decrease in radius. The static pressure
increases radially outward because of the centrifugal
eld induced by vortex ow in the hydrocyclone.
This static pressure head is primarily determined by
the distribution of the tangential velocities within the
ow. Therefore the tangential velocity distribution can
be estimated from the simple measurements of the
radial static pressure distribution.
Since the outer and the inner layers move in oppo-
site vertical directions (i.e., the ow in the outer vortex
moving down and the inner vortex moving up), there is
a well-dened locus of zero vertical velocity between
the two vortices. This locus forms an invisible bound-
ary, which plays an important role in particle separa-
tion (see the equilibrium orbit theory discussed in the
next section).
The radial velocity is the smallest of the three velo-
city components. It has been found that the inward
radial ow (toward the center of the vessel) is a max-
imum near to the wall and it diminishes with decrease
in radius until it is zero at the air core interface.
5.2.4 Theories of Separation
A number of physical models have been proposed for
the separation process in a hydrocyclone (Driessen
MG, 1951; Bradley and Pulling, 1959; Fahlstrom,
1960; Kelsall, 1952; Rietema, 1961; and Schubert and
Neesse, 1980). Among these, dierent phenomenologi-
cal approaches have led to the development of two
basic theories: the equilibrium orbit theory and the
residence time theory.
Equilibrium Orbit Theory. The general concept
that particles of a given size reach an equilibrium
radial orbit position in the hydrocyclone forms the
basis of equilibrium orbit theory. The ne particles
reach equilibrium at small radii where the ow is
moving upwards and transports nes to the overow,
while the coarse particles nd equilibrium position at
large radii where the ow is moving downwards and
carries these particles to the underow outlet (apex).
The dividing surface is the locus of zero vertical velo-
city (LZVV). The size of the particles that nd equili-
brium radius on LZVV will be the cut size that has
an equal chance to nish in either overow or under-
ow.
In developing the equilibrium orbit theory, a key
assumption made by Bradley and Pulling (1959) is
the existence of a mantel in the hydrocyclone,
which precludes inward radial velocity in the region
immediately below the vortex nder. Furthermore,
the LZVV is assumed to be in the form of an imaginary
cone whose apex coincides with the apex of the hydro-
cyclone and whose base is at the bottom of the mantle.
Based on these assumptions, the equilibrium orbit
theory has led to the development of empirical correla-
tions for determining the cut size and pressure drop in
hydrocyclone operation.
The major deciency of the equilibrium orbit theory
lies in its lack of consideration of the eect of turbu-
lence ow on particle separation and the residence time
of the particles in the hydrocyclone (as not all particles
are able to nd equilibrium orbits within their resi-
dence time). In spite of such weaknesses, it proves to
be a reasonable approach for determining the hydro-
cyclone performance at low solids concentrations
(2.0% by volume).
Residence Time Theory. Residence time theory is
based on the consideration of whether a particle will
reach the cyclone wall in a given residence time. In
the development of this theory, the distribution of all
particles across the inlet is assumed to be homo-
geneous (Rietema, 1961). The cut size will be the
size of those particles that enter the center of the
inlet pipe and just reach the wall within the residence
time. Using this theory coupled with extensive experi-
mental test data, Rietema was able to establish a set
of empirical correlations and suggest a criterion (a
characteristic cyclone number) for optimum design
of hydrocyclones.
Despite the very dierent approaches and assump-
tions, the forms of correlations obtained by the equili-
brium orbit theory and residence time theory are
similar. For specic hydrocyclone designs, both
theories provide their respective empirical equations
for determining the cut size and pressure drop in
terms of three dimensionless groups, the Stokes
number at cut size, Stk
50
, the Euler number, Eu, and
the Reynolds number, Re (see discussions in Sec. 5.4
below):
Stk
50

d
2
50
r
s
rn
18mD
34
Eu
p
n
2
r=2
35
Re
Dnr
m
36
where d
50
is the cut size, r
s
the solid density, r the
liquid density, v the supercial velocity of liquid, p
the pressure drop, D the hydrocyclone diameter, and
m the liquid viscosity.
5.3 Design and Operation Variables
In practice the performance of a hydrocyclone is
aected by two groups of variables. These are: (1)
the operating variables that are related to the operating
conditions but independent of hydrocyclone size and
proportions; and (2) the design variables that are
dependent on the physical dimensions and proportions
of the hydrocyclone.
5.3.1 Operating Variables
The most important operating variable is the pressure
drop, p, which is proportional to the square of volu-
metric feed rate, Q (which is directly related to the
tangential velocity):
p / Q
2
37
As the operating pressure drop increases, the tangen-
tial velocity increases. Consequently, the higher
separation eciency or smaller cut size d
50
results.
Generally, this relationship only holds to be true
with the operating pressure up to 2 bar. Beyond that,
further increase in the operating pressure has little
eect on the separation eciency.
The second important operating variable is the feed
solid concentration. With increasing feed solid concen-
tration, the separation eciency falls o rapidly owing
to its eect on liquid ow pattern and the interaction
among solid particles. Therefore hydrocyclones are
usually operated with dilute feed solids concentrations
(<2% by volume).
The adjustment of the underow orice (apex)
opening is also a very important operating variable.
Correct adjustment of this opening is vital for the
best operation of the hydrocyclone. In general, an
increase in the underow orice size causes an increase
in operating capacity, but it tends to reduce the cut size
and the underow solids concentration.
5.3.2 Design Variables
The key design variables are dened as those that are
associated with the hydrocyclone dimensions, includ-
ing cyclone diameter, D, inlet diameter, D
i
, outlet
(apex) diameter, D
u
, and vortex nder diameter, D
o
.
In addition, the total length of the cyclone, L, the
length of the cylindrical section, l, and the cone
angle, , are also important. They all aect the operat-
ing performance of the hydrocyclone.
The cyclone diameter or the diameter of the base of
the cone is a primary design variable, and all other
dimensions are usually related to it. For a given feed
ow rate, the eects of cyclone diameter on the
operating performance of the hydrocyclone can be
described by proportionalities
d
50
/ D
x
38
and
P / D
y
39
where x ranges from 1.36 to 1.52 and y from 3:6 to
4:1. In other words, at a constant feed ow rate, the
larger the cyclone diameter the larger the cut size and
the smaller the operating pressure drop. If the feed
ow rate is varied to maintain a constant pressure
drop, the values of x will then vary from 0.41 to 0.50.
The inlet diameter, D
i
, aects both the capacity and
the separation eciency, since it controls the inlet velo-
city and therefore the tangential velocities inside the
cyclone. Based on the theoretical analysis, the cyclone
resistance coecient (in terms of Euler number)
increases exponentially with a decrease in D
i
:
Eu / D
n
i
40
where n is in the range of 1.2 to 4 with an average of
about 2.
At constant operating pressure drop, the feed ow
rate is proportional to the square root of 1=Eu.
Therefore the capacity of the cyclone, Q, is roughly
proportional to D
i
. In addition, the eect of D
i
on
the cut size d
50
is signicant. According to Bradley
(1965), the cut size d
50
is related to inlet diameter by
the relationship
d
50
/ D
z
i
41
where z varies from 0.6 to 0.68.
5.4 Design and Scale-Up
A reliable scale-up and performance prediction of the
hydrocyclone is limited to the low solids concentration
(<2% by volume). Under dilute conditions the ow
pattern in hydrocyclones is unaected by the presence
of solid particles, and the particleparticle interaction
is not signicant. In contrast to low feed concentration
conditions, quantitative prediction of the hydrocyclone
performance under higher feed concentrations is
uncertain.
Generally, the scale-up of hydrocyclones is based on
the concept of cut size d
50
on the grade eciency curve,
because the shape of the grade eciency curve remains
the same for a family of geometrically similar hydro-
cyclone designs. Dimensional analysis coupled with
theoretical considerations gives two functional rela-
tionships for a hydrocyclone (Svarovsky, 1990):
Stk
50
Eu C
1
42
and
Eu K
p
Re
n
p
43
where the constants C
1
, K
p
, and n
p
are empirical
performance constants for geometrically similar
hydrocyclones.
Table 9 summarizes the dimensions and experimen-
tally determined performance constants for several
known hydrocyclone designs (Svarovsky, 1984). In
the absence of actual test results, Eqs. (42) and (43),
using constants given in Table 9, can serve as useful
guides for estimating the performance of an existing
hydrocyclone or selecting a proper hydrocyclone size
for a given duty.
Table 9 Summary of Several Known Hydrocyclone Designs
Type and size of
hydrocyclone
Geometrical proportions Scale-up constants
Running cost
criterion
a
D
i
=D D
0
=D l=D L=D Cone angle, y, degrees Stk
50
Eu K
p
n
p
(Stk
50
4=3
Eu
Rietemas design
D 0:075 m
0.28 0.34 0.40 5 20 0.0611 316 0.134 2.12
Bradleys design
D 0:038 m
0.133 0.20 0.33 6.85 9 0.1111 446.5 0.323 2.17
Mozley cyclone
D 0:022 m
0.154 0.214 0.57 7.43 6 0.1203 6381 0 3.20
Mozley cyclone
D 0:044 m
0.160 0.25 0.57 7.71 6 0.1508 4451 0 4.88
Mozley cyclone
D 0:044 m
0.197 0.32 0.57 7.71 6 0.2182 3441 0 8.70
Warman 3
00
Model R
D 0:076 m
0.29 0.20 0.31 4.0 15 0.1079 2.618 0.8 2.07
R W 2515 (AKW)
D 0:125 m
0.20 0.32 0.8 6.24 15 0.1642 2458 0 6.66
a
Running cost criterion is directly proportional to the pressure drop.
Source: Svarovsky L, 1984, p 60, Table 5.1.
NOMENCLATURE
C
1
empirical constant
C
f
solid concentration in the feed
C
u
solid concentration in the underow
d
50
cut size
D hydrocyclone diameter
D
i
inlet pipe diameter
D
o
diameter of the vortex nder
D
u
underow (apex) diameter
E
T
total eciency
E
0
T
reduced total eciency
Eu Euler number
F
f
x Cumulative percentage of solids with particle
size x in the feed
F
o
size x in the overow
F
u
size x in the underow
Gx grade eciency
G
0
x reduced grade eciency
K
p
empirical constant
l length of the cylindrical section
L total length of the hydrocyclone
n empirical constant
n
p
empirical constant
p operating pressure drop
Q feed volumetric ow rate
R radius
Re Reynolds number
R
f
underow to throughput ratio
Stk
50
Stokes number of cut size
U underow volumetric ow rate
v supercial velocity
V tangential velocity
x empirical constant
y empirical constant
z empirical constant
Greek Letters
y cone angle
r density dierence between solids and liquid
m liquid viscosity
REFERENCES
Bradley D. The Hydrocyclone. London: Pergamon Press,
1965.
Bradley D, Pulling DJ. Flow patterns in the hydraulic
cyclone and their interpretation in terms of performance.
Bretney E. Water Purier. U.S. patent No. 453,105, May 26,
1891.
Driessen MG. Theory of ow in cyclone. Rev Ind Mining,
Special Issue 4: pp 449461, 1951.
Fahlstrom PH. Discussion (on hydrocyclone) Proceedings of
International Mining Processing Congress. Institute of
Mining and Metallurgy, 1960, pp 632643.
Fontein DF. Cyclone separator of solid mixtures of dierent
grain size and specic gravity. U.S. patent No. 2,550,341,
1951.
Kelsall DF. A study of the motion of solid particles in a
hydraulic cyclone. Trans Inst Chem Eng 30:87104, 1952.
Rietema K. Performance and design of hydrocyclones. Parts
I to IV. Chemical Engineering Science 15:298325, 1961.
Schubert H, Neesse T. A hydrocyclone separation model in
consideration of turbulent multi-phase ow. In: Proc
International Conference on Hydrocyclones. Cambridge,
BHRA Fluid Engineering, Craneld, 13 October 1980,
Paper 3, 2336.
Svarovsky L. Hydrocyclones. New York: Technomic, 1984,
pp 3660.
Svarovsky L, ed. SolidLiquid Separation Processes and
Technology. Amsterdam: Elsevier, 1985, pp 1827.
Svarovsky L. Hydrocyclones. In: Svarovsky L, ed. Solid
Liquid Separation, 3d ed. UK: Butterworths, 1990, pp
203248.

Handbook of Fluidization and Fluid-Particle Systems

Uploaded by

Copyright:

Available Formats

Handbook of Fluidization and Fluid-Particle Systems

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Handbook of Fluidization and Fluid-Particle Systems

Uploaded by

Copyright:

Available Formats

HANDBOOK of

Department of Chemical Engineering, Tokyo University of Agriculture and Technology,

in one continuous pattern was

. The angles of repose for commonly occurring

. The angles of internal friction for com-

ber die Stokessche Formel und u ber die

angle and the triangular or the simple rhombic layer of

(see Fig. 1). The vertical stacking of

) and simple rhombic layer (60

= vector form of gravitational acceleration

= vector quantity of volumetric ow rate per

ber Kapillare Leitung des Wassers im Boden.

C and found that both the frequency of bubble

C and found a reduction

C in bubbling beds but becomes

C, this was supported by the subsequent study of

C, radiation heat transfer is also an important

C. Average heat transfer

ditions, Toulouse, 1973, pp 207222.

included angle) gas jet

stergaard, A So rensen, eds. Fluidization V.

stergaard K, So rensen A, eds. Fluidization V. New

Technical University Hamburg-Harburg, Hamburg, Germany

Current aliation: BMHClaudius Peters, Buxtehude,

= particle size independent rate constant,

C. At higher temperatures, radiant

C. This increase was not more rapid,

C, where radiation is signicant. It was also found

In the above equation, the term S

s.=sublimation, v.=vaporization, a.=adsorption.

C) of the combustor, calcination of the sorbent

and are very

size range. This powder is generated by a pre-

C and fed into a uidized bed cata-

C. Heat transfer tubes are usually arranged

C and 1.32 bar. The

C. In order to avoid agglomeration and sheeting,

Jiang et al. (1999)

Pr = Prandtl number ( n,a)

= nondimensional gas velocity

= empirical constant in Eq. (40)

F) and pressures up to 689 kPa (100

C. (From Glicksman et al., 1994.)

F. Depending on the type of

charge from a paran molecule at a Lewis acidic

The rst term on the right-hand side is the catalyst-to-

F, can cause severe mechanical damage to the

C. Fuel gas ammonia content may be a particular

C for most coals) is controlled by the char

C) and the solids circulation

C, within an almost uniform temperature

C) much lower than conventional combus-

C, generates a lowheating value

C above the dense bed temperature.

C at about 1.7 atmosphere, with a reactor con-

C at about 1 to 5 atmospheres. Main reac-

C at 2.5 to 6 atmospheres. For

C before the catalyst nes were removed.

C and about 20 atmos-