Physics of Gas-Liquid Flow

Physics of Gas–Liquid Flows
Presenting tools for understanding the behavior of gas–liquid flows based on the ways
large-scale behavior relates to small-scale interactions, this text is ideal for engineers
seeking to enhance the safety and efficiency of natural gas pipelines, water-cooled
nuclear reactors, absorbers, distillation columns, and gas lift pumps. The review of
advanced concepts in fluid mechanics enables both graduate students and practicing
engineers to tackle the scientific literature and engage in advanced research.
The text focuses on gas–liquid flow in pipes as a simple system with meaningful
experimental data. This unified theory develops design equations for predicting drop
size, frictional pressure losses, and slug frequency, which can be used to determine flow
regimes, the effects of pipe diameter, liquid viscosity, and gas density. It describes the
effect of wavy boundaries and temporal oscillations on turbulent flows, and explains
transition between flow regimes, which is key to understanding the behavior of gas–
liquid flows.
Thomas J. Hanratty is Professor Emeritus at the University of Illinois at Urbana-

Champaign, and was a leader in establishing industrially important multiphase flow as
a new academic discipline, by relating macroscopic behavior to small-scale interactions.
His research has been recognized by nine awards from the American Institute of
Chemical Engineers (AIChE), the American Society for Engineering Education, Ohio
State University, Villanova University, and University of Illinois. He was the inaugural
winner of the International Multiphase Flow Prize. Hanratty was named as one of the
influential chemical engineers of the modern era at the AIChE centennial celebration
in 2008. He has been elected to the National Academy of Engineering, the National
Academy of Sciences, and the American Academy of Arts and Sciences.
“This authoritative and impressive monograph, written by a widely acclaimed pioneer in
the field, is an excellent resource for new students as well as seasoned practitioner. It
develops the concepts systematically and packages many decades worth of literature in
this field lucidly, giving the readers a chance to understand appreciate the evolution of
this field.”
Sankaran Sundaresan, Princeton University
“Physics of gas-liquid flows” is a must read for graduate students, researchers and
engineers seeking a solid basis or wanting to update their knowledge in the dynamics
of gas-liquid systems. The book is an authoritative reference, mainly built around the
results of the author, a leading expert in the field during several decades. It presents our
current experimental and theoretical understanding at both the local and global scales
with an original contribution to wave phenomena as they appear in film stratified and
annular flows that may be a source of inspiration for researchers and teachers in the years
to come.”
Jean Fabre, Institut National Polytechnique, Toulouse
“A lifetime of probing research and deep thinking about gas-liquid flows is enclosed
between the covers of this book. Starting from simple analyses – the style of which will
be familiar to many undergraduates – the author moves gradually to more advanced
topics building a succinct yet exhaustive picture of the present understanding of these
important flows. Practicing engineers and researchers alike will find many gems in this
book.”
Andrea Prosperetti, Johns Hopkins University
Physics of Gas–Liquid Flows
THOMAS J. HANRATTY
University of Illinois at Urbana-Champaign
University Printing House, Cambridge CB2 8BS, United Kingdom
Published in the United States of America by Cambridge University Press, New York
Cambridge University Press is part of the University of Cambridge.
It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9781107041202
© Thomas J. Hanratty 2013
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2013
Printed and bound in the United Kingdom by TJ International Ltd. Padstow Cornwall
A catalog record for this publication is available from the British Library
Library of Congress Cataloging in Publication data
Hanratty, Thomas J.
Physics of gas–liquid flows / Thomas J. Hanratty, University of Illinois at Urbana-Champaign.
pages cm
ISBN 978-1-107-04120-2 (hardback)
1. Multiphase flow. 2. Gas–liquid interfaces. I. Title.
TA357.5.M84H36 2013
5320 .56–dc23
2013013373
ISBN 978-1-107-04120-2 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Contents
Preface page xiii

List of symbols xv
Chapter 1 One-dimensional analysis 1

1.1 Introduction 1
1.2 Single-phase flow 2
1.2.1 Flow variables 2
1.2.2 Momentum and energy theorem 2
1.2.3 Steady flow in a duct of constant area 3
1.2.4 Choking 5
1.2.5 Flow of an ideal gas in a pipe of constant area 6
1.2.6 Mechanical energy balance 8
1.3 The homogeneous model for gas–liquid or vapor–liquid flow 9
1.3.1 Basic equations 9
1.3.2 Choking 10
1.3.3 Calculation of wall shear stress 12
1.3.4 Comments about the reliability of the homogeneous model 12
1.4 Separated flow model for gas–liquid flow 12
1.4.1 Basic equations 12
1.4.2 Choking 14
1.5 The Lockhart–Martinelli analysis 15
1.5.1 Evaluation of homogeneous and separated flow models 15
1.5.2 Frictional pressure gradient 15
1.5.3 Rationale for the Lockhart–Martinelli plots 16
1.5.4 Lockhart–Martinelli correlation 18
1.6 Steady flow in a nozzle 19
1.6.1 Preliminary comments 19
1.6.2 Single-phase flow 19
1.6.3 Two-phase flow in a nozzle 22
1.6.4 One-dimensional two-fluid model 24
Chapter 2 Flow regimes 27

2.1 Need for a phenomenological understanding 27
2.2 Flow regimes in horizontal pipes 28
2.2.1 Flow regimes for air and a low-viscosity liquid 28
2.2.2 Wavy stratified flow in air–water systems 28
vi Contents
2.2.3 Annular flow 29

2.2.4 Plug and slug flow 30
2.2.5 Stability of a slug 31
2.2.6 Transition to slug/plug flow 32
2.2.7 Pseudo-slugs 33
2.2.8 Bubble/foam regime 33
2.2.9 Flow at large gas and liquid velocities 34
2.2.10 Effect of pipe diameter on flow regimes 34
2.3 Large-wavelength Kelvin–Helmholtz waves / viscous large-wavelength
instability 35
2.4 Effect of liquid viscosity on flow regimes in a horizontal pipe 37
2.5 Effect of gas density on flow regimes in a horizontal pipe 40
2.6 Effect of drag-reducing polymers 41
2.7 Vertical upflows 42
2.7.1 Flow patterns 42
2.7.2 Mechanisms for the breakdown of the bubbly flow pattern 44
2.7.3 Mechanism for the slug flow to churn flow transition 45
2.7.4 Mechanism for the churn/annular transition 45
2.7.5 Mechanism for the transition to dispersed bubbly flow 46
2.8 Vapor patterns in evaporating and condensing flows / critical heat flux 47
2.9 Downflows in vertical pipes 49
2.10 Microgravity flows 50
2.10.1 General comments 50
2.10.2 Flow regimes 50
2.10.3 Comparison of flow regimes for microgravity and 1g flows 52
2.11 Capillaries and microchannels 52
2.11.1 Capillary tubes 52
2.11.2 Microchannels 53
Chapter 3 Film flows 58

3.1 Free-falling layer 58
3.1.1 Laminar flow 58
3.1.2 Turbulent flow 59
3.1.3 Interpolation formula 60
3.2 Gas–liquid flows 60
3.2.1 Horizontal flow in a channel 60
3.2.2 Film flow in a vertical pipe 62
3.2.3 Flow reversal and flooding 64
3.3 Interfacial stress for vertical flows 65
3.3.1 Waves in annular flows 65
3.3.2 Scaling 66
3.3.3 Correlation for fi using dt/2 as a scaling factor 67
3.3.4 Correlation using the viscous length for the gas phase 69
Contents vii
Chapter 4 Inviscid waves 71

4.1 Inviscid, incompressible, irrotational flow 71
4.2 Propagation of small-amplitude two-dimensional waves 74
4.3 Dispersion relation for propagating waves 77
4.4 Propagation of waves at the interface of stationary fluids 79
4.5 Kelvin–Helmholtz instability 81
4.5.1 General conditions for a KH instability 81
4.5.2 Gas–liquid flows with deep gas and liquid layers 82
4.5.3 Effect of hG 83
4.6 Experimental verification of the KH instability 84
4.7 Group velocity 85
4.7.1 Physical interpretation of group velocity 85
4.7.2 Wave resistance 88
4.7.3 Waves which decay in space, rather than time 89
4.8 Uses of the notion of a KH instability 90
4.8.1 Initiation of slugging 90
4.8.2 Rate of atomization 90
4.8.3 Interfacial stress in horizontal stratified flows 92
Chapter 5 Stratified flow 94

5.1 Scope 94
5.2 Stratified flow in a rectangular channel 94
5.3 Stratified flow of gas and liquid in a pipe 95
5.3.1 Geometric relations 95
5.3.2 Laminar flow 96
5.3.3 Laminar liquid–turbulent gas flow in a pipe 96
5.3.4 Model of Taitel & Dukler 97
5.4 Determination of the interfacial friction factor 100
5.4.1 Measurements of fi/fS 100
5.4.2 Equations for interfacial friction factor and wave steepness 102
5.4.3 Design relations for a horizontal pipe 103
5.5 Inclined pipes 104
5.5.1 General comments about the effect of inclination 104
5.5.2 Upflows 105
5.5.3 Downward inclinations 106
Chapter 6 Influence of viscosity on large Reynolds number interfacial waves; effect

of spatially and temporally induced oscillations on a turbulent flow 111
6.1 Introductory comments 111
6.2 Equation for small disturbances in a viscous fluid 112
6.3 Boundary conditions 114
6.4 Solution of the Orr–Sommerfeld equation 116
viii Contents
6.4.1 General solution 116

6.4.2 Dispersion relation 117
6.5 Special solutions for constant liquid velocity 119
6.5.1 Solution of (6.19) for constant liquid velocity 119
6.5.2 Free-surface waves 119
6.6 Wave generation by a gas flow 122
6.6.1 Dispersion relation for large Reynolds numbers 122
6.6.2 Influence of P ~ and T~ 125
6.6.3 Prediction of P ^ and T^ using the quasi-laminar assumption 126
6.6.4 Changes in the quasi-laminar theory to include wave-induced
turbulence 126
6.7 Flow over solid wavy surfaces 128
6.7.1 Measurements of shear stress along a solid wavy surface 128
6.7.2 Equations for flow over solid waves, in boundary-layer coordinates 129
6.7.3 Evaluation of rîj using a modified van Driest model 130
6.7.4 Testing turbulence models against wall shear stress measurements 134
6.8 Response of turbulence to imposed temporal oscillations 136
6.8.1 Motivation 136
6.8.2 Measurements 137
6.8.3 Comparison with the quasi-steady solution 139
6.9 Prediction of wave generation 141
6.10 Dissolution patterns 142
Chapter 7 Large-wavelength waves; integral equations 148

7.1 Prologue 148
7.2 Integral equation used to analyze the behavior of liquid layers 149
7.2.1 Approach 149
7.2.2 Dispersion relation 149
7.2.3 Relation between the wall shear stress and ua 151
7.3 Laminar free flow 154
7.4 Kinematic waves 156
7.5 Roll waves 158
7.5.1 Outline 158
7.5.2 Free turbulent flows on an inclined plane 158
7.5.3 Generation of roll waves by a gas flow 160
7.6 Model for a fully developed roll wave 164
7.7 Ripples generated on a liquid film at high gas velocities 166
7.8 Roll waves in pipelines 170
Chapter 8 Bubble dynamics 173

8.1 Prologue 173
8.2 Free-fall velocity of a solid sphere 173
8.3 Stokes law for a solid sphere 174
Contents ix
8.4 Rise velocity of bubbles at low ReP 178

8.5 Measurements of CD 178
8.5.1 Solid spheres 178
8.5.2 Gas bubbles 179
8.6 Gaseous volume fraction for bubbly flow 180
8.7 Cap bubbles 181
8.8 Taylor bubbles / vertical slug flow 183
8.9 Benjamin bubble 184
8.10 Gas lift pumps 186
8.11 Bubble columns 189
8.11.1 Volume fraction of gas in bubble columns 189
8.11.2 Behavior of a bubble column 190
8.11.3 Analysis of a bubble column 194
8.12 Bubbly gas–liquid flow in vertical pipes, distribution of bubbles 194
8.13 Performance of downcomers in tray columns 195
8.14 Sedimentation and fluidization 197
8.14.1 Focus 197
8.14.2 Sedimentation 197
8.14.3 Fluidization 199
9 Horizontal slug flow 202

9.1 Prologue 202
9.2 Slug stability 202
9.2.1 Necessary conditions for the existence of slugs 202
9.2.2 Transition to plug flow 206
9.2.3 Measurements of the shedding rate 207
9.2.4 Equations for the bubble velocity and the shedding rate 208
9.2.5 Calculations of the stability of a slug 211
9.3 Modeling of slug flow in a horizontal pipe 212
9.4 Frequency of slugging 216
9.4.1 Shallow liquid inviscid theory 216
9.4.2 Regular slugging 217
9.4.3 Stochastic slugging 221
9.4.4 Stochastic models 228
9.5 Triggering of slug formation 229
10 Particle dispersion and deposition 232

10.1 Prologue 232
10.2 Dispersion of fluid particles in homogeneous isotropic turbulence 232
10.3 Equation of motion of a particle or a bubble 235
10.4 Lift forces and lateral drift 237
10.4.1 Inviscid lift forces 237
10.4.2 Saffman viscous lift force 237
x Contents
10.5 Characterization of fluid turbulence 238

10.6 Relation of particle turbulence to fluid turbulence 240
10.7 Dispersion of particles in a turbulent gas which is homogeneous and
isotropic 243
10.7.1 No effect of gravity 243
10.7.2 Effect of gravity; heavy particles 244
10.8 Measurements of particle turbulence for suspension flow in a pipe 245
10.8.1 The experiments of Lee et al. 245
10.8.2 The experiments of Young & Hanratty 248
10.9 Turbophoresis 249
10.9.1 The concept of turbophoresis 249
10.9.2 Formulation in cylindrical polar coordinates 250
10.9.3 Formulation in Cartesian coordinates 252
10.9.4 Physical interpretation of turbophoresis; effect on concentration
profiles 252
10.10 Deposition of particles 253
10.11 Stochastic representation of fluid turbulence 257
10.11.1 Prologue 257
10.11.2 The use of a modified Langevin equation in channel flow 257
10.11.3 Identification of mechanisms for deposition 259
10.11.4 Concentration profiles 263
11 Vertical annular flow 268

11.1 Prologue 268
11.2 Distribution function for drop size 269
11.3 Drop size 272
11.4 Rate of deposition 279
11.5 Experiments in turbulence generated by a DNS 281
11.6 Rate of atomization 285
11.7 Entrainment in vertical flows 286
12 Horizontal annular flow 292

12.1 Prologue 292
12.2 Drop distribution and local gas velocity 293
12.2.1 Measurements 293
12.2.2 Maps of gas velocity and droplet flux 293
12.2.3 Stochastic calculations for a 2-D rectangular channel 297
12.2.4 Formulation of the diffusion equation for particles in a pipe 301
12.2.5 Concentration distributions measured in a pipe 302
12.3 Distribution of liquid in the wall layer of a horizontal pipe 304
12.3.1 Description of the wall film 304
12.3.2 Asymmetric wall layer model 305
12.3.3 Distribution of the liquid in an asymmetric wall layer 310
Contents xi
12.4 Entrainment in horizontal pipes 315

12.4.1 Entrainment theory 315
12.4.2 Entrainment measurements 317
12.4.3 Calculations for a natural gas pipeline 320
12.5 Pool model 322
12.6 Transition to the annular regime 324
12.6.1 Air–water flow 324
12.6.2 Natural gas pipelines 327
Index 331
Preface
Gas–liquid flows are ubiquitous in industrial and environmental processes. Examples are
the transportation of petroleum products, the cooling of nuclear reactors, the operation of
absorbers, distillation columns, gas lift pumps. Quite often corrosion and process safety
depend on the configuration of the phases. Thus, the interest in this area should not be
surprising.
The goal of this book is to give an account of scientific tools needed to understand the
behavior of gas–liquid systems and to read the scientific literature. Particular emphasis is
given to flow in pipelines.
The following brief historical account is taken from a plenary lecture by the author at
the Third International Conference on Multiphase Flow, Lyon, France, June 8–12, 1998.
(Int. J. Multiphase Flow 26, 169–190, 2000):
A symposium held at Exeter (P. M. C. Lacey) in 1965 brought together 160 people with a wide range
of interests. Discussions at the 42 presentations indicated, to me, that something special was
happening and that future directions of work on multiphase flow were being defined. This thrust
was continued in conferences at Waterloo, Canada, in 1968 (E. Rhodes, D. S. Scott) and at Haifa,
in 1971 (G. Hetsroni). Intellectual activity in ensuing years is exemplified by more focused
conferences on Annular and Dispersed Flows held at Pisa, 1984 (S. Zanelli, P. Andreussi, T. J.
Hanratty) and in Oxford, England, in 1987 (G. F. Hewitt, P. Whalley, B. Azzopardi), the Symposium
on Measuring Techniques at Nancy (J. M. Delhaye, 1983) and the Conference on Gas Transfer at
Heidelberg (Jähne, 1995). However, the 350 papers presented at the Second International
Conference on Multiphase Flow in 1995 (A. Serizawa, Y. Tsuji) manifested a new level of activity.
A fair question is what happened between 1965 and 1995. My own assessment is that
major successes came about, mainly through efforts that relate macroscopic properties of
multiphase systems to small-scale behavior. An outcome of this approach is the possible
emergence of a new field. This is evidenced in many ways, of which the establishment
of the International Journal of Multiphase Flow (Gad Hetsroni, 1973) and the Japan
Society of Multiphase Flow (A. Akagawa, T. Fukano, 1987) are examples. The following
excerpt from a talk by R. T. Lahey at the inauguration of the Japan Society would indicate
that my observations are not original: “I believe that this new field will become as widely
accepted in the future as other emerging fields . . .”.
Physics of Gas–Liquid Flows addresses both graduate students and practitioners.
The treatment is based on a course, taught at the University of Illinois, which required
only that the students had taken one undergraduate course in fluid dynamics. As a
consequence, attention is given to topics that are usually bulwarks in graduate courses
xiv Preface
in fluid dynamics, such as ideal flow theory, the Navier–Stokes equations and interfacial
waves.
In reporting the results from published research I have retained the units (metric or
imperial) as used in the original work.
This work has a kinship to One Dimensional Two-Phase Flow by Graham Wallis,
published by McGraw-Hill in 1969. One should recognize that much has happened since
the publication of this book and that my way of presenting the material could be different
from that of that of Dr. Wallis.
Physics of Gas–Liquid Flows leans heavily on contributions by researchers in my
laboratory. A verbal summary of these works is contained in an account of the Research
of Thomas J. Hanratty which was deposited with the University of Illinois in the Illinois
Digital Environment for Access to Learning and Scholarship (IDEALS). It can be
visited at http://hdl.handle.net/2142/9132. The illustrations were prepared by Dorothy
Loudermilk. Taras V. Pogorelov provided a helping hand in transmitting the manu-
scripts to Cambridge University Press.
Cover
The photograph featured in the cover was obtained in the laboratory of Thomas J.
Hanratty by James B. Young. It captures the trajectories of particles in a turbulent liquid
flowing down a vertical pipe. Cross-sections at several locations were illuminated by thin
sheets having different colors. Axial viewing photography was used to capture the paths
of the particles. The color of a particle gives its axial location.
Symbols
A Area
A Measure of the thickness of the viscous wall layer, used by van Driest
AG Area of the gas space
AL Area of the liquid space
ALS Area of the liquid in a slug
AL1 Area of the liquid layer in front of a slug
APi Acceleration of a particle in the i-direction
AL0 Critical area for a liquid layer to sustain a stable or growing slug
At Area of a tube or pipe
a Amplitude of a wave
av Interfacial area per unit volume
B Height of a rectangular channel, or the height of the gas space
C Concentration
C0 Concentration at y = 0
CB Bulk concentration
CD Drag coefficient
CL Lift coefficient
CP Heat capacity at constant pressure
CV Heat capacity at constant volume
CW Concentration at the wall
c Velocity of sound; molecular velocity
c Complex wave velocity = cR+icI
c0 Wave velocity for stagnant fluids
cF Velocity of the liquid at the front of a slug
cB Velocity of the bubble behind a slug
cg Group velocity, the speed at which wave energy is transmitted
cG Sound velocity in the gas phase of a gas–liquid flow
cKW Kinematic wave velocity, defined by (7.51)
cL Sound velocity in the liquid phase of a gas–liquid flow
ct Turbulent concentration fluctuation
D Molecular diffusion coefficient
dm Maximum drop diameter
dt Pipe diameter
xvi Symbols
dh Hydraulic diameter = 4A/wetted perimeter dvμ

dvμ Volume median diameter
dqm Drop diameter defined by Mugele & Evans
d10 Number mean diameter
dP Diameter of a bubble or a drop
d32 Sauter mean diameter
E Entrainment
E Energy associated with one wavelength = P + T
EM Maximum possible entrainment
E(n) Spectral density function
e Energy per unit mass
ec Capture efficiency
!
F Force
F Mechanical energy lost per unit mass of fluid due to friction
F Flux of particles
FD Resisting force on a particle due to fluid drag
FD Flux due to diffusion
FB Buoyancy force
!
FL Lift force
f Fanning friction factor
f Factor that accounts for the volume of fluid dragged along by a particle
fn(dP) Number distribution function
fv(dP) Volume distribution function
fS Friction factor for a smooth surface
fS Frequency of slugging
fi Friction factor for gas flow over a gas–liquid interface
fPi Force of the fluid on a particle
G Mass velocity defined by (1.4)
GC Mass velocity at the choking condition
GL Mass velocity of the liquid
GG Mass velocity of the gas
GLE Mass velocity of drops entrained in the gas
!g Acceleration of gravity
g̑ Defined by (3.20) and (3.21)
g+ Equal to g=v
H Submergence of injector in a gas-lift pump
h Enthalpy
h Height above a datum plane
hH Enthalpy of a gas–liquid mixture
hL Enthalpy of the liquid
hG Enthalpy of the gas
hL Height of liquid layer in a channel
hL1 Height of the liquid layer in front of a slug
Symbols xvii
hL In a pipe, the length of the bisector of the liquid layer

hþL Dimensionless height of the liquid layer ¼ hL v =
hG Height of the gas layer in a channel
hL0 Critical height of the liquid layer needed for slugs to appear
hcL Location of the centroid of an area
hLB Height of the liquid at the bottom of a horizontal pipe
hW Distance of the top of the waves from the wall
hLS Height of the liquid layer if the flow is steady
Δh Wave height
I Intermittency, fraction of the time that disturbance waves are present
I1 Integral defined by (12.63)
I2 Integral defined by (12.64)
i Internal energy per unit mass
k Wave number
kM Wave number at which wave growth is a maximum
kS Sand roughness
kD Deposition coefficient
L Length of pipe or channel
L Length through which liquid is lifted in a gas lift pup
LD Pipe length needed for slugs to develop
LS Slug length
LU Length of a region of unstable stratified flow
l Length in the flow direction
ℓ Mixing length; characteristic viscous length
M Molecular weight
MLG Mass transfer rate per unit area from the liquid to the gas
MGL Mass transfer rate per unit area from the gas to the liquid
m Average height of the liquid layer around the circumference
mP Mass of a particle
m+ Dimensionless height based on the friction velocity and the viscosity
mþ G Ratio of the film height to a gas-phase length scale
mþ c Dimensionless film height, based on vc and the viscosity of the liquid
N Local rate of mass transfer
n Frequency, cycles per second
P Perimeter
P Potential energy associated with one wavelength
PG Length of the pipe perimeter in contact with the gas
PL Length of the pipe perimeter in contact with the liquid
PS Pressure imposed on a solid or liquid surface by a flowing fluid
~
P Pressure made dimensionless with ρLgdt
p Pressure
pa Ambient pressure
pG Pressure along a sinusoidal wavy interface = pGR+ipGi
xviii Symbols
pGR Pressure component which has a minimum at the wave crest

pGI Pressure component which is in phase with the wave slope
pGi Gas pressure at the interface
pLi Liquid pressure at the interface
QA Volumetric rate at which a liquid layer is atomizing
QG Volumetric flow of gas
QL Volumetric flow of the liquid
Qsh Volumetric flow at which slugs shed liquid
q Velocity of the fastest-moving particle on the interface of a wave
q Magnitude of the velocity
q Volumetric flow per unit breadth ≈ hua
q2 Equal to u2x þ u2y þ u2z
q_ Time rate at which heat is added to a control volume
qW Rate of heat addition at a wall per unit area
q_ rev Heat transfer rate if changes are occurring reversibly
qWL Heat transfer rate per unit area from a wall to a liquid
qWG Heat transfer rate per unit area from a wall to a gas
qGL Heat transfer per unit area from a gas to a liquid
qLG Heat transfer per unit area from a liquid to a gas
R Wave resistance defined by (4.116) and (4.117)
R Molar gas constant
R Radius of curvature of a surface
RD Mass rate of deposition of particles per unit area
RA Mass rate of atomization of the wall layer per unit area
R1,2 Principal radii of curvature of a surface
Rij The Reynolds stress divided by the density
Rpath
i Correlation coefficient of fluid velocity fluctuations seen by a particle as it moves
around in a turbulent field
RLi Lagrangian correlation coefficient for turbulent fluid particles
RLPi Lagrangian correlation coefficient for particles in a turbulent field
RE Eulerian correlation coefficient
rt Radius of a tube or a pipe
rP Particle or bubble radius
rij Wave-induced variation of the Rij
S Slip ratio, equal to gas velocity divided by the liquid velocity
S Projected area of a particle
Si Length of the interface in an idealized stratified flow
s Entropy
s Sheltering coefficient, defined by Jeffreys
T Temperature
T Kinetic energy in one wavelength
TS Tangential stress imposed on the interface by a flowing fluid
t Time
Symbols xix
tC Time constant reflecting the rate of growth of waves

UP Particle velocity
UA Velocity of atomizing drops in the flow direction
UD Velocity of depositing drops in the flow direction
UB Velocity of a bubble in a stationary liquid; bulk velocity
US∞ Rise velocity of a bubble in an infinite stagnant liquid
U1 Convection velocity of a homogeneous, isotropic field
U Settling velocity of a suspension of particles
!C
US Relative velocity between a particle (or a bubble) and a fluid
u One-dimensional velocity
!u Fluid velocity vector
uc Fluid velocity at the center of a pipe or channel
utt Turbulent velocity component
u+ Velocity made dimensionless using the friction velocity
un Component of the velocity normal to a surface
ut Component of the velocity tangent to a surface
uH Velocity of a homogeneous mixture, defined by (1.48)
uL Velocity of the liquid
uL1 Velocity of the stratified layer in front of a slug
uL3 Velocity of the liquid in the body of a slug
uslug Velocity of a slug
uG Gas velocity
uGc Critical gas velocity for the initiation of atomization
uGM Maximum value of uG
um Mixture velocity = uGS + uLS
ua Spatially averaged liquid velocity
uS Liquid velocity at the interface
uga Spatially averaged gas velocity
uGSt Critical superficial gas velocity at which Kelvin–Helmholtz waves appear
uGS Superficial gas velocity = volumetric flow of gas divided by A
uLS Superficial liquid velocity = volumetric flow of liquid divided by A
uo Rise velocity of bubbles
uG∞ Rise velocity of single bubbles in infinite media
V Volume
VD Mean velocity of particles striking a wall
VDL Drift velocity associated with a lift force
VDrift Drift velocity, defined by (8.29)
VD Mean velocity with which particles are depositing
VW Average velocity with which particles strike a boundary
VP Volume of a particle or a bubble
VT Terminal free-fall velocity
VR Average velocity of entrained particle in the radial direction
V 0i Velocity with which particles enter the field
xx Symbols
Vtp Turbophoretic velocity

V GS Dimensionless group defining the initiation of annular flow (11.1)
v Specific volume
vG Specific volume of a gas
vL Specific volume of a liquid
v* Friction velocity, using the shear stress at the wall or the interface
vc Critical frictional velocity
vG Friction velocity based on gas density and stress at the interface
vL Friction velocity based on the liquid density and the stress at the wall
vc Friction velocity based on liquid density and the characteristic stress
!v Velocity of a fluid particle
!vP Velocity of an entrained particle or bubble
vtPi Turbulent velocity fluctuation
W Weight rate of flow
WL Weight flow rate of liquid
WLP Mass flow of liquid in the “pool” at the bottom of the pipe
WG Weight flow rate of gas
WW Mass flow of liquid on the wall
WLE Mass flow of entrained drops
WLF Mass flow rate in the wall layer
WLFC Critical flow below which atomization does not occur
w_ Rate at which fluid in a volume is doing work on surroundings
w Width of a control volume
X 2P Mean square displacement of particles in a turbulent field
x Mixture quality, mass fraction that is a gas or vapor
x Coordinate in the flow direction in a Cartesian coordinate system
y Coordinate that is perpendicular to a wall
yfl Location at which particles start a free-flight to the wall
y+ Distance from the wall made dimensionless with the friction velocity
y0 Average location of the interface
z Coordinate in the direction of flow in a pipe flow
z Distance from reference plane
Greek symbols
α Volume fraction of gas in a flowing mixture

α Ratio of the fastest moving particle at the interface to wave velocity
αW Ratio of wave height to what would be observed on a deep liquid
α Parameter
representing
1 shape of velocity profile
α Equals τ path
i
αP Volume fraction of particles or bubbles
αm+ Dimensionless quantity defined by (3.28)
β Reciprocal of the inertial time constant of an entrained particle
Symbols xxi
β~ Reciprocal time constant for a particle in a liquid (10.50)

Γ Gamma function
Γ Volumetric flow per unit length in the spanwise direction
Γc Critical film flow below which atomization does not occur
Γ c Modified critical film flow (12.74)
ΓS A shape factor for the velocity profile (7.6)
γ Ratio of the heat capacities at constant pressure and constant volume
δ Thickness of a viscous boundary layer
δij Delta function
ε Rate of dissipation of mechanical energy per unit volume
ε Lagrangian turbulent diffusivity of fluid particles
ε Ratio of the fastest-moving particle in the interface to the wave velocity
εh Diffusivity representing turbulent mixing in the wall layer
εL Liquid holdup = (1 − α)
εP Lagrangian turbulent diffusivity of entrained particles
εEP Eulerian diffusivity of entrained particles
η Displacement of the interface from its average location
θ Inclination angle to the horizontal
θ Phase for an oscillating quantity
θ Angular location on the wall of a pipe, where θ = 0 can be the top or bottom
of the pipe
θg Orientation angle of flow to the gravitational vector (Figure 1.2)
Λ Lagrangian length scale
λ Wavelength
λm Wavelength which is growing the fastest
λT Lagrangian micro-time-scale
μ Viscosity
μG Gas viscosity
μL Liquid viscosity
μP Viscosity of the fluid inside a drop or bubble
μt Turbulent viscosity
ρ Density
ρL Liquid density
ρG Gas density
ρP Particle density
ρf Fluid density
ρS Density of a dissolving solid surface
ρH Density of a homogeneous mixture
ρG ρG cothðkhG Þ
ρL ρL cothðkhL Þ
σ Surface tension
σP Root-mean-square of the turbulent velocity fluctuations of particles
σ Root-mean square of a fluctuating quantity
xxii Symbols
τ Magnitude of a shear stress

τc Time constant characterizing the rate of growth of waves
τc Characteristic stress in a liquid layer = (2/3)τW + (1/3)τi
τP Inertial time constant of a particle
τP Effective stress due to atomization and deposition
τ path
i Time constant characterizing Rpath
i
τL Lagrangian time constant
τE Eulerian time-scale
τPB Volume-average inertial time constant
τPS Time constant for a particle obeying Stokes law
τW Shear stress at a wall
τWG Resisting stress at the wall on a gas
τWL Resisting stress at the wall on a liquid
τi Shear stress at an interface
τij Component of a stress tensor
τ tij Component of a turbulent stress tensor
τDA Net momentum flux due to atomization and deposition
Kinematic viscosity
L Kinematic viscosity of the liquid
G Kinematic viscosity of the gas
t Turbulent kinematic viscosity
ϕ Potential function describes velocity for an irrotational field
ϕ2G Dimensionless frictional pressure gradient defined by (1.100)
ϕ2L Dimensionless frictional pressure gradient defined by (1.101)
ω Circular frequency = kc
!
ωP Angular rotation of a particle or bubble
ψ Stream function defined by (6.20)
Dimensionless groups
β~ ¼ 3C D ρf j!
u ! v P j=4rP ð2ρP þ ρf Þ
dh
Bo Bond number =
½σ=ðρL ρG Þg 1=2
rffiffiffiffiffi
γðReLF Þ L ρL
F ¼ is a parameter in (3.41)
Re0:9
G G ρG
FH F for a horizontal pipe (12.48)
Fr Froude number = uGS =ðgd t Þ1=2
Ma Mach number = ratio of the fluid velocity to the velocity of sound
Re Reynolds number
ReG Gas-phase Reynolds number = dtWG/ArμG for a pipe
ReLF Film Reynolds number = 4G=L
ReL0 Liquid Reynolds number based on the velocity at the interface
Symbols xxiii
ReP Particle Reynolds number = d P ρU S =μ

Sc Schmidt number
θ Number reflecting the effect of viscosity on wave growth
We Weber number
X Martinelli parameter
Other symbols
h i Spatial average, ensemble average or phase average

( )i Quantity evaluated at the interface
j j Signifies an absolute value
D
Substantial derivative. Time change seen in a framework of a moving fluid
Dt
particle (defined by equation (4.13))
N Overbar indicates a time-average
N0 Prime indicates a fluctuating quantity
Nt A turbulent fluctuation
^
N Indicates complex amplitude of a fluctuating quantity induced by sinusoidal
waves
~
N ^ divided by the wave amplitude, a
Tilde indicates N
~
~
N Difference between the phase-average and the time-average
N̑
Defined by (3.20), (3.21)
dP
Frictional contribution to the pressure gradient
dz
F
dP
Frictional pressure gradient for gas flowing along a pipe
dz
GS
dP
Frictional pressure gradient for liquid flowing along a pipe
dz LS
1 One-dimensional analysis
1.1 Introduction
The “simplest” models for gas–liquid flow systems are ones for which the velocity is
uniform over a cross-section and unidirectional. This includes flows in a long straight
pipe and steady flows in a nozzle.
A treatment of pipe flow with a constant cross-section is initiated by reviewing
analyses of incompressible and compressible single-phase flows. A simple way to use
these results is to describe gas–liquid flows with a homogeneous model that assumes
the phases are uniformly distributed, that there is no slip between the phases and
that the phases are in thermodynamic equilibrium. The volume fraction of the gas, α,
is then directly related to the relative mass flows of the phases. However, the assumption
of no slip, S = 1, can introduce considerable error. This has prompted a consideration
of a separated flow model, where uniform flows of gas and liquid are pictured as
moving parallel to one another with different velocities and to be in thermodynamic
equilibrium.
The two-fluid model develops equations for the interaction of two interpenetrating
streams. It does not require the specification of S or the assumption of equilibrium
between the phases. However, it introduces several new variables.
An interesting feature of the single-phase analysis of a compressible fluid is the
existence of the choking phenomenon whereby there is a maximum flow which can be
realized for a system in which the pressure at the pipe inlet and the pressure in the receiver
are controlled. The homogeneous model predicts much smaller choking velocities for
gas–liquid flows than would exist for gas flowing alone.
The separated flow model requires equations for the stress at the wall, τW, and α
(or the slip ratio). The homogeneous model requires only an equation for τW. The
specification of these quantities is a continuing interest and is a focus in future chapters.
A starting point for this pursuit is the widely used Lockhart–Martinelli analysis
discussed in this chapter.
Steady flow in a nozzle is different from flow in a straight pipe in that changes in inertia
override the importance of wall resistance and of gravity. Equations that use the separated
flow model are developed in this chapter for gas–liquid flows in a nozzle.
The materials in Section 1.6 on flow in a nozzle and on the two-fluid model can
be ignored in a first reading since they are not needed to follow the main narrative of
this book.
1.2 Single-phase flow
1.2.1 Flow variables

When a single phase is flowing, the field is described by the variables pressure, p,
velocity, !
u , density, ρ and temperature, T. Four equations are needed: conservation of
momentum, conservation of energy, conservation of mass, and a relation among state
variables. Examples of the last are the perfect gas law and tables of thermodynamic
properties.
1.2.2 Momentum and energy theorem

For flow systems, it is convenient to formulate Newton’s second law of motion as a
momentum theorem, which focuses on a fixed volume in space rather than on particles
moving through a field. A sketch of an arbitrary volume is shown in Figure 1.1. Note that,
on the boundary, the velocity vector, !
u , can be represented by components perpendicular,
un, and tangent, ut, to the boundary. The theorem is stated as follows: “The time rate of
change of momentum in a fixed volume in space plus the net flow of momentum out of the
volume equals the net force acting on the volume.”
ð ð
∂ðρ!uÞ !
dV þ ρun !
u dA ¼ F ð1:1Þ
∂t
Thermodynamic variables are defined for systems which are at equilibrium. The appli-
cation of conservation of energy and the equation of state to a flowing system, therefore,
usually involves the assumption that the adjustment of molecular properties to a change
of the environment occurs much more rapidly than the change of the flow field.
(Examples where this assumption might not be valid are flow through a shock wave or
the stagnation of flow at a very small impact tube.) The energy of a flowing fluid is
defined as
juj2
e¼iþ ð1:2Þ
2
u
un
ut
Figure 1.1 An arbitrary volume in space.

1.2 Single-phase flow 3
dz z
τW θg
dp
p+
p g
τW
Flow
Figure 1.2 Control volume in a pipe.
where i is the internal energy per unit mass and |u|2/2 is the kinetic energy per unit mass.
The first law of thermodynamics is formulated for a flow system as the energy
theorem: “The time rate of change of energy in a fixed volume in space plus the net
flow of energy out of that volume equals the rate at which heat is added, q_, minus the rate
at which the fluid in the volume is doing work on the surroundings, w_.”
ð ð
∂ðρeÞ
dV þ ρun e dA ¼ q_ w_ ð1:3Þ
∂t
where ρe is the energy per unit volume.
1.2.3 Steady flow in a duct of constant area

The equations defining the steady flow in a pipe of constant area, A, can be derived by
considering the fixed control volume shown in Figure 1.2. The pipe is inclined at an angle
θg to the direction of the gravitational vector. The fluid is flowing in the z-direction with
an average velocity, u. Equations are to be developed describing the changes of u, p, ρ,
and T with z.
Under steady flow conditions, the mass flow, ρuA, does not vary along the pipe. The
equation for conservation of mass is
AG ¼ uρA ¼ constant ð1:4Þ
where term G is called the mass velocity.

The flux of momentum along the pipe is given as GAu. Since the flow is steady, the
time rate change in the control volume, indicated by the dashed lines in Figure 1.2, is zero
so the momentum theorem states that the net flow of momentum out of a fixed volume
equals the net force. The flows of momentum into and out of the control volume are given
by ρu2A and ρu2A + d(ρu2A), so the net flow of momentum out is d(ρu2A).
The forces due to pressure acting on the front and back faces are pA and –(p + dp)A, so
the net force is –(dp)A. The force of gravity is − ρg cos θg Adz. The wall resists the flow
with a stress, τW, acting on the side walls of the control volume and the net force due to
this stress is −τWPdz where P is the perimeter of the duct. The momentum theorem gives
d ðAGuÞ ¼ ðdpÞA τ W Pdz ρAg cos θg dz ð1:5Þ

The wall shear stress is usually represented as

1
τ W ¼ ρu2 f ð1:6Þ
2
where the Fanning friction factor, f, needs to be determined empirically for turbulent
flows. Conservation of mass indicates that d(GA) = 0. If this is used in (1.5), the following
is obtained after substituting (1.6).
f PG2
G du ¼ dp dz ρg cos θg dz ð1:7Þ
2Aρ
From conservation of mass, if A is constant, du = G d(1=ρ) so that

f PG2
G2 dð1=ρÞ ¼ dp dz ρgcosθg dz ð1:8Þ
2Aρ
This shows that pressure changes result from balancing the frictional drag, the gravita-
tional force and the acceleration of the fluid.
Equation (1.8) can be solved directly for p if ρ is constant, that is, if there is no
change in velocity in the direction of flow. If ρ is not constant the relation ρ (p) is
needed. This can be obtained from the energy theorem and the equation of state. The
rate of energy addition to the control volume can be calculated from the rate of heat
addition at the wall per unit area as qWP dz. The rate at which the fluid in the control
volume does work on the surroundings consists of two terms. The rate at which
the fluid at z works to get into the control volume is puA. The rate at which the fluid
at z + dz does work to get out of the control volume is (puA + d(puA)). The net rate of
work on the surroundings due to these effects is d(pua). The fluid in the control volume
works against gravity to lift the fluid through a distance cos θg dz. This contributes a
rate of work of GAg cos θg dz.
Thus, for the steady flow depicted in Figure 1.2, the energy theorem gives
ju2 j
AG di þ AG d ¼ qW P dz dðpAuÞ GAg cosθg dz ð1:9Þ
2
The enthalpy per unit mass is given as

p
h¼iþ ð1:10Þ
ρ
where 1/ρ is the volume per unit mass. The kinetic energy per unit mass is ju2 j=2. If (1.10)
and u = G/ρ are substituted into (1.9) and A is constant,

1 1 q P
dh ¼ G2 d 2 þ g cos θg dz ¼ W dz ð1:11Þ
2 ρ AG
Equation (1.11) defines the change of enthalpy. For a given fluid, the thermodynamic
state is defined by any three state variables, so
h ¼ f ðp; ρÞ ð1:12Þ
Therefore, a table or a diagram representing the thermodynamic variables provides

a relationship between ρ and p if h(z) is known. Equation (1.11) gives a relation
between p and p along the z-axis. This can be used in (1.8) to eliminate either p(z)
or ρ(z).
1.2.4 Choking
Consider a length of pipe for which the fluid enters at a pressure p1 and discharges into a
chamber whose pressure, p2, is changed. Equation (1.8) can be solved to obtain G for a
given p2/p1. The solution shows that G initially increases as p2/p1 decreases. Eventually
it gives a maximum and the impossible result that G decreases with decreasing p2/p1.
The maximum in G represents a choking condition for which further decreases of the
pressure in the discharge chamber have no effect on the mass flow and the flow along
the pipe remains unchanged. This can be understood when it is realized that the
equations predict that a decrease in p2 is accompanied by an increase in the velocity
and a decrease in the density. These have opposite effects on the mass velocity since
G = ρu. Thus, for p2/p1 less than the critical value the equations predict that decreases in
ρ offset the increases in u.
At choking, the fluid velocity equals the velocity with which small disturbances (sound
waves) propagate in a fluid. Thus, information about the change in the pressure at the
outlet cannot be transmitted upstream and the behavior in the pipeline is not affected by
what is happening in the receiving chamber. The fluid discharging from the pipeline
expands supersonically and forms shock waves in the receiver.
This interpretation is illustrated, in a direct way, by considering the momentum
balance, equation (1.8). The term on the left side of (1.8) may be written as

1 G2 dρ
G2 d ¼ 2 dρ ¼ u2 dp ð1:13Þ
ρ ρ dp
where dp/dρ is the change of pressure with density along the pipe. Substituting (1.13) into
(1.8) gives

dp u2 f G2 P
1 2 ¼ ρg cos θg ð1:14Þ
dz c 2Aρ
where the velocity of sound, c, is given by

dp
c2 ¼ ð1:15Þ
dρ
It is noted that the coefficient of dp/dz changes sign when the Mach number, Ma = u/c,
assumes values greater than unity. This suggests that the frictional pressure gradient
changes from a negative to a positive value. A positive frictional pressure gradient is an
impossibility so Ma cannot change from a value less than unity to a value greater than
unity in a pipeline of constant area. The flow chokes when u = c at the outlet.
The entropy of a fluid can be defined by the following thermodynamic relation
dp
T ds ¼ dh ð1:16Þ
ρ
Thus, the change of entropy along the pipeline can be calculated if h(z), p(z) and ρ(z) are
known. This calculation shows that, for an adiabatic flow, s increases with z because of
the irreversible effects associated with wall friction. At the choking condition ds = 0 so,
from (1.15), the disturbance velocity is given as

∂p
c ¼
2
ð1:17Þ
∂ρ s
This is a thermodynamic quantity which is the classical definition of the velocity of

sound. When the flow is not adiabatic, qW ≠ 0, this is not the case. For example, if the
frictional heating is balanced by heat losses, the temperature might be kept approximately
constant so that

∂p
c2 ¼ ð1:18Þ
∂ρ T
1.2.5 Flow of an ideal gas in a pipe of constant area

An ideal gas is defined with the equations
p R
¼ T ð1:19Þ
ρ M
di ¼ C V dT ð1:20Þ
dh ¼ C P dT ð1:21Þ
where R is the molar gas constant, M is the molecular weight, CV is the heat capacity
at constant volume and CP is the heat capacity at constant pressure. Thus, for an
ideal gas

∂p p
¼γ ð1:22Þ
∂ρ s ρ

∂p R
¼ T ð1:23Þ
∂ρ T M
where γ = CP /CV and

R
¼ CP CV ð1:24Þ
M
The application of the equations developed in Section 1.2.3 will be illustrated by

considering the adiabatic flow of an ideal gas in a pipe of constant area. Since a gas is
being considered, gravitational effects can be neglected and the momentum balance
equation is written as
ρ dρ 2f
2
dp þ dz ¼ 0 ð1:25Þ
G ρ dt
where dt is the pipe diameter. This can be integrated if the dependency of p on ρ is known.
Conservation of energy for qW = 0 gives

1 1
dh þ Gd 2 ¼ 0 ð1:26Þ
2 ρ
Since conservation of mass gives G = constant, the equation can be integrated to give
G2 G2
ð h h1 Þ ¼ 2
2 ð1:27Þ
2ρ1 2ρ
where subscript 1 indicates inlet conditions. From (1.21),

ðh h1 Þ ¼ C P ðT T 1 Þ ð1:28Þ
If (1.19) and (1.24) are used in (1.28)

R γ γ p p1
h h1 ¼ ðT T 1 Þ ¼ ð1:29Þ
M γ1 γ 1 ρ ρ1
The following relation between p and ρ is obtained if (1.27) is substituted into (1.29):
p ðγ 1Þ G2 ρ1 ðγ 1Þ G2 ρ ρ
¼ þ þ ð1:30Þ
p1 γ 2p1 ρ1 ρ γ 2p1 ρ1 ρ1 ρ1
Equation (1.30) is used to eliminate p from (1.25). The following relation for ρ is
obtained if f is assumed to be constant.

ðγ þ 1Þ 1 ρ ðγ 1Þ 1 ρ2 p1 ρ1 ρ2 2f ðz z1 Þ
ln 1 1 ¼0 ð1:31Þ
γ 2 ρ1 γ 4 ρ1 2
2G ρ1
2 2 dt
If the density variation is calculated with (1.31), the pressure variation is obtained from
(1.30) and the velocity can be obtained from conservation of mass
u ρ
¼ 1 ð1:32Þ
u1 ρ
The temperature is obtained from (1.19).

T p ρ1
¼ ð1:33Þ
T 1 p1 ρ
From (1.16), (1.11), (1.19) the following equation is obtained for the entropy

R G2
ds ¼ dρ dp ð1:34Þ
M p ρ2
This can be integrated to obtain s − s1, using the equations for p and ρ developed above.
If an isothermal flow is assumed

p ρ
¼ ð1:35Þ
p1 ρ 1
This condition will hold for a high molecular weight gas for which CV is quite large and
γ ¼ ðC V þ RÞ=C V ≈ 1. If γ = 1 is substituted into (1.30), it is seen that (1.35) results.
However, in general, isothermal conditions can be maintained only if heat is exchanged
with the surroundings; that is, the flow would not be adiabatic.
Equation (1.25) can be integrated directly if (1.35) is substituted for ρ
" #
p1 ρ1 p 2 p 2f ðz z1 Þ
2 1 þ ln ¼0 ð1:36Þ
2G p1 p1 dt
For an isothermal flow of an ideal gas, the enthalpy h is constant so the energy equation
gives
qW P G2 G2
ðz z1 Þ ¼ 2 ð1:37Þ
AG 2ρ 2ρ1 2
The heat that is added to the flow to maintain a constant temperature must increase with
increases in kinetic energy. Since dh = 0, equation (1.16), describing the change in
entropy, gives
1 dp
ds ¼ ð1:38Þ
T ρ
This can be integrated, using (1.19), to yield

R p
s s1 ¼ ln ð1:39Þ
M p1
The entropy will increase as long as p decreases. It need not attain a maximum at the
value of p/p1 for choking to occur. This difference from the adiabatic case arises since the
system cannot be considered as isolated so that a reversible process need not correspond
to one for which the entropy is constant.
1.2.6 Mechanical energy balance

The first law of thermodynamics defines a change in internal energy as

q_ P 1
di ¼ rev dz pd ð1:40Þ
GA ρ
where q_ rev is the heat needed for a reversible operation. If this is substituted into the
energy balance, equation (1.9),
2
dp u P
þd þ g cos θg dz ¼ ðqW q_ rev Þ dz ð1:41Þ
ρ 2 AG
1.3 The homogeneous model for gas–liquid or vapor–liquid flow 9
Define F as the rate of mechanical energy loss per unit mass of fluid, where
dF ¼ ðq_ rev qW Þ ð1:42Þ
Thus (1.41) can be rewritten as

2
dp u
þd þ g cos θg dz ¼ dF ð1:43Þ
ρ 2
A comparison of (1.43) with the momentum balance equation gives

dF τ W P 4 τ W
¼ ¼ ð1:44Þ
dz ρA dt ρ
1.3 The homogeneous model for gas–liquid or vapor–liquid flow
1.3.1 Basic equations

The simplest approach to flows involving a gas and a liquid is to treat the mixture
the same as a single phase. This requires assuming (1) zero slip between the phases,
(2) uniform flow and (3) equilibrium between the phases (one can use thermodynamic
tables when the vapor and liquid are in contact). Define the quality, x, as the mass fraction
of the flowing mixture that is a gas or vapor and a void fraction, α, as the fraction of the
cross-section that is occupied by the gas. For flows such as air–water, where phase
changes are not occurring, x does not vary along the pipeline. For flows with a fluid pair
such as steam–water, vaporization can occur, so x can change.
Define a mixture density as the ratio of the mass flow to the volume flow:
GA
ρH ¼ ð1:45Þ
AxGvG þ ð1 xÞGAvL
where
1 1
vG ¼ vL ¼ ð1:46Þ
ρG ρL
Thus
1 x ð 1 xÞ
¼ xvG þ ð1 xÞvL ¼ þ ð1:47Þ
ρH ρG ρL
A mixture velocity is defined as

G
uH ¼ ð1:48Þ
ρH
For the condition uG = uL

QG xGAvG vG ρ
α¼ ¼ ¼x ¼x H ð1:49Þ
QG þ QL xGAvG þ ð1 xÞGAvL vH ρG
Also
α x ρL
¼ ð1:50Þ
1 α ð1 xÞ ρG
The momentum balance, equation (1.8), is rewritten as

ρH dρ ρ2 τWρ P
dp þ H H2 g cos θg dz 2H dz ¼ 0 ð1:51aÞ
G 2 ρH G G A
or, alternatively, as
dp dρ1 τWP
¼ G2 H þ ρH g cos θg þ ð1:51bÞ
dz dz A
The conservation of energy equation (1.11) is given as

1 1 q P
dhH þ G2 d 2 þ g cos θg dz ¼ W dz ð1:52Þ
2 ρH AG
where
hH ¼ xhG þ ð1 xÞhL ð1:53Þ
The enthalpy of the gas, hG, and the enthalpy of the liquid, hL, are obtained from
thermodynamic correlations.
For two-component systems such as air and water, the quality, x, does not vary along
the pipeline and is fixed by the inlet conditions. The enthalpies, hL and hG, are fixed
for a given p and ρH (which relate to ρG and ρL through equation (1.47)). For a single-
component system, such as steam–water, the quality, x, and the densities, ρL and ρG, are
fixed for a given pressure and enthalpy, if equilibrium is assumed between the phases.
Thus, (1.52) can be used to provide a relation between ρH and p. This can be used to
integrate (1.9). Equation (1.52) and thermodynamic data can be used to calculate the
variation of p, ρH, G, x with distance along the pipe.
1.3.2 Choking
The momentum balance for a homogeneous flow can be rewritten as

G2 ðdρH =dpÞ ρ2 τW P
dp 1 H2 g cos θg dz dz ¼ 0 ð1:54Þ
ρH
2
G A
Choking occurs when the term in the brackets changes sign

1 dvH
¼ ð1:55Þ
G2C dp
From (1.15), define

1 dρH dv1 dvH
2
¼ ¼ H ¼ ρ2H ð1:56Þ
c dp dp dp
1.3 The homogeneous model for gas–liquid or vapor–liquid flow 11
Substitute (1.48) and (1.56) into (1.54) to get

u2H τWP
dp 1 2 ¼ dz ð1:57Þ
c A
Choking is predicted when u2H =c2 ¼ 1.

Using (1.47) for vH, one obtains

dvH dvG dvL dx
¼x þ ð 1 xÞ þ ð vG vL Þ ð1:58Þ
dp dp dp dp
From (1.56) and (1.58)
1 x ð 1 xÞ dx
¼ þ 2 2 þ ð vG vL Þ ð1:59Þ
ρ2H c2 ρ2G c2G ρL cL dp
where cG and cL are the sound velocities in the gas and in the liquid

1 dp
cG ¼ 2
2
ð1:60Þ
ρG dvG

1 dp
c2L ¼ 2 ð1:61Þ
ρL dvL
Since, for a homogeneous flow, x = α

α ð1 α Þ ðvG vL Þ dx
c2 ¼ ½αρH þ ð1 αÞρL þ þ ð1:62Þ
ρG c2G ρL c2L vH dp
Take a path of dx/dp = 0. Note that
c2 ¼ c2G α¼1 ð1:63Þ
c2 ¼ c2L α¼0 ð1:64Þ
For α ≠1 and ρL much greater than ρG
ρG c2G
c2 ¼ ð1:65Þ
ρ L ðα Þð1 α Þ
This has a minimum at α = 1/2. The important point to be made is that the homogeneous
model predicts smaller sound velocities than would exist in the gas alone. Thus choking
could occur at a smaller exit velocity for a gas–liquid mixture than for the pure gas.
For example, Wallis (1969) points out that homogeneous theory predicts choking
velocities as low as 2.1 m/s for air–water flows. This is to be compared with a value of
about 340 m/s in air at room temperature.
1.3.3 Calculation of wall shear stress

The usual practice in making calculations with a homogeneous model is to define a
friction factor based on ρH.
f
τ W ¼ ρH u2H ð1:66Þ
2
Single-phase results are used to obtain f. This requires a definition of viscosity which is
dependent on the viscosities of both the gas and the liquid.
This approach does not have a sound basis. The use of friction factors defined by (1.66)
for flows which have drops entrained in a gas flow can produce predictions of the
pressure drop which show an effect of liquid flow which is stronger than observed.
Thus, Henstock & Hanratty (1976) defined friction factors for annular flows based on the
gas density. Likewise, the prediction of pressure drops in the bubbly flow that constitutes
the body of a slug (Fan & Hanratty, 1993) are best represented by using the density of the
liquid, rather than ρH, in (1.66).
1.3.4 Comments about the reliability of the homogeneous model

There are a number of problems in using the homogeneous model. The assumption
of a homogeneous distribution of the phases is sometimes a poor approximation. The
use of single-phase relations for τW with ρH substituted for ρ and a viscosity which is
a combination of the liquid and gas viscosities is not substantiated. An appreciable
slip can exist between the phases, so the equating of α to x can be incorrect and lead
to errors in calculating the contribution of the pressure drop associated with hydro-
static head. Also, the assumption of equilibrium between vapor and liquid phases
can be in error.
1.4 Separated flow model for gas–liquid flow
1.4.1 Basic equations

The separated flow model differs from the homogeneous model by recognizing that the
velocities of the two phases are usually different. It is a special one-dimensional version
of the two-fluid model, which pictures the phases to be two separate streams.
A slip ratio between the phases is defined as
uG
S¼ ð1:67Þ
uL
and the quality of the mixture, x, retains its definition as the ratio of the mass flow of the
gas to the mass flow of the mixture. The volume flows of the gas and the liquid are given
as GxvG and G(1 – x)vL so the velocities are
1.4 Separated flow model for gas–liquid flow 13
GAx
uG ¼ ð1:68Þ
ρG Aα
GAð1 xÞ
uL ¼ ð1:69Þ
ρL Að1 αÞ
where α and (1 − α) are the fractions of the cross-section, A, occupied by the gas and the
liquid. From (1.67), (1.68) and (1.69) the slip ratio is given as
x ð1 α Þ ρ L
S¼ ð1:70Þ
ð1 xÞ α ρG
Thus, the relation between x and α for a homogeneous model, equation (1.50), is
modified as indicated in (1.70).
The momentum flux includes the contributions of both the gas and liquid flows
Momentum flux ¼ GAxuG þ GAð1 xÞuL ð1:71Þ
The substitution of (1.68) and (1.69) into (1.71) gives
Gx G ð 1 xÞ
Momentum flux ¼ GAx þ GAð1 xÞ ð1:72Þ
ρG α ρL ð1 αÞ
As in Section 1.2.2, the momentum theorem is applied to a differential length along the
pipe. The net flow of momentum out of the control volume, (dz)A, equals
!
d G2 Ax2 G2 Að1 xÞ2
þ dz
dz αρG ð1 αÞρL
The force of gravity on the control volume equals
g cos θg ½ρL ð1 αÞ þ ρG αA dz
In a liquid–vapor mixture, an increase in x is associated with an increase in the momen-

tum of the vapor and a decrease in the momentum of the liquid given by
AGðuG uL Þdx
The momentum balance for the separated flow model, therefore, gives
" #
G2 Ax2 G2 Að1 xÞ2
d þ ¼ Adp g cos θg ½ρG α þ ð1 αÞρL Adz Phτ W idz
αρG ð1 αÞρL
þ AGðuG uL Þdx
ð1:73Þ
This equation states that the change in the momentum flux equals the pressure force, the
force of gravity, the force of the wall on the fluid, where hτ W i is the average wall shear
stress around the periphery, the change in the momentum associated with phase change.
Unlike the momentum balance for homogeneous flow, (1.51a) and (1.51b), both x and α
(or x and S) appear. Since uG and uL are given by (1.68) and (1.69), the last term in (1.73)
can be subtracted from the momentum flux term to give

dp Phτ W i duG duL
¼ þG x þ ð 1 xÞ þ g cos θg ½ρG α þ ð1 αÞρL ð1:74Þ
dz A dz dz
The third term in (1.74) represents a weighted average of the accelerations of the gas and
the liquid.
The flow of kinetic energy (KE) also has contributions from both the gas and liquid.

Flow of KE x GAx 2 ð1 xÞ GAð1 xÞ 2
¼ þ
Mass flow 2 ρG αA 2 ρL ð1 αÞA
" # ð1:75Þ
1 G 2 x3 G 2 ð 1 xÞ 3
¼ þ
2 α2 ρ2G ð1 αÞ2 ρ2L
The energy theorem gives

" # 2
dhhi 1 d G2 x3 G2 ð1 xÞ3 qW P uG u2L dx
þ þ þ g cos θ g ¼ þ ð1:76Þ
dz 2 dz α2 ρ2G ð1 αÞ2 ρ2L AG 2 2 dz
hhi ¼ ð1 xÞhL þ xhG ð1:77Þ
The last term in (1.76) is the change in kinetic energy associated with phase change.
Since uG and uL are given by (1.68) and (1.69),

dhhi 1 du2G 1 du2L q P
þ x þ ð1 x Þ þ g cos θg ¼ W ð1:78Þ
dz 2 dz 2 dz AG
Note that the kinetic energy term represents the weighted average of the kinetic energy of
the gas and the kinetic energy of the liquid.
1.4.2 Choking
The momentum equation (1.73) for a separated flow can be written as
" !#
dp d G2 x2 G2 ð1 xÞ2 Phτ W i
1þ þ ¼ þ g cos θg ½ρG α þ ð1 αÞρL ð1:79Þ
dz dp αρG ð1 αÞρL A
since d/dz = d/dp(dp/dz)−1. Again, choking is defined as the condition for which the term
multiplying dp/dz changes sign.
!
1 d x2 ð 1 xÞ 2
¼ þ ð1:80Þ
G2C dp αρG ð1 αÞρL
It is not convenient to use (1.80) to calculate GC. A direct approach is more appro-
priate. Usually, the upstream conditions are fixed. Then, (1.73) and/or (1.76) are
solved (quite often by making further approximations)

to obtain the downstream
pressure, p2, for a given G. The ratio p2 p1 is decreased until a maximum value of
G is calculated. This is GC.
Usually, the balance equations are written with S, rather than α. This is done by using
(1.70), that is,
1 S ð1 xÞ ρG
¼1þ ð1:81Þ
α x ρL
to eliminate α. The integration requires a relation between x, ρG, ρL and p. This can
be obtained by assuming equilibrium between the phases and using thermodynamic
relations. If S is assumed to be independent of z the void fraction is fixed if x is fixed.
This simplification is called a separated flow with constant slip. The separated flow
model probably gives a better approximation for the choking condition than the homo-
geneous model (S = 1) does, since it allows for an assessment of the effects of slip (which
can be quite large).
1.5 The Lockhart–Martinelli analysis
1.5.1 Evaluation of homogeneous and separated flow models

The homogeneous and separated flow assumptions are not complete in that they do
not give models for τW and α. Their chief merit is that they estimate the contributions
of acceleration and kinetic energy to the pressure change and the choking condition.
Empirical relations are needed for τW and α. The approach that has been taken is to use
measurements of the pressure gradient and the void fraction under conditions that the
flow is not changing significantly in the flow direction, so the contribution of acceleration
to the pressure gradient can be ignored.
Empirical correlations of these measurements are aided by using dimensional analysis.
An important step in this approach is to discover the relevant dimensionless groups.
A widely used correlation is due to Lockhart & Martinelli (1949). It is now known that
this approach provides only a rough estimate, but its simplicity makes it a useful tool in
obtaining a first approximation.
1.5.2 Frictional pressure gradient

The momentum balance, equation (1.5), can be written in the following form for a
passage with constant area:
dp d ðGAÞ τ W P
¼ þ þ ρg cos θg ð1:82Þ
dz dz A
This equation indicates that the pressure gradient is due to fluid acceleration, frictional
effects associated with the wall drag and the overcoming of gravity. The frictional
pressure gradient is given as

dp τW P
¼ ð1:83Þ
dz F A
or, for a circular pipe, as

dp 4τ W
¼ ð1:84Þ
dz F dt
where dt is the pipe diameter. Lockhart and Martinelli chose to develop correlations for
the frictional pressure gradient rather than for τW. If τW is represented by the friction factor
relation (1.6),

dp 2ρu2
¼ f ð1:85Þ
dz F dt
For a single-phase flow, the friction factor can be approximated (Knudsen & Katz, 1958) by
f ¼ 0:046ðReÞ0:2 ð1:86Þ
for a turbulent flow and by
f ¼ 16ðReÞ1 ð1:87Þ
for a laminar flow, where the Reynolds number is defined as

d t uρ
Re ¼ ð1:88Þ
μ
A similar result is obtained for gas–liquid flows. For example, (1.84) gives the frictional
contribution to the pressure drop as PτW/A. It is noted that the gravitational contribution
g cos θg ½ρG α þ ρL ð1 αÞ requires a knowledge of the void fraction, α.
1.5.3 Rationale for the Lockhart–Martinelli plots

A rationale for using the method of Lockhart and Martinelli is obtained by using
a separated flow model which pictures the gas and liquid as flowing in parallel
pipes of areas of αA and (1 − α)A, as shown in Figure 1.3. Liquid and gas are
flowing in a pipeline with a diameter of dt and an area of A. The mass velocity is
given as
G ¼ GL þ GG ð1:89Þ
where GL and GG are the contributions by the liquid and the gas. The fictitious parallel
pipes have areas (1 − α)A and αA and diameters of dG and dL. Thus
GG, Aα dG
GL + GG, A dt
GL, A(1– α)dL
Figure 1.3 Rationale for using the method of Lockhart & Martinelli.
d 2G d 2L
α¼ ð1 αÞ ¼ ð1:90Þ
d 2t d 2t
From (1.85) and (1.86) the frictional pressure loss for a turbulent flow scales as

dp u1:8
/ 1:2 ð1:91Þ
dz F dt
For the gas-filled pipe, the following is obtained if the flow is turbulent
1:8 1:2
dp dp uG dt
¼
dz F dz GS uGS dG
ð1:92Þ
dp 1:8 0:6
¼ α α
dz GS
where (dp/dz)GS is the pressure gradient for the gas flowing alone at a velocity uGS in a
pipe with a diameter dt, and dG is defined in Figure 1.3. For a liquid-filled pipe

dp dp
¼ ð1 αÞ2:4 ð1:93Þ
dz F dz LS
Since (dp/dz)F is presumed to be equal in the two fictitious pipes, it can be eliminated
between (1.92) and (1.93) to give
,
α 2:4 dp dp
¼ ¼ X 2tt ð1:94Þ
1α dz LS dz
GS
The term Xtt is called the flow factor for a turbulent liquid and a turbulent gas. This
suggests that the void fraction is a function of X, which can be Xtt, Xtv, Xvt, Xvv depending
on whether the flow is turbulent (Re > 1000) or laminar (Re < 1000). Thus,
0:046ρL u2L Re0:2
X 2tt ¼ L
ð1:95Þ
0:046ρG u2G Re0:2
G
16Re1
L ρL uL
2
X 2vt ¼ 0:2
ð1:96Þ
0:46ReG ρG u2G
0:046ρL u2L Re0:2

X 2tv ¼ L
ð1:97Þ
16Re1 ρ
G G G u 2
16ρL u2L Re1

X 2vv ¼ L
ð1:98Þ
16ρG u2G Re1
G
For example, for a turbulent liquid and a turbulent gas

0:1 0:5
μL ρG W L 0:9
X tt ¼ ð1:99Þ
μG ρL WG
where WL and WG are the weight rates of flow for the liquid and the gas. This model, thus,
suggests that measurements of the void fraction, α, should be correlated as a function of
X, which depends on whether the phases are turbulent or laminar.
A dimensionless frictional pressure gradient can be defined as
ðdpF =dzÞ
ϕ2G ¼ ð1:100Þ
ðdp=dzÞGS
or as
ðdpF =dzÞ
ϕ2L ¼ ð1:101Þ
ðdp=dzÞLS
From (1.92) and (1.93), ϕG and ϕL are functions of α. Thus, the above equations suggest
that measured frictional pressure losses can be correlated by plotting ϕ2G or ϕ2L as a
function of X.
1.5.4 Lockhart–Martinelli correlation

Lockhart & Martinelli (1949) have exploited the above correlations in a landmark paper.
The application of this correlation has been thoroughly discussed by Hewitt (1978) and
by Wallis (1969). A simple fit to their correlations by Chisholm (1967) is as follows:
C 1
ϕ2L ¼ 1 þ þ ð1:102Þ
X X2
ϕ2G ¼ 1 þ CX þ X 2 ð1:103Þ
where C = 20 for Xtt, C = 12 for Xvt, C = 10 for Xtv, C = 5 for Xvv. Flow factor Xvt refers to a
viscous liquid and a turbulent gas. Note that X is zero and ϕ2G is unity for a liquid flow that
is zero. Both X and ϕG increase with increasing liquid flow. Note also that for zero gas
flow X = ∞. Thus ϕL is unity under this condition. With increasing gas flow, X decreases
and ϕL increases.
Wallis (1969) gives the following representation of the plot for void fraction developed
by Lockhart & Martinelli (1949):
0:375
α ¼ 1 þ X 0:8 ð1:104Þ
where X can be defined for any of the four regimes.

1.6 Steady flow in a nozzle 19
The data represented by (1.102), (1.103) and (1.104) are mainly for gas–liquid flows.
Martinelli & Nelson (1948) and Thom (1964) developed a modified approach for steam/
water. Detailed discussions of these works are presented by Hewitt (1978) and by Wallis
(1969).
1.6 Steady flow in a nozzle
1.6.1 Preliminary comments

Another situation where the one-dimensional analysis can be used is steady flow in a
nozzle. This differs from steady flow in a pipe in that the cross-sectional area, A, is
varying in the flow direction. Usually the changes in velocity are large enough that drag
on the wall and gravitational effects can be ignored compared to inertial changes. The
analysis presented here makes the simplifying assumption that the flow is uniform over a
cross-section.
1.6.2 Single-phase flow

A differential control volume in a contracting flow is depicted in Figure 1.4.
Conservation of mass gives
ρuA ¼ constant ð1:105Þ
which can be written as

du dρ dA
þ þ ¼0 ð1:106Þ
u ρ A
The momentum balance equation is obtained by applying the momentum theorem to the
differential volume in Figure 1.4. The flow of momentum into and out of the volume are
z θg
p
dp dA
p+ , A+
2 2
dU
U+
2
w
Flo
p
dp
p–
2
dA g
A–
2
dU
U–
2
Figure 1.4 Differential control volume in a contracting flow.
ρuAu and ρuA(u+du) so the net flow of momentum out is ρuAdu. The only force acting
on the volume is due to pressure since gravity and the resisting force at the wall are
neglected. The pressures on the front and back faces are pA and −(p + dp)(A + dA). In
addition there is a force acting on the slant face whose magnitude is pjdAj, where jdAj
is the magnitude of the projection of the area of the slant face on a plane perpendicular
to the flow direction. For the volume in Figure 1.4, the pressure force on the slant face
is negative, since dA is negative for a converging section. Thus, the force on the slant
face is p dA.
When all of the pressure forces are added, the net force is found to be −Adp if second-
order terms in the differential quantities are ignored. Thus, not surprisingly, the momen-
tum balance gives the Bernoulli equation
dp
udu þ ¼0 ð1:107Þ
ρ
This can be rearranged to

2
dρ du u
¼ ð1:108Þ
ρ u dp=dρ
If dρ/ρ is eliminated by using mass balance equation (1.106)
dA du
¼ 1 Ma2 ð1:109Þ
A u
where
u2
Ma2 ¼ ð1:110Þ
c2
and c2 is given as
dp
c2 ¼ ð1:111Þ
dρ
If Ma < 1, du/u has the opposite sign of dA/A, that is, the velocity increases in a
converging section. For Ma > 1 the velocity increases in an expanding section. Thus,
it is impossible to go from a subsonic (Ma < 1) to a supersonic flow in a converging
section.
If a nozzle is located between two reservoirs where the pressures are p1 and p2, the flow
will initially increase as p2/p1 decreases. However, at low enough p2/p1 the flow will
choke and the velocity at the throat will be sonic. Further decreases in p2/p1 will result in a
supersonic expansion and the creation of shock waves in the receiver. If (1.106) is
combined with (1.107) one obtains
2
dp G =2ρA2 ðdA=dzÞ
¼ ð1:112Þ
dz 1 G2 ðdv=dpÞ
where v is the specific volume. This gives a choking condition at the throat dA=dz ¼ 0
when the mass velocity at the throat is
1
dp dv
GC ¼
2
¼ ð1:113Þ
dv dp
The mass flow under this condition is the product of GC and the area of the throat.
The energy balance gives
2
u q
dh þ d ¼ W ð1:114Þ
2 W
where W is the weight rate of flow. If ðu2 =2Þ is eliminated between (1.107) and (1.114)
dp qW
dh ¼ ð1:115Þ
ρ W
For qW = 0 the flow is adiabatic and reversible; that is, it has a constant entropy. Thus
dp
dh ¼ ð1:116Þ
ρ
and

∂p
c2 ¼ ð1:117Þ
∂ρ S
It is noted that, for an adiabatic situation, changes of the flow in a converging

section are defined by (1.107) and (1.114) with qW = 0. The flow can be calculated
either with the energy or the momentum balance equation and the condition of
constant entropy.
For a perfect gas, (1.116) can be written as
dp
C P dT ¼ ð1:118Þ
ρ
By using (1.19) and (1.24) to eliminate T from (1.118),

γ p dp
d ¼ ð1:119Þ
γ1 ρ ρ
This can be written as

γ dp γ 1 dp
þ pd ¼ ð1:120Þ
ðγ 1Þ ρ ðγ 1 Þ ρ ρ
Multiplying by ðγ 1Þ and rearranging

dp 1
¼ γpd ð1:121Þ
ρ ρ
Integration gives
p
¼ constant ð1:122Þ
ργ
or
γ
p ρ
¼ ð1:123Þ
p1 ρ1
This can be substituted into the momentum equation (1.107) to eliminate either p or ρ.
Note that the relationship between p and ρ for a straight pipe, equation (1.31), differs from
(1.123) because it is for an irreversible flow that includes effects of wall friction.
1.6.3 Two-phase flow in a nozzle

Two-phase flow in a nozzle is pertinent to a number of practical problems. For example,
the calculation of choking is of particular interest in considering the performances of
relief systems and of rocket nozzles.
The influence of wall drag and gravity are ignored so the application of the momentum
theorem to the differential volume, shown in Figure 1.4, gives
" #
dp d G2 Ax2 G2 Að1 xÞ2
A ¼ þ ð1:124Þ
dz dz αρG ð1 αÞρL
where from conservation of mass

GA ¼ constant ð1:125Þ
Conservation of energy is given by

" #
dhhi 1 d G2 x3 ð1 xÞ3 G2
þ þ ¼0 ð1:126Þ
dz 2 dz α2 ρ2G ð1 αÞ2 ρ2L
with hhi defined by (1.77). The velocities of the gas and liquid streams are given by (1.68)
and (1.69). Equation (1.81) is used to eliminate α from (1.124) and (1.126) so that the
terms in the brackets then contain the slip, S, and x rather than α and x.
By combining (1.125) and (1.124) one obtains
( " #) " #
2 2
dp d x 2
ð 1 x Þ dA x 2
ð 1 x Þ
1 þ G2 þ ¼ G2 þ ð1:127Þ
dz dp αρG ð1 αÞρL dz αρG ð1 αÞρL
Note that (1.127) predicts a change in the sign of dp/dz in a converging section (dA/dz has
negative value) when the term multiplying dp/dz changes sign. This is an impossibility so
the flow chokes at the throat (dA=dz ¼ 0) where G = GC, given by (1.80). This indicates
that the choking for flow in a converging nozzle is given by the same condition as for flow
in a long straight pipe.
As already discussed in Section 1.4.2, the calculation of GC requires the solution of

(1.124) and/or (1.127), with α eliminated using (1.81). These integrations are used to
determine G as a function of the ratio of the pressures at the outlet and the inlet.
Equilibrium is usually assumed to exist between the phases so that thermodynamic
correlations can be used to calculate x. The slip ratio varies with z but, as an approx-
imation, it is usually assumed to have a constant value. A maximum in G indicates that
choking is occurring.
An extensive review of various approaches to the calculation of GC has been given by
Henry (1981), including the separated flow model that uses the assumptions of a constant
slip ratio and equilibrium between the phases. From data for air–water he suggests that
the use of S = 3.2 provides a good approximation.
An interesting solution has been presented by Moody (1965) for a fluid which is
undergoing phase changes as it flows through a nozzle. At the inlet the fluid is a vapor–
liquid mixture or a saturated vapor. The velocity is small enough for stagnation con-
ditions (po, ρo, ho, so, uo = 0) to exist in the entering fluid. The two terms in the brackets in
(1.126) are u2G x and u2L ð1 xÞ. The integration of (1.126) from the inlet to the outlet,
designated by the subscript 2, gives
1 G22 x32 1 G22 ð1 x2 Þ3
hh2 i hho i þ þ ¼0 ð1:128Þ
2 α22 ρ2G2 2 ð1 α2 Þ2 ρ2L2
where α is eliminated by using an equation presented earlier

1 S ð1 xÞ ρG
¼1þ ð1:129Þ
α x ρL
Moody assumed the flow to be isentropic, so that hs2 i ¼ hso i. Thus, for a given p2,
thermodynamic relations give hh2 i; x2 ; ρG2 ; ρL2 . If po and ρo are known, hho i and hso i can
be obtained from thermodynamic relations. Equations (1.128) and (1.129) give hh2 i in
terms of G2, the slip ratio, S, and downstream conditions. Thus, if S is fixed, G2 can be
calculated for different values of p2/po. The value of p2/po at which G2 is a maximum
represents the choking condition.
This critical mass velocity depends on the choice of S, so a homogeneous model (S = 1)
predicts lower GC than does a separated flow model. Moody (1965) suggested that the
maximum possible GC, calculated from the separated flow model, is given by
1=3
ρL
S¼ ð1:130Þ
ρG
He made calculations for the steam/water system using (1.130).

However, there is no physical reason to expect that this is the right value. The actual
S which characterizes the evolution of a given system to a choking condition must
be established either empirically or by a hydrodynamic model which depends on
an assumed distribution of the phases. Clearly, there is a need for more research in
this area.
1.6.4 One-dimensional two-fluid model

The assumption of equilibrium between the phases used in previous sections is often
inaccurate. This is especially a concern in rapidly varying flows such as exist in choking
situations. The application of a two-fluid model to represent such systems is presented by
Boure (1978). Only a brief discussion is presented here.
This model represents the flow as consisting of interpenetrating gas and liquid streams.
Separate equations are developed for the gas and the liquid. The velocity then emerges in
a natural way so correlations for slip are not introduced. Thermodynamic relations are
used for the separate phases but equilibrium between phases is not assumed. Thus, x is
not obtained from thermodynamics and the temperatures of the two phases need not be
equal. The equations presented here are for a simplified version of the model in that a
one-dimensional flow is considered; that is, the flow at different locations in the cross-
sections are assumed to be the same.
The equations of conservation of mass for steady flow in a duct with constant area, A,
can be written as
d
ðAαρG uG Þ M LG Aav ¼ 0 ð1:131Þ
dz
d
½Að1 αÞρL uL M GL Aav ¼ 0 ð1:132Þ
dz
where MGL is the mass transfer rate per unit area from the gas to the liquid, MLG is
the mass transfer rate from the liquid to the gas and av is the interfacial area per unit
volume. Thus
M LG ¼ M GL ð1:133Þ
and
dx
M LG ¼ GA ð1:134Þ
dz
By adding (1.131) and (1.132), one gets

dG
¼ 0 or G ¼ constant ð1:135Þ
dz
The momentum balance equations for the gas and for the liquid for the case of constant A
and G are
d dp Pτ WG
ρG u2G α ¼ M LG av uG α ρG α cos θg þ av τ LG ð1:136Þ
dz dz A
d 2 dp Pτ WL
ρL uL ð1 αÞ ¼ M GL av ul ð1 αÞ ρL gð1 αÞ cos θg av τ LG
dz dz A
ð1:137Þ
The liquid can exert a force on the gas at the interface. The component of this force in the
flow direction is represented by τLGavAdz where τLG is an average effective stress and
avAdz is the interfacial area in the control volume. Similarly, the force of the gas on the
liquid is given by τGLavAdz. The other stresses in (1.136) and (1.137), τWG and τWL,
represent the resisting stresses at the wall on the gas and on the liquid.
If evaporation is occurring the term MLG is positive so MLGavuG and MGLavuLrepresent
increases and decreases of momentum associated with the phase change. The sum of
these two terms is the net increase of the momentum of the flow system already discussed
in Section 1.4.1. The sum of (1.136) and (1.137) gives (1.74).
Conservation of energy for the gas and for the liquid gives

d u2 d u2 P
uG αρG G þ uG αρG hG ¼ M GL av G þ qLG av ρG uG αg cos θg þ qWG
dz 2 dz 2 A
ð1:138Þ

d u2L
d u 2
uL ð1 αÞρL þ uL 1 αÞρL hL ¼ M GL av L þ qGL av p
dz 2 dz 2
ρL uL ð1 αÞg cos θg þ qWL av
ð1:139Þ
where hG and hL are the enthalpies of the gas and the liquid, qLG and qGL are the heat
transfer rates per unit area from the liquid to the gas or from the gas to the liquid, qWG and
qWL are the heat transfer rates per unit area from the wall to the gas or from the wall to the
liquid. The first terms on the right side of (1.138) and (1.139) represent changes of the
kinetic energy of the gas or the liquid. The sum of (1.138) and (1.139) gives (1.78).
The void fraction, α, can be calculated from conservation of mass if uG and uL are
known and if G is fixed
GA ¼ uL ð1 αÞA þ uG αA ð1:140Þ
The two-fluid model thus avoids the need to specify a slip between the phases and the
void fraction because the velocities of the two phases are calculated. Also, equilibrium
between the phases need not be assumed.
However, new variables, which need to be evaluated, are introduced. These include
the rate of mass transfer, MLG, the rate of heat transfer, qLG, between the phases and the
interfacial drag, τLG. Also, a theory needs to be developed to predict the interfacial area
per unit volume, av. This quantity introduces, in a direct way, new concepts into the one-
dimensional balances, in that it requires a knowledge of the structure of the phases. For
example, it might require a specification of a bubble size distribution.
Drew & Flaherty (1994) and others have shown that solutions of the one-dimensional
two-fluid equations are unstable for the simple case of an adiabatic flow with no phase
change. (This is not the case for the homogeneous or separated flow models.) This suggests
that more attention needs to be given to the formulation of the two-fluid model and the
method of solution. The use of finite difference methods to solve the differential equations
introduces numerical viscosity which can artificially stabilize the equations.
Three-dimensional two-fluid equations have been successfully used to study adiabatic
bubbly flows (Lance & Lopez de Bertodano, 1994) and solid–fluid flows.
References
Boure, J. A. 1978 Constitutive equations for two-phase flows. In Two-phase Flows and Heat
Transfer with Applications to Nuclear Reactor Design Problems, ed. J. J. Ginoux,
Washington, DC: Hemisphere, pp. 169–175.
Chisholm, D. 1967 A theoretical basis for the Lockhart–Martinelli correlation for two-phase flow.
Nat. Eng. Lab., UK, Report no. 310 (also Int. J. Heat Mass Transfer 10, 1767–1778).
Drew, D. A. & Flaherty, J. E. 1994 Analysis of multiphase flow. Multiphase Science and
Technology 8, 207–255.
Fan, Z. & Hanratty, T. J. 1993 Pressure profiles for slugs in horizontal pipelines. Int. J. Multiphase
Flow 19, 421–437.
Henry, R. E. 1981 Calculational techniques for two-phase critical flow. In Two-Phase Fluid
Dynamics (Papers presented at the Japan–US Seminar, held July 31–August 3, 1979, Kansei,
Kobe, Japan), eds. A. E. Bergles & S. Ishigai, Washington, DC: Hemisphere, pp. 415–438.
Henstock, W. H. & Hanratty, T. J. 1976 The interfacial drag and the height of the wall layer in
annular flows. AIChE Jl 22, 990–1000.
Hewitt, G. F. 1978 Simple momentum and energy balances and their related empirical correlations.
In Two–phase Flows and Heat Transfer with Applications to Nuclear Reactor Design Problems,
ed. J. J. Ginoux, Washington, DC: Hemisphere.
Knudsen, J. G. & Katz, D. L. 1958 Fluid Dynamics and Heat Transfer, New York: McGraw-Hill,
pp. 89, 173.
Lance, M. & Lopez de Bertodano, M. 1994 Phase distribution phenomena wall effects in bubbly
two-phase flows. Multiphase Science and Technology 8, 207–255.
Lockhart, R. W. & Martinelli, R. C. 1949 Proposed correlation of data for isothermal two-phase,
two-component flow in pipes. Chem. Eng. Progr. 45, 39–48.
Martinelli, R. C. & Nelson, D.B. 1948 Prediction of pressure drop during forced-circulation
boiling of water. Trans. ASME 70, 695–702.
Moody, F. J. 1965 Maximum flow rate of a single component, two-phase mixture. J. Heat Transfer
87, 134–142.
Thom, J. R. S. 1964 Prediction of pressure drop during forced circulation of boiling water. Int.
J. Heat Mass Transfer 7, 709–724.
Wallis, G. B. 1969 One-dimensional Two-phase Flow. New York: McGraw-Hill.
2 Flow regimes
2.1 Need for a phenomenological understanding
The one-dimensional analysis and the correlations for frictional pressure drop and void
fraction (presented in Chapter 1) have been widely used as a starting point for engineer-
ing designs. However, these correlations have the handicap that the structure of the phase
boundaries is ignored. As a consequence, they often give results which are only a rough
approximation and overlook phenomena which could be of first-order importance in
understanding the behavior of a system.
It is now recognized that the central issue in developing a scientific approach to gas–
liquid flows is the understanding of how the phases are distributed and of how the
behavior of a multiphase system is related to this structure (Hanratty et al., 2003). Of
particular interest is the finding that macroscopic behavior is dependent on small-scale
interactions. An example of this dependence is that the presence of small amounts of high
molecular weight polymers can change an annular flow into a stratified flow by damping
interfacial waves (Al-Sarkhi & Hanratty, 2001a).
A goal of this book is to develop an understanding of the basic scientific tools needed to
describe gas–liquid flows. This chapter opens this discourse by describing flow patterns that
have been defined. Detailed discussions of the theory will be presented in later chapters.
Four systems are considered: flow in horizontal pipes, flow in vertical pipes, micro-
gravity flows, flow in capillaries and microchannels. Discussions of the flow regimes in
horizontal pipes and of the mechanisms for transitions from one regime to another
provide the motivation for a large portion of this book.
The Kelvin–Helmholtz (inviscid) instability of a stratified flow plays an important role
in understanding gas–liquid flows. Therefore, it is discussed in detail in Chapter 4. The
results of this theory are previewed in this chapter since they arise in much of the
discussion of the behavior of gas–liquid flows. The theory is also of historic interest
because of its early use to explain the effect of pipe diameter on the transition from a
stratified flow to an intermittent (slug or plug) flow.
Chapter 5 shows that the orientation of a pipe can have a strong effect on the flow
pattern because the role of the component of the gravitational force in the direction
of flow cannot be ignored. The present chapter discusses only the patterns observed in
horizontal flows and in vertical flows.
Changes in fluid properties can influence the flow configuration. The sections in this
chapter on the effects of liquid viscosity and gas density address this issue. Increases in
28 Flow regimes
2.0
2.54 cm pipe, L/dt = 600
1.0 9.53 cm pipe, L/dt = 260
Slug/plug
Pseudo-slug
ULS (m/s)
0.10
Annular
Stratified
0.01
Onset of
atomization
0.003
1.0 10.0 100 300
UGS (m/s)
Figure 2.1 Flow regime map for air and water flowing in horizontal 2.54 cm and 9.53 cm pipes. Lin &
Hanratty, 1987b.
gas density cause increases in the kinetic energy of the gas flow and of energy transfer
from the gas flow to interfacial waves. The generation of waves is easier. Thus, at large
enough densities, the appearance of slugs depends on their stability, rather than the
stability of stratified flows.
2.2 Flow regimes in horizontal pipes
2.2.1 Flow regimes for air and a low-viscosity liquid

Figure 2.1 presents a Mandhane (Mandhane et al., 1974) plot for air and water flowing in
horizontal pipes with diameters of 2.54 cm and 9.53 cm and lengths of 15.5 m and 24.6 m
under atmospheric conditions (Lin & Hanratty, 1987b). These observations were made at
the downstream end of pipes, where the flow is approximately fully developed. The
abscissa and ordinate are the superficial gas and liquid velocities, defined as the volu-
metric flows of the gas and liquid divided by the area of the pipe. A detailed consideration
of this figure provides an introduction to the physical challenges that are presented in
relating the behavior of a gas–liquid flow to interfacial structure.
2.2.2 Wavy stratified flow in air–water systems

Consider, first, the results for low gas and liquid velocities in a 9.53 cm pipe, for which
the flow is stratified. In this pattern, all of the liquid moves along the bottom of the pipe.
At very low gas and liquid velocities the interface is smooth. However, at larger gas
velocities waves appear.
Three types of waves have been identified: Jeffreys waves, Kelvin–Helmholtz
waves and roll waves. The first interfacial disturbances observed in air–water flow
with increasing gas velocity are Jeffreys waves (Jeffreys, 1925) for which a balance
exists between the energy fed to the waves by the air flow and dissipation in the liquid
(Hanratty & Engen, 1957; Cohen & Hanratty, 1965; Andritsos & Hanratty, 1987).
This transition is observed to occur at uGS = 4 m/s to 0.5 m/s as uLS varies from
0.001 m/s to 0.1 m/s. Note that the area occupied by the gas decreases with increasing
uLS. Thus, the actual critical air velocity at transition is roughly equal to 4 m/s at all
water velocities.
The presence of waves induces pressure variations along the interface which, for very
small amplitude sinusoidal waves, can be represented as the sum of a component which is
a maximum where the wave slope is a maximum, pGI, and a component, pGR, which is a
minimum (a suction) where the wave height is a maximum. Energy is fed to Jeffreys
waves through pGR.
The magnitude of the induced gas phase pressure variations at the interface increases
with increasing gas velocity. Irregular large-amplitude waves appear at large gas veloc-
ities for which the destabilizing effect of pGR becomes large compared with the stabiliz-
ing effects of gravity and surface tension (Andritsos & Hanratty, 1987; Andreussi &
Persen, 1987). For air–water flow, they are observed at uGS ≈ 14 m/s to 5 m/s as uLS varies
from 0.002 m/s to 0.1 m/s. These disturbances are called Kelvin–Helmholtz waves since
they are initiated by a Kelvin–Helmholtz instability. At uGS approximately twice that
needed to initiate irregular waves, drops are observed to be ripped from the crests of
waves (see Figure 2.1).
Roll waves (Hanratty & Engen, 1957; Hanratty & Hershman, 1961) appear over a
range of conditions. These are large-wavelength disturbances with steep fronts and
gradually sloping backs. They may be considered as surges of the flow in the liquid
layer. The surface of the surges can be smooth (Figure 2.2b) or wavy (Figure 2.2d),
depending on the interfacial condition of the stratified flow on which these waves are
imposed.
2.2.3 Annular flow

The drops created by atomization of irregular waves can deposit at the top of the
pipe. If the rate of deposition is large enough a turbulent film forms. This pattern is
called an annular flow (Lin & Hanratty, 1987b), whereby part of the liquid flows
along the wall and part flows as drops entrained in the gas (see Figure 2.2a and c).
Waves aid in bringing liquid up the pipe wall so that the gas velocity needed to form an
annular flow decreases with increasing uLS. For air–water flow in a 9.53 cm pipe, an
annular pattern is initiated at gas velocities of 30–50 m/s (see Figure 2.1). For very
low uLS, atomization does not occur in the air–water system, so an annular flow is not
observed.
30 Flow regimes
Flow
dt dt
μL = 70 cP (a) (b)
dt = 9.53 cm
dt dt
μL = 1 cP (c) (d)
Annular flow Large–amplitude waves ("roll")
dt = 2.52 cm
Figure 2.2 Annular flows and large roll waves for 1 cP and 70 cP liquids. Andritsos & Hanratty, 1987.
2.2.4 Plug and slug flow

At large enough uLS, waves that touch the top of the pipe can form a liquid bridge. At uSG
< ~ 0.6 m/s and large uLS a plug flow occurs whereby these bridges are part of a train of
elongated bubbles which move along the top of the pipe (Ruder & Hanratty, 1990). At larger
gas velocities, the bridges form liquid slugs, sketched in Figure 9.1. These aerated blocks
of liquid move over a slowly flowing stratified layer at approximately the same velocity as
the gas. Slugs can be quite long, so they can have large momenta and can cause vibrations
when they impact on surfaces. Therefore, it is usually desirable to design a piping system so
as to avoid slugging. Pictures of the front and back of a slug are given in Figure 2.3b, c.
The transition from a slug to a plug flow has been defined by Ruder et al. (1989) as the
condition for which a sudden increase in liquid height (a hydraulic jump) in front of a
slug cannot exist. The jump moves with a velocity cF. An equation for this condition
(see Figure 9.1) is derived by arguing that dissipation of mechanical energy accompanies
the sudden increase in height. (See Chapter 9 for a derivation.) For a rectangular channel,
this concept yields
ðcF uL1 Þ2
>1 ð2:1Þ
gB
where hL1 and uL1 are the height and velocity of the liquid in front of the slug and B is the
channel height. The term (cF − uL1) is the velocity of the slug relative to the velocity of the
liquid layer in front of the slug. For a pipe with diameter dt, the critical condition
is predicted to be
ðcF uL1 Þ2
> 0:95 1:21 ð2:2Þ
gd t
(a)
(b)
(c)
(d)
Figure 2.3 (a) Regular waves observed 180 diameters from the entry of a 9.53 cm pipe for uLS = 0.06 m/s,
uGS = 4.3 m/s, uL = 20 cP. (b) Front of a slug that was formed 100 diameters from the entry of a
9.53 cm pipe and observed at 180 diameters. (c) Tail of the slug in (b). (d) A slug that has just
formed at 30 diameters from the entry of a 9.53 cm pipe for uL = 0.033 m/s, uGS = 7 m/s, uL = 100
cP. Andritsos et al., 1989.
where the term on the right depends on the value of hL1/dt. (See Ruder et al., 1989, Table
1.) Ruder & Hanratty (1990) carried out visual studies for air and water flowing
in a 9.53 cm pipe and defined the transition from plug to slug flow as occurring when
ðcF uL1 Þ2
> 1:2 ð2:3Þ
gd t
Hurlburt & Hanratty (2002) showed how this condition can be used to predict the
critical uGS.
Barnea et al. (1980) have differentiated between plug and slug flow by considering
whether the liquid blockage contains gas bubbles. This appears to be equivalent to the
condition cited above for air and water flowing in horizontal pipes since the hydraulic
jump in the front of the slug occludes gas at high gas velocities.
2.2.5 Stability of a slug

An understanding of the transition to intermittent (plug or slug) flow requires a consid-
eration of the stability of a slug, as well as the stability of a stratified flow. Slugs moving
along a stratified layer pick up liquid at their front at a rate given by
32 Flow regimes
Rate of pickup ¼ AL1 ðcF uL1 Þ ð2:4Þ
where AL1 is the area of the stratified liquid flow in front of the slug, uL1 is the velocity in
this layer and cF is the velocity of the front of the slug. The volumetric rate at which slugs
shed liquid at their back is specified as Qsh, so slugs grow if
AL1 ðcF uL1 Þ > Qsh ð2:5Þ
Define AL0 and hL0 as the area and height of the stratified flow under conditions for
which the pickup rate equals the shedding rate. When hL1 > hL0 a slug will grow; when
hL1 < hL0, it will decay. Thus, (2.5) defines the critical height of a stratified flow, below
which it is impossible to sustain stable or growing slugs.
A detailed derivation and a discussion about the implementation of the above ideas are
given in Section 9.2 of the chapter on slug flow.
2.2.6 Transition to slug/plug flow

For air–water flow under atmospheric conditions in a horizontal 9.53 cm pipe, slugs
are initiated at low gas velocities, uGS < ~ 5 m/s, by regular waves that reach the top wall
of the pipe. Jeffreys waves with a wavelength of approximately 8.5 cm are generated in
the early part of the pipeline. These grow in height and increase in velocity to a point
where they can resonate with very small amplitude waves having a length of 17 cm.
Energy is fed from the 8.5 cm wave to the 17 cm wave. These larger wavelength waves
can grow and tumble or, if the liquid flow is large enough, strike the top wall to form a
slug. (See Fan et al., 1993 and Figure 2.4.)
L/dt = 10
(a)
L/dt = 120
(b)
L/dt = 260
(c)
Figure 2.4 Development of waves on water, along a 9.53 cm pipe. Fan et al., 1993.
Flow direction
UGS = 3.7 m/s ULS = 0.14 m/s

Side view
Figure 2.5 Formation of slugs by coalescing of roll waves in a 9.53 cm pipe. Lin & Hanratty, 1987b.
For large superficial gas velocities, irregular Kelvin–Helmholtz waves appear. These
can coalesce to form a high enough wave that it touches the top wall (see Figure 2.5).
The height of the stratified layer decreases with increasing gas velocity so it can become
too small to support a stable slug (Woods & Hanratty, 1996). Then, a consideration of the
stability of slugs provides a prediction of the possibility of forming a slug flow.
It is possible for slugs to exist at sub-critical conditions (as defined by the stability of a
stratified flow) if disturbances are introduced to the system and conditions are such that
slugs are stable. (Terrain slugging is an example.)
2.2.7 Pseudo-slugs
At uGS > 9 m/s a pseudo-slug region can exist in a 9.53 cm pipe carrying air and water
(see Figure 2.1). The lower boundary of this region defines locations where large
amplitude waves are initiated. These waves can coalesce and touch the top of the pipe
to form a possible slug. The higher boundary of the pseudo-slug region is defined by the
stability of a slug. Visual observations could, mistakenly, label these large-amplitude
waves as slugs. However, they do not have the basic properties of slugs. That is, they do
not have velocities which are approximately equal to the gas velocity and they are not
coherent over a long length of pipe. Instrumental methods have been developed by Lin &
Hanratty (1987a) to differentiate pseudo-slugs from slugs.
2.2.8 Bubble/foam regime

At very large liquid velocities and low gas velocities a bubble/foam regime exists. The
transition (not shown in Figure 2.1) from a bubbly flow to a flow with elongated bubbles
(plug flow) has been defined as occurring at uLS = 4–5 m/s (Mandhane et al., 1974). In
34 Flow regimes
this bubble region, the gas phase exists as bubbles which have Sauter mean diameters
of 2–5 mm in a 5.03 cm pipe. Because of gravitational effects, they tend to move to the
top of the pipe. A maximum in the void fraction occurs at about one bubble radius from
the wall (Kocamustafaogullari & Wang, 1991). The distribution of void fraction is related
to a balance of the gravitational force on the bubbles and forces due to liquid phase
turbulence. When the void fraction at the peak reaches 0.60–0.65, coalescence ensues
and a transition to intermittent flow occurs. Taitel & Dukler (1976) suggested that the
transition occurs because a decrease in liquid flow reduces the effectiveness of turbulence
in suspending the bubbles. There appears to be a hysteresis effect in that transition from
a plug/slug flow to a bubbly flow occurs at a slightly lower superficial liquid velocity
( uLS = ~ 2–4 m/s) than transition from a bubbly flow to a plug/slug flow.
2.2.9 Flow at large gas and liquid velocities

The mechanism for the transition from a pseudo-slug or a slug regime to an annular
regime is not well defined. At low gas velocities, slugs occlude gas at their front. The gas
escapes at their back. The suggestion has been made that, at large gas velocities, gas is
occluded at the rear and that the slug is eventually destroyed by blow-through. (However,
this mechanism has not been substantiated.)
The flow pattern that exists at high gas and liquid velocities, also, is not understood.
As one increases the liquid velocity in the annular regime, there is evidence for the
appearance of large structures, formed by drop coalescence, which have many of the
characteristics of slugs. These could be considered as slugs with a continuous gas phase
(Dukler & Hubbard, 1975).
2.2.10 Effect of pipe diameter on flow regimes

The chief differences to be noted in comparing the mappings for air–water flow in a
2.54 cm pipe and in a 9.53 cm pipe are the strong effect of pipe size on the critical liquid
velocity needed to generate slugs at low gas velocities, the increase in the size of the
pseudo-slug region with a decrease in pipe diameter and the need for larger gas velocities
to generate annular flow in large diameter pipes.
Some of these effects can be explained if the transition is predicted by an instability
of stratified flow to a disturbance whose wavelength is large compared with the heights
of the gas space and the liquid layer. The presence of these waves causes contractions
and expansions of the gas space. These are accompanied by changes in the area of the
gas space and, through the Bernoulli effect, by decreases in the pressure at the crest.
This destabilizing effect, which is counterbalanced by the effect of gravity, increases
with decreasing pipe size since the size of the gas space decreases. Thus, the initiation of
roll waves occurs at smaller liquid flows in a smaller pipe.
Coalescence of waves can lead to the formation of a larger-amplitude wave that
temporarily touches the top wall (a pseudo-slug), or to a slug flow. The ability of
coalescing waves to reach the top wall is enhanced in smaller-diameter pipes, so the
pseudo-slug region is larger. The transitions to slug flow for 2.54 cm and 9.53 cm pipes at
2.3 Kelvin–Helmholtz waves / viscous large-wavelength instability 35
large liquid velocities appears to be associated with the stability of slugs (equation (2.5)),
so they are observed at approximately the same uLS in the two pipes. Detailed discussions
of this explanation have been given by Soleimani & Hanratty (2003) and by Hurlbert &
Hanratty (2002).
As mentioned in the previous section, both deposition of drops and the transport of
liquid around the circumference of the pipe by liquid phase turbulence associated with
disturbance waves (shown in Figure 2.2c) are responsible for the initiation of annular
flow. The ability of waves to wet the top of the pipe increases with decreasing pipe
diameter so the transition to annular flow can be observed at smaller uGS in a 2.54 cm pipe
than in a 9.53 cm pipe.
The initiation of annular flow by deposition requires that the gas velocity is large
enough for drops to form by atomizing the liquid layer. However, it also requires
that the gas phase turbulence is sufficient to carry drops from the bottom of the
pipe, where a thick liquid layer exists, to the top of the pipe. Drop mixing might
become a more important factor in pipes with very large diameters. Thus, from a
consideration of the data reported by Wu et al. (1987) for a natural gas/condensate
pipeline with a diameter of 20.3 cm, Baik & Hanratty (2003a) point out that larger
uGS (than would be expected from experiments summarized in Figure 2.1) are needed
to facilitate a transition to annular flow. This topic is covered in greater detail in
Chapter 12.
2.3 Large-wavelength Kelvin–Helmholtz waves / viscous

large-wavelength instability
Since theoretical considerations of the initiation of slugs/plugs at low gas velocities has
centered on the stability of a stratified flow to large-wavelength disturbances, a brief
historical account is appropriate. More detailed discussions of the theory are given in
later chapters and in the paper by Hurlburt & Hanratty (2002).
Kordyban & Ranov (1970) and Wallis & Dobson (1973) explored an inviscid stability
mechanism whereby the slugs arise from the growth of infinitesimal disturbances at the
interface of a stratified flow. Since the spacing between slugs is large it was natural to
restrict the analysis to wavelengths that are large compared with the height of the gas
space. If viscous effects are neglected and the waves are considered to have small
amplitudes, the following relation between the wave velocity and wave number (which
is derived in Chapter 4, equation (4.73)) is obtained for a rectangular channel of height
B inclined at an angle θ to the horizontal
kρL ðuL cÞ2 coth khL þ kρG ðuG cÞ2 coth khG ¼ g cos θðρL ρG Þ þ σk 2 ð2:6Þ
In this equation, k is the wave number, σ is the surface tension, uG is the gas velocity, uL is
the liquid velocity, hG is the height of the gas space, hL is the height of the liquid space, ρG
is the gas density, ρL is the liquid density and g is the acceleration of gravity. The wave
number is defined as k = 2 π/λ, where λ is the wavelength.
36 Flow regimes
If (2.6) is solved for the wave velocity, it is found, for large enough uG, that c is complex,
thus indicating an instability. The critical condition at which this occurs is defined by the
following equations:

2 hG tanhðkhG Þ ρG tanhðkhL Þ
ð uG u L Þ ¼ g cos θðρL ρG Þ þ k σ 2
þ ð2:7Þ
ρG khG ρL khL
uG ρ G hL þ uL ρ L hG
c¼ ð2:8Þ
ρL hG þ ρG hL
For long waves (khL <<1, khG <<1), surface tension effects can be ignored. Thus, stability
conditions (2.7) can be written as
hG
ðuG uL Þ2 ¼ ðgρL cos θÞ ð2:9Þ
ρG
Taitel & Dukler (1976) reformulated (2.9) and (2.8) as follows for a pipe with diameter dt,
area At and ρGhL/ρLhG small:
AG
ρG ðuG cÞ2 > ρL g cos θ ð2:10Þ
Si
c ¼ uL ð2:11Þ
where AG is the area occupied by the gas and Si is the length of the interface; that is,
2 2
Si hL
¼1 2 1 ð2:12Þ
dt dt
where hL is the height of the liquid along a vertical diameter.

Equations (2.9) and (2.10) define critical hL/B or hL/dt for a given (uG − uL). The term
uG is the actual gas velocity so that
hG
uGS ¼ uG ð2:13Þ
B
for a channel flow and
AG
uGS ¼ uG ð2:14Þ
At
for a pipe flow, where At is the area of the empty pipe. Equations (2.9) and (2.10) for
inviscid flows are consistent with observations that transition from a stratified flow to
a slug flow at small uG is predicted to occur at a larger hL/dt in a larger diameter pipe for
air–water flows. However, comparisons with measurements indicate that the critical hL/dt
predicted by Taitel and Dukler is larger than what is observed; that is, the system is
predicted to be more stable (Wallis & Dobson, 1973).
A physical explanation of this deficiency of (2.9) and (2.10) has been given by Lin &
Hanratty (1986) and by Wu et al. (1987), who associated it with the neglect of viscous
effects. They retained the assumption that the wavelength is long compared with hG and
hL, but abandoned the assumption of inviscid flow by including the effects of the drag of
2.4 Effect of liquid viscosity on flow regimes in a horizontal pipe 37
the gas and the resisting stress of the wall on the liquid. Equations (2.9) and (2.10) still
hold provided the velocities in the gas and the liquid can be approximated as uniform
flows. The principal difference found in their viscous large-wavelength analysis is
that the wave velocity is not given by (2.08) or (2.11). Rather, it is the kinematic wave
velocity defined by Lighthill & Whitham (1955).
The first term on the left side of (2.6) represents the destabilizing effects of liquid
inertia. For air–water flow at atmospheric conditions, pG/pL = 1.2 × 10−3, so (2.8)
simplifies to c = uL for the range of conditions over which transition is observed.
Consequently, the inviscid analysis, equation (2.6), predicts no influence of liquid inertia
on stability. It represents a static instability. In contrast, the viscous large-wavelength
analysis gives non-zero values of (c/uL − 1) at the critical condition. As a result, an
important destabilizing effect of liquid inertia is considered. The inclusion of viscous
stresses has the surprising effect of causing the air–water system to be more unstable,
in that the initiation of large-wavelength interfacial disturbances is predicted to occur
at lower gas velocities (consistent with experimental observations) than is given by an
inviscid analysis.
Viscous large-wavelength theory has been used by Hanratty & Hershman (1961) to
describe the initiation of roll waves. Also, as indicated in this subsection, it has been used
to predict a critical height for the transition from a stratified air–water flow to a slug/plug
flow at uSL = 10 m/s in a 9.53 cm pipe and at uLS = 4 m/s in a 2.54 cm pipe (Hurlburt &
Hanratty, 2002). Experiments with air–water by Soleimani & Hanratty (2003) show that
the theory correctly predicts the initiation of slugs/plugs at low gas velocities and the
initiation of pseudo-slugs at high gas velocities. It is of interest to note that the observed
mechanism for transition to slug/plug flow at low gas velocities (see Section 2.2.6 and
Figure 2.4) is quite different from what is conjured by theory, that is, the monotonic
growth of very large-wavelength waves.
As is illustrated in the discussions in this subsection, theoretical constructs that
represent the transition to slug/plug flow provide a critical hL/dt or hL/B for a given
uGS. An important issue in utilizing these theories is the specification of uLS. This requires
a proper model for a stratified flow, a topic to be covered in Chapter 5. A discussion about
the sensitivity of predicting the critical uLS, to determining interfacial drag, to modeling
the liquid flow and to predicting gradients of hL in the flow direction, is presented by
Hurlburt & Hanratty (2002). The effect of an underdeveloped flow is particularly
important at low gas velocities, for which changes of hL in the flow direction can have
effects comparable to the pressure gradient.
2.4 Effect of liquid viscosity on flow regimes in a horizontal pipe
The influence of liquid viscosity on flow regimes has been studied by Taitel & Dukler
(1987), Andritsos & Hanratty (1987) and Andritsos et al. (1989). Results for a 9.53 cm
pipe, with L/dt = 250, for air and 1 cP water, and for air and a 100 cP solution are given
in Figure 2.6. Results for a 2.54 cm pipe, with L/dt = 400, for a 70 cP solution are
shown in a paper by Andritsos & Hanratty (1987). Figure 2.7 presents a graph, from
38 Flow regimes
1.0
Slug/plug transition
Pseudo-slug
0.1
ULS (m/s)
Annular
Atomization
Stratified KH
0.01 waves
100 cP liquid
1 cP liquid
0.001
1.0 10.0 100.0
UGS (m/s)
Figure 2.6 Comparison of flow regime maps for air–water and an air–glycerine solution in a horizontal pipe
with a diameter of 9.53 cm. Andritsos et al., 1989.
0.9
0.8 dt = 9.53 cm, Air - 100 cP
dt = 2.52 cm, Air - 70 cP
0.7
0.6
hL/dt
0.5
0.4
0.3 dt = 9.53 cm, air–water
dt = 2.52 cm, air–water
0.2 dt = 7.63 cm, air–water θ = –0.5°
0.1
1 10
UGS (m/s)
Figure 2.7 Critical hL/dt for transition from stratified to slug/plug flow for dt = 9.53 cm, dt = 7.63 cm and
dt = 2.52 cm. Hurlburt & Hanratty, 2002.
Hurlburt & Hanratty (2002), which plots the critical hL/dt versus superficial gas velocity
for the transition from a stratified flow to a slug flow.
Despite the differences between water and high-viscosity glycerine–water solutions,
the curves in the flow regime map, in Figure 2.6, representing the initiation of atom-
ization, the formation of annular flows, the appearance of Kelvin–Helmholtz waves and
the transition to slug flow at high gas velocities, are similar. The chief differences are
associated with the initiation of slugs at low gas velocities. This can be understood by
considering Figure 2.7.
Of particular interest is the observation that the critical height is the same for high-
viscosity liquids in 9.53 cm and 2.52 cm pipes, and for flow in a 7.63 cm pipe declined
2.4 Effect of liquid viscosity on flow regimes in a horizontal pipe 39
dt dt
dt
dt
0.30
hL/dt
0.20 0.10
hL/dt
0.10 0.00
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
Time (s) Time (s)
fi fi
(a) UGS = 8.8 m/s, = 5.0 (b) UGS = 12.0 m/s, = 7.5
fG fG
dt = 9.53 cm, ULS = 0.005 m/s
dt dt
0.10
hL/dt
0.10 0.05
hL/dt
0.00 0.00
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
Time (s) Time (s)
fi f
(c) UGS = 14.0 m/s, = 9.0 (d) UGS = 24.0 m/s, i = 15.0
fG fG
dt = 9.53 cm, ULS = 0.005 m/s
Figure 2.8 Large-amplitude waves on a viscous liquid. Andritsos & Hanratty, 1987.
at an angle of 0.5° to the horizontal (discussed in the chapter on stratified flow). The
common feature of all three of these studies is that the interface is smooth at transition.
The results for declined flows are interpreted in Section 5.5.3, Figures 5.8 and 5.9 and,
in more detail, in a paper by Woods et al. (2000). Only effects of liquid viscosity in
horizontal pipes are considered in this section.
Consider a stratified flow of a very viscous liquid. The liquid rate is fixed. The interface
is smooth. The gas velocity is increased (so hL/dt decreases). The first interfacial
disturbances that are observed, with increasing gas velocity, in stratified high-viscosity
liquids have small wavelengths and small amplitudes. They are regular two-dimensional
waves (see Figure 2.3a). With just a slight increase in gas velocity, they give way to a few
large-amplitude roll waves with steep fronts, smooth troughs, and spacings that can vary
from a few centimeters to a meter. Occasionally, several small two-dimensional waves
40 Flow regimes
can be observed in front of the large waves (see Figure 2.8a). Eventually, the waves
become cell-like (see Figure 2.8c). Under these conditions the spacing is 1–2 cm.
Slugs can evolve very rapidly from the large-amplitude roll waves, shown in Figure
2.8a, if hL/dt is large enough. For the air–water system, the appearance of slugs at low
gas velocities is associated with a viscous large-wavelength instability of a stratified
flow (as discussed above). Consistent with this interpretation, Figure 2.7 shows an effect
of pipe diameter on the transition to slug flow in the air–water system. For viscous
liquids, transition to slug flow is caused by an inviscid small-wavelength KH instability,
described by equation (2.7), with khL and khG of the order of unity. (see Chapter 4). Thus,
the critical hL/dt is not influenced by liquid viscosity (as indicated in Figure 2.7).
For high gas velocities, slugs are formed by coalescence of roll waves, both for the
air–water and for air and fluids with large viscosities. Because of this, the transition to
slugging is independent of pipe diameter for both systems. However, the more viscous
liquids are found to be more stable.
At high gas velocities, transition is described by considering the stability of slugs, that
is, equation (2.5). The height of the liquid needed for the existence of stable or growing
slugs is dictated by the rate of shedding from the slug, Qsh. This shedding rate is
commonly interpreted by assuming that the back of the slug (see Figure 9.1b) may be
considered to be a bubble which is moving at a velocity cB. The shedding rate is then
given as Qsh = (cB − um) A1, if aeration is ignored. Hurlburt & Hanratty (2002) provide the
following speculative explanation for the influence of liquid viscosity on measurements
of the critical hL/dt at large uGS, seen in Figure 2.7. They point out that the velocity of
the bubble behind the slug is given by cB ≈ 1.2um for air–water and suggest that the
slug tail is laminar for glycerine–water. Measurements of Sam & Crowley (1986)
show that cB = 2um for laminar flows. This leads to a larger Qsh. From (2.5), a larger
critical hL/dt is needed for stable slugs.
2.5 Effect of gas density on flow regimes in a horizontal pipe
Well-defined results on the effect of gas density on flow regimes are not plentiful. Studies
by Crowley et al. (1986) in a 17.8 cm pipe and by Wu et al. (1987) in a 20.3 cm pipe
are discussed by Hurlburt & Hanratty (2002) and by Andritsos et al. (1992), who
emphasize the importance of considering slug stability in determining transition at a
large gas density. The study by Crowley et al. involved the use of freon gas and water.
The gas density was 32.5 kg/m3 (compared with a density of 1.2 kg/m3 for air at
atmospheric conditions). The study by Wu et al. involved a natural gas/condensate.
The gas density was 65 kg/m3 and the surface tension was 0.0118 N/m (compared with
0.07 N/m for water). The studies of the transition from slug flow to annular flow by
Reimann & Seeger (1981) for steam–water and air–water are also of interest.
The effect of gas flow on generating instabilities in stratified flows, in initiating
atomization, and on influencing the flow rate in a stratified liquid should be represented
roughly through variations of the gas phase kinetic energy. Thus, many investigators
have suggested that flow regime maps of the type shown in Figure 2.1 should use ρ0:5 G uGS
2.6 Effect of drag-reducing polymers 41
as the abscissa, rather than uGS. This approach is partially consistent with the researches
cited above.
Since instability of a stratified flow varies roughly as ρ0:5 G uGS , it is possible that
mechanisms which consider the instability of a stratified flow (such as the viscous
large-wavelength and the Kelvin–Helmholtz analyses) will predict the appearance of
waves at too low a gas velocity for the instability to evolve into a stable slug. Then,
transition is controlled by slug stability, rather than the instability of a stratified flow.
The transition from slug/plug flow to annular flow clearly shows an influence of ρG
whereby the critical uGS decreases with ρnG with n ≈ 0.4. (Reimann and Seeger 1981).
2.6 Effect of drag-reducing polymers
One of the more intriguing advances in studies of single phase turbulence is the finding
that the introduction of small amounts of long-chain polymers into a liquid can cause
decreases in the fluid turbulence and the resistance of the wall to the flow (Toms, 1948).
This motivated a number of studies of the effects of drag-reducing polymers on gas–lquid
flows. The main focus was the change in the pressure drop. A review of work in this
area by Manfield et al. (1999) suggests that an understanding of the influence of
drag-reducing polymers on multiphase flow was not available in 1999. That is, changes
in the pressure gradient could not be directly related to the drag reduction observed in
single-phase flows.
More recent research has focused on the effect of drag-reducing polymers on the
configuration of the phases. Thus, Al-Sarkhi & Hanratty (2001a,b) have studied the
influence of long-chain polymers on annular flows. They found that disturbance waves
are destroyed and therefore the production of drops is discontinued. The flow changed to
a stratified configuration. Drag-reductions of 48% were realized for a 9.53 cm pipe and
63% for a 2.54 cm pipe.
Soleimani et al. (2002) studied the effects of drag-reducing polymers on interfacial
drag and on the initiation of slugs for air and water flowing in a 2.54 cm pipe over a range
of superficial gas velocities of 1–100 m/s. The interfacial drag can be represented
as τ i ¼ f i ρG u2G . Soleimani et al. found that added polymers cause a decrease in interfacial
activity, and consequently a greatly reduced interfacial friction factor, fi. However, for
stratified flows, decreases in the interfacial friction factor are accompanied by increases
in the liquid height, and therefore increases in the gas velocity that counterbalance the
effect of the decrease in interfacial roughness. Soleimani et al. showed that the addition
of polymers to stratified flows causes large decreases in fi but modest decreases in the
pressure drop.
A more interesting result is the finding that polymers delay the transition from a
pseudo-slug to a slug flow regime, observed with increases in the superficial velocity
of the liquid. This is consistent with the finding, discussed in Section 2.2.6, that transition
from a pseudo-slug regime to slug flow with increasing liquid flow is dictated by a
consideration of the stability of a slug. An increase in liquid viscosity (see Section 2.4)
dampens turbulence and decreases the stability of slugs by increasing the shedding rate.
42 Flow regimes
Baik & Hanratty (2003b) extended the work of Soleimani et al. by studying the effect
of drag-reducing polymers on stratified flows of air and water in a 9.53 cm pipe. They
also observed a damping of interfacial waves. This led to decreases in the interfacial
stresses. Drag-reductions as high as 19% were observed.
A second effect observed by Baik & Hanratty was a modification of the mechanism
for the transition from stratified flow to slug flow at very low gas velocities. For air–water
flows, this transition results from the growth of Jeffreys waves into slugs by the
bifurcation process described in the first paragraph of Section 2.2.6 and illustrated in
Figure 2.4. The addition of polymers inhibits this bifurcation (by mechanisms which are
not completely understood). Drag reductions as large as 40% were realized.
2.7 Vertical upflows
2.7.1 Flow patterns

Several investigations of patterns for upward flow of air and water have been reported in
the literature. Significant differences exist, even for the same system at roughly the same
gas and liquid flows. There are a number of reasons for these apparent contradictions:
The patterns have been defined in different ways. A number of techniques have been
used to provide criteria to define transitions (Dukler & Taitel, 1986; Delhaye, 1981). The
presence of surfactants and the method for introducing the gas have an effect (Hewitt &
Hall-Taylor, 1970, p. 13).
For a fixed liquid flow, the following regimes typically appear with increasing gas
velocity.
At low liquid flows a bubble pattern is observed (at low gas flows) for which the liquid
phase is continuous and the gas is broken into bubbles by inertial forces. The bubbles
can have a distribution of sizes. The largest, called cap bubbles, have a hemispherical
shape. The bubbles are not distributed uniformly over the pipe cross-section so the
voidage can vary with location. Clusters can form. Void fraction waves can exist under
certain conditions.
At higher gas velocities a slug flow can appear. It consists of a progression of bullet-
shaped bubbles (called Taylor bubbles) that fill most of the pipe cross-section. The fronts
resemble cap bubbles and the backs are approximately flat. The slugs of liquid between
the Taylor bubbles can be aerated. Large pressure fluctuations are observed. The liquid
between the gas and the wall moves downward. Thus, to an observer moving with the
Taylor bubble, the liquid flows around the bubble.
With further increases in gas velocity, slugs break down and a chaotic oscillatory
motion of a highly aerated liquid appears. This pattern is called churn flow. (In some
studies, it is described as a foam flow.) It occurs over a range of gas and liquid velocities
which increases with increasing pipe diameter. At low gas flows, a churn pattern may be
looked upon as a breakup of slugs by an occasional bridging across the pipe by the liquid
phase. At high gas flows “it may be considered as a degenerate form of annular flow with
the direction of the wall film changing and very large waves being formed at the interface.
2.7 Vertical upflows 43
5
Dispersed bubble
3
Liquid superficial velocity (m/s)
2
Bubble
Churn
1
0.5 Annular
0.3
Slug
0.2
0.03 0.1 0.3 1 3 10 30
Gas superficial velocity (m/s)
Figure 2.9 Representative map for upward flow of air and water. (constructed by Omebere-Iyari & Azzopardi,
2000)
In this range, the term semi-annular flow has sometimes been used” (Omebere-Iyari &
Azzopardi, 2000).
At still larger gas flow (uGS > ~ 10 m/s) an annular pattern exists whereby part of
the liquid flows upward as a film along the wall and part, as drops entrained in the gas.
There is an exchange of liquid between the gas and the flowing film. For air–water flows
there is a critical film flow, defined as ReLF =4 ¼ W LF =μL πd t (Andreussi et al., 1985),
below which drops cannot be torn from the film. For a fixed gas velocity, the volume
fraction of drops in the gas increases with increasing uLS after the critical liquid flow in
the film is reached. At large uGS and large enough uLS the drops reach a concentration
such that large structures appear in the gas flow. This pattern, called wispy annular flow,
is described by Bennett et al. (1965, 1967).
At very large liquid flows and small gas flows, a pattern called a dispersed bubble
regime emerges (Taitel et al., 1980). Turbulence in the liquid causes the breakup of the
gas into bubbles. This regime is similar to the bubbly pattern observed in horizontal gas–
liquid flow. Thus, the critical liquid velocity needed for this type of bubbly flow to appear
is approximately the same for these two orientations.
A representative flow pattern map is shown in Figure 2.9 for upward flow of air and tap
water in a 29 mm pipe. This was constructed by Omebere-Iyari & Azzopardi (2000) from
correlations for the bubble/slug (Taitel et al., 1980), for the bubble/dispersed bubble (Taitel
et al., 1980), for the slug/churn (Brauner & Barnea, 1986), and for churn/annular
(McQuillan & Whalley, 1985) transitions. It should be noted that the slug pattern was
not observed by Cheng et al. (1998) in a pipe with a diameter of 150 mm, by Kytömaa &
Brennan (1991) in a pipe with a diameter of 102 mm or by a McMaster University research
group in a pipe with a diameter of 100 mm (cited by Omebere-Iyari & Azzopardi, 2000).
44 Flow regimes
An article by Omebere-Iyari et al. (2008) summarizes data from a number of sources.

These show that a churn-turbulent pattern replaces slug flow in large-diameter pipes.
2.7.2 Mechanisms for the breakdown of the bubbly flow pattern

Discussions of early work on the change from a bubble pattern to a slug flow are given in
books by Hewitt & Hall-Taylor (1970) and by Govier & Aziz (1972). If the inlet mixing
process creates bubbles, it is evident that the appearance of slugs depends on coalescence
of the bubbles and on the stability of the slugs that are formed. Griffith & Snyder (1964)
suggested that bubbly flow cannot be sustained for bubble fractions greater than 35%.
This observation has been supported by calculations of Radovich & Moisis (cited by
Hewitt & Hall-Taylor, 1970, p. 12), which show that the number of collisions increases
rapidly with void fraction in the range of α = 0.30. Dukler & Taitel (1986) argue that
collisions cannot be a factor at α = 0.52, which is characteristic of a close packed cubic
lattice, since the bubbles would not have enough freedom of motion. They assumed a
criterion for transition of α = 0.25 and used this, along with the rise velocity of bubbles
given as
U S ¼ uG uL ð2:15Þ
to predict transition. After substituting

uGS
uG ¼ ð2:16Þ
α
uLS
uL ¼ ð2:17Þ
ð1 αÞ
where α is the void fraction, one gets

ð1 α Þ
uLS ¼ uGS ð1 αÞU S ð2:18Þ
α
Dukler & Taitel (1986) used the following equation developed by Harmathy (1960) for
single large bubbles in an infinite medium:

σg ðρL ρG Þ 1=4
U S∞ ¼ 1:53 ð2:19Þ
ρ2L
(See Govier & Aziz, 1973, p. 369.) The rise velocity in a swarm of bubbles decreases
with increasing α so an equation developed by Zuber & Hench (cited by Dukler & Taitel,
1986) was used to relate US to US∞ :
U S ¼ U S∞ ð1 αÞ1=2 ð2:20Þ
Thus, (2.19) and (2.20), with α = 0.25, can be substituted into (2.18). It is noted that these
relations give a transition condition which is independent of pipe diameter.
The notion that the transition can be explained solely by considering the gradual
coalescence in a uniform swarm of bubbles has been questioned. This classical explanation
2.7 Vertical upflows 45
implies that a transition can always occur if a pipe is long enough. Cheng et al. (1998) quote
studies which show a small change in bubble properties for an air/tap water flow along a
pipe, and no effect of pipe length on the void fraction. These results suggest that transition
occurs throughout the column by a stability mechanism, rather than by gradual coalescence.
2.7.3 Mechanism for the slug flow to churn flow transition

The main thrust of theories on the transition from a slug flow to a churn flow is the
consideration of the stability of slugs. The interpretation most favored is a breakdown of
the film between Taylor bubbles and the wall, whereby it changes from a net downward
to a net upward flow. Large (flooding) waves carry liquid upward and the liquid film
between the flooding waves carries fluid downward (Jayanti & Brauner, 1994; Hewitt &
Hall-Taylor, 1970).
However, Brauner & Barnea (1986) suggest that instability arises from excessive
aeration in the liquid slugs between the Taylor bubbles. Jayanti & Brauner (1994) compare
these two approaches in their review of research on churn flow. Little, if any, attention has
been given to the possibility of associating this transition with the breakdown of the front
of a Taylor bubble. Yet, this could be an important consideration in explaining why slug
flow is not observed in large diameter pipes. (See Omebere-Iyari et al., 2008.)
2.7.4 Mechanism for the churn/annular transition

Transition from churn flow to annular flow can be explained by considering the con-
ditions needed for the existence of an annular flow, such as the necessity for all of the
liquid in the film to be moving upward or for the droplets to be carried upward. Most of
the attention has been given to the first of these. A reversal of the direction of film flow is
consistent with the existence of a zero velocity gradient (or a zero shear stress) at the wall.
A force balance on a small length of film gives the following:
0 ¼ πd t τ W dz þ πd t τ i dz ρL ghL πd t dz ð2:21Þ
where the thickness of the film, hL, is assumed to be small compared with dt. If τW = 0 and
τ i ¼ f i 12 ρG u2G ,
1
d t f i ρG u2G ¼ ρL ghL ð2:22Þ
2
at transition. Wallis (1969) gives the following rough approximation for the friction
factor in an annular flow

hL
f i ¼ 0:005 1 þ 300 ð2:23Þ
dt
If (2.23) is substituted into (2.22) the following relation is obtained if uG ≈ uGS

1=2
uGS ρG
1=2 1=2
¼C ð2:24Þ
g1=2 d t ρL
46 Flow regimes
where C is approximately equal to 0.87. McQuillan & Whalley (1985) used (2.24) with
C = 1 to correlate data on the transition to annular flow. The agreement is approximate
but quite good, considering that the data covered a range of gas densities from 1.29 to
107.0 kg/m3, tube diameters of 0.01 to 0.105 m, surface tensions of 0.0074 to 0.072 N/m,
liquid viscosities of 0.096 to 1.06 N s/m2 × 103.
2.7.5 Mechanism for the transition to dispersed bubbly flow

Dukler & Taitel (1976) and Taitel et al. (1980) point out that the breakup of bubbles by
turbulence at large liquid velocities has been observed in a number of situations and
recommended the use of the following correlation developed by Hinze to calculate the
bubble size, dP, created by turbulence:
d P ε2=5
ðσ=ρL Þ3=5
where σ is the surface tension and ε is the rate of dissipation of mechanical energy per unit
volume in the liquid. Since ε = ρMF, where F, the frictional loss per unit mass of liquid,
can be calculated with (1.44).
The notion behind (2.25) is that turbulent velocity fluctuations may be viewed as
containing a number of scales (or wavelengths). Spatial variations of the pressure fluctua-
tions that have a scale close to the bubble diameter can exert a force on the bubbles. This is
balanced by surface tension forces, characterized as σ/dP. For bubbles in equilibrium with
the turbulence field, these forces will be equal so that
ρL v2λ¼d P ¼ Cσ=d P ð2:26Þ
where vλ is the characteristic velocity of fluctuations with a wavelength λ and C is a

constant. It is assumed that λ is small enough to be independent of the method by which
turbulence is produced and large enough for the fluctuations to be independent of
viscosity. Komogoroff has proposed that, in this range of wavelengths, the velocity
fluctuations in a homogeneous isotropic field depend only on ε and λ. Thus, from
dimensional reasoning,
vλ ≈ ðελÞ1=3 ð2:27Þ
provided λ/dt << 1 and λvλ/L >> 1. From (2.26) and (2.27), the equilibrium bubble size
depends on σ/ρL, ε, dP (with dP being substituted for λ). Equation (2.25) results from
dimensional analysis. It is presumed to be valid for all turbulent liquid flows. Taitel
et al. (1980) argue that, in addition, the dispersed bubbles must keep a spherical shape.
They use this criterion and (2.25) to provide a criterion for transition from a dispersed
bubble flow to a bubbly pattern or a slug pattern Thus, in Figure 2.9, the dispersed
bubble regime represents conditions for which bubbles are created by liquid turbu-
lence, and the bubble regime represents conditions for which bubbles are created by
liquid inertia.
2.8 Vapor patterns in evaporating and condensing flows / critical heat flux 47
Suppression of nucleate boiling Locus of burnout
X Y Z
Z
Y
X=6
A B C D E F G H I J K L M N O
ditto X
Heat flow increased Onset of
Subcooled water in equal steps nucleate boiling
(constant flow rate)
Figure 2.10 Patterns for upward flow of a liquid in a heated tube. Figure 2.11 in Hewitt & Hall-Taylor, 1970.
2.8 Vapor patterns in evaporating and condensing flows / critical heat flux
An important application involving a gas–liquid mixture is the evaporation of an

upwardly flowing liquid which is supplied with a constant heat flux at the wall, shown
in Figure 2.10 (taken from Hewitt & Hall-Taylor, 1970). The sketches depict situations
with a constant mass flow rate and increasing heat flux. Boiling first appears as a bubbly
pattern at the downstream end of the pipe. With increasing heat flux, the location at which
boiling is initiated moves upstream and more vapor is generated. Thus, the bubbly pattern
can evolve into a slug or churn flow. At large enough heating, an annular flow occurs. At
still larger heat fluxes, the liquid in the wall film evaporates and a vapor/droplet flow
exists for which the pipe wall is in contact with the vapor flow. The light curves represent
constant qualities defined as the fraction of the mass flow that is vapor.
Hewitt & Hall-Taylor describe Figure 2.10 along the following lines: The dark curve
labeled X in the sketch represents the location at which boiling is initiated. In order
for this to occur, the liquid at the wall must be superheated. However, over most of the
channel cross-section the liquid is below its boiling point, so that bubbles generated
at the wall collapse after they leave the wall. Under these conditions, the average
temperature of the flowing liquid is below the boiling point and the evaporation process
is known as subcooled boiling. Eventually, phase change occurs not only by boiling
at the wall but also by evaporation at the interfaces. The latter process becomes
more important as the fluid progresses along the pipe so that the liquid at the wall is
not sufficiently superheated for boiling to occur. In the annular flow region, vapor
generation occurs primarily by evaporation at the interface of the wall film. Curve Y
indicates the loci at which vapor generation at the wall ceases. Curve Z indicates
48 Flow regimes
Mass velocity (kg m–2s–1) 4000

Wispy
3000 annular
Burnout limit
Pure for 3.66 m-long
water 12.7 mm-ID tube
2000
Annular
1000
0 Bubble Slug Churn

–10 0 10 20 30 40 50 60 70 80 90 100
Quality (percentage by weight)
Figure 2.11 Flow pattern data collected by Bennett et al. (1965) for steam–water flow at 6.89 MP pressure.
Figure 2.6 in Hewitt & Hall-Taylor, 1970.
whether dryout at the wall has occurred. Since, after dryout, the liquid blanket at the
wall is no longer available the wall temperature increases rapidly. In situations for
which the heat input is constant, this can lead to a deterioration of the integrity of the
pipe wall. Operating conditions (for example, heat flux) need to be selected so as to
avoid this burnout.
A critical heat flux need not be associated with a dryout of the wall film in annular flow.
For example, burnout can also occur at low flows and high fluxes. The number of boiling
sites at the wall can then increase to such an extent that a film of gas forms at the wall.
This film boiling is not able to remove as much heat at the wall as does nucleate boiling,
so the wall temperature increases rapidly. This is similar to the boiling crisis observed for
pool boiling.
The analysis of evaporating flows is very difficult. Flow maps, such as shown in
Figure 2.11 (Hewitt & Hall-Taylor, 1970), can be useful. These measurements were made
for the flow of steam and water up a 3.66 m length of 12.7 mm pipe subjected to a
constant heat flux. The mass flow was kept constant and the heat flux was increased.
From (1.11) and the assumption that the liquid is uniformly distributed over the pipe
cross-section, the following energy balance can be written for a flowing mixture if the
contributions to the enthalpy from changes in kinetic energy and work against gravity are
negligible compared with the heat flux:
ðh hi ÞGA ¼ qW Pl ð2:28Þ
where hi is the enthalpy at the inlet. The term in parentheses in (2.28) represents the
enthalpy change over a pipe length l for a given heat flux at the wall and Pl is the inside
area, with P equal to the pipe perimeter. Now, if equilibrium is assumed to exist between
the phases the quality can be calculated from tables of thermodynamic functions. (Hewitt
and Hall-Taylor (1970) call this a thermodynamic quality.) A quality obtained in this
manner is plotted as the abscissa in Figure 2.11. The important role of the annular flow
region is emphasized.
2.9 Downflows in vertical pipes 49
This method for representing flow patterns is a way to use gas–liquid flow data (with no
phase change) to approximate, very roughly, flow regimes in evaporative and condensing
flows. This would simply involve plotting ðρG uGS Þ=ðρG uGS þ ρL uLS Þ as the abscissa.
2.9 Downflows in vertical pipes
Bubbles in a liquid experience a buoyancy force given by

!
F b ¼ V P ðρL ρG Þ!
g ð2:29Þ
This is the major contributor to the slip velocity of bubbles, (uG − uL). In a shear flow,
bubbles also experience a lateral force perpendicular to the mean flow direction described by
duz
F Lr ¼ C L ðuG uL ÞρL ð2:30Þ
dr
where CL is an empirically determined lift coefficient and VP is the volume of the bubble
(see Section 10.4).
In an upward dispersed flow, the slip velocity is positive and duL/dr is negative, so
bubbles will tend to move toward the wall. However, for downward flow, the gravita-
tional force is such that bubbles move in a direction opposite to the flow. Thus, the slip is
negative. The lift force is away from the wall, so bubbles will tend to move toward the
center of the pipe.
Another difference between upflows and downflows is that, at zero gas velocity, one
can have a falling film along the wall, which resembles an annular flow.
Oshimoto & Charles (1974) report on the coring described above for downflows. They
identified the appearance of the following patterns with increasing gas flow: (1) bubble-
coring, (2) bubble-slug, (3) falling film, (4) froth, (5) annular flow.
Slug flow is initiated by coalescence of the coring bubbles. The Taylor bubbles appear
to be the same as observed in upward flow. However, they are upside-down and move
counter to the liquid flow. Thus, the nose of the bubble points upstream.
The falling film identified by Oshimoto & Charles is thicker than the film observed in
annular flows. Bubbles occluded in the film escape and bridges of liquid form across the
pipe. The froth flow is similar to the froth flow observed in vertical upflows, except that
the mixture of gas and liquid is less turbulent. The annular flow that appears at very large
gas velocities is the same as observed in upflows.
Sekoguchi et al. (1996) and Ishii (2004) studied air and water flowing down 2.58 cm
and 5.08 cm pipes. Air was mixed in the flowing liquid by using a centrally located
sparger. A specially designed array of needle conductivity probes enabled an observation
of the configuration of the air–water interfaces. At small uGS, the gas in the bubble was
found to be moving upstream relative to the liquid (a negative slip velocity). Under these
circumstances the large bubble is asymmetric in the flow direction. At intermediate uGS, a
range of conditions exist for which the slip velocity is close to zero and the front and back
of the bubble are the same.
50 Flow regimes
2.10 Microgravity flows
2.10.1 General comments

Microgravity effects on gas–liquid flows are an important consideration in space appli-
cations. Moreover, they supply a test for theories developed from 1g data. Situations for
which g = 0 are considered in this section. Gas–liquid flows change in several ways:
(1) The pattern is not affected by the orientation of the conduit or by the direction of
flow, as is found when g is not close to zero. (2) The flows tend to be axisymmetric, so
that stratified patterns do not exist. (3) The rise velocity of bubbles, US, is zero, so that
local gas and liquid velocities are equal.
However, the spatially averaged gas, 〈uG〉, and liquid, 〈uL〉, velocities need not be
equal if the gas is not uniformly distributed, even though US = 0 locally. Zuber & Findlay
(1965) suggest that, for air–liquid flow at 1g,
huG i ¼ Cum þ U S ð2:31Þ
where
um ¼ uLS þ uGS ð2:32Þ
and C can vary with flow pattern. Nicklin et al. (1962) suggest that C =1.2 for vertical slug
flows at 1g (see Section 8.5). Colin et al. (1991) claim C =1.1 for small bubbles at 1g.
2.10.2 Flow regimes

Measurements of gas–liquid flow under microgravity conditions have been made for a
range of conditions by Dukler et al. (1988), Colin et al. (1991), Bousman et al. (1996),
Zhao & Rezkallah (1993). These studies were carried out in drop towers and in flights of
jet aircraft. By following the parabolic path of a jet, periods of microgravity of 15–25 s
were realized. Photographs of the different flow patterns are provided by Bousman et al.
(1996) and by Colin et al. (1991). All of these experiments show that under microgravity
conditions the pattern is independent of orientation.
Measurements by Colin et al. (1991) in a microgravity environment confirm that
US = 0 and give C = 1.2 in (2.31) both for bubbly and slug flow. By definition,
uLS ¼ huL ih1 αi ð2:33Þ
uGS ¼ huG ihαi ð2:34Þ
Substitution of (2.31) into (2.34), using US = 0, yields

uGS
¼ hαiC ð2:35Þ
uLS þ uGS
Equation (2.35) can be used to calculate hαi as a function of uLS and uGS. Values obtained
in this way, with C = 1.2, agree with measurements for bubbly and slug flows in
microgravity.
2.10 Microgravity flows 51
Three regimes have been defined for microgravity flows. These are described by Colin
et al. (1991) as follows:
(1) “bubbly flow at low gas rates, where the bubble size is typically smaller than the pipe
diameter,”
(2) “slug flow for moderate gas and liquid rates, consisting of Taylor bubbles longer than
1dt and a liquid slug in which smaller bubbles may be dispersed; these bubbles travel
axially without significant drift with respect to the Taylor bubbles,”
(3) “annular flow for high gas rates, in which a wavy liquid film exists at the wall, on
which roll waves propagate.”
The transition from bubbly flow to slug flow with increasing gas rate is more gradual
than is observed at 1g, so a transitional regime is defined for a range of conditions.
The bubbly pattern is similar to what has been observed in downflows at 1g, in that
coring is observed. This could explain why the constant, C, in (2.31) is the same for slug
and bubbly flow, in that the gas accumulates in the central region of the pipe for both
regimes. Coring also seems to affect bubble coalescence.
Bousman et al. (1996) present flow regime maps for water (μ = 1 cP, σ = 72 dynes/cm),
a 50% water–glycerol solution (μ = 6 cP, σ = 63 dynes/cm) and a surfactant solution (μ =
1 cP, σ = 21 dynes/cm). Tubes with diameters of 12.7 mm and 25.4 mm were used. The
initiation of annular flow for these systems occurred at roughly the same conditions.
The transition from bubbly to slug flow is given by uGS = uLS for all of the experiments
in a 12.7 mm tube and for air and water–glycerine solutions in a 25.4 mm tube.
Experiments with water and with surfactant solutions in a 25.4 mm tube are reported to
occur at a smaller uGS for a given uLS, that is,
1
uGS ¼ uLS ð2:36Þ
2:5
The difference in the results for air and water in a 25.4 mm tube from those in a 12.7 mm
tube, cited above, should be noted. Some support for this finding on the effect of pipe
diameter is obtained from the study of air–water in a 40 mm tube by Colin et al. (1991).
Their critical gas velocities are represented by
1
uGS ¼ uLS ð2:37Þ
3:2
which, within experimental error, is the same as (2.36).

The result uGS = uLS for the transition from bubbly to slug flow, obtained with a
12.7 mm tube, corresponds to a critical 〈α〉 of about 0.40, that is, a closely packed bubble
swarm. The results (2.36) and (2.37) obtained for air and water in 25.4 mm and 40 mm
tubes correspond to an 〈α〉 of about 0.20. Bousman et al. (1996) examined movie films of
the air–water system near transition. In the 25.4 mm tube:
the gas–liquid bubble interfaces were in a continuous state of fluctuations while those in the 12.7 mm
tube under the same flow conditions were more stable. These oscillations in the large tube, which
can be attributed to turbulence, give the bubbles a larger effective diameter, thus increasing the
probability of contacting nearby bubbles. The result is a transition to slug flow at a lower void
52 Flow regimes
fraction. This effect was not observed in the air–water/glycerine experiments in the large tube due to
a reduction in turbulence. . .
2.10.3 Comparison of flow regimes for microgravity and 1g flows

The influence of gravity is most noticeable in horizontal flows where it can have a direct
effect on the distribution of liquid. The stratified pattern is not observed and the slug pattern
is quite different under microgravity conditions. It would seem that comparisons should be
made with vertical flows. However, even for these flows, differences can be observed.
In vertical upflow, gravity opposes the flow. Thus, churn flow, which involves inter-
mittent reversals close to the wall due to gravity, does not exist in microgravity flows.
In vertical downflows, gravity aids the flow, so that annular flows at 1g are more easily
formed and can be observed over a wider range of gas velocities. This is not the case for
microgravity conditions.
However, since the absence of gravity leads to a zero value of the rise velocity, there
are a number of less obvious reasons to expect differences in flow regime boundaries. As
already discussed, this can lead to differences in 〈α〉, in the spatial distribution of bubbles
(because of lift forces), in the interaction of bubbles with one another, and in the wake
structure of small bubbles in the liquid.
Comparisons of experiments with and without gravity can be complicated because
they might not involve identical systems and might involve the use of different methods
to mix the liquid and the gas. Thus, the microgravity experiments by Colin et al. (1991) in
a 40 mm tube are welcome. They show a transition from bubbly to slug flow which is
roughly the same for microgravity and for upflow in a 1g environment.
Transition to annular flow in microgravity is observed at a lower uGS than is found for
upward flow at 1g. This can be explained because transition at 1g involves a change from
a churn flow. At g = 0, it involves a breakdown of a slug flow. At large uGS (say 10 m/s)
the annular flow resembles what is usually found in vertical 1g flows. That is, a thin wall
film is agitated by disturbance waves. There is an exchange of drops between the gas
flow and the wall film. Annular flow in microgravity at uGS < 10 m/s is characterized by a
relatively thick wall layer with large-amplitude waves and a gas flow that contains only a
small amount of entrained drops. Sometimes these differences are acknowledged by
defining a new flow regime.
2.11 Capillaries and microchannels
2.11.1 Capillary tubes

Microgravity flows are situations where surface tension effects are much larger than
gravitational effects. The relative importance of these two forces is characterized by the
Bond number
dh
Bo ¼ ð2:38Þ
½σ=ðρL ρG Þg1=2
2.11 Capillaries and microchannels 53
where λ ¼ ½σ=gðρL ρG Þ1=2 is the Laplace length scale and dh is the hydraulic diameter
of a passage, equal to 4 times the cross-sectional area divided by the wetted perimeter. For
a circular tube dh = dt. Thus, surface tension can dominate gravitational effects for small
dh, as well as for small g. Serizawa & Feng (2000) point out that λ = 6.6 mm for boiling
water and λ = 2.7 mm for air–water at atmospheric conditions.
Studies by Damianides & Westwater (1988) in horizontal capillaries with dt = 1–5 mm
and by Fukano and his co-workers (Fukano & Kariyasaki, 1993) show that stratified
flows do not exist for air–water flows in horizontal tubes with dt ≤ 5 mm. Furthermore,
Fukano & Kariyasaki (1993) show that bubbles have a zero rise velocity in a still liquid
for dt ≤ 5 mm.
Photographs, as well as flow regime maps of the configurations of the phases, are given
by Triplett et al. (1999) and by Fukano & Kariyasaki (1993). Three principal patterns
bubbly, slug (intermittent) and annular, are identified. The maps are observed to be roughly
the same for tubes with diameters of 1–5 mm and to be independent of orientation.
The annular flow pattern and the transition from slug flow to annular flow are similar to
what is found for microgravity situations.
However, the range of conditions associated with the bubbly pattern is much
smaller for air–water flows in capillary tubes (dt = 1 – 5 mm) than in microgravity flows
(dt = 12.7 mm, 25.4 mm, 40 mm). It appears that the small-diameter capillary tubes
promote the formation of slugs so that bubbly flows of the type discussed in Section
2.7.5 do not exist. However, at large liquid velocities the dispersed bubbly pattern is
observed. Thus, in the air–water studies of Fukano & Kariyasaki (1993), patterns with
small bubbles occupy only a small region (roughly, uLS > 0.8 m/s, uGS < 1 m/s). Since the
properties of dispersed bubbles are believed to be governed by liquid phase turbulence,
the picture presented here could change in more viscous liquids (which would be
laminar).
2.11.2 Microchannels
Tubes with diameters much smaller than 1 mm, say 20–100 microns (μm), are called
microchannels. Gas–liquid flows in these tiny conduits are of interest in the design of
compact heat exchange devices that involve boiling. Studies in microchannels include
air–water flow in 25 micron tubes by Serizawa & Feng (2000, 2001) and in 100 micron
tubes by Kawahara et al. (2002). The gas and liquid were admitted at the inlet to the test
section where inertia and liquid phase turbulence mixes the two phases.
The dispersed bubble pattern defined for capillary tubes is not observed. As pointed
out by Serizawa & Feng (2000), the decrease in channel size can result in the existence of
a laminar environment. This could explain the absence of the dispersed regime which
depends on the interaction of liquid phase turbulence and bubbles. Thus, the main flow
regimes are slug flow (elongated bubbles) and a separated regime in which all of the
liquid flows along the wall as a smooth or wavy layer, or as a sequence of rings. These
rings could result from the breakdown of a slug flow which has very short bridges.
The paper by Serizawa & Feng (2000) suggests the existence of an annular pattern
with a thin liquid film and droplets entrained in the gas at large gas flows. They show
54 Flow regimes
photographs for steam–water flow in a 50 micron heated silicon tube. Tiny steam bubbles
generated at the wall are observed close to the wall at a low heating rate. By increasing
the heating rate the following patterns are observed: bubbles of the same size as the
tube, short elongated bubbles, long bubbles (slug flow), a liquid ring flow, and annular-
dispersed flow. A discussion of forced flow boiling in microchannels is given by
Ghiaassiaan (2003).
References
Al-Sarkhi, A. & Hanratty, T.J. 2001a. Effect of drag-reducing polymers on annular gas–liquid flow
in a horizontal pipe. Int. J. Multiphase Flow 27, 1152–1162.
Al-Sarkhi, A. & Hanratty, T.J. 2001b Effect of pipe diameter on the performance drag-reducing
polymers in annular gas–liquid flows. Trans IChemE, 79 A, 402–408.
Andreussi, P. & Persen, L.N. 1987 Stratified gas–liquid flow in downwardly inclined pipes. Int.
J. Multiphase Flow 13, 565–575.
Andreussi, P., Asali, J.C. & Hanratty, T.J. 1985 Initiation of roll waves in gas–liquid flows. AIChE
Jl 21, 119–126.
Andritsos, N. & Hanratty, T.J. 1987 Interfacial instabilities for horizontal gas–liquid flows in
pipelines. Int. J. Multiphase Flow 13, 583–603.
Andritsos, N., Williams, L. & Hanratty, T.J. 1989 Effect of liquid viscosity on the stratified-slug
transition in horizontal pipe flow. Int. J. Multiphase Flow 15, 877–892.
Andritsos, N., Bontozoglou, V. & Hanratty, T.J. 1992 Transition to slug flow in horizontal pipes.
Chem. Eng. Comm. 118, 361–385.
Baik, S. & Hanratty, T.J. 2003a Concentration profiles of droplets and prediction of the transition
from stratified to annular flow in horizontal pipes. Int. J. Multiphase Flow 29, 329–338.
Baik, S. & Hanratty, T.J. 2003b Effects of a drag-reducing polymer on stratified gas–liquid flow in a
large diameter horizontal pipe. Int. J. Multiphase Flow 29, 1749–1757.
Barnea, D., Shoham, O., Taitel, Y. & Dukler, A.E. 1980 Flow pattern transition for gas–liquid flow
in horizontal and inclined pipes. Int. J. Multiphase Flow 6, 387–397.
Bennett, A.W., Hewitt, G.F., Kearsey, H.A., Keeys, R.K.F. & Lacey, P.M.C. 1965 Flow visual-
ization studies of boiling at high pressure. AERE-R4874.
Bennett, A.W., Hewitt, G.F., Kearsey, H.A. & Keeys, R.K.F. 1967 Heat transfer to steam–water
mixtures flowing in uniformly heated tubes in which the critical heat flux has been exceeded.
Proc. Inst. Mech. Eng. 180 (3C): 1–11. AERE-R15373, presented in Hewitt & Hall-Taylor 1970.
Bousman, W.S., McQuillen, J.B. & Witte, L.C. 1996 Gas–liquid flow patterns in microgravity:
effects of tube diameter, liquid viscosity and surface tension. Int. J. Multiphase Flow 22,
1035–1053.
Brauner, N. & Barnea, D. 1986 Slug/churn transition in upward gas–liquid flow. Chem. Eng. Sci.
40, 159–163.
Cheng, H., Hills, J.H. & Azzopardi, B.J. 1998 A study of the bubble to slug transition in vertical
gas–liquid flow in columns of different diameter. Int. J. Multiphase Flow 24, 431–452.
Cohen, L.S. & Hanratty, T.J. 1965 Generation of waves in the concurrent flow of air and a liquid.
AIChE Jl 11, 138–144.
Colin, C., Fabre, J. & Dukler, A.E. 1991 Gas liquid flow at microgravity conditions:1 Dispersed
bubble and slug flow. Int. J. Multiphase Flow 17, 533–544.
References 55
Crowley, C.J., Sem, R.G. & Rothe, P.H. 1986 Investigation of two-phase flow in horizontal and
inclined pipes at large pipe size and high gas density. Report TN-399 for the Pipeline Research
Committee, American Gas Associates, Hanover, NH.
Damianides, C.A. & Westwater, J.W. 1988. Two-phase flow patterns in a compact heat exchanger
and in small tubes. Second UK National Conference on Heat Transfer, Glasgow, 14–16
September. London: Mechanical Engineering Publications, pp. 1257–1268. (See also
Damianides, C. A. 1987 Horizontal two-phase flow of air–water mixtures in small diameter
tubes and compact heat exchangers, Ph.D. thesis, University of Illinois at Urbana-Champaign.)
Delhaye, J.M. 1981 Two-phase flow patterns. In Two-Phase Flow and Heat Transfer in the Power
and Process Industries, eds A.E. Bergles, J.G. Collier, J.M. Delhaye, G.F. Hewitt, & F.
Mayinger, Washington, DC: Hemisphere.
Dukler, A.E. & Hubbard, M.G. 1975 A model for gas–liquid slug flow in horizontal tubes. Ind.
Eng. Chem. Fundamentals 14, 337–347.
Dukler, A.E. & Taitel, Y. 1986 Flow pattern transitions in gas–liquid systems: Measurement and
modelling. Multiphase Sci. Technol. 2, 1–94.
Dukler, A.E., Fabre, J.A., McQuillen, J.B. & Vernon, R. 1988 Gas–liquid flow at microgravity
conditions: flow patterns and their transitions. Int. J. Multiphase Flow 14, 389–400.
Fan, Z., Lusseyran, F. & Hanratty, T.J. 1993 Initiation of slugs in horizontal gas–liquid flows.
AIChE Jl 39, 1742–1753.
Fukano, T. & Kariyasaki, A. 1993 Characteristics of gas–liquid two-phase flow in a capillary tube.
Nucl. Eng. Des. 141, 59–68.
Ghiaasiaan, M. 2003 Gas–liquid two-phase flow and boiling in microchannels. Multiphase Sci.
Technol. 15, 323–333.
Govier, G.W. & Aziz, K. 1972 The Flow of Complex Mixtures in Pipes. New York: VanNostrand
Reinhold.
Griffith, P. & Snyder, G.A. 1964 The bubbly-slug transition in a high velocity two-phase flow. MIT
Report 5003–29 (TID-20947).
Hanratty, T.J. & Engen, J.M. 1957 Interaction between a turbulent air stream and a moving water
surface. AIChE Jl 3, 294–304.
Hanratty, T.J. & Hershman, A. 1961 Initiation of roll waves AIChE Jl 7, 488–497.
Hanratty, T.J., Theofanous, T., Delhaye, J.-M., Eaton, J., McLaughlin, J., Prosperetti, A., Sundaresan,
S. & Tryggvason, G. 2003 Workshop on scientific issues in multiphase flow. Int. J. Multiphase
Flow 29, 1042–1116.
Harmathy, T.Z. 1960 Velocity of large drops and bubbles in media of infinite or restricted extent.
AIChE Jl 6, 281–288.
Hewitt. G.F. & Hall-Taylor, N.S. 1970 Annular Two-Phase Flow. Oxford: Pergamon Press.
Hurlburt, E.T. & Hanratty, T.J. 2002 Prediction of the transition of from stratified to slug and plug
flow for long pipes. Int. J. Multiphase Flow 28, 707–729.
Ishii, M., Pavanjope, S.S., Kim, S. & Sun, X. 2004 Interfacial structure and interfacial area
transport in downward two-phase bubbly flow. Int. J. Multiphase Flow 30, 779–801.
Jayanti, S. & Brauner, N. 1994 Churn flow. Multiphase Sci. & Technol. 8, 471–521.
Jeffreys, H. 1925 On the formation of water waves by wind. Proc. R. Soc. Lond. A107,
189–206.
Kawahara, A., Chung, P.M.-Y., & Kawaji, M. 2002 Investigation of two-phase flow pattern and
pressure drop in a microchannel. Int. J. Multiphase Flow 28, 1411–1435.
Kocamustafaogullari, G. & Wang, Z. 1991 An experimental study on local interfacial parameters in
a horizontal bubbly two-phase flow. Int. J. Multiphase Flow 17, 553–572.
56 Flow regimes
Kordyban, E.S. & Ranov, T. 1970 Mechanism of slug formation in horizontal two-phase flow.
J. Basic Eng. 92, 857–864.
Kytömaa, H.K. & Brennan, C.E. 1991 Small amplitude kinematic wave propagation in two-
component media. Int. J. Multiphase Flow 17, 13–26.
Lighthill, M.J. & Whitham, G.B. 1955 On kinematic waves. 1. Flood movement in long rivers; 2.
Theory of traffic flow on long narrow roads. Proc. R. Soc. Lond. A229, 281–345.
Lin, P.Y. & Hanratty, T.J. 1986 Prediction of the initiation of slugs with linear stability theory.
Int. J. Multiphase Flow 12, 79–98.
Lin, P. Y. & Hanratty, T.J. 1987a Detection of slug flow from pressure measurements.
Lin, P.Y. & Hanratty, T.J. 1987b Effect of pipe diameter on flow patterns for air–water flow in
horizontal pipes. Int. J. Multiphase Flow 13, 549–563.
Mandhane, J.M., Gregory, G.A. & Aziz, K.A. 1974 A flow pattern map for gas–liquid flow in
Manfield, C.J., Lawrence, C. & Hewitt, G. 1999 Drag-Reduction with additives in multiphase flow.
Multiphase Sci. Technol. 11, 197–201.
McQuillan, K.W. & Whalley, P.B. 1985 Flow patterns in vertical two-phase flow. Int. J. Multiphase
Flow 11, 161–176.
Niklin, D.J., Wilkes, J.O. & Davidson, J.F. 1962 Two-phase flow in vertical tubes. Trans. IChem E
40, 61–68.
Omebere-lyari, N.K. & Azzopardi, B.J. 2000 Links across flow-patterns in gas–liquid flow in
vertical pipes. Paper presented at the 2nd Japanese-European Two-Phase Group Meeting,
Tsukuba, 25–29 September.
Omebere-lyari, N.K., Azzopardi, B.J., Lucas, D., Beyer, M. & Prasser, H-M. 2008 The character-
istics of gas/liquid flow in large risers at high pressures. Int. J. Multiphase Flow 34, 461–476.
Oshimoto, T. & Charles, M.E. 1974 Vertical two-phase flow, Part 1. Flow pattern correlations. Can.
J. Chem. Engng 52, 25–35.
Reimann, J. & Seeger, J.W. 1981 Transition to annular flow in horizontal air–water and steam–
water flow. Report of Kernforschungszentrum, Karlsruhe, KfK 3198.
Ruder, Z. & Hanratty, T.J. 1990 A definition of gas–liquid flow in horizontal pipes. Int.
Ruder, Z., Hanratty, P.J. & Hanratty, T.J. 1989 Necessary conditions for the existence of stable
slugs. Int. J. Multiphase Flow 15, 209–226.
Sam, R.G. & Crowley, C.J. 1986. Investigation of two-phase processes in coal slurry/hydrogen
heaters. Creare Report TH-1085 for Department of Energy.
Sekoguchi, K., Mori, K. & Masuo, K. 1996 Interfacial profiles and flow characteristics in vertical
downward two-phase plug and foam flows. Chem. Eng. Comm. 141, 415–441.
Serizawa, A. & Feng, Z.P. 2000 Reviews of two-phase flow in microchannels. In The US-Japan
Seminar on Two-Phase Flow Dynamics, June 4–9, Santa Barbara, CA.
Serizawa, A. & Feng, Z.P. 2001 Two-phase flow in microchannels. In Proceedings of the 4th
International Conference on Multiphase Flow, May 27–June 1, New Orleans, LA.
Soleimani, A. & Hanratty, T.J. 2003 Critical liquid flows for the transition from pseudo-slug and
stratified patterns to slug flow. Int. J. Multiphase Flow 29, 51–67.
Soleimani, A., Al-Sarkhi, A. & Hanratty, T.J., 2002. Effect of drag-reducing polymers on pseudo-
slugs, interfacial drag and transition to slug flow. Int. J. Multiphase Flow 28, 1911–1927.
Taitel, Y. & Dukler, A.E. 1976 A model for predicting flow regime transitions in horizontal and
near-horizontal gas–liquid flow. AIChE Jl 22, 47–55.
References 57
Taitel, Y. & Dukler, A.E. 1987 Effect of pipe length on the transition boundaries for high viscosity
liquids. Int. J. Multiphase Flow 13, 577–581.
Taitel, Y., Barnea, D. & Dukler, A. 1980 Modelling flow transitions for steady upward gas–liquid
flow in vertical tubes. AIChE Jl 26 345–354.
Toms, B.A. 1948 Some observations on the flow of linear polymer solutions through straight tubes
at large Reynolds numbers. In Proceedings of the 1st International Conference on Rheology, Vol.
2, Amsterdam: North Holland, 135–141.
Triplett, K.A., Ghiaasiaan, S.M., Abdel-Khalik, S.I. & Sadowski, D.L. 1999 Gas–liquid two-phase
flow in microchannels Part 1: two-phase flow patterns. Int. J. Multiphase Flow 25, 377–394.
Wallis, G.B. 1969 One Dimensional Two-phase Flow. New York: McGraw-Hill.
Wallis, G.B. & Dobson, J.E. 1973 The onset of slugging in horizontal stratified air–water flow. Int.
Woods, B.D. & Hanratty, T.J. 1996 Relation of slug stability to shedding rate. Int. J. Multiphase
Flow 22, 809–828.
Woods, B.D., Hurlburt, E.T. & Hanratty, T.J. 2000 Mechanism of slug formation in downwardly
inclined pipes. Int. J. Multiphase Flow 26, 977–998.
Wu, H.L., Pots, B.F.M., Hollenburg, J.F. & Meerhof, R. 1987. Flow pattern transitions in two-
phase gas/condensate flow at high pressre in an 8-inch horizontal pipe. In Proceedings of the 3rd
BHRA Conference on Multiphase Flow, The Hague, The Netherlands, 13–21.
Zhao, L. & Rezkallah, K.S. 1993 Gas–liquid flow patterns at microgravity conditions. Int.
Zuber, N. & Findlay, J.A. 1965 Average volumetric concentration in two phase flow systems.
J. Heat Transfer 87, 453–468.
3 Film flows
3.1 Free-falling layer
3.1.1 Laminar flow

The simplest example of a wall film is the free-fall of a liquid along a wall. Consider the
case of a flat wall where y is the distance from the wall and z is the distance in the direction
of flow. Liquid moves down the wall under the force of gravity. It is desired to calculate
the thickness of the film as a function of the volumetric flow.
Consider the differential element, shown in Figure 3.1, of size dzdy, and of length l in
the z-direction. The time-rate of change of momentum in the volume is zero so, from the
momentum theorem, the net force is zero. That is, the force of gravity is balanced by
shear stresses defined by τ and τ + dτ.
The force of gravity is +ρgldzdy. The convention is used that the shear stress is positive
when the forces on the two sides of a plane have the same sign as the normals to the plane.
Thus, from Figure 3.1, the force due to the shear stresses equals −τldz + (τ + dτ)ldz, so
0 ¼ ρgldydz τldz þ ðτ þ dτ Þldz
or
dτ
¼ ρg ð3:1Þ
dy
This can be integrated to give

τ ¼ ρgy þ c
where c is the constant of integration. Since τ = 0 at y = m, c = ρgm and

τ ¼ ρgy þ ρgm ð3:2Þ
where m is the height of the liquid layer. From (3.2), the stress at the wall is τW = ρgm.
For laminar flow τ is given by Newton’s law of viscosity,
du
τ ¼ μL ð3:3Þ
dy
where μL is the viscosity of the liquid. Substituting (3.3) into (3.2) and integrating
twice gives
3.1 Free-falling layer 59
y
z
τ τ+τ
Figure 3.1 Differential volume in a free film of height m on a vertical wall.
ðm
gm3
Γ¼ udy ¼ ð3:4Þ
0 3L
where L = μL/ρL is the kinematic liquid viscosity, and Γ is the volumetric flow per unit
length in the spanwise direction. The hydraulic diameter, defined as 4 times the area
divided by the wetted perimeter, is given as 4 m. A film Reynolds number is defined as
ReLF = 4Γ/L. Since τW = ρgm, (3.4) can be written as
m2
μL Γ ¼ τW ð3:5Þ
3
or as
mþ ¼ 0:866 Re0:5
LF ð3:6Þ
where m+ = mv*/L and a friction velocity is defined as v* = (τW/ρ)1/2.

In Section 3.2.2, a characteristic stress, τc = (2/3)τw + (1/3)τi, is defined. A friction
velocity based on this stress is designated as vc ¼ ðτ c =ρÞ1=2 . For the case being consid-
ered, the interfacial stress, τi, is zero, so that an alternative form of equation (3.6), for a
free-falling film, is
mþ
c ¼ 0:707ReLF
0:5
ð3:7Þ
where mþ
c ¼ mvc =L .
For axisymmetric flow, such as would exist for flow down the outside of a tube of
radius rt, the above analysis gives
" #
gm3 m 3 m 2 1 m 3 1 m 4
Γ¼ 1þ þ þ þ ð3:8Þ
3L rt 20 rt 40 rt 140 rt
3.1.2 Turbulent flow

At high Reynolds numbers the flow is turbulent, so momentum can be transferred both by
molecular and turbulent motions. Thus, the stress is defined as
60 Film flows
τ du du
¼ þ t ð3:9Þ
ρ dy dy
A number of models for the turbulent kinematic viscosity, t, have been proposed. These
are discussed in the book by Fulford (1964). Measurements at large Reynolds numbers
are fitted by
mþ
c ¼ 0:031ReLF
0:90
ð3:10Þ
3.1.3 Interpolation formula

The following interpolation relation for free-falling films, suggested by Henstock and
Hanratty (1976), can be used for all Reynolds numbers.
h i1=5
0:5 5 0:9 5
mþ
c ¼ 0:707Re LF þ 0:031Re LF ð3:11Þ
Note that it has the correct behavior for ReLF → 0 and for ReLF → ∞.
3.2 Gas–liquid flows
3.2.1 Horizontal flow in a channel

For gas–liquid flow in a horizontal rectangular channel, the liquid is pulled along by
the drag of the gas at the interface, τi, which is not zero. For a fully developed film flow,
the momentum theorem indicates that the net force on a control volume, such as shown
in Figure 3.1, is zero. The distance from the wall is given as y and the coordinate in
the direction of flow is z. The gas flow has a pressure gradient, dp/dz, and imposes
an interfacial stress, τi. Since the flow is fully developed, there is no mean flow in the
y-direction and the pressure gradient in the gas flow is the same as the pressure gradient in
the liquid film.
The volume l dzdy is acted on by the pressure gradient (which is negative) and shear
stresses. Thus a force balance gives
ðτ þ dτ Þldz τldz þ pldy ðp þ dpÞldy ¼ 0
This can be rearranged to give

dτ dp
¼ ð3:12Þ
dy dz
where, as indicated above, dp/dz is a constant for a fully developed flow.

Equation (3.12) can be integrated from y to m to give
dp
τ i τ ¼ ð m yÞ ð3:13Þ
dz
3.2 Gas–liquid flows 61
so that at the wall (y = 0)

dp
τW ¼ τi m ð3:14Þ
dz
Note that τW > τi since dp/dz is negative. (For a horizontal gas–liquid flow τw ≈ τi and
τc ≈ τW. This is not the case for vertical flows, to be considered in Section 3.2.2.)
Colburn & Carpenter (1949) assumed that a film behaves the same as single-phase
turbulent flow. This implies that the law of the wall relation
uþ ¼ f ð y þ Þ ð3:15Þ
with y+ = yv*ρ/μ and u+ = u/v*, can be used to describe the velocity profile. The friction
velocity is defined with the shear stress at the wall, v* = (τW/ρ)1/2. The film Reynolds
number is obtained by integrating (3.15) between y+ = 0 and m+. Since ReLF = 4Γ/ and
Ðm
Γ ¼ 0 udy
ð mþ
ReLF
¼ uþ dyþ ð3:16Þ
4 0
The Colburn–Carpenter hypothesis gives (Henstock & Hanratty, 1976)

h 2:5 2:5 i0:4
mþ
c ¼ 0:707Re0:5
LF þ 0:0379Re0:9
LF ð3:17Þ
Equation (3.17) is similar to (3.11), which was developed for free-falling films. It has the
same or similar behavior at high and low liquid Reynolds numbers. However, the
transition from laminar to turbulent flow is more gradual.
When the interface is smooth, the flow is laminar. When three-dimensional pebbled
waves are introduced, the Colburn–Carpenter hypothesis becomes valid and equation
(3.17) can be used to calculate m+. This is illustrated in the paper by Cohen & Hanratty
(1966), which shows that m+ is only a function of ReLF.
Studies of horizontal gas–liquid flows are described in a number of papers
(see, for example, Hanratty & Engen, 1957; Cohen & Hanratty, 1966; Andreussi
et al., 1985). These consider the case of a liquid layer flowing along the bottom of
an enclosed channel and air flowing concurrently with it. For a fixed ReLF, the
interface is smooth at low gas velocities. As the gas velocity increases, a critical
condition is reached at which two-dimensional waves appear at the interface
(see Figure 6.2). At higher gas velocities, a three-dimensional wavy interface is
observed. At still higher gas velocities, roll waves are observed. (Of interest is the
finding that equation (3.17) also holds when roll waves are present.) With increasing
gas velocity the average height of the film decreases and the intermittent flow surges
change their appearance in that they are a complicated collection of capillary waves.
They have the appearance of patches of turbulence. Hewitt & Hall-Taylor (1970) call
them “disturbance waves.” At high enough gas velocities, capillary waves are
entrained in the gas flow.
62 Film flows
3.2.2 Film flow in a vertical pipe

Consider the common case of a film on the wall of a pipe through which a gas is flowing.
Equation (3.8) and the paper by Henstock & Hanratty (1976) show that the film may be
treated as a two-dimensional flow, similar to what would be found for flow along a flat
wall. The y-axis is perpendicular to the wall. The z-axis is in the direction of the mean
flow.
A force balance on a differential volume yields the following equations
dτ dp
¼ þ ρg upflow ð3:18Þ
dy dz
dτ dp
¼ ρg downflow ð3:19Þ
dy dz
since the film is acted upon both by the pressure gradient and by the force of gravity.
Thus, the negative pressure gradient is augmented in downflow and opposed in upflow
by the force of gravity. Because of the gravitational force, there can be a large spatial
variation of the shear stress in the film (not seen in horizontal flows), so the choice of a
characteristic stress for the liquid film is not obvious. For convenience, define

1 dp
g̑ ¼ þ g ð3:20Þ
ρ dz
for downflow and

1 dp
g̑ ¼ g ð3:21Þ
ρ dz
for upflow, where |dp/dz| signifies the absolute value of the pressure gradient.
The integration of (3.18), (3.19) gives
τ τ W ¼ ρg̑ y ð3:22Þ
The stress at the interface is

τ i τ W ¼ ρg̑ m ð3:23Þ
Therefore, from (3.22)

du
μ ¼ τ W þ ρg̑ y ð3:24Þ
dy
for a laminar flow. Integration of (3.24) gives

y2
μL u ¼ τ W y þ ρg̑ ð3:25Þ
2
The volumetric flow is obtained by integrating (3.25) and by using (3.23) to eliminate
ρ~
gm
3.2 Gas–liquid flows 63

m2 2 1
μL Γ ¼ τW þ τi ð3:26Þ
2 3 3
Note that equation (3.5) is recovered for τi = 0. Thus, the use of a characteristic stress of
τ c ¼ ð2=3Þτ W þ ð1=3Þτ i ð3:27Þ
represents both a free-falling film and a film in the presence of a gas flow for laminar
films.
The velocity field and the volumetric flow can be calculated for a turbulent field
by substituting (3.9) into (3.22) and integrating. The Colburn–Carpenter hypothesis
can be employed by assuming that the spatial variation of the kinematic turbulent
viscosity is the same as for single-phase flows. The calculated velocity profiles can be
given as uþ þ
c ¼ u=vc versus yc ¼ yvc =. Only one additional parameter is needed to
characterize the profiles, if m/rt is small enough. This is
g̑ m
αmþ ¼ ð3:28Þ
v2c
where α ¼ ĝ =v3 c . For typical annular flows, the friction velocity is large so αm is
+
close to zero.
For upward flows, αm+ is a positive quantity. For increasing αm+, the profile is
distorted by the gravitational force and at αm+ = 3 the velocity gradient at the wall is
zero. For αm+ larger than 3, the calculated average velocity shows negative values close
to the wall. Downward flowing liquid films in an upwardly flowing gas have negative
values of αm+. For free-falling layers, τi = 0 and αm+ = −3/2. For downward film flows the
interfacial stress imposed by the countercurrent gas velocity retards the flow so that
the maximum velocity in the film is not at the outer edge. For still larger gas flows the
velocity at the interface can assume a zero value. Thus, between αm+ = −3/2 and αm+ = 0
the flow changes from a free-falling downward flow to an upwardly flowing film. The
critical condition for which the flow starts to reverse its direction is reached at an
intermediate negative value of αm+.
Henstock & Hanratty (1976) have used the equation for the eddy viscosity developed
by van Driest to integrate (3.22) and (3.9). The calculated relation between the volumetric
flow and the height of the film is plotted as mþ c versus ReLF with αm as a parameter in
+
Figure 3.2. Because of the use of τc, rather than τW, as a characteristic stress, the plot is
independent of αm+ at small ReLF where the film is laminar. For larger ReLF, the
parameter αm+ has a smaller effect than would be expected, considering its large
influence on the shape of the velocity profile. Measurements available to Henstock &
Hanratty did not include conditions with extreme values of αm+, so they recommended
equation (3.17), which agrees approximately with their calculation for αm+ = 0.
Thus, equation(3.17) provides the value of mþ
1=2

c ¼ mvc = for a given flow rate of the

film. Since vc ¼ τ c =ρ and τc are given by equation (3.27), one needs a relation for
the drag on the interface, τi, to calculate m for a given film Reynolds number. This matter
is addressed later in Section 3.3.
64 Film flows
αm+ = 0.6
103 αm+ = 1.2
αm+ = 1.8 αm+ = 0
αm+ = 2.4 αm+ = −0.75
αm+ = 3.0 αm+ = −1.5
102
mc+
101
100
101 102 103 104 105 106
ReLF
Figure 3.2 Plot of mþ

c versus ReLF, calculated using a van Driest model for the mixing length with A = 26.
Parameter αm+ is defined by equation (3.28). Henstock & Hanratty, 1976.
3.2.3 Flow reversal and flooding

Hewitt (1982), Delhaye (1981) and others have commented on flow reversal and flooding
as examples of situations for which gravity has a strong effect on the behavior of a
vertically flowing film.
Consider the case of an upward flow of a gas and a liquid. Equation (3.18) presents a
force balance for the liquid film. Integration of this equation yields
dp
τi τ ¼ ðm yÞ þ ρgðm yÞ ð3:29Þ
dz
or
dp
τ ¼ τi ðm yÞ ρgðm yÞ ð3:30Þ
dz
Since dp/dz is negative, the gas phase interfacial stress and pressure gradient supply the
motive forces pulling the film upward. These are opposed by gravity. As the gas velocity
decreases, the interfacial stress decreases and it is no longer able to keep the entire film
moving upward. From (3.30), the shear stress at the wall is
dp
τW ¼ τi m ρgm ð3:31Þ
dz
where τW = μL(du/dy)W.
The gas velocity at which τW = 0 represents a situation for which the average velocity
gradient at the wall is zero. A plot of pressure gradient versus gas flow rate presented in
3.3 Interfacial stress for vertical flows 65
Hewitt (1982) shows a minimum at τW = 0. For gas velocities below this critical condition
a net negative flow would exist in the region close to the wall. Figure 2.1.20 in the paper
by Hewitt (1982) shows that beyond this critical condition the pressure drop increases
with decreasing flow rate, the flow is haphazard and alternating its direction. At a small
enough gas flow the film completely changes its direction. Section 2.4.4 of the Hewitt
paper uses the criterion of τw = 0 to correlate data on the transition from annular flow to
churn flow.
Flooding is observed when the fluid in a downward flow changes direction. It is
suggested in Section 3.2.2 that this is initiated when the liquid velocity in the film at the
interface changes direction. Measurements indicate that at gas velocities larger than this
critical value an intermittent pattern exists, for which large-amplitude upwardly moving
waves alternate with a downwardly flowing film.
As indicated by Hewitt (1982) and by Delhaye (1981), a knowledge of when
this occurs is important in many applications. Flooding can limit the operation of
reflux condensers and mass transfer operations involving falling film absorbers or
packed columns. In a loss-of-coolant accident in a nuclear reactor, flooding can limit
the rate at which emergency cooling water can enter the reactor. Slug flow in vertical
tubes involves the passage of elongated bubbles which have a thin downwardly
flowing film at the wall. The breakdown of this pattern as the gas velocity increases
has been explained as resulting from breakdown of the film by a process similar to
flooding.
The analysis of Henstock & Hanratty (1976) can explain changes in the shape of the
velocity profiles in a film and the dependence of flow reversal and flooding on system
variables.
3.3 Interfacial stress for vertical flows
3.3.1 Waves in annular flows

This section focuses on annular flows in vertical pipes for gas velocities that are high
enough for flow reversal or flooding not to be considered. The behavior is similar for
vertical and for horizontal films. At very large gas flows and very small liquid flows, the
film is covered with long crested ripples that have steep fronts. Between these ripples the
interface is smooth and the flow appears laminar.
At sufficiently large liquid flows, disturbance waves (Section 3.2.1) appear on the film.
They travel at much larger velocities than the ripples in the base film and are coherent in
that they keep their identity over a long length of pipe (see Figure 3.3). The disturbance
waves are several times thicker than the base film over which they move.
At high gas velocities, this transition depends mainly on the liquid Reynolds number;
that is, it is only weakly dependent on the gas velocity. For flow in a horizontal channel
(Andreussi et al., 1985), this critical liquid Reynolds number (defined as 4WLF/μLP with
P signifying the pipe perimeter) is equal to 280. The same critical Reynolds number is
found for flow in vertical pipes (Andreussi et al., 1985).
66 Film flows
Figure 3.3 Photograph of annular flow in a vertical pipe that shows the intermittent disturbance waves.
3.3.2 Scaling
The interfacial stress, τi, represents the drag on the liquid film. Negative τi indicates the
resistance of the film to the gas flow. It is desired to predict the dependence of τi on the
flows of the gas and of the liquid film. The solution of this problem is central to predicting
the behavior of gas–liquid flows. The interfacial stress is determined by measuring the
gas-phase pressure gradient and the height of the liquid layer under conditions of a fully
developed symmetric flow.
Consider a control volume with a length, l, extending from the center of the pipe to
the edge of the wall layer, which has a height m. Liquid can be injected into the gas
flow from disturbance waves at a velocity, UA, at a rate per unit area of RA. Drops deposit
at the wall with a velocity, UD, at a rate per unit area of RD. Under fully developed
conditions, RA = RD. These processes can lead to a net transfer of momentum equal to
(RDUD − RAUA)2π((dt/2) − m)l.
Applying the momentum theorem to the gas-phase control volume defined above
yields
2
dt dt dp dt
2πl m ðRD U D RA U A Þ ¼ π m l 2π m τil ð3:32Þ
2 2 dz 2
3.3 Interfacial stress for vertical flows 67
The term −τi represents the resisting stress of the film to the gas flow. The first term on the
right side is the force on the control volume due to the pressure gradient.
The following equation is obtained from (3.32):

dt 1 dp
ðτ i þ τ P Þ ¼ m ð3:33Þ
2 2 dz
where
τ P ¼ ðRD U D RA U A Þ ð3:34Þ
represents an effective stress due to deposition and atomization. This term has been
considered by several investigators (see, for example, Moeck & Stachiewicz, 1972).
A difficulty in using (3.34) is the evaluation of UD and UA. A number of suggestions
have been made. For example, UA has been set equal to the velocity of the disturbance
waves and UD has been represented as the average axial velocity of the drops in the
gas space.
The interfacial stress is usually described in terms of a friction factor, fi, so that
1
τ i ¼ f i u2G ð3:35Þ
2
Calculations by Fore & Dukler (1995) and by Schadel & Hanratty (1989) suggest that τP
is of the order of 0.1 − 0.2τi. Because of the uncertainties in evaluating τP, it is not treated
separately in this chapter. Its influence is included in the specification of fi.
If the interface is smooth, fi is the same as is found for flow over smooth solid
surfaces, designated as fS. However, the presence of waves at the interface can lead to
very large values of fi/fS, which are related to the amplitudes of the waves. Two
approaches have been taken to represent this effect. One argues that the wave amplitude
varies as the thickness of the wall layer, m, and that this, in turn, varies as the dimension
of the conduit, dt.
fi m
/ ð3:36Þ
fS dt
The other uses the viscous length in the gas phase, G/v*, where G is the kinematic
viscosity of the gas and v* = (τi/ρG)1/2. Thus
fi mv*
/ ð3:37Þ
fS G
3.3.3 Correlation for fi using dt/2 as a scaling factor

For a smooth surface,
f S ¼ 0:046 Re0:20
G ð3:38Þ
68 Film flows
where ReG = dtWG/AtμG, WG is the mass flow of the gas, At is the tube area, μG is the
viscosity of the gas. A difficulty in using (3.36) is that the fundamental equation for film
height (3.17) contains the unknown τi. Henstock & Hanratty (1976) proposed a method
that avoids a trial and error solution.
Equation (3.17) can be written in the form
mþ
c ¼ γðReLF Þ ð3:39Þ
where γ(ReLF) is defined by the right side of (3.17). This can be rearranged to
mðτ c =τ i Þ1=2 ðτ i =ρL Þ1=2

¼ γðReLF Þ ð3:40Þ
L
After some algebra Henstock & Hanratty obtained

pffiffiffiffiffiffiffiffiffiffi
m 6:59F τ i =τ c
¼ pffiffiffiffiffiffiffiffiffiffiffi ð3:41Þ
dt f i =f S
where
rffiffiffiffiffi
γðReLF Þ L ρL
F¼ ð3:42Þ
Re0:9
G G ρG
From (3.36) and (3.41), they suggested that fi/fS is strongly dependent on F and weakly
dependent on αm+
fi
¼ gðF; αmþ Þ ð3:43Þ
fS
The advantage of using (3.43) is that F can be calculated directly if ReLF and ReG are
known.
Henstock & Hanratty chose to correlate measurements of fi/fS by using (3.43).
They obtained the following relation for upflows and downflows at high gas velocities:
fi
1 ¼ 1400F ð3:44Þ
fS
For downflows with moderate and small gas velocities, the effect of αm+ should be
considered. They suggested the following approximate equation

fi τi

¼ 1 þ 1400F 1 exp ð3:45Þ
fS ρL gm
For both upflows and downflows, the following equation was obtained for m:
m 6:59F
¼ ð3:46Þ
d t ð1 þ 1400F Þ1=2
References 69
3.3.4 Correlation using the viscous length for the gas phase
Asali et al. (1985) used additional measurements of interfacial stress for upward vertical
annular flows and chose a scaling based on the viscous length for the gas phase. Particular
attention was given to low liquid rates and high gas velocities for which the film can have
thicknesses of 30 to 200 μm. The following relation for the friction factor was obtained at
uG > 25 m/s
fi
1 ¼ 0:45 mþ
G 4 ð3:47Þ
fS
where mþ
G ¼ mvG =G and vG ¼ ðτ i =ρG Þ
1=2
. Surprisingly, the same equation represents
measurements under conditions that roll waves are present. Measurements for downward
flow by Andreussi & Zanelli (1978), Charvonia (1959), Chien & Ibele (1964) and Webb
(1970) for pipe diameters ranging from 2.4 cm to 6.35 cm are represented by the relation
fi
1 ¼ 0:45 mþ
G 5:9 ð3:48Þ
fS
If the wave height varies as the thickness of the film, m, equations (3.47) and (3.48) have
a kinship to what has been found for sand-roughened solid boundaries. That is, the wavy
interface is similar to a sand-roughened surface in that it may be considered to be
hydraulically smooth if the dimensionless roughness, k þ S , is small enough.
The evaluation of fi/fS from either (3.47) or (3.48) requires the specification of mþ G . Asali
et al. (1985) suggest approximations for mþ G both in the ripple and disturbance wave regimes.
References
Andreussi, P. & Zanelli, S. 1978 Downward annular and annular-mist flow of air–water mixtures.
Paper presented at International Seminar, Dubrovnik, Yugoslavia, September 4–9.
Jl 31, 119–126.
Asali, J.C., Hanratty, T.J. & Andreussi, P. 1985 Interfacial drag and film height for vertical annular
flow. AICh Jl 31, 895–902.
Charvonia, D.A. 1959 A study of the mean thickness of the liquid film and the characteristics of the
interfacial surface in annular two-phase flow in a vertical pipe. Jet Propulsion Center Report No.
I-59–1, Purdue University and Purdue Research Foudation, Lafayette, IN.
Chien, S.F. & Ibele, W. 1964 Pressure drop and liquid film thickness in annular mist flows. J. Heat
Transfer 86, 164.
Cohen, L.S. & Hanratty, T.J. 1966 Height of a liquid film in a horizontal concurrent gas–liquid
flow. AIChE Jl 12, 290–292.
Colburn, A.P. & Carpenter, F.G. 1949 Heat transfer lecture. Nucl. Energy USAEC 2, 105.
Delhaye, J.M. 1981 Two-phase flow patterns. In Two-phase Flow and Heat Transfer in Power and
Process Industries, Section 1–8, eds A.E. Bergles, J.G. Collier, J.M. Delhaye, G.F. Hewitt, and
F. Mayinger. Washington, DC: Hemisphere.
70 Film flows
Fore, L.B. & Dukler, A.E. 1995 The distribution of drop size and velocity in gas–liquid annular
flow. Int. J. Multiphase Flow 21, 137–149.
Fulford, G.D. 1964 The flow of liquids in thin films. In Advances in Chemical Engineering, Vol. 5.
New York: Academic Press, 151–236.
Henstock, W.H. & Hanratty, T.J. 1976 The interfacial drag and the height of the wall layer in
Hewitt, G.F. 1982 Flow regimes, In Handbook of Multiphase Systems, ed. G. Hetsroni.
Washington, DC: Hemisphere, pp. 2–25.
Hewitt, G.F. & Hall-Taylor, N.S. 1970 Annular Two-Phase Flow. Oxford: Pergamon Press.
Moeck, E.O. & Stachiewicz, J.W. 1972 A droplet interchange for annular-dispersed, two-phase
flow. Int. J. Heat Mass Transfer 15, 673–683.
Schadel, S.A. & Hanratty, T.J. 1989 Interpretation of atomization rates of the liquid film in
gas–liquid annular flow. Int. J. Multiphase Flow 15, 893–900.
Webb, D. 1970 Studies of the characteristics of downward annular two-phase flow. AERE Report
R6426.
4 Inviscid waves
4.1 Inviscid, incompressible, irrotational flow
A Reynolds number characterizing waves at an interface is λc/, where λ is the

wavelength, c is the wave velocity and is the kinematic viscosity. If this has a
large magnitude the assumption of an inviscid irrotational flow field is a useful
approximation.
An irrotational flow is one for which fluid particles are not rotating. It is defined by
curl !
u¼0 ð4:1Þ
where !
u is the velocity given by
! ! ! !
u ¼ i x ux þ i y uy þ i z uz ð4:2Þ
! ! !
and i x , i y , i z are unit vectors along the x, y, z coordinates. The definition of the curl is

! ! ∂uz ∂uy ! ∂ux ∂uz ! ∂uy ∂ux
curl u ¼ ix iy iz ð4:3Þ
∂y ∂z ∂z ∂x ∂x ∂y
Equation (4.1) is satisfied if the three amplitudes are respectively zero.

∂uz ∂uy ∂ux ∂uz ∂uy ∂ux
¼ ¼ ¼ ð4:4Þ
∂y ∂z ∂z ∂x ∂x ∂y
These equations can be satisfied if !

u is given by a potential function, ϕ(x, y, z, t), whereby
! ! ∂ϕ ! ∂ϕ ! ∂ϕ
u ¼ grad ϕ ¼ i x iy iz ð4:5Þ
∂x ∂y ∂z
so that
∂ϕ ∂ϕ ∂ϕ
ux ¼ uy ¼ uz ¼ ð4:6Þ
∂x ∂y ∂z
This can be seen simply by substituting (4.6) into (4.4).

The equation of conservation of mass is

∂ρ ∂ðρux Þ ∂ ρuy ∂ðρuz Þ
¼ þ þ ð4:7Þ
∂t ∂x ∂y ∂z
72 Inviscid waves
This equates the time change of mass in a fixed volume to the net flow of mass out of the
volume. For an incompressible flow, ρ = constant, so equation (4.7) simplifies to
∂ux ∂uy ∂uz
0¼ þ þ ð4:8Þ
∂x ∂y ∂z
The substitution of (4.6) into (4.8) gives

∂2 ϕ ∂2 ϕ ∂2 ϕ
0¼ þ þ ð4:9Þ
∂x2 ∂y2 ∂z2
This equation describes the velocity field for an incompressible, irrotational flow.
Equation (4.9) cannot satisfy the conditions that at a solid boundary both the normal
and tangential velocity components are equal to the velocity of the boundary. The usual
practice is to satisfy only the condition on the normal component. Thus, the tangential
velocity of the fluid can be different from that of the boundary; that is, there can be
slippage.
The equation of conservation of momentum needs to be considered in order to
calculate the pressure. For an inviscid flow
!
∂u ∂!
u ∂!
u ∂!
u
ρ þ ux þ uy þ uz ¼ grad p þ ρ! g ð4:10Þ
∂t ∂x ∂y ∂z
where !
g is the acceleration of gravity
! ! ! !
g ¼ i x gx þ i y gy þ i z gz ð4:11Þ
! ∂p ! ∂p ! ∂p
grad p ¼ i x þ iy þ iz ð4:12Þ
∂x ∂y ∂z
The left side of (4.10) represents the time rate of change of momentum in a fixed
differential volume in space plus the net flow of momentum out of that volume. The
right side gives the net force on the differential volume due to pressure and gravity.
The operator on the left side of (4.10)
D ∂ ∂ ∂ ∂
¼ þ ux þ uy þ uz ð4:13Þ
Dt ∂t ∂x ∂y ∂z
is the substantial derivative. It gives the change of a property of a field with time, as seen
by an observer moving with a fluid particle. This can be understood by defining f(x, y, z, t)
as a property of the field. Then
∂f ∂f ∂f ∂f
df ¼ dt þ dx þ dy þ dz ð4:14Þ
∂t ∂x ∂y ∂z
Divide by dt to get the change of f over any arbitrary path.

df ∂f ∂f dx ∂f dy ∂f dz
¼ þ þ þ ð4:15Þ
dt ∂t ∂x dt ∂y dt ∂z dt
4.1 Inviscid, incompressible, irrotational flow 73
Suppose the path is that of a fluid particle. Then

dx dy dz
ux ¼ uy ¼ uz ¼ ð4:16Þ
dt dt dt
and (4.14) is the same as (4.13). Thus (4.10) can be written as

D!
u
ρ ¼ grad p þ ρ!
g ð4:17Þ
Dt
That is, the terms on the left side of (4.10) that are enclosed in parentheses represent the
acceleration of a fluid particle.
Equation (4.10) has x-, y-, z-components which can be written as

∂ux ∂ux ∂ux ∂ux ∂p
ρ þ ux þ uy þ uz ¼ þ ρgx ð4:18aÞ
∂t ∂x ∂y ∂z ∂x

∂uy ∂uy ∂uy ∂uy ∂p
ρ þ ux þ uy þ uz ¼ þ ρgy ð4:18bÞ
∂t ∂x ∂y ∂z ∂y

∂uz ∂uz ∂uz ∂uz ∂p
ρ þ ux þ uy þ uz ¼ þ ρg z ð4:18cÞ
∂t ∂x ∂y ∂z ∂z
where
∂h ∂h ∂h
gx ¼ g gy ¼ g gz ¼ g ð4:19Þ
∂x ∂y ∂z
and h is the height above some datum plane. From (4.4), ∂uy/∂x, ∂uz/∂y can be substituted
for ∂ux/∂y, ∂uy /∂z in (4.18a) to give

∂2 ϕ 1 ∂q2 ∂p ∂h
ρ þ ¼ ρg ð4:20Þ
∂x∂t 2 ∂x ∂x ∂x
where (4.6) is substituted for ux and

q2 ¼ u2x þ u2y þ u2z ð4:21Þ
Equation (4.20) can be integrated with respect to x, keeping y, z and t constant to

give
∂ϕ q2 p
þ þ þ gh ¼ Fðy; z; tÞ ð4:22Þ
∂t 2 ρ
In the same way, (4.18b,c) can be integrated with respect to y and z, keeping x, z, t and
x, y, t constant. These produce (4.22) with the right sides replaced by F(x, z, t) and
F(x, y, t). This indicates that
F ðy; z; t Þ ¼ F ðx; z; t Þ ¼ F ðx; y; t Þ ¼ F ðt Þ ð4:23Þ

74 Inviscid waves
uG
y
hG
c B
η
x
hL
λ uL
Figure 4.1 Sketch of a propagating wave.
Thus
∂ϕ q2 p
þ þ þ gh ¼ FðtÞ ð4:24Þ
∂t 2 ρ
This is the Bernoulli equation for an incompressible fluid. It neglects viscous effects and
assumes an irrotational flow.
4.2 Propagation of small-amplitude two-dimensional waves
Consider a two-dimensional wave propagating on a liquid layer in a channel, as depicted

in Figure 4.1. The heights of the channel, the liquid space and the gas space for the
undisturbed flow are B, hL, hG. Plug flows with velocities uL and uG are also assumed.
The coordinate y is perpendicular to the wall and y0 = hL is the average location of the
interface. The displacement from y0 is designated by η. It is convenient to represent η by
the real part of
¼ a exp ik ðx ctÞ ð4:25Þ
The amplitude a and the wave number, k = 2π/λ are real. The wave velocity is complex
c ¼ cR þ icI ð4:26Þ
where cR is the velocity with which the wave is propagating downstream.

If (4.26) is substituted into (4.25)
¼ ½a exp ik ðx cR t Þ½exp kcI t ð4:27Þ
It is noted that if cI is positive the disturbance is growing in time. Now

exp ikðx cR tÞ ¼ cos kðx cR tÞ þ i sin kðx cR tÞ ð4:28Þ
Therefore, the real part of (4.27) is

¼ ½a cos k ðx cR t Þ expðkcI tÞ ð4:29Þ
This represents a wavy disturbance with wave number k and a propagation velocity cR
that is growing or decaying in time.
We are interested in calculating the velocity field introduced by the waves. The
assumption of an inviscid, incompressible, irrotational flow is made. The velocities in
the gas and liquid are represented by
4.2 Propagation of small-amplitude two-dimensional waves 75
!
u L ¼ uL þ !
u L0 ð4:30Þ
u ¼u þ!
! u0 ð4:31Þ
G G G
uL0 , !
where uL , uG are the averages and ! uG0 are the disturbances. Equation (4.9) can be used
to calculate the two-dimensional velocity so that
ϕ ¼ ϕ þ ϕ0 ð4:32Þ
and
∂2 ϕ0L ∂ϕ0L
þ 2 ¼0 ð4:33Þ
∂x2 ∂y
∂2 ϕ0G ∂2 ϕ0G
þ ¼0 ð4:34Þ
∂x2 ∂y2
Two of the boundary conditions are that the velocities of the gas and liquid normal to the
walls are zero:
∂ϕ0G
¼ 0 at y ¼ hG ð4:35Þ
∂y
∂ϕ0 L
¼0 at y ¼ hL ð4:36Þ
∂y
Two other conditions at the interface are needed.

One of these is that the normal velocities of the fluids on both sides of the interface are
equal to the normal velocity of the interface. The substantial derivative follows the
change of a property of a fluid particle with time. The time change of the height of a
particle at the interface is the normal velocity of that particle. Thus
0
D 0 ∂ϕ L
¼ ðuLy Þi ¼ ð4:37Þ
Dt ∂y i
where, by definition, η is a function only of x and t. Thus, from (4.13)

0
∂ ∂ ∂ ∂ϕL
þ uL þ u0L ¼ ð4:38Þ
∂t ∂x ∂x ∂y i
This is known as the kinematic condition at the interface. The assumption of

small-amplitude waves is made so that η / a, as do ϕ0 and u0Lx . Thus, all the terms in
(4.38) vary with a, except the third term on the left side, which varies as a2. Since a is
assumed to be a small number (4.38) simplifies to
0
∂ ∂ ∂ϕL
þ uL ¼ ð4:39Þ
∂t ∂x ∂y i
Since the normal velocity of the gas at the interface equals the normal velocity of the
liquid,
76 Inviscid waves
0
∂ ∂ ∂ϕG
þ uG ¼ ð4:40Þ
∂t ∂x ∂y i
One would like to apply (4.39) and (4.40) at a fixed location, y = 0, rather than at the
interface. A Taylor series expansion yields
0 0 2 0
∂ϕL ∂ϕL ∂ϕ
¼ þ þ ... ð4:41Þ
∂y i ∂y y¼0 ∂y2 y¼0
If η is a small number the second term on the right side of (4.41) can be neglected.
The linearized form of the kinematic conditions
0
∂ ∂ ∂ϕL
þ uL ¼ ð4:42Þ
∂t ∂x ∂y 0
0
∂ ∂ ∂ϕG
þ uG ¼ ð4:43Þ
∂t ∂x ∂y 0
specifies the derivatives of ϕ0L and ϕ0G at y = 0. Thus the disturbed velocity fields in the
liquid and in the gas can be calculated with equations (4.33) and (4.34) using boundary
conditions (4.42), (4.43), (4.35), (4.36).
Assume a solution of (4.33) of the form
^ ðyÞexp ikðx ctÞ
ϕ0L ¼ ϕ ð4:44Þ
L
^ needs to be linearly dependent on amplitude a in order to satisfy (4.42). After

Note that ϕ L
^L ðyÞ is obtained.
substituting into (4.33), an equation for ϕ
^
d2 ϕ L ^ ¼0
k2ϕ ð4:45Þ
L
dy2
If (4.25) is substituted into the kinematic boundary conditions, (4.42) and (4.43)
!
^
∂ϕ
i a k ðc uL Þ ¼ L
ð4:46Þ
∂y
0
!
^G
∂ϕ
i a k ðc uG Þ ¼ ð4:47Þ
∂y
0
Two fundamental solutions of (4.45) are exp(ky) and exp(−ky). These are used if an
infinite field is considered. For a finite field, it is more convenient to use hyperbolic
functions
1
sinh ðkyÞ ¼ ½exp ðkyÞ exp ðkyÞ ð4:48Þ
2
which equals 0 at y = 0 and
4.3 Dispersion relation for propagating waves 77
1
cosh ðkyÞ ¼ ½expðkyÞ þ exp ðkyÞ ð4:49Þ
2
which equals 1 at y = 0. Thus the solution for the flow field in the liquid is
^ ¼ ½AsinhðkyÞ þ ½BcoshðkyÞ
ϕ ð4:50Þ
L
In order to satisfy the boundary conditions (4.46) and (4.36)

½A ¼ iðuL cÞa ð4:51Þ
½B ¼ iðuL cÞ cothðkhL Þa ð4:52Þ
Following the same procedure, equation (4.34) can be solved to give

^G exp ikðx ctÞ
ϕ0G ¼ ϕ ð4:53Þ
^ ¼ ½DsinhðkyÞ þ ½E coshðkyÞ
ϕ ð4:54Þ
G
where
½D ¼ iðuG cÞa ð4:55Þ
½E ¼ iðuG cÞcothðkhG Þa ð4:56Þ
4.3 Dispersion relation for propagating waves
An important use of the results in Section 4.2 is the derivation of the relation between
the wave velocity, c, and the wave number, k. A starting point is a consideration of the
Laplace formula, which specifies the pressure difference across an interface separating
two media (Landau & Lifshitz, 1959).
Take, for example, a gas–liquid interface. Define R1and R2 as the principal radii of
curvature of the surface, where R is positive for a convex liquid surface and negative for a
concave liquid surface. According to the Laplace formula the pressure change across the
interface is

1 1
pLi pGi ¼ σ þ ð4:57Þ
R1 R2
where σ is the surface tension. For example, a bubble with radius rP has radii of curvature
that are negative and equal to −rP so that
2σ
pGi pLi ¼ ð4:58Þ
rP
Thus, the pressure in the gas is greater than the pressure in the liquid. This can be
understood by considering the sketch of one half of a bubble shown in Figure 4.2. The
force due to the surface tension by the half which is not pictured is σ2πrP. This force is
78 Inviscid waves
rp
Figure 4.2 Sketch of one half of a bubble.
balanced by the force due to the pressure difference between the gas and the liquid,
that is, the product of Δp and the projected area, Dpπr2P . Equating these two forces
gives (4.58).
For a wavy interface, such as shown in Figure 4.1, radii of curvature can be approxi-
mated for small wave amplitudes as indicted below (Landau & Lifshitz, 1959):
2
∂ ∂2
pLi pGi ¼ σ 2 2 ð4:59Þ
∂x ∂z
For a two-dimensional wave this simplifies to

∂2
pLi pGi ¼ σ ð4:60Þ
∂x2
The pressures pLi and pGi can be calculated by using the Bernoulli equation (4.24).
Let the reference level y0 be equal to zero. At t = 0, ∂ϕ ∂t = 0 and the fluid is at rest.
Define p0 = p at y = 0 and t = 0. Then, from the liquid and gas phase Bernoulli equations,
p0 u2L
F L ðt Þ ¼ þ ð4:61Þ
ρL 2
p0 u2G
F G ðt Þ ¼ þ ð4:62Þ
ρG 2
The Bernoulli equation (4.24) for the liquid can then be written as

∂ϕ 1 1 p p0 u2
0 ¼ þ u2Lx þ u2Ly þ gy þ L L ð4:63Þ
∂t 2 2 ρL 2
1 2 1 2 u2 u02
uLx ¼ uL þ u0Lx ¼ L þ uL u0Lx þ Lx ð4:64Þ
2 2 2 2
02
uLy ¼ uLy
2
ð4:65Þ
Because of the assumption of small-amplitude waves, uLx02 and uLy02 are negligible so
(4.63) can be rewritten as
∂ϕ0L p0 p0
0¼ þ uL u0Lx þ gy þ L ð4:66Þ
∂t ρL
4.4 Propagation of waves at the interface of horizontal stationary fluids 79
Similarly, for the gas

∂ϕ0G p0 p0
0¼ þ uG u0Gx þ gy þ G ð4:67Þ
∂t ρG
where, from (4.6)

∂ϕ0L ∂ϕ0G
u0Lx ¼ u0Gx ¼ ð4:68Þ
∂x ∂x
The interface is located at y = η. The substitution of (4.66) and (4.67) into (4.60) yields
0 0 0 0
∂2 ∂ϕL ∂ϕG ∂ϕL ∂ϕG
σ 2 ¼ ρL ρG þ ρL uL ρG uG ðρL ρG Þg y
∂x ∂t i ∂t i ∂x i ∂x i
ð4:69Þ
Because of the assumption of small-amplitude waves

0 0 0 0
∂ϕL ∂ϕL ∂ϕG ∂ϕG
¼ ¼ ð4:70Þ
∂t y¼ ∂t y¼0 ∂t y¼ ∂t y¼0
0 0 0 0
∂ϕL ∂ϕL ∂ϕG ∂ϕG
¼ ¼ ð4:71Þ
∂x y¼ ∂x y¼0 ∂x y¼ ∂x y¼0
Equations (4.25), (4.44) and (4.53), with ϕ ^ and ϕ ^ given by (4.50) and (4.54), are
L G
substituted into (4.69). Using the approximations given by (4.70) and (4.71), the follow-
ing relation is obtained:
σk 2 a ¼ ρL ikc½B þ ρG ikc½E þ ρL uL i½B ρG uG ik ½E ðρL ρG Þg y a ð4:72Þ
Since constants [B] and [E] are given by (4.52) and (4.56) the following equation, relating
c to k, is obtained:
σk 2 þ ðρL ρG Þgy ¼ ρG ðuG cÞ2 k cothðkhG Þ þ ρL ðuL cÞ2 k cothðkhL Þ ð4:73Þ
which is equation (2.6) in Chapter 2.
4.4 Propagation of waves at the interface of horizontal stationary fluids
Consider (4.73) for the case of uL ¼ 0, uG ¼ 0 and ρG much smaller than ρL.

σk g
c ¼
2
þ tanhðkhL Þ ð4:74Þ
ρL k
where the subscript y on the gravity term is dropped since a horizontal channel is being
considered. Equation (4.74) is the classical relation between wave velocity and the
wavelength. It describes a situation for which the stabilizing effects of gravity and surface
tension balance destabilizing effects of inertia.
80 Inviscid waves
30
29
28
Theory
27
c (cm/s)
26
25
24 Measured using a dipper

Measured using an
23 electric wave generator
22
0 10 20 30 40 50 60
Frequency = kc (s–1)
Figure 4.3 Dispersion relation for distilled water at 23.8 °C.
For khL ≥ 2; tanh ðkhL Þ ¼ 1, so

kσ g
c2 ¼ þ ð4:75Þ
ρL k
This is called the deep liquid assumption, even though khL= 2 corresponds to λ/hL = π.
Equation (4.75) is plotted in Figure 4.3 for air–water at atmospheric pressure. Agreement
between theory and the experiments of Kim (1968) is noted. Small wavelengths
for which kσ/ρL is much greater than g/k are called capillary waves. The wave velocity
increases with decreasing wavelength or increasing frequency. Gravity waves, for which
the wave velocity increases with increasing wavelength, or decreasing frequency, are
observed for g/k is much greater than kσ/ρL. Thus, there is a wavelength at which a
minimum in the wave velocity occurs. As seen in Figure 4.3, this occurs at λ = 1.78 cm
and c = 23.5 cm/s for an air–water interface. The Reynolds number for this system at
the minimum is λc/L = 4200, which supports the assumption that viscous effects can
be ignored.
For small khL and tanh(khL) =1
2
k hL σ
c ¼
2
þ ghL ð4:76Þ
ρL
These waves are known as shallow liquid waves. If effects of surface tension can be
ignored,
c2 ¼ ghL ð4:77Þ
The velocity of shallow gravity waves increases with liquid height and is independent of
wavelength.
4.5 Kelvin–Helmholtz instability
4.5.1 General conditions for a KH instability

The wave velocity, calculated above, for small-amplitude waves at a gas velocity of zero
is real. That is, the system is neutrally stable. This is not the case for high gas velocities,
for which c can be complex. This situation is called a Kelvin–Helmholtz instability,
which can have a number of outcomes. It has been used widely to explain phenomena in
multiphase systems.
Define c0 as the velocity when uL ¼ 0; uG ¼ 0:
σk 2 þ ðρL ρG Þg
c20 ¼ ð4:78Þ
k ρL þ ρG
ρL ¼ ρL cothðhk L Þ ρG ¼ ρG cothðkhG Þ ð4:79Þ
It is noted that (4.78) differs from (4.75) because the upper fluid need not be a gas. Thus,
the density ρG is assumed to be smaller than ρL but not negligible.
The general relation (4.73) can be written as follows:

ρG þ ρL c20 ¼ ρG u2G 2uG c þ c2 þ ρL u2L 2uL c þ c2 ð4:80Þ
This can be solved for c to get

" #1
ρG uG þ ρL uL ρ
ρ ð u G u L Þ 2 2
c¼ c20 G L 2 ð4:81Þ
ρG þ ρL ρ þ ρ G L
From (4.78), c20 is a positive number so long as ρL > ρG. From (4.81), it is seen that c is a
complex number if
ρG ρL ðuG uL Þ2

c20 5 2 ð4:82Þ
ρG þ ρL
Equation (4.82) defines the ðuG uL Þ needed to initiate a Kelvin–Helmholtz instability.

The time constant characterizing the growth rate of the wave, defined by (4.27), is
" #12
ρG ρL ðuG uL Þ2
kcI ¼ k c0
2
ð4:83Þ
ðρG þ ρL Þ2
The velocity of unstable waves, defined by the real part of (4.81), is

ρG uG þ ρL uL
cR ¼ ð4:84Þ
ρG þ ρL
82 Inviscid waves
pG
σ σ
g
Figure 4.4 Forces acting at a wave crest.
4.5.2 Gas–liquid flows with deep gas and liquid layers

A consideration of a Kelvin–Helmholtz instability of waves at an interface dividing deep
gas and liquid layers with ρG much smaller than ρL provides a good example. For this
case, ρG ¼ ρG and ρL ¼ ρL . Equation (4.81) simplifies to
12
ρ uG þ ρ L uL σk g ρG
c¼ G þ ð uG u L Þ 2 ð4:85Þ
ρL ρL k ρL
At small ðuG uL Þ2 the wave velocity is real; it decreases with increasing ðuG uL Þ.
Eventually a critical condition is reached where
ρG σk g
ð uG uL Þ 2 ¼ þ ð4:86Þ
ρL ρL k
The wave velocity, c, will be complex for large ðuG uL Þ.

This can be interpreted as follows: The streamlines representing the gas flow are
squeezed at the crest. This leads to an increase in the gas velocity and, from a consid-
eration of the Bernoulli equation, a decrease in the gas pressure. As depicted in
Figure 4.4, the left side of (4.86) represents the magnitude of the destabilizing effect of
this suction. The right side represents the stabilizing effects of gravity and surface
tension. The imbalance of these forces leads to an instability and a rapidly growing
wave. At neutral stability and for all super-critical conditions
ρG uG þ ρL uL
cR ¼ ð4:87Þ
ρL
ρG σk 3
k 2 c2I ¼ k 2 ð uG uL Þ 2 gk ð4:88Þ
ρL ρL
Figure 4.5 sketches the above results for a deep liquid in a plot of wavelength, λ, versus
ðuG uL Þ. The solid curve indicates neutral stability. The domain inside this curve
represents conditions for which cI > 0. The critical ðuG uL Þ below which cI = 0 for
all wavelengths is indicated by the solid vertical line. This can be obtained by taking the
derivative of (4.86) and setting d ðuG uL Þ=dk ¼ 0 to find the extremum
ρ g 12
k critical ¼ L ð4:89Þ
σ
hL
B
6.66 m/s
λ (cm)
cI > 0
1.7 cm
Maximum
growth
(uG – uL) m/s
Figure 4.5 Sketch of the neutral stability curve for air flowing over water.
The critical ðuG uL Þ corresponding to kcritical is obtained by solving (4.86).

For air–water at atmospheric conditions
λcritical ¼ 1:7 cm ð4:90Þ
ðuG uL Þcritical ¼ 6:66 m=s ð4:91Þ
As can be seen from (4.27), the growth rate is represented by kcI. From (4.88), the wave
number at which the growth rate is a maximum for a given ðuG uL Þ is
ρG 3σk M g
ρL 2ρL 2k M
The loci of maximum growth are sketched in Figure 4.5 by the dotted curve. The
argument is usually given that for ðuG uL Þ larger than the critical the waves that appear
are the most rapidly growing ones given by (4.92).
4.5.3 Effect of hG
For very long wavelengths (coth khL = 1khL), the deep liquid assumption is not valid. The
critical condition predicted by KH theory simplifies to
ρL
ðuG uL Þ2critical ¼ ghL ð4:93Þ
ρG
Thus, the neutral stability curve shown in Figure 4.5 for coth khL = 1 needs to be modified
by including the dashed curves which are sketched on the figure, where hL/B is a
parameter and σ/ρL, g, ρG/ρL, B are specified. For small λ, the plot is independent of
hL/B. However, for large λ the neutral stability changes so that (4.93) is satisfied.
As discussed in Section 2.3, equation (4.93) has been used to predict the initiation of
slugging. The predicted critical gas velocity is found to be too large. Better agreement
with observation is obtained if viscous effects are considered for very large wave-
lengths (viscous large-wavelength theory). The dashed curves then move to the left
(see Andritsos et al., 1989).
84 Inviscid waves
4.6 Experimental verification of the KH instability
There are not many clear-cut experiments that display the Kelvin–Helmholtz instability.
This is because waves are usually present on the interface before the KH critical gas
velocity is attained. Thus, the discussion of waves in an air–water stratified flow in
Section 2.2.2 characterizes the critical condition as a change of the character of the waves
from “regular” to “irregular” for air–water flows.
However, as pointed out in Chapter 2 “regular waves” observed for air–water flow are
not observed for high-viscosity liquids. Therefore, KH critical conditions are realized
when the liquid interface is smooth. This observation was exploited by Andritsos et al.
(1989), who used horizontal pipelines with diameters of 2.52 cm and 9.53 cm to study
stratified flows of air and water–glycerine solutions. For a 70 cP liquid, the interface was
smooth at low gas velocities. At a critical gas velocity, sinusoidal disturbances with a
wavelength of 1–2 cm appeared. This is to be compared with KH theory which predicts
λcritical = 1.7 cm at the critical gas velocity of 6.66 m/s for air–water. Depending on the
height of the stratified liquid layer, the KH instability leads to tumbling large-amplitude
waves or to slugs.
Figure 4.6 is taken from the paper by Andritsos et al. (1989). The data indicate either
the transition to slugs (large hL/dt) or the initiation of large-amplitude waves (small hL/dt).
Good agreement is found between the KH analysis and experiment. It is noted that both
experiment and theory indicate that transition is independent of pipe diameter (as already
shown in Figure 2.7).
0.9
dt = 9.53 cm, 100 cP
dt = 2.52 cm, 70 cP
0.7
hL / dt 0.5
Kelvin–Helmholtz
instability
0.3
0.1
0.2 0.5 1.0 2.0 5.0 10 20 50
uGS (m/s)
Figure 4.6 Comparison of Kelvin–Helmholtz stability calculations with observed initiation of slugs or
large-amplitude waves on 70 cP (dt = 2.52cm) and 100 cP (dt = 9.53cm) liquids. Andritsos &
Hanratty, 1989.
Earlier experiments by Francis (1954, 1956) used an oil with a viscosity of 2.2 Poise
and a surface tension of 34 dynes/cm. Air was blown over the oil surface. Both the air and
oil layers were deep. For this liquid, KH theory predicts a critical air velocity of 5.16 m/s,
a critical wavelength of 1.27 cm and a critical wave velocity of 0.67 cm/s. A transition
from a smooth interface to one with small-amplitude waves with λ ≈ 1cm and cR= 1 cm/s
was observed when the air velocity was 9.67 m/s at 8 cm above the interface. This critical
velocity is higher than predicted by KH theory.
However, the very large liquid viscosity used in the experiments of Francis would
greatly decrease the growth rates of waves. Therefore, a much longer channel might be
needed to observe the critical condition. (It should be pointed out that Francis first
observed waves at the end of his channel.)
4.7 Group velocity
4.7.1 Physical interpretation of group velocity

Large energies can be associated with wave motion at gas–liquid interfaces. The group
velocity, cg, is related to the speed with which this energy is transported downstream.
It can be different from the wave velocity and is important in understanding many
phenomena. Two-dimensional gravity waves, in a situation for which there is no mean
flow and the waves have a small amplitude, will be discussed. A unit width in the
spanwise direction is considered. Waves can have kinetic energy, T, associated with
fluctuations in the velocity and a potential energy, P, associated with the displacement of
the interface in a gravitational field.
The wave motion being considered is depicted in Figure 4.1, with uL ¼ 0 and
uG ¼ 0. The displacement of the interface from its average location, hL, is given as
¼ a cos ðkx ωtÞ ð4:94Þ
where ω =kc is the circular frequency. Consider a liquid element, dxdy, in the region
where η is positive. The net force due to gravity and buoyancy is −g(ρL −ρG)dxdy for a
unit width in the z-direction. The work done to lift this element to a height y above the
average location of the interface is g(ρL −ρG)ydxdy. Integration with respect to y in the
region where the interface has a positive displacement gives g(ρL −ρG)(η2/2)dx.
Similarly, in regions where the interface has negative displacements the net force
on a gaseous element due to gravity and buoyancy is given by g(ρL −ρG)dxdy. Since y
will have negative values, the work needed to move a gaseous particle to a location y is
–gy(ρL −ρG)dydx. Integration between y =η and y = 0 gives g(ρL −ρG)(η2/2)dx. Therefore,
the potential energy associated with one wavelength is
ðλ
2 1
P ¼ gðρL ρG Þ dx ¼ a2 gρL λ ð4:95Þ
0 2 4
where ρG has been neglected and η is a function of x and t, given by (4.94).

The kinetic energy in one wavelength is given by
86 Inviscid waves
ðh ðλ
ρL q2 1
T¼ dydx ¼ a2 gρL λ ð4:96Þ
0 0 2 4
where
q2 ¼ u2x þ u2y ð4:97Þ
and ux, uy are defined with (4.6), (4.44), (4.50). The total energy is
1
E ¼ P þ T ¼ a2 gρL λ ð4:98Þ
2
This result is independent of hL so it is valid both for shallow and deep layers.
Now consider the rate at which fluid to the left of a y–z plane at a given x is doing work
on the fluid to the right side of the plane. Again, a unit slice in the z-direction is
considered. The force on a small area perpendicular to the direction of propagation is
pdy. The rate of work on this differential area is pux dy. The rate of work over the plane at
a given time can be obtained by integrating the force over the area
ð hL þ
Rate of work ¼ pux dy ð4:99Þ
0
Integrate over one period to obtain a time-averaged rate of work. The Bernoulli equation
(4.24) gives p = ∂ϕL/∂t if terms of second order are neglected. Since ux is given by (4.6),
the average rate of work over one period is given by
ð ð
ω 2π=ω yL ∂ϕL ∂ϕL
Average rate ¼ dydt ð4:100Þ
2π 0 0 ∂t ∂x
Since
^ cos ðkx ωtÞ
ϕL ¼ ϕ ð4:101Þ
L
equation (4.100) can be written as

ð ð
ω 2π=ω yL ^2 ðyÞdydt
Average rate ¼ fsin 2 ðkx ωt Þgkωϕ L ð4:102Þ
2π 0 0
where ^ϕL ðyÞ is given by (4.50). The following result is obtained:

1 2khL
Average rate ¼ gρL a2 c 1 þ ð4:103Þ
4 sinh 2khL
where the relation

2cosh khL sinh khL ¼ sinh 2khL ð4:104Þ
has been used (see Lamb, 1953, p. 383, equation 11).


The energy per length in the spanwise direction is 12 a2 gρL . Define a group velocity as
the velocity with which this energy propagates. Thus

1 1 2khL
cg a2 gρL ¼ gρL a2 c 1 þ ð4:105Þ
2 4 sinh 2khL
Solve for cg.

c 2khL
cg ¼ 1þ ð4:106Þ
2 sinh 2khL
Thus, for gravity waves on a deep liquid

c
cg ¼ ð4:107Þ
2
For a shallow liquid (small khL)
cg ¼ c ð4:108Þ
Thus, the group velocity is less than the phase velocity for gravity waves on a deep liquid.
Consider the packet of gravity waves on a deep liquid shown in Figure 4.7. The packet
would move downstream at a velocity cg. The waves in the packet have a velocity c,
which is greater than cg, so the waves are formed at the front of the packet and disappear
at the back. If the packet contains more than one wavelength (or phase velocity), the
group of waves could break up to form two or more packets. Experiments with wave
packets have been reported by Kim (1970) and by Kim & Hanratty(1971). Waves were
generated at the center of a long tank. A brass rod with a diameter of 3/16 inch that
extended over the whole width of the tank was placed 5 mm above the water surface. A
constant 9 kV voltage with a superposed AC signal between the electrode and the water
was applied. By modulating the AC voltage, a wave packet with a maximum steepness of
about 0.03 could be formed.
A general relation for group velocity is given in Lamb (1953, p. 381) as
dω dc
cg ¼ ¼cþk ð4:109Þ
dk dk
cg
c gravity waves
c capillary waves
Figure 4.7 Motion of a packet of waves at the interface of a deep liquid.

88 Inviscid waves
since ω = kc. This is valid for all types of waves. Sound waves are non-dispersive, that is,
c is independent of k, so
cg ¼ c ð4:110Þ
For capillary waves on deep water

kσ
c2 ¼ ð4:111Þ
ρL
and
3
cg ¼ c ð4:112Þ
2
Thus, for deep capillary waves, the group velocity is larger than the phase velocity.
Capillary waves in a packet would be formed at the back of the packet and disappear at
the front.
4.7.2 Wave resistance

A boat moving through a liquid with velocity c forms gravity waves in its rear. The
consequence is that more power is needed. That is, the waves may be considered to
provide an additional resistance.
A sketch of a very simplified example is shown in Figure 4.8. The energy of the waves
per unit length is given by (4.98). The length of the wave train produced per unit time is
the velocity of the object, if the train is attached. Thus, the energy per unit time needed to
create this wake is

1
energy=time ¼ c ρL ga 2
ð4:113Þ
2
The energy flowing through the dashed reference plane shown in Figure 4.8 is
Energy
= cg 12 ρLga 2
Time
Figure 4.8 Wave train behind a partially submerged body moving with a velocity c.

1 2
energy=time ¼ cg ρL ga ð4:114Þ
2
The work needed to produce the waves can be written as the product of the velocity of the
object and a wave resistance, R.
1 1
c ρL ga2 ¼ cg ρL ga2 þ Rc ð4:115Þ
2 2
Thus, the wave resistance is given by
c cg 1
R¼ ρ ga2 ð4:116Þ
c 2 L
For gravity waves,

1 2khL
R ¼ ρL ga 1
2
ð4:117Þ
4 sinhðkhL Þ
If the body moves faster than the maximum wave velocity, (ghL)1/2, it cannot have an
attached train. Wave resistance vanishes.
4.7.3 Waves which decay in space, rather than time

The wave height has been given as
¼ a exp ikðx ctÞ ð4:118Þ
¼ a exp iðkx ωtÞ ð4:119Þ
where ω is the circular frequency and is equal to kc. The analysis in this chapter
represents the interfacial displacement given by (4.118) with k real and c complex. As
shown in (4.29) this represents a wave pattern growing or decaying in time.
The more usual observation is for waves growing or decaying as they propagate.
This is represented by (4.118), (4.119) with k complex and c or ω real. The analysis
with k complex is more difficult than the one with c complex. Fortunately, work
by Huerre & Monkewitz (1985) and by Gaster (1962) show that, for small growth
rates,
k R cI
kI ¼ ð4:120Þ
cg
where cg is the group velocity. Thus, the analysis for temporal changes (c complex)
can be used to calculate spatial changes (k complex). This result will be used in
Chapter 6.
90 Inviscid waves
4.8 Uses of the notion of a KH instability
4.8.1 Initiation of slugging

The usual explanation for the initiation of slugging is the growth of an unstable wave.
However, the final stage in the closing of the air space to form a slug occurs when the gas
velocity above the crest of the wave reaches the critical value for a KH instability. This is
particularly evident for transitions in downwardly inclined pipes, Section 5.5.3, for
which the instability is initiated in a stratified configuration at gas flows less than the
critical KH velocity. It also determines the critical height at the inlet of an upwardly
inclined pipe. For this situation, reversed flow causes an intermittent buildup at the
upstream end of the pipe (Section 5.5.2).
4.8.2 Rate of atomization

Taylor (1940) argued that air flowing over a liquid surface can atomize it by removing
small ripples through a KH instability. This has been supported by photograpic studies of
Woodmansee & Hanratty (1969). See Chapter 11.
Taylor suggested that the volumetric rate of removal varies as the wavelength cubed
and the time constant characterizing the rate of growth of the waves
λ3 1
QA / / ð4:121Þ
tc k 3 tc
The mass rate per unit area then satisfies the following:
ρL ρ
RA / / L ð4:122Þ
k tc λ
3 2 kt c
The time-scale is defined as

t 1
c / kcI ð4:123Þ
where, for capillary ripples,

ρG kσ
c2I ¼ ð uG uL Þ 2 ð4:124Þ
ρL ρL
(See equation (4.88).) Consider small uL . Then

1
ρL ρG 2 k
RA / k uG 1 ð4:125Þ
k ρL k0
or

1=2 1=2 k
RA / ρL ρG uG f ð4:126Þ
k0
4.8 Uses of the notion of a KH instability 91
where
ρG u2G
k0 ¼ ð4:127Þ
σ
Taylor included effects of liquid viscosity, which would dampen the wave growth, so that

RA k
1=2 1=2
¼ f ; θ ð4:128Þ
ρL ρG uG k0
where
ρL σ 2
θ¼ ð4:129Þ
ρG μ2L u2G
Increasing θ causes a decrease in RA. For θ >10 the influence of viscosity can be
0:5 1
neglected since f in (4.128) varies approximately as 1 þ θ . Then

RA kσ
¼f ð4:130Þ
1=2 1=2
ρ L ρ G uG ρG u2G
For deep water, one can argue that k should represent the fastest growing wave. From
(4.92)
ρG u2G
kM / ð4:131Þ
σ
so kσ=ρG u2G of the atomizing capillary waves is constant and, for θ < 10,
RA
1=2 1=2
ρ G ρ L uG
This result can be applied to annular flows where the wall film is characterized by the
intermittent appearance of disturbance waves. The photographic studies by Woodmansee
& Hanratty (1969) show that atomization occurs by the removal of waves in these
disturbances. The wavelengths of the atomizing waves are less than λmax calculated for
thick films since viscous effects associated with the presence of the wall limits the
maximum possible wavelength that can exist. Therefore, measured RA are less than
predicted by (4.132). Tatterson (1975) and Schadel & Hanratty (1989) argued that the
unstable waves have lengths that scale with the film height so that

RA ρG u2G hL
¼ If ð4:133Þ
1=2 1=2
ρ ρ uG σ
L G
where I is an intermittency factor which represents the fraction of the time that disturb-
ance waves are present. This equation is complicated to use and the assumptions made to
apply Taylor’s theory to annular flows have not been substantiated. Therefore, it has not
been successfully used for annular flows.
92 Inviscid waves
The empirical equation below (Pan & Hanratty, 2002) has been employed in flows
for which the rate of atomization varies linearly with the mass flow rate of the wall
layer
RA ðW LF W LFC Þ
1=2 1=2
¼ kA ð4:134Þ
ρG ρL u G Pσ
Equation (4.134) has similarities to (4.133) in that the left sides are the same and RA
varies inversely with surface tension. The representation of the effect of σ as given in
(4.134) has been motivated by the papers of Lopez de Bertodano et al. (1997) and Assad
et al. (1998).
4.8.3 Interfacial stress in horizontal stratified flows

The drag for stratified flows can be larger than what is found for a smooth surface if
waves are present. This is particularly evident at velocities larger than required for a
Kelvin-Helmholtz instability (uG = 6–7 m/s for air–water flow at atmospheric condi-
tions). This is because large amplitude irregular waves evolve from this instability
(see Chapter 5).
References
Andritsos, N., Williams, L. & Hanratty, T. J. 1989 Effect of liquid viscosity on the stratified-slug
Assad, A., Jan, C. S., Lopez de Bertodano, M. & Beus, S. 1998 Scaled entrainment measurements
in ripple-annular flow in a small tube. Nucl. Eng. Des. 184, 437–447.
Francis, J. R. D. 1954 Wave motions and the aerodynamic drag on a free oil surface. Phil. Mag., Ser.
7, 45: 366, 695–702.
Francis, J. R. D. 1956 Wave motions on a free oil surface. Phil. Mag., Ser. 8, 47: 7, 685–688.
Gaster, M. 1962 A note on the relation between temporally-increasing and spatially-increasing
disturbances in hydrodynamic instabililty. J. Fluid Mech. 14, 222–224.
Huerre, P. & Monkewitz, P. A. 1985 Absolute and convective instabilities in free shear layers.
J. Fluid Mech. 159, 151–168.
Kim, Y. Y. 1968 Wave motion on viscoelastic fluids. M.Sc. thesis, University of Illinois.
Kim, Y. Y. 1970 Non-linear effects in wave motion. Ph.D. thesis, University of Illinois.
Kim, Y. Y. & Hanratty, T. J. 1971 Weak quadratic interactions of two-dimensional waves. J. Fluid
Mech. 50, 107–132.
Lamb, H. 1953 Hydrodynamics. 6th edn, Cambridge: Cambridge University Press.
Landau, L. B. & Lifshitz, E. M. 1959 Fluid Mechanics. Reading, MA: Addison-Wesley, p. 231.
Lopez de Bertodano, M. A., Jan, C. S. & Beuss, G. G. 1997 Annular flow entrainment rate experi-
ment in a small vertical pipe. Nucl. Eng. Des. 178, 61–70.
Pan, L. & Hanratty, T. J. 2002 Correlation of entrainment for annular flow in vertical pipes. Int.
Schadel, S. A. & Hanratty, T. J. 1989 Interpretation of atomization rates of the liquid film in
gas–liquid annular flow. Int. J. Multiphase Flow 15, 893–900.
References 93
Tatterson, D. F. 1975 Rates of atomization and drop size in annular two-phase flow. Ph.D. thesis,
University of Illinois.
Taylor, G. I. 1940 Generation of ripples by wind blowing over a viscous fluid. Reprinted in The
Scientific Papers of Sir Geoffrey Ingram Taylor, Vol. 3, ed. G. K. Batchelor. Cambridge:
Cambridge University Press (1963), pp. 244–252.
Woodmansee, D. E. & Hanratty, T. J. 1969 Mechanism for the removal of droplets from a liquid
surface by a parallel air flow. Chem. Eng. Sci. 24, 299–307.
5 Stratified flow
5.1 Scope
This chapter describes the stratified pattern observed in gas–liquid flows, for which liquid
flows along the bottom of a conduit and gas flows along the top. The gas exerts a shear
stress on the surface of the liquid. It is desired to calculate the average height of the liquid
layer and the pressure gradient for given liquid and gas flow rates. The flow is considered
to be fully developed so that the height of the liquid is not changing in the flow direction
and the pressure gradient is the same in the gas and liquid flows.
In order to consider stratified flow in circular pipes, the simplified model of the flow
pattern, presented by Govier & Aziz (1972), is exploited. The interface is pictured to be
flat. At large gas velocities, some of the liquid can be entrained in the gas. This pattern is
considered in Section 12.5 entitled “the pool model” for horizontal annular flow.
The approach taken in utilizing the simplified model is to consider the interfacial stress
as given and, then, to calculate the behavior of the liquid phase. It will turn out that the
prediction of the interfacial stress is the principal problem to be addressed.
At high gas velocities, the liquid layer thins out and climbs up the wall of the pipe. The
interface is not flat, so the simplified model could be a poor one. Under these circum-
stances, the flow could display a behavior intermediate between a stratified and an
annular flow. Thus, the treatment of the liquid distribution in the wall layer of a horizontal
annular flow (discussed in Chapter 12) could be a good starting point for considering
stratified flows with curved interfaces.
Kelvin–Helmholtz waves can cause the interfacial stress to be much larger than would
be realized if the interface were smooth. Experiments show that this increase in stress is
related to the steepness of the waves. Results from inviscid flow theory are used to
develop an equation for the wave height – and, therefore, to determine the parameters
affecting interfacial stress.
Small inclinations of a pipeline can have a strong effect on the behavior of a stratified
flow. Thus, separate subsections on inclined and declined pipes are presented.
5.2 Stratified flow in a rectangular channel
The main focus will be on circular pipes. However, because of its simplicity, the
rectangular channel will first be treated. Materials from Chapter 3 on film flows are used.
5.3 Stratified flow of gas and liquid in a pipe 95
The channel is assumed to be wide enough for a two-dimensional flow to be consid-

ered. As shown in Chapter 3, the variation of shear stress in the liquid layer is given by
dτ dp
¼ ρL g sin θ ð5:1Þ
dy dz
where θ is the inclination angle of the channel to the horizontal and z is the coordinate in
the direction of flow (Figure 5.5), ρL is the density of the bottom fluid and y is the distance
from the wall. This can be integrated to give the stress distribution

dp
τ ¼ τ i ð hL y Þ ρL g sin θ ð5:2Þ
dz
By using a Boussinesq approximation,

duL duL
τ ¼ μL þ μt ð5:3Þ
dy dy
where μL is the molecular viscosity and μt is the turbulent viscosity. Substituting (5.3)
into (5.2),

duL dp
ð μL þ μt Þ ¼ τ i ð hL y Þ ρL g sin θ ð5:4Þ
dy dz
Integrating from the wall where y = 0
ðy
τ i ðhL yÞ dpdz ρL g sin θ
uL ¼ dy ð5:5Þ
0 ð μL þ μt Þ
The volumetric flow per unit width can be obtained by integrating (5.5) from y = 0 to hL.
The interfacial stress is obtained from empirical relations or by considering a momen-
tum balance for the upper fluid for given flows of the gas and the liquid:

t duG dp
ð μG þ μ Þ ¼ τ W2 ðB yÞ ρG g sin θ ð5:6Þ
dy dz
where B is the height of the channel, τW2 is the stress on the top wall and y is the distance
from the bottom wall. Equations (5.5) and (5.6) are solved using the boundary conditions
and that velocities uG and uL are zero at the two walls, and that uG = uL and
μG ðduG =dyÞ ¼ μL ðduL =dyÞ at the interface.
For the case of laminar flows in the upper and lower phases, μt =0. The solution is
given in Bird, et al. (1960, p. 54).
5.3 Stratified flow of gas and liquid in a pipe
5.3.1 Geometric relations

For flow in a pipe, the simplifying assumption is usually made that the interface is flat.
Geometric parameters AG, AL are the areas of the pipe occupied by the gas and the liquid,
96 Stratified flow
Si is the length of the interface, PG is the length of the pipe wall in contact with the gas
phase, PL is the length of the pipe wall in contact with the liquid phase and hL is the
distance from the center of the interface to the wall.
The following relations are presented by Govier & Aziz (1972, pp. 563–565), where γ
is the angle sustended by two radii to the ends of Si

2hL
γ ¼ 2cos1 1 ð5:7Þ
dt
" 2 #1=2
Si hL hL
¼2 ð5:8Þ
dt dt dt
PL γ
¼ ð5:9Þ
πd t 2π
PG ¼ πd t PL ð5:10Þ
AL 1
¼ ðγ sin γÞ ð5:11Þ
At 2π
AG ¼ At AL ð5:12Þ
5.3.2 Laminar flow

The situation of laminar–laminar flow in a horizontal pipe is usually of interest for two
immiscible liquids. For a fully developed pipe flow, the assumption is made that secondary
flows can be ignored. Then, the streamwise velocities vary in two directions because the
geometries of the cross-sections require a variation of the velocity in two dimensions.
Thus, for a fully developed flow, the velocity fields in the upper and lower phases are
described by
∂2 u L ∂2 u L 1 dp
þ 2 ¼ ð5:13Þ
∂y 2 ∂x μL dz
∂2 u G ∂2 u G 1 dp
þ 2 ¼ ð5:14Þ
∂y2 ∂x μG dz
These equations are usually solved numerically. The interface is smooth and flat. The
boundary conditions are that the fluid velocities are zero at the walls and that the shear
stresses and velocities are equal at the interface, uL ¼ uG ; μL ð∂uL =∂yÞ ¼ μG ð∂uG =∂yÞ.
Examples are given in Govier & Aziz (1972, pp. 562–569).
5.3.3 Laminar liquid–turbulent gas flow in a pipe

The case of a laminar liquid and a turbulent gas in a pipe has been considered by Russel
et al. (1974). The simplified geometry described in Section 5.3.1 was used. In addition,
the stress was assumed to be constant along the interface (consistent with ignoring
secondary flows).
5.3.4 Model of Taitel & Dukler

The widely used correlation for frictional pressure drop and holdup developed by
Lockhart & Martinelli, discussed in Chapter 1, can overpredict pressure drops by as
much as 100% (Baker, 1954; Hoogendorn, 1959). Direct calculations of stratified flow in
a pipe are confronted with a number of problems, in addition to the modeling of turbulent
transport. The interface is not flat. Attempts to deal with this problem have been
described by Chen et al. (1997), Hart et al. (1989), Grolman & Fortuin (1997),
Vlachos et al. (1999). The interfacial stress is not constant along the interface. Suzanne
(1985) has shown that secondary flows can exist in the liquid. For cases in which the
liquid contains sediment, the secondary flow is evident by the appearance of a streak of
deposited particles along the bottom of the pipe.
Consequently, the simplified approach formulated by Taitel & Dukler (1976a,b) is
commonly used in design calculations for stratified flows. The geometric configuration
described in Section 5.3.1 is employed and the stress along the interface is assumed to be
constant. For fully developed flow in a pipe, one-dimensional momentum balance
equations are used to represent the pressure gradients in the gas and the liquid.

dp
AG τ WG S G τ i S i ρG AG g sin θ ¼ 0 ð5:15Þ
dz G

dp
AL τ WL S L τ i S i ρL AL g sin θ ¼ 0 ð5:16Þ
dz L
where the four terms represent forces in the z-direction due to the pressure gradient
(which is a negative quantity), the wall stress, the interfacial stress and gravity. Since the
flow is fully developed, ðdp=dzÞG ¼ ðdp=dzÞL ¼ dp=dz. The elimination of dp/dz
between (5.15) and (5.16) gives

τ WG S G τ WL S L 1 1
þ τiSi þ ðρL ρG Þg sin θ ¼ 0 ð5:17Þ
AG AL AL AG
ðAL þ AG Þ ¼ A ð5:18Þ
Equations (5.17) and (5.18) define AL and AG, that is, the holdup
εL ¼ AL =ðAL þ AG Þ ð5:19Þ
which is related to the void fraction, in that α ¼ 1 εL . The dimensionless height, hL / dt,
can also be calculated if AL is known. (See equations (5.7) to (5.12).) Either (5.15) or
(5.16) gives the pressure gradient.
Taitel & Dukler represented the stresses with friction factor relations
ρG u2G
τ WG ¼ f G ð5:20Þ
2
ρ u2
τ WL ¼ f L L L ð5:21Þ
2
f i ρ ð uG uL Þ 2
τi ¼ G ð5:22Þ
2
98 Stratified flow
Turbulent flows through non-circular passages are treated by using data for single-phase
flow through a circular pipe to evaluate the friction factor. This is done by defining a
hydraulic diameter as four times the area divided by the wetted perimeter. Thus, for the
gas phase,
4AG
dh ¼ ð5:23Þ
PG þ S i
(This relation was tested by Andritsos et al. (1989) by placing inserts on the bottom of a
pipe to create passages with the same shape as described in equations (5.7) to (5.11).) At
the interface, the velocity gradient in the liquid is much smaller than the velocity gradient
in the gas. Thus, the flow in the liquid could be considered to be the same as for an open
channel. Taitel & Dukler suggested that
4AL
dh ¼ ð5:24Þ
SL
Figure 5.1 presents calculations of hL / dt using (5.16) and (5.17), for different values of
fi / fS, where fS is the friction factor for a smooth interface. The abscissa is the Martinelli
factor (Section 1.5.3)

ðdp=dzÞLS 0:5
X ¼ ð5:25Þ
ðdp=dzÞGS
where ðdp=dzÞLS is the pressure gradient if the liquid were flowing alone in a smooth pipe
with diameter dt and ðdp=dzÞGS is the pressure drop if the gas were flowing alone in a
smooth pipe with diameter dt. Parameter X increases with an increasing ratio of liquid to
gas velocity. The top half of Figure 5.1 is for a laminar liquid and a turbulent gas. (This
might be realized for a very viscous liquid.)
The bottom half of Figure 5.1 presents calculations for a turbulent gas and a turbulent
liquid, for which (5.25) is equation (1.95) in Chapter 1. The friction factors ( f G ; f L ; f S )
are given as
f ¼ 0:046Re0:2 ð5:26Þ
Note that the calculations in Figure 5.1 are sensitive to changes in fi / fS, the ratio of the
interfacial friction factor, fi, to what would be realized for a smooth surface. Taitel &
Dukler assumed that fi = fS . However, if waves are present, f i =f S > 1.
Figure 5.2 is a plot of the pressure gradient against the Martinelli factor. Here, ϕ2G ¼
ðdp=dzÞ=ðdp=dzÞGS is the ratio of the actual pressure gradient to the pressure gradient
which would exist if the gas were flowing alone in the pipe (see Section 1.5.3). Again, a
sensitivity to fi / fS is noted.
A limitation of the calculations presented in Figures 5.1 and 5.2 is the calculation of
τWL with (5.21). Because of the sheared interface, the liquid flow can have a kinship with
the flows described in Section 3.2.1, rather than with free flow in a conduit. Furthermore,
as mentioned earlier in this section, the flow could contain secondary patterns. These
issues have not been addressed adequately.
1.0
Turbulent gas/laminar liquid
0.5
0.2
hL 0.1
dt fi / fS
0.05
1
35
812 20
0.02
0.01
0.005
0.005 0.01 0.02 0.05 0.1 0.2 0.5 1.0 2.0 5.0 10.0
Δp Δp 0.5
X= [( ) ( ) ]
ΔL LS
/
ΔL GS
1.0
Turbulent gas / turbulent liquid
0.5
0.2 fi / fS
1
hL 0.1 3
5
dt 8
0.05 12
20
0.02
0.01
0.005
0.005 0.01 0.02 0.05 0.1 0.2 0.5 1.0 2.0 5.0 10.0
ReLS –0.1 uLS ρL 0.5
X= (Re ) ( u ) ( ρ )
GS GS G
Figure 5.1 Effect of fi / fS on calculations of the layer height, hL / dt, for flow in a pipe. Henstock &
Hanratty, 1976.
Thus, Andritsos & Hanratty (1987) adopted the approach taken by Cheremisinoff &
1=2
Davis (1979), who developed a correlation between hþ L ¼ hL ð τ c ρ Þ =L and
ReL ¼ d h ρL uL =μL , where τc is a characteristic stress in the liquid. Andritsos &
Hanratty defined

2 hL 1
τ c ¼ τ WL 1 þ τi ð5:27Þ
3 dt 3
for the liquid layer. They found that

hþ
L ¼ 1:082ReL
0:5
ð5:28Þ
100 Stratified flow
20.0
Turbulent gas / laminar liquid
10.0
φG 5.0
fi /fS
2.0
20 12
8 5
3 1
1.0
0.005 0.01 0.02 0.05 0.1 0.2 0.5 1.0 2.0 5.0 10.0
Δp Δp 0.5
X= [ ( Δz ) / ( Δz ) ]
LS GS
20.0
Turbulent gas / turbulent liquid
10.0
( ReG > 2000 )
ReL > 2000
φG 5.0
fi/fS
2.0 20 12
8 5
3
1
1.0
0.005 0.01 0.02 0.05 0.1 0.2 0.5 1.0 2.0 5.0 10.0
ReLS –0.1 uLS ρ 0.5
X= (
ReGS ) ( )( ρ )
uGS
L
G
Figure 5.2 Effect of f i =f S on calculations of the pressure gradient for stratified flow in a pipe. Henstock &
Hanratty, 1976.
for laminar flow and

0:5
hL
hþ
L ¼ 0:098Re0:85
L 1 ð5:29Þ
dt
for turbulent flow. The following interpolation formula provides a representation of both
laminar and turbulent flow:
8 9
> " , 0:5 #5 =0:2
< h
5
hþ
L
L ¼ ð1:082Re0:5
L Þ þ 0:098ReL 1 ð5:30Þ
0:85
>
: dt ;
5.4 Determination of the interfacial friction factor
5.4.1 Measurements of fi/fS

The critical issue in analyzing stratified flows is the determination of the interfacial stress,
τi, or the interfacial friction factor, fi, defined by (5.22). The gas flow creates waves and
20
dt = 2.52 cm dt = 9.53 cm
μL (cP) μL (cP)
1 1
15 16 80
70
(fi / fS)–1
10
0
0 2 4 6 8 10
(Δh/λ) × 100
Figure 5.3 Measurements of the effect of wave amplitude on the dimensionless friction factor fi/fS. Andritsos
& Hanratty, 1987.
these, in turn, can present a roughened surface to the flow. Thus, fi is related to the
structure of the waves.
Measurements of the liquid height provide an average, hL , and a deviation from the
2
average, hL hL , which can be characterized by a standard deviation, σ 2 ¼ hL hL .
A characteristic height of the waves is defined by assuming a sinusoidal wave pattern,
pffiffiffiffiffi
that is, Dh ¼ 2 2σ . A characteristic wavelength is defined as the product of the wave
velocity and the wave frequency.
Values of fi (defined by equation (5.22)) are determined by measuring the pressure
drop and the height of the liquid layer, hL , and calculating τi with (5.15). Terms AG, SG,
Si are calculated from the knowledge of hL . (Thus, the measurements of τi are tied into
the use of the simple model of a stratified flow defined by equations (5.7)–(5.12).)
Figure 5.3, from Andritsos & Hanratty (1987), presents measured values of fi/fS,
where fS is what would be obtained if the interface were smooth. It is noted that fi/fS
can be much larger than unity. The abscissa in Figure 5.3 is the ratio of the wave
height, Dh, to the wavelength. It is seen that the interfacial friction factor correlates
approximately with the steepness of the waves and that values of fi/fS as large as 16
can be realized.
A common way to scale wave heights in engineering practice is to assume that they are
proportional to the liquid height. This notion has been applied with reasonable success
to very thin films that exist in annular gas–liquid flows (see Chapter 3). However, it does
not work for stratified flows. The data of Andritsos & Hanratty (Figure 5.3), for example,
show values of Dh=hL ranging from 0.09 to 1.7.
102 Stratified flow
5.4.2 Equations for interfacial friction factor and wave steepness

The data shown in Figure 5.3 are part of an extensive study that involved pipelines
with diameters of 2.52 cm and 9.53 cm, liquids with viscosities of 1 cP, 4.5 cP, 12 cP,
16 cP, 70 cP, 80 cP, hL/dt taking values in the range 0.03–0.22. Andritsos & Hanratty
(1987) present the measurements in three graphs, where values of fi/fS are plotted
against uGS for height ranges of hL/dt = 0.03–0.05, hL/dt = 0.08–0.12, hL/dt = 0.17–0.22.
The characteristic wave numbers of the waves observed in these experiments are roughly
equal to the wave number at which the growth rate of Kelvin–Helmholtz waves is a
maximum, kM. (See equation (4.92) and Figure 4.5.) For air–water, this is approximately
3.7 cm−1. Under the conditions of the experiments described in Figure 5.3,
ðk M =gÞ1=2 ðρG =ρL Þ1=2 is almost constant. Andritsos & Hanratty (1987) show that the
minimum gas velocity at which KH waves appear, uGt , can be approximated by the
equation

k M ρG 0:5
ð uG u L Þ t ¼1 ð5:31Þ
g ρL
Figure 5.3 shows that fi/fS is approximately equal to 1 at low uGS, where the interface is
smooth or has regular Jeffreys waves. For uG =uGt > ~1, the measurements increase
rapidly with increases in uGS. On the basis of these observations, Andritsos & Hanratty
recommended the following approximate equations:
f i =f S ¼ 1 for uGS ≤ uGSt ð5:32Þ
0:5
fi hL uGS uGSt
¼ 1 þ 15 for uGS ≥ uGSt ð5:33Þ
fS dt uGSt
where uGSt is the critical superficial gas velocity at which KH waves appear (see
Section 4.5).
This problem has been revisited by Bontozoglou & Hanratty (1989), who were
motivated by Figure 5.3 to use inviscid flow theory to develop an expression for Δh/λ.
(A solution for infinitesimal waves is presented in Chapter 4.) The following equations
are derived for the appearance of KH waves:
ρG σk g
ρL ρL k
ρ uG þ ρL uL
cR ¼ G ð5:35Þ
ρL
where cR is the wave velocity, k is the wave number and σ is the surface tension.
Saffman & Yuen (1982) extended this linear analysis so as to include finite-amplitude
waves. They obtained the following relation for gravity waves:
" 2 #
k 2 kρG 2 ð2π Þ2 Dh 2 k 2
c þ ð uG c R Þ 1 ¼ 2 cR 1 þ 1 ð5:36Þ
g R gρL 8 λ g
In order to establish a criterion for defining wave height, Saffman & Yuen (1982) and
Bontozoglou & Hanratty (1988) assumed that it corresponds to the geometric limit,
which occurs when the calculated wave slope becomes unphysical. A criterion which is a
measure of closeness to breaking is the ratio, q/c, of the horizontal velocity of the fastest-
moving particles on the interface, q, to the wave velocity, c. There is evidence that this
ratio is related to the geometric limit. Free-surface gravity waves, for example, are limited
in height by the formation of a sharp peak at the crest when ε = q/c = 1 (Stokes, 1847). The
value of ε increases with an increase in wave height for waves in the presence of an air
flow and ε = 1 is associated with a sharp peak at the crest or with an infinite slope
elsewhere along the profile (Holyer, 1979). Because of numerical difficulties, the wave
height corresponding to a given value ε, rather than ε = 1, was calculated.
Bongtozoglou & Hanratty (1989) used numerical methods developed by Saffman &
Yuen to calculate the change of gravity waves in the presence of an air flow. They chose
ε = 0.22 so that the calculated wave steepness is in the range represented by Figure 5.3.
Geometrically limited free (no gas flow) surface waves on liquids of arbitrary height have
been calculated by Cokelet (1977). For a freely flowing liquid layer, these show that
αW ¼ Dh=Dhdeep (where Δhdeep is the height of waves on a deep liquid) increases with
increasing khL and reaches a maximum value of unity at khL ≈ 2:5. The calculations of
Bontozoglou & Hanratty show that the same functionality is obtained for waves gen-
erated by a gas flow (see Figure 5.4). They also show that the dependence on ρG/ρL is
weak. The following result was obtained:

Dh uG uL
¼ 0:079ðαW Þ 1 ð5:37Þ
λ uGt
where αW is the function given by the solid curve in Figure 5.4 and uGt is the Kelvin–
Helmholtz critical velocity predicted by linear theory. Thus, from Figure 5.3, the
following relation for the friction factor is obtained:

fi u G uL
1 ≈ 13 ðαW Þ 1 ð5:38Þ
fS uGt
A comparison of (5.38) with (5.33) shows agreement between the two proposals. The
chief difference is the representation of the influence of khL with an analytical (rather than
an empirical) expression.
5.4.3 Design relations for a horizontal pipe

The simplest method for calculating the liquid holdup and the frictional pressure gradient
is to use Figures 5.1 and 5.2. The value of fi/fS could be approximated either with (5.33)
or (5.38) with αW ¼ DhL =Dhdeep defined by Figure 5.4. An iterative approach is needed.
A value of fi/fS is assumed. Figure 5.1 is used to calculate hL =d t . Equation (5.33) or (5.38)
is used to calculate fi/fS. The procedure is repeated until the calculated friction factor
agrees with the assumed friction factor.
Figures 5.1 and 5.2 indirectly use the assumption that τ WL can be calculated using
(5.21). An alternative approach is to use (5.30) instead of Figures 5.1 and 5.2 to model
104 Stratified flow
1.0
0.8
0.6
Δh/Δhdeep
0.4
ρG / ρL uG
0.000 0
0.2 0.001 0.9 uGt
0.100 0.9 uGt
0.0
0.0 0.5 1.0 1.5 2.0 2.5
Dimensionless liquid depth khL
Figure 5.4 Comparison of the calculations of Cokelet (1976) for free-surface gravity flows (the solid curve) with
calculations by Bontozoglou for gas–liquid flows, where uGt is the critical gas velocity at which a
Kelvin–Helmholtz instability occurs. Bontozoglou & Hanratty, 1989.
z
τi
Gas hL
id θ
Liqu τW
g
Figure 5.5 Sketch of an inclined channel.
flow in the liquid. Again, this would involve an iterative procedure, which is initiated by
assuming hL =d t and checking this against (5.30).
5.5 Inclined pipes
5.5.1 General comments about the effect of inclination

The inclination of a pipeline to the horizontal at small angles can have a strong effect on the
flow patterns. This is particularly evident for upward flows where slugging can be induced.
The design of such pipelines needs to be carefully executed so as to avoid slugging. This
effect can be understood by considering the momentum balance for the liquid phase.
Assume a fully developed flow in a rectangular channel, as indicated in Figure 5.5. The
average height of the liquid layer is designated as hL , the drag stress at the interface as τi,
the resisting stress at the wall as τW, and the acceleration of gravity as g. The channel is
inclined with an angle θ to the horizontal so the component of gravity in the direction of
flow is -g sin θ; that is, the gravitational force opposes the flow. For the case of a declined
flow the component of gravity in the direction of flow is g sin θ, so the gravitational force
augments the flow.
A force balance on the liquid layer is given as follows for a fully developed liquid
flow:
dp
0 ¼ hL þ τ i hL ρL g sin θ τ W ð5:39Þ
dz
The first term represents the force of the pressure gradient; τi is the drag force of the gas;
the third term is the force of gravity; τW is the resisting stress at the wall.
Since τi varies as ρG u2GS the influence of gravity can be large if u2GS ρG =hL gj sin θjρL
is small. For upflows, this leads to a decrease in uL for a given hL . At small enough
uGS, a reversal of flow can occur. For downflows, gravity causes an increase in uL for a
given hL .
5.5.2 Upflows
Simmons & Hanratty (2001) observed upflows of air and water at inclination angles
of 0 ; 0:05 ; 0:2 ; 0:4 , and 1.2° in a pipe with a diameter of 7.63 cm. They found that
reversed flows occur at conditions that are close to those for which τW = 0. Figure 5.6
presents measurements of the critical conditions for a transition from a steady stratified
flow to a slug flow. The thick lines at the bottom of the graph indicate the critical gas
velocity below which a stratified flow cannot exist for a given inclination.
0.30
θ = 0° Slug stability
θ = 0.05°
0.25 θ = 0.2° Slug stability
θ = 0.4° Slug stability
θ = 1.2° Slug stability
0.20
uLS (m/s)
0.15
0.10
0.05
uGSC uGSC uGSC
0.00
0 1 2 3 4 5 6 7 8 9 10 11 12
uGS (m/s)
Figure 5.6 Effect of inclination on the transition from stratified flow to slug flow, where uGSC is the critical
superficial gas velocity below which slugs always appear. Simmons & Hanratty, 2001.
106 Stratified flow
For very small uGS and uLS, the stratified liquid finds its own level in a pipeline when
gas drag is unable to counterbalance gravitational effects. If the angle is very small,
pooling could extend over the whole pipeline. The hydraulic gradient causes the liquid to
flow out of the pipeline. This behavior will be avoided if the pipeline is long enough for
the outlet to be completely above the pipe inlet. This can be realized for
sin θ5d t =L ð5:40Þ
where L is the length of the pipe. An angle of 0.19° would be required for the system
studied by Simmons & Hanratty (2001). Below this angle, measurements would not
correspond to what occurs in a very long pipe. Thus, the results for 0.05°, presented in
Figure 5.6 for small uGS, do not represent what would be observed in a longer pipe.
Another effect of positive inclination is the existence of reversed flows that occur
when the influence of gravity is large compared with the drag of the gas on the liquid. The
thick lines on the bottom of Figure 5.6 are the critical superficial gas velocities, uGSC,
calculated for the realization of τW = 0 at a given inclination. The region to the left of the
dashed curves represents conditions for which a calculated steady stratified flow would
show reversed flow close to the wall. Observations indicate that an intermittent flow
exists in these regions.
The intermittent pattern that occurs in a horizontal pipe is initiated downstream of the
entry by growth or coalescence of waves. For the reversed flow regions associated with
upflows, transition occurs close to the entry, where slugs or plugs are formed intermit-
tently. These slugs propagate downstream and leave a depleted layer at the inlet. This
depleted layer is built up by the inflow and eventually reaches a level at which a Kelvin–
Helmholtz instability leads to the formation of slugs or plugs. These carry all of the liquid
out of the pipe. The stratified layer that exists between or behind the slugs moves
backward.
Simmons & Hanratty (2001) present measurements of the time-varying liquid holdup for
two experiments at an inclination of 0.4° (uGS = 1.54 m/s, uLS = 0.064 m/s; uGS = 0.54 m/s,
uLS = 0.078 m/s), which would be characterized as slug and plug flow by investiga-
tors, who use gas occlusion to characterize an intermittent pattern. The holdup
measurements agree with this characterization in that they indicate a small amount
of aeration in the first case and negligible aeration in the second case. However, the
shapes of the holdup patterns suggest that the intermittent flow contains slugs in
both cases.
5.5.3 Downward inclinations

Figure 5.7 shows a map for the transition from a stratified to an intermittent flow,
developed by Woods et al. (2000), for air and water flowing in a declined 7.63 cm pipe
with a length of 23 m. Large effects of declination are observed for small gas velocities.
However, for uGS > 8 m=s (where large-amplitude waves and annular flows exist) the
declined flows are roughly the same as observed for horizontal pipes. Two dramatic
changes from the flow patterns, shown in Figure 2.1 for air–water flow in a horizontal
pipe are observed: (1) At declination angles of 0.2° and greater, regular Jeffreys waves
1
θ = 0.0°
θ = 0.2°
θ = 0.5°
θ = 0.8°
uLS (m/s)
0.1
1 10
uGS (m/s)
Figure 5.7 Effect of declination on the transition from a stratified flow to a slug flow. Woods et al., 2000.
are not observed in the stratified flows (see Andreussi & Persen, 1987). (2) The
transition to slug flow at low gas velocities occurs at much larger superficial liquid
velocities than for a horizontal pipe and, from Figure 2.6, at larger hL/dt. Theoretical
discussions of these effects are given by Woods et al. (2000) and by Hurlburt &
Hanratty (2002).
As mentioned in Section 5.5.1, the liquid velocities can be much larger for a declined
flow, at a given liquid height, if u2GS ρG =hL gρL j sin θj is small. The damping of Jeffreys
waves in declined flows can be associated with this increase in the liquid velocity, which
would result in an increase in the wave velocity. Theoretical studies of Jeffreys waves
by Miles (1957) and Benjamin (1959) indicate that the transmission of energy from
the gas to the waves can depend on the location of the critical layer where the wave
velocity equals the gas velocity. This could account for the influence of declination on
this process, but this is not an established explanation.
Woods et al. (2000) show that the transition to slug/plug flow in declined pipes is
predicted by a viscous large-wavelength analysis, as had been found for horizontal
air–water flows. However, the mechanism is different from what is described in
Section 2.2.6, since Jeffreys waves are not present at the interface on transition.
Figure 5.8 provides an example of the generation of slugs for air–water flow in a
pipe which is declined at an angle of 0.5° (Woods et al., 2000). The superficial gas
velocity was 2.4 m/s, the superficial liquid velocity was 0.59 m/s and the pipe diameter
was 0.0763 m. The interface is disturbed by waves in the early part of the pipe. These
are damped; they are not observed at L=d t ¼ 106. A large-wavelength disturbance
with a wavelength of ~12 m (approximately one-half of the pipe length) and a period
of ~ 8 s appears in the early part of the pipe. It grows with distance downstream and is
visible at L/dt = 60. At L/dt = 106 (approximately midway in the pipe) disturbances,
which are precursors of slugs, appear. At L/dt = 190, well formed growing slugs are
observed.
The eight photographs in Figure 5.9 (flow is from right to left) were obtained
sequentially at 1/30 s intervals under the same conditions represented by Figure 5.8.
The photographs shown cover a much shorter length of the pipe than the wavelength, so
108 Stratified flow
0.7 (a) L/dt = 35

hL/dt
0.6
0.5
90 95 100 105 110 115 120 125 130 135 140
Time (s)
0.7 (b) L/dt = 60

hL/dt
0.6
0.5
90 95 100 105 110 115 120 125 130 135 140
Time (s)
1 (c) L/dt = 106
0.8
hL/dt
0.6
0.4
0.2
0
90 95 100 105 110 115 120 125 130 135 140
Time (s)
1 (d) L/dt = 190

0.8
hL/dt
0.6
0.4
0.2
0
90 95 100 105 110 115 120 125 130 135 140
Time (s)
Figure 5.8 Generation of slugs for air–water flow in a pipe declined at 0.5°. Woods et al., 2000.
the interface appears to be flat. The level of this segment of the interface increases as a
wave passes (Figure 5.9c and d). The interface in Figure 5.9a is approximately smooth. A
small instability, observed at the far right of the second frame is seen to grow in amplitude
in Figure 5.9c, d and e. This disturbance eventually touches the top wall to form a slug in
Figure 5.9f, g and h.
Slugs are thus seen to be the result of a local KH instability that develops at the crest
of a large-wavelength disturbance. This type of behavior has also been observed by
Kordyban (1985) and by Kordyban & Ranov (1970), who photographed the initiation of
slugs triggered by mechanically generated waves.
Woods et al. (2000) show that viscous large-wavelength theory correctly predicts the
uSL – uSG curve that describes the initiation of slugs in a 0.0763 m pipe declined at an
angle of 0.05° and measurements of Shoham (1982) in 2.52 cm and 5.1 cm pipes declined
at 1.0°.
The results of studies of flow regimes in pipes declined at angles of 0° to 80° have
been reported by Barnea et al. (1982). These show an increasing importance of the
stratified regime and a decreased importance of the slug regime with increasing
angle.
References 109
Flow direction
(a)
dt = 7.63 cm
(b)
(c)
(d)
(e)
(f)
(g)
(h)
Figure 5.9 Photographs of slug formation for the conditions in Figure 5.8. Woods et al., 2000.
References
Andreussi, P. & Persen, L. N. 1987 Stratified gas–liquid flow in downwardly inclined pipes. Int.
Andritsos, N. & Hanratty, T. J. 1987 Influence of interfacial waves in stratified gas–liquid flows.
AIChE Jl 33, 444–454.
Andritsos, N., Williams, L. & Hanratty, T. J. 1989 Effect of liquid viscosity on the stratified-slug
Baker, O. 1954 Simultaneous flow of oil and gas. Oil Gas J. 53, 185.
Barnea, D., Shoham, O. & Taitel, Y. 1982 Flow pattern transition for downward inclined two-phase
flow: horizontal to vertical. Chem. Eng. 37, 735–740.
Benjamin, T. B. 1959 Shearing flow over a wavy boundary. J. Fluid Mech., 37, 161–205.
Bird, R. B., Stewart, W. E. & Lightfoot, E. N. 1960 Transport Phenomena. New York: John Wiley.
Bontozoglou, V. & Hanratty, T. J. 1988 Effects of finite depth and current velocity on large
amplitude Kelvin–Helmholtz waves. J. Fluid Mech. 196, 187.
Bontozoglou, V. & Hanratty, T. J. 1989 Wave height estimation in stratified gas–liquid flows.
AIChE Jl 35, 1346–1350.
110 Stratified flow
Chen, T., Cai, X. D. & Brill, J. P. 1997 Gas–liquid stratified wavy flow in horizontal pipelines.
J. Energy Resources 119, 209–216.
Cheremisinoff, N. P. & Davis, E. J. 1979 Stratified turbulent–turbulent gas liquid flows. AIChE Jl
25, 48.
Cokelet, E. D. 1977 Steep gravity waves in water of arbitrary uniform depth. Phil. Trans. R. Soc.
Lond. A286, 183–230, 183.
Govier, G. W. & Aziz, K. 1972 The Flow of Complex Mixtures in Pipes. New York: Van Nostrand
Reinhold.
Grolman, E. & Fortuin, M. H. 1997 Gas–liquid flow in slightly inclined pipes. Chem. Eng. Science
52, 4461–4471.
Hart, J., Hamersa, P. J. & Fortuin, M. H. 1989 Correlations predicting frictional pressure drop
and liquid holdup during horizontal gas–liquid pipe flow with a small liquid holdup. Int.
annular flows. AIChE JL 22, 990–1000.
Holyer, J. Y. 1979 Large amplitude progressive waves. J. Fluid Mech. 93, 433.
Hoogendorn, C. J. 1959 Gas–liquid flow in horizontal pipes. Chem. Eng. Sci. 9, 205.
Hurlburt, E. T. & Hanratty, T. J. 2002 Prediction of the transition to slug and plug flow. Int.
Kordyban, E. S. 1985 Some details of developing slugs in horizontal two-phase flow. AIChE Jl 31,
802–806.
Kordyban, E. S. & Ranov, T. 1970 Mechanism of slug formation in horizontal two-phase-flow.
J. Basic Eng. 192, 857–864.
Miles, J. W. 1957 On generation of surface waves by shear flows. J. Fluid Mech. 3, 185–204.
Russel, T. W. F., Etchels, A. W., Jensen, R. H. & Arruda, T. J. 1974 Pressure drop and holdup in
stratified gas–liquid flow. AIChE Jl 20, 664.
Saffman, O. & Yuen, H. C. 1982 Finite-amplitude interfacial waves in the presence of a current.
J. Fluid Mech. 123, 459–469.
Shoham, O. 1982 Flow pattern transitions and characteristics in horizontal pipes. Ph.D. thesis,
Tel-Aviv University.
Simmons, M. J. H. & Hanratty, T. J. 2001 Transition from stratified to intermittent flows in small
angle upflows. Int. J. Multiphase Flow 27, 599–616.
Stokes, G. G. 1847 On the theory of oscillatory waves. Trans. Camb. Phil. Soc. 8, 441.
Suzanne, C. 1985 Structure de l’ ecoulemont stratifie de gaz et de liquide en canal rectangulaire.
These d’ Etat, Mecanique de Fluides, L’Institut National Polytechnique de Toulouse.
Taitel, Y. & Dukler, A.E. 1976a A theoretical approach to the Lockhart–Martinelli correlation for
stratified flow. Int. J. Multiphase Flow 2, 591.
Taitel, Y. & Dukler, A. E. 1976b A model for predicting flow regime horizontal and near horizontal
gas–liquid flows. AIChE Jl 22, 47–55.
Vlachos, N. A., Paras, S. V. & Karabelas, A. J. 1999 Prediction of holdup, axial pressure gradient
and wall shear stress in wavy/stratified and stratified/atomization gas–liquid flow. Int.
Woods, B. D., Hurlburt, E. T. & Hanratty, T. J. 2000 Mechanism of slug formation in downwardly
6 Influence of viscosity on large Reynolds
number interfacial waves; effect of
spatially and temporally induced
oscillations on a turbulent flow
6.1 Introductory comments

Equation (6.19), developed in this chapter, considers the flow field associated with wave
propagation on a viscous liquid. If the field is represented by a length scale equal to the
height of a liquid layer, hL , and a velocity scale equal to c ua , the following group
represents the ratio of the inertia terms to the viscous terms: ½hL ρL ðc ua Þ=μL ½khL .
Interfacial waves with small or moderate values of ðkhL Þ are usually characterized by a
large value of this dimensionless group. This leads to an expectation that the disturbed
velocity over all or most of the field can be described by inviscid equations.
However, if viscous effects are confined to thin layers in the liquid, of thickness δ, a
second length scale is needed. These layers are characterized by ½δρL ðc ua Þ=μL ½khL ,
so that viscous effects cannot be ignored in these layers. The relation between the wave
velocity and the wavelength developed in Chapter 4 remains valid. (This is not the case
for large-wavelength waves considered in Chapter 7.) There are a number of situations
where these viscous effects need to be considered. One involves the damping of waves
downstream of a wavemaker. Another is the generation of waves by a gas flow, discussed
in Section 2.2.1. In the second example, energy is fed to the liquid flow through pressure
and shear forces exerted by the gas on interfacial waves. This energy is dissipated in
viscous layers close to the interface and close to the wall. A stable situation can be
realized if the energy fed to the waves is balanced by this dissipation.
The evaluation of the interfacial stresses is a central problem in analyzing gas–liquid
flows. This has been attacked by solving for the disturbed velocity field in the gas caused
by flow over a small-amplitude wavy interface. Initial efforts have involved the use of a
quasi-laminar theory whereby turbulence is taken into account by using a mean velocity
profile which is characteristic of a turbulent flow. This approach has been improved by
recognizing that turbulent flow over a wavy boundary experiences spatial disturbances of
the turbulence, as well as of the mean flow. A turbulence model developed by Loyd et al.
(1970) has been tested with measurements of the wall shear stress variation along solid
wavy surfaces over which a turbulent liquid is flowing, and by studies of the initiation of
waves at a gas–liquid interface.
The behavior of a turbulent flow subjected to an imposed temporal oscillation is similar
to flow over a wavy boundary, where spatial oscillations are introduced. Thus, analytical
methods developed to describe flow over solid waves are applicable to this problem.
112 Influence of viscosity on interfacial waves
Techniques used to analyze flow over small-amplitude waves can also be used to
analyze dissolution patterns found on the underside of river ice and in caves. Here, the
interest is to predict the wavelength characterizing the pattern.
Small values of ðhL ρc=μL ÞðkhL Þ can be realized for very small values of ðkhL Þ, that is,
very large wavelengths. The flow can be either turbulent or laminar. Viscous and
turbulent stresses introduced by the wave motion, then, can extend over the entire liquid
layer. These type flows are considered in Chapter 7.
6.2 Equation for small disturbances in a viscous fluid
As in Chapter 4, the assumption is made that the waves are of small enough amplitude, a,
that a linearization of the defining equations is justified. A first step is the development of
an equation for small disturbances in the liquid velocity field that includes the effect of
viscosity. The assumption of incompressibility is made.
A consideration of viscous effects in the momentum balance requires the use of the
Navier–Stokes equations (Schlichting, 1968):

∂ui ∂ui ∂τ ij
ρ þ uj ¼ þ ρgi ð6:1Þ
∂t ∂xj ∂xj
where i is the component of the velocity vector being considered and the Einstein
convention of summation over repeated indices is used. Thus (6.1) represents three
equations for which i = x, y or z. The term τ ij represents the i component of the stress
on a face perpendicular to the j-axis. The convention is used that τ ij is positive if i and j
have the same sign. Since pressure, p, is compressive, it represents a negative contribu-
tion to the normal stress. Making use of the Einstein convention,
∂τ ij ∂τ ix ∂τ iy ∂τ iz
¼ þ þ ð6:2Þ
∂xj ∂x ∂y ∂z
For a Newtonian incompressible fluid (Schlichting, 1968)

∂ui ∂uj
τ ij ¼ δij p þ μ þ ð6:3Þ
∂xj ∂xi
where δij is a delta function which is unity when i = j and zero otherwise. For μ = 0,
τ xx ¼ τ yy ¼ τ zz ¼ p and all the shear stresses are zero, so (4.10) is recovered.
The two-dimensional system to be considered is flow in a rectangular channel which
is infinitely wide. The flow is fully developed, so the mean velocity varies only in the
y-direction. The equations describing an incompressible two-dimensional viscous fluid
are the equation for conservation of mass
∂ux ∂uy
þ ¼0 ð6:4Þ
∂x ∂y
and the equations for conservation of momentum

6.2 Equation for small disturbances in a viscous fluid 113

Dux 1 ∂p μ ∂2 ux ∂2 ux
¼ þ þ 2 þ gx ð6:5Þ
Dt ρ ∂x ρ ∂x2 ∂y

Duy 1 ∂p μ ∂2 uy ∂2 uy
¼ þ þ þg y ð6:6Þ
Dt ρ ∂y ρ ∂x2 ∂y2
where the x-axis is in the mean flow direction. Since the flow is fully developed, the
average of (6.5) produces the following equation for the mean velocity in a non-turbulent
flow:
2
1 ∂p μ ∂ ux
0¼ þ þ gx ð6:7Þ
ρ ∂x ρ ∂y2
The velocity is given as the sum of the average and the fluctuating component imposed
by a two-dimensional wave motion at the interface.
ux ¼ ux þ u0x ð6:8Þ
uy ¼ uy þ u0y ð6:9Þ
0
p¼pþp ð6:10Þ
where uy = 0. Equations (6.8), (6.9) and (6.10) are substituted into (6.5), (6.6). Equation
(6.7) is subtracted from (6.5). The imposed waves are assumed to have a small amplitude
so second-order terms in the fluctuating quantities are neglected. The following equations
for u0x, u0y and p0 are obtained:
∂u0x ∂u0y
þ ¼0 ð6:11Þ
∂x ∂y
2 0
∂u0x ∂u0 dux 1 ∂p0 μ ∂ ux ∂2 u0x
þ ux x þ u0y ¼ þ þ ð6:12Þ
∂t ∂x dy ρ ∂x ρ ∂x2 ∂y2
2 0 !
∂u0y ∂u0y 1 ∂p0 μ ∂ uy ∂2 u0y
þ ux ¼ þ þ 2 ð6:13Þ
∂t ∂x ρ ∂y ρ ∂x2 ∂y
The interfacial disturbance is defined by (4.25), (4.29) as a sinusoidal wave for which the
displacement from the average location of the interface is given as ¼ aexp ikðx ctÞ.
Since the system is linear, the solution is of the form
u x0 u y0 p0
¼ ¼ ¼ expik ðx ct Þ ð6:14Þ
^
ux u ^y ^
p
^x ðyÞ, u
where u ^y ðyÞ, p
^ðyÞ are complex and proportional to the wave amplitude, a. The
wave velocity c is complex; the wave number k is real. Thus (6.11) and (6.14) give
dûy
0 ¼ ik^
ux exp ik ðx ct Þ þ exp ik ðx ct Þ ð6:15Þ
dy
The following relation between ^

ux and ûy is obtained from (6.15):
i dûy
^
ux ¼ ð6:16Þ
k dy
since i2 ¼ 1. The substitution of (6.14) into (6.12) and (6.13) yields

dux k μ d 2 ûx
ikc^ ux þ ^
ux þ ikux ^ uy ¼ i ^p þ k ûx
2
ð6:17Þ
dy ρ ρ dy2

1 d^p μ d 2 ûy
ikcûy þ ikux ^
uy ¼ þ k 2
^
u y ð6:18Þ
ρ dy ρ dy2
The pressure amplitude is eliminated between (6.17) and (6.18) after taking the derivative
of (6.17) with respect to y. The amplitude ûx is eliminated from the resulting equation by
substituting (6.16). The following relation for ûy is obtained:
2
d ûy d 2 ux i μ d 4 ûy 2d ^
2
uy
ðux cÞ k ^2
uy ûy 2 ¼ 2k þ k ûy
4
ð6:19Þ
dy2 dy k ρ dy4 dy2
This is the Orr–Sommerfeld equation used to calculate the stability of viscous flow fields
(Schlichting, 1968). The solution of (6.19) can be substituted into (6.16) and (6.17) to
give ^
ux ðyÞ and ^
pðyÞ.
6.3 Boundary conditions
The inclusion of the effects of viscosity requires a consideration of the terms on the right
side of (6.19). Thus, the equation changes from second order to fourth order if the right
side is not zero. The solution of the inviscid equation (Chapter 4) needs the specification
of two boundary conditions: the assumption that ûy ¼ 0 at the stationary wall and the
specification of ^ uy at the interface by using the kinematic condition. The solution of
(6.19) requires two additional boundary conditions. One of these is the assumption of
zero tangential velocity at the wall (the no slip condition)
ux ¼ 0 at y ¼ hL ð6:20Þ
where y = 0 is the average location of the interface. The other, to be satisfied at the
interface, is the equality of the wave-induced variations of the shear stresses in the liquid
to the wave-induced shear stresses in the gas and tangential forces due to the variation of
tensile stresses, such as surface tension, in the interface (caused by temperature variations
or by variation of the concentration of a surface active agent).
Gas flowing over a wave, defined by (4.25), exerts a normal stress, P, and a
tangential stress, T, on the liquid as indicated in Figure 6.1a. These are defined in a
6.3 Boundary conditions 115
(a) P
T
(b) Y
θ y
X
θ
x
Figure 6.1 (a) Sketch of waves with vectors indicating pressure and shear stress at the surface. (b) Relation
between boundary-layer coordinates and Cartesian coordinates.
coordinate system, X, Y, that is imbedded in the interface. Fluctuations of the imposed

stresses and of the interfacial tension, σ, can be represented as given below, since these
terms vary linearly with a.
T 0 ¼ aT~ exp ik ðx ct Þ ð6:21Þ
P 0 ¼ aP~ exp ik ðx ct Þ ð6:22Þ
0
σ ¼ a~
σ exp ik ðx ct Þ ð6:23Þ
where the linear dependence of T^ , P

^ and σ^ on wave amplitude, a, is recognized and T~ , P
~
and σ~ can be complex.
Using (4.28), the real parts of (6.21) and (6.22) are given as
T 0 ¼ aT~ R cos k ðx cR t Þ aT~ I sin k ðx cR t Þ ð6:24Þ
P0 ¼ a P~ R cos kðx cR tÞ aP ~ I sin k ðx cR t Þ ð6:25Þ
where P ~ R; T~ R are components in phase (or 180° out of phase) with the wave height
~ I , T~ I are in phase with the wave slope.
and P
The equality of the shear stresses and the tangential forces at the interface can be
expressed as
∂σ
Ti þ ¼ ðτ X Y Þi ð6:26Þ
∂X
where ðτ X Y Þi is the shear stress at the interface of the liquid in the X,Y coordinate system.
We want to express these liquid stresses in laboratory coordinates x,y as shown in Figure
6.1b. To do this, the transformation for a Cartesian tensor (Long, 1963) is used:
τ X Y ¼ τ kl aX k aY l ð6:27Þ
where aX k ¼ cos ðX ; xk Þ and aY l ¼ cos ðY ; xl Þ, with x1 ¼ x, x2 ¼ y and x3 ¼ z.

For a very small amplitude wave the angle θ in Figure 6.1b is small enough that
cos ðX ; xÞ ¼ cos ðY ; yÞ ≈ 1, where cos ðX ; yÞ ≈ ∂=∂x, cos ðY ; xÞ ≈ ∂=∂x. Thus, by
summing with respect to k and l, equations (6.26) and (6.27) give
2
∂ ∂ ∂
τ X Y ¼ τ xx þ τ xy þ τ yy τ yx ð6:28Þ
∂x ∂x ∂x
Represent the stresses in (6.26) as the sum of averages and wave-induced fluctuations and
substitute (6.28)
T i ¼ τ xy ð6:29Þ
∂σ0
∂ ∂
þ T 0i ¼ τ xx þ τ 0xy þ τ yy ð6:30Þ
∂x ∂x i ∂x
where second-order terms in the fluctuations have been neglected. Again, for small-
amplitude waves,
τ xx ¼ τ yy ¼ p ð6:31Þ
so
0
∂σ 0 0 ∂u x ∂u y0
þT ¼μ þ ð6:32Þ
∂x ∂y ∂x i
where, from (6.3),

0 ∂u x0 ∂u y0
τxy ¼μ þ ð6:33Þ
∂y ∂x
For small-amplitude waves, (6.32) can be evaluated at y = 0 so

0
∂σ 0 ∂u x ∂u y0
þ T0 ¼ μ þ ð6:34Þ
∂x ∂y ∂x 0
By substituting (6.21), (6.23) and (6.14) and using (6.16) to eliminate ûx the following
equation is obtained from (6.34):
2
~ i d ~uy
0 ¼ ik~
σþT μ þ ik~uy ð6:35Þ
k dy2 y¼0
6.4 Solution of the Orr–Sommerfeld equation
6.4.1 General solution

The solution of (6.19) for ^
uy is given by
^
uy ¼ A^
uy1 þ Bûy2 þ Cûy3 þ Dûy4 ð6:36Þ
6.4 Solution of the Orr–Sommerfeld equation 117
where ûy1 ; ^
uy2 ; ^
uy3; ^
uy4 are four independent solutions. The constants A, B, C, D can be
obtained by using the four constraints on the system provided by the boundary conditions,
d^
uy
^
uy ¼ 0 ¼0 at y ¼ hL ð6:37Þ
dy
equation (6.35) and the kinematic condition, which relates the normal velocity in the
liquid at the interface to the wave amplitude; see (4.42).
^
uy0 ¼ aik ðux0 cÞ ð6:38Þ
6.4.2 Dispersion relation

The principal interest is the development of a relation for the complex wave velocity, so
the calculation of the velocity field is bypassed. As was shown in Section 4.3, this can be
realized by relating the normal stresses on both sides of the interface
∂2
P ¼ ðτ Y Y ÞY ¼ σ ð6:39Þ
∂x2
or, for small-amplitude waves

∂2
P ≈ ðτ Y Y ÞY ¼0 þ ρg σ ð6:40Þ
∂x2
where P is the pressure at the interface. The term ρg takes account of the pressure
change from y ¼ to y ¼ 0 due to hydrostatic head. In Chapter 4, the pressures on both
sides of the interface were related to the velocity fields by using the Bernoulli
equation. This cannot be done if viscous effects are included.
The fluid stress τ Y Y is related to stresses in the x–y laboratory coordinate system by
using (6.27)
∂ ∂
ðτ Y Y Þ0 ¼ τ xy 0 þ τ yy 0 τ yx 0 ð6:41Þ
∂x ∂x
From (6.3),

∂uy
τ yy ¼ p þ 2μ ð6:42Þ
∂y
Substituting (6.41) into (6.40) and using (6.42) gives the following equation for the
wave-induced fluctuating quantities, where second-order terms have been neglected.
0
0 0 ∂u y ∂ ∂2
ρG g P ¼ p ð0Þ þ 2μ 2T þ ρL g σ 2 ð6:43Þ
∂y 0 ∂x ∂x
Substitute (6.22) for P0 and evaluate p0 ð0Þ with (6.17). Use (6.14) and, after some algebra,
(6.43) gives

~
2
1 d 3 ûy dûy
0 ¼ aP þ aðρ ρG Þg þ aσk a2ikT þ μ 2 3 3
k dy dy 0 ð6:44Þ
i d^
uy i duy
ρðux0 cÞ þ ρûy ð0Þ
k dy y¼0 k dy y¼0
The conditions on the shear stress at the interface, (6.35), and the pressure, (6.44), at the
interface can be written as

H 0 ûy ¼ 0 ð6:45Þ

G0 ûy ¼ 0 ð6:46Þ
If (6.37), (6.45) and (6.46) are substituted into (6.36), the following equations for
constants A, B, C, D are obtained:

0 ¼ A ðûy1 ÞhL þ B ^ uy2 hL þ C ûy3 hL þ D ûy4 hL ð6:47Þ

d^
uy1 d^
uy2 dûy3 dûy4
0¼A þB þC þD ð6:48Þ
dy hL dy hL dy hL dy hL

0 ¼ AH 0 ^ uy1 þ BH 0 ûy2 þ CH 0 ûy3 þ DH 0 ûy4 ð6:49Þ

0 ¼ AG0 ^ uy1 þ BG0 ûy2 þ CG0 ûy3 þ DG0 ûy4 ð6:50Þ
The zero subscripts indicate that the terms are evaluated at the average location of the
interface. Note that the above equations are homogeneous in that the left sides are zero.
From the theory of determinants, a solution for A, B, C, D can exist only if the
determinant of the coefficients in these equations equals zero; that is,

ð^
u ÞhL ð^
u ÞhL ðû ÞhL ðû ÞhL
y1 y2 y3 y4
dûy1 d^
uy2 dûy3 dûy4

0 ¼ dy hL dy hL dy hL dy hL ð6:51Þ
H ^
H 0 ^

H 0 ûy3

H 0 ûy4
0 uy1 uy2
G ^ G0 ^ G0 ûy3 G0 ûy4
0 uy1 uy2
Equation (6.51) gives a relation between the complex wave velocity, c ¼ cR þ cI , and the
wave number, k. The real and imaginary parts of the expansion produce two equations
that can be solved for cR and cI . This calculation is simplified by using the assumption
that ðc u0 Þ=ðμ=ρÞk is large.
Note that

iμ d 2 ^ uy σ~ iT~
H ûy ¼ þ iμk þ ûy ð6:52Þ
k dy2 ðux0 cÞ k ðux0 cÞ
uy
d3^ dûy
G ^
uy ¼ μ 3 þ 3k 2 μ ikρðux cÞ
dy dy
ð6:53Þ
ik^
uy ~ dux
þ P þ ðρ ρG Þg þ σk 2ikT þ ρ
2
ðux0 cÞ
ðux0 cÞ dy 0
where (6.38) is incorporated by using

^
uy0
¼1 ð6:54Þ
aik ðux0 cÞ
6.5 Special solutions for constant liquid velocity
6.5.1 Solution of (6.19) for constant liquid velocity

Solutions of (6.19) for cases where ux is not varying spatially can be obtained by
substituting exp(ny). It is found that this provides four solutions since the expression
for n is a quartic. These are exp(ky), exp(−ky), exp(my), exp(−my), where
iðux cÞk
m2 ¼ k 2 þ ð6:55Þ
μ=ρ
Thus ^
uy can be represented either as
uy ¼ Aeky þ Beky þ Cemy þ Demy
^ ð6:56Þ
or as
^
uy ¼ A cothðkyÞ þ B sinhðkyÞ þ C cothðmyÞ þ D sinhðmyÞ ð6:57Þ
6.5.2 Free-surface waves

The case of zero values of the mean gas and liquid flows will now be considered; that
is, ux ¼ 0, T ¼ 0, P ¼ 0. The interface is assumed to be clean so the surface tension is
constant. The goal is to derive a relationship for wave damping due to viscosity.
The solution for a deep liquid is considered. Wave-induced disturbances decay to zero
at large distances from the interface (large negative values of y). Thus, B and C in (6.56)
may be taken as zero. The constants A and D are calculated using (6.52) and (6.53)
~ T , T~ , σ~ equal to zero. For the case considered, the shear condition, equation (6.35)
with P,
or (6.52), gives

0 ¼ A H 0 eky þ DH 0 ðemy Þ ð6:58Þ
where
2
i d ûy
^y ¼ μ
H u þ μikûy ð6:59Þ
k dy2
The condition on the normal stress gives

0 ¼ AG0 eky þ DG0 ðemy Þ ð6:60Þ
The only way for a solution to exist is that the determinant of the coefficients A and D in
(6.58) and (6.60) equals zero

0 ¼ H 0 ðemy ÞG0 eky G0 ðemy ÞH 0 eky ð6:61Þ
(This equation can also be obtained by solving (6.58) and (6.60) for A/D and equating the
two results.)
The expansion of the secular determinant (6.61) gives

2k ðμ=ρÞ μ c 2k ðμ=ρÞ 2k ðμ=ρÞ
c 1
2
¼ c0
2
2k 1
3
þ 2k m2
ic ρ ik 2 ic ic
ð6:62Þ
where
gðρ ρG Þ σ
c20 ¼ þ k ð6:63Þ
ρk ρ
Solving (6.62) for c gives

" 2 #1=2
μ μ
c ¼ i2k c0 þ 4km
2
ð6:64Þ
ρ ρ
For small μ=ρ (or large c=k, where is the kinematic viscosity), equation (6.64)
simplifies further to
μ
c ¼ i2k c0 ð6:65Þ
ρ
where c ¼ cR þ icI . Equation (6.65), therefore, gives

2kμ
cI ¼ ð6:66Þ
ρ
cR ¼ c0 ð6:67Þ
Thus, from (6.66) and (6.67), the real part of (4.27) is

¼ a cos k ðx c0 t Þexp 2k 2 t ð6:68Þ
This represents a disturbance with wave number k propagating with velocity c0 and
decaying exponentially with time due to viscous damping.
The fluctuating normal velocity, u0y , is given as

u0 y ¼ ^
uy cos k ðx c0 t Þ exp 2k 2 t ð6:69Þ
where, from (6.56),

uy ¼ Aeky þ Demy
^ ð6:70Þ
Terms eky and emy are, respectively, the inviscid and viscous parts of the solution. Both
terms in (6.70) show an exponential decrease with decreasing y, where y ¼ 0 is the
average location of the interface. The characteristic length is 1/k for the inviscid
solution. For small μ=ρ, the real part of m is
pffiffiffi
ðcR uÞx k 1=2 2
mR ¼ ð6:71Þ
2
The characteristic length of the viscous solution, for small , is

1=2

l¼ ð6:72Þ
ðc ux Þk
For large Reynolds numbers, the ratio of the viscous length to the inviscid length is small
so that viscous effects are confined to a thin boundary layer close to the interface.
For a finite layer of height, hL , it is convenient to use (6.57) as the solution. The
dispersion relation is obtained by solving the complete determinant (6.51), for small
viscosity. This yields
c2R ¼ c20 tanhðkhL Þ ð6:73Þ
ðkÞ1=2 c0 tanh3=4 ðkhL Þsech2 ðkhL Þ
1=2
cΙ ¼ 2k þ 3=2
ð6:74Þ
ð 2Þ
where c20 is defined by (6.63). The two terms on the right side of (6.74), respectively,
represent the damping in thin viscous boundary layers at the interface and at the wall.
For very deep liquid layers (large khL ), sechðkhL Þ ¼ 0; equation (6.66) is recovered.
As khL decreases sech2 ðkhL Þ and wave damping increase. Note that the second term on
the right side of (6.74) varies as 1=2 (rather than ) so, for small , dissipation in the
boundary layer at the wall can be dominant.
A derivation of (6.73) and (6.74) was given by Kim (1968). He verified these equations
by generating two-dimensional waves with an oscillating dipper. Wave damping was
studied by measuring the decrease in amplitude downstream of the dipper, that is,
damping in space, rather than in time. Thus, the wave height is represented by
¼ a exp iðkx ct Þ ð6:75Þ
where c is real and k is complex

k ¼ k R þ ik I ð6:76Þ
Thus,
¼ a exp iðk R x ctÞexpðk I xÞ ð6:77Þ
so that k I is the spatial damping coefficient. As discussed in Chapter 4, Gaster (1962) has
shown that k I can be related to the temporal coefficient, k R cI , by using the equation
k R cI
kI ¼ ð6:78Þ
cg
where cI is calculated with (6.74) and cg is the group velocity, defined as

dcR
cg ¼ k ð6:79Þ
dk
Wave velocity cR is given by (6.73). Thus

3σk 2 þ ρg tanhðkhL Þ þ ρgk þ σk 3 sech2 ðkhL Þ
cg ¼ ð6:80Þ
2ρcR k R
Kim (1968) found good agreement between measurements of k I and cR with (6.73),
(6.74), if (6.78) is used to relate spatial damping to temporal damping.
6.6 Wave generation by a gas flow
6.6.1 Dispersion relation for large Reynolds numbers

As discussed in Section 2.2.1, the first disturbances that appear at the interface when air
flows over water are regular two-dimensional waves of the type shown in Figure 6.2. The
liquid will not have a uniform velocity so that dux =dy ≠ 0.
The analysis presented in this subsection was given by Cohen (Cohen & Hanratty,
1965; Cohen, 1964). It uses the height of the liquid, hL , and the velocity of the liquid at
the interface, ux0 , to make variables dimensionless. A low viscosity, that is, a large
Reynolds number, is assumed. The dimensionless wave number is of the order of
unity, so ðhL ρc=μÞðkhL Þ is also large. The equation for the disturbed flow created by
the interfacial waves has two inviscid and two viscous solutions, as shown in Section 6.5.
Figure 6.2 Photograph of two-dimensional waves generated by an air flow. Hanratty & Engen, 1957.
6.6 Wave generation by a gas flow 123
The inviscid solutions are obtained by considering only the left side of (6.19). For the
case to be considered, dux =dy ≠ 0. However,
d 2 ux =dy2
l¼ ð6:81Þ
k2c
is usually a small number so the two inviscid solutions are

uy 1 ¼ eky ûy2 ¼ eky ð6:82Þ
As discussed earlier in this section, the viscous solutions can be simplified since viscosity
is important only in thin boundary layers at the wall and near the interface, where the
velocity profiles are, respectively,

dux
u¼ ð hL þ y Þ ð6:83Þ
dy W

dux
u ¼ ux0 þ y ð6:84Þ
dy 0
with y = 0, the average location of the interface. Asymptotic solutions of (6.19) for ûy3
and ^
uy4 , in the limit of small viscous effects, have been presented by Cohen & Hanratty
(1965) and by Craik (1966). These were obtained by scaling the distances from the
wall and from the interface with viscous lengths characteristic of the sizes of the
viscous boundary layers. The solutions of the scaled differential equations for small
disturbances were determined by assuming that they can be represented by power
series of the form
^ uðy0Þ þ ðkReL0 Þ1=2 ûðy1Þ þ ðkReL0 Þ1 ûðy2Þ þ . . .

uy ¼ ^ ð6:85Þ
As mentioned above, all terms in the solution are scaled with u0 and hL . The expansion
parameter is kReL0 , where ReL0 is a Reynolds number based on the mean velocity at the
interface, ux0 , the height of the liquid layer, and k is made dimensionless with hL .
The main interest is to develop a dispersion relation that satisfies boundary conditions
(6.45), (6.46) and (6.37). The viscous solutions at the wall and at the interface are chosen
so that they decay very rapidly with distance away from the boundary. Therefore, the
solutions for the viscous boundary layers at the wall and at the interface are, respectively,
zero at the interface and at the wall. The determinant (6.51) simplifies to
j j
ð^
u ÞhL ð^
u ÞhL ðû ÞhL 0
y1 y2 y3
dûy1 dûy2 dûy3
0
0¼ dy dy hL dy hL ð6:86Þ
h L
H0 ûy1 H 0 ûy2 0 H 0 ûy4

G0 ûy1 G0 ûy2 0 G0 ûy4
For the case being considered, all terms in the equations for the H ðÞ operator and the GðÞ
operator are considered except that σ~ is set equal to zero.
The imaginary and real parts of the determinant (6.86) are separately set equal to zero.
These two equations give the real and imaginary parts of the wave velocity, cR and cI ,
which, respectively, define the wave velocity and the critical condition for waves to
appear. The imaginary part of (6.86) gives
" 2 #1=2
ðdux =dyÞ0 ðdux =dyÞ0 1 ~R
P
cI ¼ 1 tanh k tanh k þ tanh k kWe þ þ
2k 2k kFr2 k
ð6:87Þ
Equation (6.87) is the same as was obtained for inviscid flows, except that a term
involving ðDux Þ0 appears, where D ¼ d=dy. The Froude number and the Weber number
in (6.87) are given as Fr2 ¼ u2x0 =ghL and We ¼ σ=ρu2x0 hL .
For neutral stability, cI ¼ 0, the real part of (6.86) then gives

~ ~ ðDux Þ0 4ðDux Þ0 k 2 4k 3 ðc 1Þcoth k
PI þ T R coth k þ ¼ þ
k ð c 1Þ kReL0 kReL0
2
k ðc 1Þ coth 2 k 1
2
þ ð6:88Þ
ð2kcReL0 Þ12
Since P ~ I and T~ R are functions of the gas velocity, (6.88) gives the gas flow needed to
generate waves.
All of the terms in (6.87) and (6.88) have been made dimensionless with u0 and hL , so
the wave number k is the product of the dimensional wave number and hL . The term
multiplying T~ R is equal to the ratio of the amplitudes of the wave-induced streamwise
and normal velocity fluctuations, j^ uy0 j. If (6.88) is multiplied by ûy0, the terms on the
ux0 =^
left side represent the energy transferred from the gas to the liquid, while those on the
right side represent energy dissipated at the interface and at the wall. The shear stress
amplitude, T~ R , is expected to be an order of magnitude smaller than the pressure
amplitude, P ~ I , because the gas viscosity is small. However, for very small heights of
the liquid and/or large wavelengths, ^ ux0 >> ûy0 , so the contribution by the shear stress
could then be dominant.
In order to examine the growth of surface waves, the wave velocity is considered to be
complex. The time constant representing the growth rate, kcI , is given by the following
equation if cR >> cI and c2I is small
"
Dux0 ~ ~ Dux0
kcI 2ðcR 1Þcoth k þ ¼ PI þ T R coth k þ
k ð c R 1Þ k ð c R 1Þ
4Dux0 2k 2 ðcR 1Þcoth k

ReL0 ReL0 #
k 3=2 ðcR 1Þ2 ðcoth 2 k 1Þ
ð6:89Þ
ð2cR ReL0 Þ1=2
This would not be valid for gas velocities near that needed for a Kelvin–Helmholtz
instability. In these cases, the growth rate of waves can be “explosively” large.
6.6.2 ~ and T~
Influence of P
The critical issue in using (6.87), (6.88) and (6.89) is the evaluation of the wave-induced
variations of the pressure and the surface shear stress. Of particular interest is the
prediction of the gas velocity at which waves first appear.
This problem was first investigated by Lord Kelvin, who considered fluids of infinite
extent. Irrotational two-dimensional flows, with constant uL and uG , were assumed (see
Chapter 4). The predicted critical gas velocity is much larger than is observed for air–water
systems. This led Jeffreys (1924, 1925) to develop his theory of sheltering for which the gas
pressure in phase with the wave height is given by Kelvin–Helmholtz theory
^ R ¼ aρG k ðuG cR Þ2
P ð6:90Þ
and for which the pressure amplitude in phase with the wave slope is given by
^ I ¼ asρG k ðuG cR Þ2
P ð6:91Þ
where s is the sheltering coefficient. A rationale of this equation is that the gas separates
just behind the crest and reattaches near the trough. This creates a dead zone of low
pressure behind the wave. From observations of the initiation of waves on a body of
water, Jeffreys suggested that s ≈ 0:3.
Several difficulties arise in using this interpretation. Since very small waves exist at
transition, flow separation might not occur. The gas velocity varies with distance from
the interface, so one needs to select an appropriate velocity. One choice is to use the
gas velocity at y = 1/k. Another is to use a spatial mean. Articles by Benjamin (1959)
and by Miles (1957) are milestones in that they show sheltering can occur without
separation.
Cohen & Hanratty (1965) tested the Jeffreys equation by measuring the gas flow at
which waves are initiated for flow of air and water (or a water–glycerine solution) in a
rectangular channel with a height of 2.54cm and a width of 30.5cm. The wave
velocity was greater than the liquid velocity, so the liquid did not receive energy
from the average velocity profile in the liquid. That is, waves were not the result of a
Tollmien–Schlichting instability in the liquid, as suggested by Feldman (1957). Since
the wave velocity was much smaller than the average gas velocity, resonance with
turbulence in the gas (Phillips, 1957; Charles & Lilleleht, 1965) could not be
responsible for the observed waves. That is, components of the turbulent pressure
fluctuations with lengths comparable to the waves at the interface would have
velocities much higher than the wave velocities, since they would be comparable to
the average gas velocity.
Thus, one could expect that the energy is supplied to the waves by air flow through
pressure and shear forces at the interface, as described in (6.89). Cohen & Hanratty
(1965) substituted (6.91) into (6.88), with s = 0.3, and defined the characteristic gas
velocity as the spatial average. Since air has a small viscosity, the contribution of T^ R was
ignored. Approximate agreement was noted between the measured and the predicted
critical air velocity for the initiation of waves.
6.6.3 Predictions of P^ and T^ using the quasi-laminar assumption

The approach proposed by Jeffreys cannot be rationalized by assuming separation at
the initiation condition. Therefore, considerable attention has been given to the develop-
ment of theoretical predictions of P ^ I and T^ R . The approach taken by Miles
(1957,1959a,b,1962a,b) and Benjamin (1959) was to use an equation such as (6.14) to
represent wave-induced variation of the gas velocity. Turbulence in the gas is taken into
account only insofar as it defines the average velocity profile (a quasi-laminar approx-
imation). The thin viscous layer in the gas, which exists close to the interface, gives rise to
phase shifts in the wave-induced velocity fluctuations much the same as is found in
Rayleigh’s solution for the flow in a viscous fluid above an oscillating plate (Schlichting,
1968, p. 85). Consequently, the maximum in the velocity does not occur at the wave crest.
This causes a phase shift from that predicted by Kelvin–Helmholtz theory and an
interfacial shear stress profile which has a maximum upstream of the wave crest. Miles
also considered viscous effects for a non-turbulent gas flow in the critical layer, where the
gas velocity equals the wave velocity.
Cohen & Hanratty (1965) compared experimental results with a solution of the quasi-
laminar disturbance equation in a Cartesian coordinate system (6.19). Τhe pressure and
the shear stress in the gas phase along the wave interface were obtained from this
calculated velocity field. They used numerical methods developed by Miles (1962a,b)
and measurements of the profile of the average gas velocity in a channel with smooth
walls. The predicted critical gas-phase Reynolds number for the appearance of waves
was somewhat larger than observed, but it was close enough to encourage further
exploration of this linear theory.
6.6.4 Changes in the quasi-laminar theory to include wave-induced turbulence

The works of Miles and Benjamin have prompted a number of studies which have
focused on the accuracy of the quasi-laminar assumption and on the influence of the
choice of a coordinate system in which to formulate the disturbance equation.
For example, Benjamin (1959) used an orthogonal curvilinear coordinate system, ξ−,
for which ¼ 0 coincides with the wave profile when the waves have a small amplitude.
Coordinates of constant correspond to the streamlines determined in the inviscid
analysis in Chapter 4. (However, it is noted that Benjamin argues that the influence of
terms introduced by the use of curvilinear coordinates have small effects.) A conformal
mapping technique (Caponi et al., 1982; McClean, 1983) has been developed to map on
a rectangular grid the region between a wavy boundary and a plane at T , located in the
log-layer of a deep turbulent boundary layer. This mapping has also been adapted
to the case where is a rigid wavy wall (Frederick, 1986; Frederick & Hanratty, 1988).
Thorsness et al. (1978) and Abrams & Hanratty (1985) have used a boundary-
layer coordinate system for which and ξ are, respectively, perpendicular and parallel to
the wave surface. Solutions in boundary-layer coordinates are matched to an outer
solution which is linearized around an average profile defined in a Cartesian coordinate
system.
In what follows, a Cartesian coordinate system is used in order to demonstrate how

the equation for small disturbances changes if the effects of turbulence are considered.
In Cartesian coordinates, the equation of conservation of mass for an incompressible fluid is
∂uGj
¼0 ð6:92Þ
∂xj

where uGj signifies the j-component of the gas velocity. If uGi ∂uGj =∂xj ¼ 0 is added to
the equation for conservation of momentum, equation (6.1) can be written as

∂uGi ∂uGj uGi ∂p ∂τ ji
ρG þ ρG ¼ þ ð6:93Þ
∂t ∂xj ∂xj ∂xj
where τ ji represents the effect of viscous stresses (see equation (6.3)).

The variables are represented as the sum of a mean and fluctuations due to turbulence.
Thus
uGi ¼ uGi þ utGi ð6:94Þ
where, by definition, uti ¼ 0. Take an average of (6.93) to get

!
∂uGi ∂utGi utGj ∂p ∂τ ji
ρG uGj þ ¼ þ ð6:95Þ
∂xj ∂xj ∂xi ∂xj
The second term on the left side represents a net flow of momentum out of the control
volume due to turbulence. It is usually brought to the right side, multiplied by the density
and called a Reynolds stress. Define
Rij ¼ utGi utGj ð6:96Þ
For an incompressible flow over a flat surface,

∂uGi ∂p ∂τ ji ∂ Rji
ρG uGj ¼ þ þ ρG ð6:97Þ
∂xj ∂xi ∂xj ∂xj
Two-dimensional interfacial waves induce two-dimensional variations of Rij , as well as

of uGi , so that
Rij ¼ Rij þ rij ð6:98Þ
For small-amplitude waves, an equation analogous to (6.14) can be used to represent rij
and u0Gj .
rij ¼ ^rij exp ik ðx ct Þ ð6:99Þ
The substitution of (6.98), (6.99) into (6.97) and the use of the assumption of small
disturbances yields two equations analogous to (6.17) and (6.18) for disturbances
introduced by the waves, with the exception that additional terms involving ^rxy , ^rxx ,
^ryy , ^ryx and their derivatives appear on the right side. The elimination of ^p between the
x- and y-momentum balance equations yields
2
d ^
uGy 2 d 2 uGx
ð uG c Þ k ^
u Gy ^
u Gy
dy2 dy2
4 2 ð6:100Þ
i d ^
uGy 2d ûGy 4
¼ G 2k þ k ^
u Gy ikℜ
k dy4 dy2
where
ℜ ¼ D2^rxy þ ikD^rxx þ k 2^ryx ikD^ryy ð6:101Þ
with D ¼ d=dy. A comparison with (6.19) shows that the effect of wave-induced
turbulence is represented by ℜ in Cartesian coordinates. The theoretical challenge is
the modeling of ℜ. A number of approaches have been explored.
6.7 Flow over solid wavy surfaces
6.7.1 Measurements of shear stress along a solid wavy surface

The testing of turbulence models by measuring stresses at a gas–liquid interface has not
been realized. Therefore, studies of turbulent flow over solid waves have been pursued.
Pressure profiles have been measured for air and water flows over solid waves by a
number of investigators (Kendall, 1970; Hsu & Kennedy, 1971; Sigal, 1970; Beebe,
1972; Zilker, 1976; Buckles et al., 1984; Kuzan, 1983). These are summarized by
Abrams & Hanratty (1985). The drawback with these studies is that the realization of
measureable pressure variations requires the use of waves with large amplitudes. Thus, a
linear behavior is usually not observed.
This motivated studies of turbulent liquid flow over solid waves with very small
amplitudes. Measurements of the profiles of wall shear stress were made by Cook
(1970), Thorsness et al. (1978), Abrams & Hanratty (1985). In these studies, the wavy
boundary formed one wall of a 2in × 24in rectangular channel through which an electrolyte
was circulated. The waves were preceded by a 27.5ft length of channel, with smooth walls,
so that studies were made with a fully developed turbulent velocity field. Wall shear stresses
were determined by using electrochemical techniques developed by Reiss & Hanratty
(1963), Mitchell & Hanratty (1966), Son & Hanratty (1969). The basic idea is to measure
the rate of mass transfer to small circular electrodes mounted flush with the wall. These
electrodes form cathodes of electrochemical cells, which have an anode of much larger
area. A large enough voltage is applied so that the current is controlled by the rate of mass
transfer to the cathode and the concentration of the reactive species is zero at the surface of
the cathode. A mass transfer coefficient, K, is defined and, from a mass balance equation,
the wall shear stress is related to the mass transfer coefficient by the equation

2KΓ ð4=3Þ 4Le μ
τW ¼ ð6:102Þ
3 D2
where D is the diffusion coefficient, Γ is the gamma function and Le is the equivalent
length of the electrode, equal to 0.82 times the diameter of the electrode.
1.4
1.2
τW 1.0
τW Re = 25 200
0.8
0.6
Wave height (in)
Flow
0.02
0.01
0
0 0.2 0.4 0.6 0.8 1.0
x
λ
Figure 6.3 Profile of boundary shear stress. Thorsness et al., 1978.
An example of measurements of the wall shear stress is given in Figure 6.3. This shows
part of a train of ten sinusoidal waves with an amplitude of 0.0114 in and a wavelength of
2 in. The Reynolds number characterizing the flow was 25 200, where Re is based on the
channel half-width and the bulk-averaged velocity.
6.7.2 Equations for flow over solid waves, in boundary-layer coordinates

The solid wave is not moving (c = 0), so that the possible influence of a critical layer,
where the fluid velocity and the wave velocity are equal, need not be considered. The line
drawn through the data in Figure 6.3 represents the best fit, in a least-squares sense, of a
sinusoidal curve with a wavelength equal to that of the waves on the surface. The results
suggest that the waves had small enough amplitudes for linear theory to be applicable;
that is, the measurements contain only one harmonic. Further evidence for linearity is that
the average shear stress over the wavy surface is the same as would be measured in a
channel with flat walls. These measurements of τ W can be represented by an amplitude
and a phase angle (the number of degrees that the maximum is upstream of the crest).
They were interpreted by Thorsness et al. (1978) by using a boundary-layer coordinate
system. The y-axis is perpendicular to the wave surface, with y = 0 corresponding to the
surface; the x-axis is parallel to the wave surface. The velocity components in the x- and
y-directions are given as
ux ¼ hux i þ ûx eikx ð6:103Þ
uy ¼ ûy e ikx
ð6:104Þ
where ux is the time-averaged velocity at a given y and hux i is the average along one
wavelength. From conservation of mass
dûy
ikûx ¼ ð6:105Þ
dy
The following equation is derived for ^ uy by using conservation of momentum:

2
d ûy d 2 hux i
hux i k 2
^
u y ^
u y ik 3 hux i2 a
dy2 dy2
ð6:106Þ
i d 4 ûy 2d ^
2
uy 2 d hux i
2
¼ 2k þ k 4
^
u y a2k ak 4
hu x i þ ℜ
k dy4 dy2 dy2
The wavelength average of ux is set equal to the average for a flat plate and

ℜ ¼ aik 3 Rxx þ a3k 2 DRxy þ ik D^rxx D^ryy þ k 2^rxy þ D2^rxy ð6:107Þ
A comparison of (6.106), (6.107) with (6.100), (6.101) shows that additional terms are
introduced when formulating the disturbance equation in curvilinear coordinates. Of
these changes, the inclusion of a centripetal acceleration term (the last term on the left
side of equation (6.106)) could be important. All of the non-homogeneous terms are
linear in the amplitude, a, as would be expected since ûy varies linearly with a because of
the assumption of small-amplitude waves.
Equation (6.106) is solved numerically using the boundary conditions of zero velocity
and zero velocity gradient at the wall
dûy
^
uy ¼ 0 ¼ 0 at y ¼ 0 ð6:108Þ
dy
and the expectation that a velocity in the fluid far from the wave surface is the same as
would be present for a flat surface:
d^
uy dux
¼ ika ð6:109Þ
dy dy
^
uy ¼ ux ika ð6:110Þ
The shear stress at the boundary is calculated from the solution for ûy as follows:

T hT i d 2 ûy
¼ eikx ð6:111Þ
ρ ρ ik dy2 0
The pressure at the wavy boundary is obtained from the x-momentum equation evaluated
at the wall.
2
P hPi d 3 ^ uy d hux i
¼ þ e ikx
aik eikx ð6:112Þ
ρ ρ k dy3 0 dy2 0
6.7.3 Evaluation of r iĵ using a modified van Driest model

From (6.97) it is seen that the equations for a turbulent fluid require the specification of
turbulent stresses
τ tij ¼ ρuti utj ¼ ρRij ð6:113Þ

in addition to viscous stresses. The simplest approach is to represent Rij by a Boussinesq

approximation (Hinze, 1975, pp. 23–24), which is analogous to (6.3).

qt2 t ∂ui ∂uj
Rij ¼ δij þ þ ð6:114Þ
3 ∂xj ∂xi
where t is a turbulent kinematic viscosity, which can vary with spatial location, and
qt2 ¼ ut2
x þ uy þ uz
t2 t2 ð6:115Þ
The first term on the right side of (6.114) is the average pressure due to turbulence.
Assume that the waves not only induce variations in the fluid velocity but also in the
turbulent viscosity so that
^t exp ik ðx ct Þt
t ¼ t þ ð6:116Þ
Neglecting higher-order terms in the wave-induced fluctuations, equation (6.114) gives

∂ûi ∂ûj ∂ui ∂uj
Rij ¼ Rij þ t þ ^t
þ þ ð6:117Þ
∂xj ∂xi ∂xj ∂xi
Thus

∂ui ∂uj ∂ûi ∂ûj
^rij ¼
^t þ þ t þ ð6:118Þ
∂xj ∂xi ∂xj ∂xi
is substituted into (6.106), (6.107) to obtain an equation for u^y in boundary-layer coor-
dinates. The theoretical issue in this Boussinesq formulation is the specification of t .
The influence of turbulence can be seen by examining measurements for fully devel-
oped flow in a channel or a pipe with flat walls. Consider a horizontal volume of unit
breadth attached to the wall. The force due to the existence of a pressure gradient is
yW Dp. This is balanced by shear stresses at the top and bottom of the volume,
τDx τ W Dx. Thus
Dp
ðτ τ W Þ yW ¼0 ð6:119Þ
Dx
For a symmetric fully developed flow in a channel, τ ¼ 0 at the half-width,

yW ¼ B=2, so
B Dp
τW ¼ ð6:120Þ
2 Dx
where Dp=Dx is negative. The friction velocity, v ¼ ðτ W =ρÞ1=2 , can be calculated if the
pressure gradient is measured. The local fluid shear stress, which is obtained from
(6.119), can be set equal to the sum of a turbulent and a viscous contribution
τ yx ¼ τ ℓyx þ τ tyx ð6:121Þ
where
- Re1 Bv *ρ
Re =
- Re2 μ
Re2 > Re1
uG+ = A + C log10y+
uG
v* 2y B ≈ 0.20
2y B ≈ 0.20
y+ ≈ 40
log10(y+W = yWv*ρμ)
Figure 6.4 Sketch of a turbulent velocity profile where B is the half-height of the channel.
dux
τ ℓyx ¼ μ ð6:122Þ
dy
τ tyx ¼ ρuty utx ð6:123Þ
Measurements of turbulent velocity profiles are sketched in Figure 6.4, where the half-
height of the channel is designated by B=2.
Velocities and distances from the wall have been made dimensionless with the friction
velocity and the friction length =v . The region yþ 55 is called the laminar sub-layer
since turbulent stresses are negligible compared to viscous stresses and the velocity is
given by
uþ ¼ y þ ð6:124Þ
For yþ > 40 viscous stresses can be neglected. In the region yþ ¼ 40 to 2y/B = 0.20, the
velocity profile is given by
uþ ¼ A þ C log10 yþ ð6:125Þ
where A = C = 5.6. As sketched in Figure 6.4, the region beyond 2y=B ¼ 0:20 is a
weak function of Re ¼ Bv =. The choice of A = 5.5 and C = 5.75 gives a good
approximation over the whole pipe cross-section, with the exception of yþ 540
(see Schlichting, 1968). The region between yþ ¼ 5 and the inner edge of the log-
layer, where both viscous and turbulent stresses are important, is known as the buffer
layer.
A frequently used approach in specifying the turbulent stresses is the mixing-length
theory of Prandtl. This is inspired by the kinetic theory of gases whereby the kinematic
viscosity is shown to be proportional to the product of the molecular velocity and the
molecular mean-free path. Thus, the kinematic turbulent viscosity is given as
t / vt ℓ ð6:126Þ
where vt is a characteristic turbulent velocity and ℓ is the mixing length.

Prandtl assumed that vt / ℓjdux =dyj. Thus, in a boundary layer

τt dux dux
¼ ℓ 2 ð6:127Þ
ρ dy dy
Consider a turbulent boundary layer on a flat surface where the pressure gradient is
zero and the fluid stress, τ ¼ τ t þ τ ℓ , is constant and equal to τ W . Beyond yþ ¼ 40
equation (6.127) gives

dux dux
v2 ¼ ℓ 2 ð6:128Þ
dy dy
where v ¼ ðτ W =ρÞ1=2 , since τ ¼ τ t ¼ τ W . From (6.125) the velocity gradient in the

log-layer is
dux v
¼ ð6:129Þ
dy κy
where κ ¼ 0:4 is the von Karman constant. Substituting (6.129) into (6.128) gives
ℓ ¼ κy ð6:130Þ
and from (6.130) and (6.126)

t ¼ κv y ð6:131Þ
Thus, both ℓ and t vary linearly with y in the log-layer.

Van Driest (Hinze, 1975, p. 622) suggested the following relation for the mixing
length

yv
ℓ ¼ κy 1 exp ð6:132Þ
A
Note that, at large y, (6.132) gives (6.130). The term in brackets represents the role of
viscosity in damping the mixing length in the buffer layer. The parameter A is a measure
of the thickness of the viscous wall region. A value of 25 fits measurements of the
velocity profile over a flat boundary, for which dp=dx ¼ 0. Loyd et al. (1970) applied
(6.132) to turbulent boundary layers with a range of imposed pressure gradients.
The thickness of the viscous wall layer varies with pressure gradient so A is given by
the function
2 !
dp dp
A ¼ A 1 þ k1 þ k2 þ . . . ð6:133Þ
dx dx
with k 1 ¼ 30 and k 2 ¼ 1:54 103 for equilibrium flows for which the pressure
gradient is not varying. For flow over a small-amplitude sinusoidal solid wave, the
quadratic term in (6.133) can be neglected, so
A ¼ A þ a A~ eix ð6:134Þ
The pressure variation is given as

p ¼ p þ a~peikx ð6:135Þ
From equation (6.133) with c ¼ 0 (since a solid wave is being considered)

~ ¼ aAk 1 ik~pð0Þ
aA ð6:136Þ
The substitution of (6.134), (6.136) into (6.132) provides an equation for the wave-induced
variation of the mixing length.
Loyd et al. (1970) have suggested that for non-equilibrium conditions, such as exist for
flow over waves, an effective pressure gradient should be used in (6.133), where

dp dp dp
dx dx eff
d dx
eff
¼ ð6:137Þ
dx kL
This introduces a lag between the imposition of a non-zero pressure gradient and a
change of the scale, for which the following relation is obtained
k 1 Aik~pð0Þ ikx
A¼Aþa e ð6:138Þ
1 þ ikk L
Loyd et al. suggested that k L ¼ 3000.
6.7.4 Testing turbulence models against wall shear stress measurements

Thorsness et al. (1978) used measurements of the shear stress variation along
small-amplitude solid waves, such as shown in Figure 6.3, to test different models for
the Reynolds stress. The profiles of wall shear stress were represented by phase angles
and amplitudes, plotted against the dimensionless wave number, as shown in Figures 6.5
and 6.6. The quasi-laminar approach (Section 6.6.3) involves the neglect of all terms in
(6.106) and (6.107) which involve ^rij or its derivatives. The resulting equation differs
from the formulation in a Cartesian system because of terms associated with the use of
curvilinear coordinates.
Thorsness et al. found that the quasi-laminar assumption in Cartesian coordinates
produces large errors. The use of the quasi-laminar assumption in boundary-layer
coordinates overpredicts the magnitude of T^ ¼ aT~ (Figure 6.6) and incorrectly predicts
that the phase angle is relatively insensitive to changes in k þ (Figure 6.5). This led
Thorsness et al. to explore whether improved results could be obtained by considering
wave-induced variations in the turbulence.
100
Abrams & Hanratty (1985)
Kendall (1970)
Hsu & Kennedy (1971) Relaxation theory
80 k1 = –33 kL = 1650
Sigal (1970)
τW ρ
60
τW
Phase angle of
40
Quasi-laminar model
Frozen turbulence model
^
l =0
20 0
0
Equilibrium turbulence model
k1 = –33 kL = 0
–20
0.00001 0.0001 0.001 0.01 0.1 1.0
k+ = kυ/v*
Figure 6.5 The phase angle characterizing the variation of the wall shear stress as a function of the
dimensionless wave number. Abrams & Hanratty, 1985.
100
Abrams & Hanratty (1985)
Kendall (1970)
Hsu & Kennedy (1971)
1.0 Sigal (1970)
0.1
Equilibrium turbulence model

τW 32 ρ 12
k1 = –33 kL = 0
τ^ W μ
0.01
Frozen turbulence model
^
l0 = 0
Relaxation theory
0.001 k1 = –33 kL = 1650
0.0001 Quasi-laminar model
0.00001
0.00001 0.0001 0.001 0.01 0.1 1.0
k+ = kυ/v*
Figure 6.6 The amplitude characterizing the variation of the wall shear stress as a function of the
dimensionless wave number. Abrams & Hanratty, 1985.
From (6.117) and (6.118), it is seen that the wave-induced variations of Rij can have
contributions associated with t and ^t . Hussain & Reynolds (1970) assumed ^t =0. This
approach did not produce results much different from the quasi-laminar approach. Thus,
^t needs to be taken into account. By using the model of Loyd et al. (1970),
the effect of
Thorsness et al. found that a good representation of the measurements could be obtained
by using k1 = −30 and k L ¼ 1500, values close to what were used by Loyd et al. for quite
different systems.
However, this fit suggested that a sharp change in phase should be experienced at
values of k þ smaller than were studied by Thorsness et al. (1978). Abrams & Hanratty
(1985) extended the experimental study to include smaller k þ by using larger flow rates
in the same equipment. Their measurements of the phase and amplitude, along with
those of Kendall (1970), Hsu & Kennedy (1971) and Sigal (1970) are plotted in
Figures 6.5 and 6.6. Note that the results of Abrams & Hanratty are predicted quite
well by relaxation theory by using k 1 ¼ 33 and k L ¼ 1650.
For very small k þ (large wavelengths) the flow is changing so slowly that an
equilibrium or pseudo-steady-state behavior exists. For large k þ (small dimensionless
wavelengths), the flow is changing so rapidly that the turbulence is frozen, ^μt ¼ 0. At
intermediate k þ , the flow is in a transition region between these two extremes.
6.8 Response of turbulence to imposed temporal oscillations
6.8.1 Motivation
The results in Figures 6.5 and 6.6 show that the neglect of turbulence effects in a
Cartesian coordinate system (Miles–Benjamin approach) or in a curvilinear system
(quasi-laminar approach) does not describe accurately the changes of the wall shear
stress along a wavy solid boundary. Some improvement is obtained by considering
wave-induced turbulence variations. Of particular interest is the accounting for the
sharp change in the phase of the shear stress profile observed over a narrow range
of wave numbers by using a turbulence model which allows for relaxation. The
picture behind the model is that the wave-induced variations of the pressure gradient
cause the viscous wall layer to change its thickness. As the wave number increases
the spatial variation of the pressure gradient is too rapid for the turbulence to respond
fully and changes in the pressure gradient are less effective in changing the
turbulence.
One way to check this model is to look at the effect of the imposition of a sinusoidal
temporal oscillation on the flow in a smooth walled pipe. Such experiments have been
carried out by Finnicum & Hanratty (1988) in a 5.08 cm pipe, by Mao & Hanratty (1986)
in a 19.4 cm pipe and by Ramaprian & Tu (1983) in a 5 cm pipe. If the amplitude of
the imposed oscillation is small enough, a linear response is obtained for which the
phase-averaged field is described as follows:
ux ¼ ux ðyÞ þ ûx ðyÞ cos ½ϖt þ θu ðyÞ ð6:139Þ
dp dp d^p
¼ þ cos ϖt ð6:140Þ
dx dx dx
with x being the distance in the flow direction, y the distance from the wall, ϖ the angular
frequency, t the time and θu , the phase relative to the pressure gradient. The time-mean
pressure gradient and the time-mean velocity of the undisturbed flow are designated by
dp=dx and ux ðyÞ. The oscillations of the wall shear stress are given by
τ W ¼ τ W þ ^τ W cos ðϖt þ θτ W Þ ð6:141Þ
where θτ W is the phase of τ W relative to the pressure gradient. Thus, for the flows being
considered, the velocity field is exposed to a time-varying pressure gradient, while flow
over a solid wavy wall is exposed to a spatially varying pressure gradient. The geometric
complications associated with a wavy wall are avoided.
6.8.2 Measurements
Figure 6.7 presents measurements of d^p=dx and of ^τ W =τ W obtained in a 5.08cm
pipe for a Reynolds number of 17100. The ratio of the oscillations of the centerline
velocity to the mean centerline velocity, a, was 0.102. The frequency of the imposed
oscillations was f = 0.50Hz. The circular frequency, made dimensionless with wall
parameters, was ϖþ = 0.0093. The curve through the data represents a best least-squares
fit of a cosine function. Note that ^τ W lags d^p=dx by about 90° . Figure 6.7 shows a
linear behavior in that the measurements are described by a single harmonic and the
mean shear stress at the wall is unchanged.
Measurements of the amplitude, ^τ W , and of the phase of τ W relative to the pressure
gradient, θτW , in a 5.08cm pipe and in a 19.4cm pipe are plotted in Figures 6.8
0.004
+
^
–dp
0.000
dx+
–0.004
0 90 180 270 360
Phase
1.25
τ^W
τW 1.00
0.75
0 90 180 270 360
Phase
Figure 6.7 Phase-averaged values of the wall shear stress and the pressure gradient at Re = 17 100, f = 0.50 Hz
and a = 0.102 centerline velocity. Finnicum & Hanratty, 1988.
100
Finnicum & Hanratty (1988) (dt = 5.08 cm)
Mao & Hanratty (1986) (dt = 19.4 cm)
Model D Re = 20 000
Model D Re = 50 000
Quasi-laminar solution
^
τ W
+
10–1 10–1
^+
u c
10–2
10–3 10–2 10–1 100
ω+
Figure 6.8 Amplitude of the wall shear stress variation. Finnicum & Hanratty, 1988.
–30
–40
–50
θ τW
–60
–70
Finnicum & Hanratty (1988) (dt = 5.08 cm)
Mao & Hanratty (1986) (dt = 19.4 cm)
–80 Model D Re = 20 000
Model D Re = 50 000
–90 Quasi-laminar solution
–100
10–3 10–2 10–1 100
ω+
Figure 6.9 Phase lag (relative to the pressure gradient) of the wall shear stress variation. Finnicum &
Hanratty, 1988.
and 6.9 against the circular frequency. The amplitude ^τ W is divided by the amplitude
of the oscillations of the centerline velocity ûc. All variables have been made
dimensionless using wall parameters, v ¼ ðτ W =ρÞ1=2 and . Measurements of Mao
& Hanratty, of Finnicum & Hanratty and of Ramaprian & Tu agree, if plotted in
this way. A striking feature is the sharp change in the phase angle when ϖþ =15 is
in the range 0.0005–0.002 (where cþ ¼ 15 is the convective velocity of turbulence in
the viscous wall layer). This represents a transition from a quasi-steady behavior
at low frequencies to a quasi-laminar behavior at high frequencies.
Designate ~ ~
ux ðy; t Þ as the difference between the phase-averaged velocity, ux ðy; t Þ,
and the time-averaged velocity, ux ðyÞ. The following equation for ~~ux is obtained from
the x-momentum balance for flow in a pipe, where τ t is the phase-averaged Reynolds
stress and τ t is the time-averaged Reynolds stress:
~
∂~
~
ux ~
~ 1∂
∂p μ∂ ∂~ux
ρ ¼ þ ½ r ðτ τ Þ þ
t t
r ð6:142Þ
∂t ∂x r ∂r ρ ∂r ∂r
In order to solve (6.142), the phase-averaged Reynolds stress, τ t , needs to be specified.

By using the eddy viscosity model and assuming small wave-induced fluctuations
~
~ ∂~ux ~t ∂ux
~τ t ¼ τ t τ t ¼ ρ t ~
þ ð6:143Þ
∂r ∂r
where the time-averaged stress, τ t , is given as

∂ux
τ t ¼ ρt ð6:144Þ
∂r
6.8.3 Comparison with quasi-steady solution

In the limit of small ϖ the inertia term on the left side of (6.142) can be ignored. Then the
relation between τ W and the centerline velocity, uc , is the same as obtained for steady flow
1
τ W ¼ f ρu2c ð6:145Þ
2

d t uc n
f ¼A ð6:146Þ

~τ W to ~
The quasi-steady solution relating ~ uc is obtained by substituting τ W ¼ τ W þ ~~τ W and
~
~
uc ¼ uc þ ~
uc into (6.145) and (6.146). The following result was presented by Finnicum &
Hanratty
~
~τ þ
W
¼ 0:339 Re−0:125 ð6:147Þ
~
~
uc
Equation (6.147) is a weak function of Reynolds number. Over the range of Reynolds
numbers covered by the experiments (10000–50000), ~~τ þ ~uþ j is a good
W ¼ 0:095j~ c
approximation. In addition, quasi-steady behavior predicts that the phase of ~~τ W would
be the same as for ~ uc (90°). Figures 6.8 and 6.9 indicate that for ϖþ ≤ 0:004 the
~
quasi-steady prediction roughly approximates amplitudes and phases of the flow-induced
variations of the magnitude and phase of the flow-induced turbulent velocity fluctuations.
Note also that ϖþ ¼ 0:004 corresponds to the frequency at which the plug flow at the
center of the pipe disappears and for which the relationship of the centerline velocity to
the favorable pressure gradient differs from (6.149).
For almost all of the ϖþ covered in Figures 6.8 and 6.9 viscous and turbulent stresses can
be ignored in the center of the pipe, so the disturbed velocity profile is flat and described by
∂~~ux ∂p~~
ρ ¼ ð6:148Þ
∂t ∂x
A solution of (6.148) gives

∂^p
ϖρjûj ¼ j j ð6:149Þ
∂x
and a phase angle relative to the favorable pressure gradient of

1
θû ¼ π ð6:150Þ
2
The region outside the core is influenced by turbulent and viscous stresses. As ϖ
increases the region of plug flow increases. Eventually, it extends to the laminar sub-layer
where a quasi-laminar model is valid
∂~
~
ux ∂~p~ μ ∂2 ~~ux
ρ ¼ þ ð6:151Þ
∂t ∂x ρ ∂y2
with y being the distance from the wall. The solution of (6.151) is presented by Uchida
(1956). It gives a shear stress at the wall which leads the pressure gradient by 45° and has
an amplitude given by
j~
~τ þ
Wj
¼ ðϖ þ Þ
1=2
ð6:152Þ
j~~ þ
u j
c
The quasi-laminar solution is indicated by the dotted lines in Figures 6.8 and 6.9. Good
agreement between predicted and measured amplitudes is observed for large ϖþ in
Figure 6.8. Agreement is also noted for the phase angle at large ϖþ shown in
Figure 6.9. The same type of relaxation observed for measurements of the spatial
variation of τ W induced by a wavy boundary over a range of dimensionless wave
numbers of 0.0005–0.002 is shown in Figure 6.9. This suggests that a similar mechanism
is operating in the two studies. These transitions can be compared numerically by
replotting Figure 6.9 using ϖþ =cþ as the abscissa, where cþ ¼ 15 is the convection
velocity characterizing turbulence in the viscous wall region.
Both Mao & Hanratty (1986) and Finnicum & Hanratty (1988) used the model
described in Section 6.7.3 to capture the behavior at intermediate frequencies. The
main difference is that the relaxation equation (6.137) is rewritten as
þ þ þ
d dpdx dxþ dxþ eff
dp dp
eff
¼ ð6:153Þ
cþ dt þ kL
Equation (6.142) has been solved numerically by using the analytical solution of (6.151)
by Uchida (1956) as the initial condition and the boundary condition that the velocity
is zero at the wall. Calculations by Finnicum & Hanratty (1988) used k 1 ¼ 25 and
k L ¼ 3500 (Model D). A good fit to the measured amplitudes is shown in Figure 6.8.
A satisfactory fit to the phase angle data is shown in Figure 6.9. The relaxation model
does a reasonable job in representing the response of turbulent shear stresses both to a
wavy boundary and to temporally induced oscillations. However, the physics behind the
model has not been established. Clearly, more work on this problem is needed.
6.9 Prediction of wave generation 141
Note that the amplitudes measured by Mao & Hanratty at high frequencies are below
the quasi-laminar solution. Their experiments were done in a larger pipe than used by
Finnicum & Hanratty. Mao pointed out that, in a 19.4cm pipe, the frequency of the
imposed oscillations is close to that of the turbulent velocity fluctuations. This suggests
that there could be a direct interaction between the imposed flow oscillations and the
turbulence.
6.9 Prediction of wave generation
Section 2.2 describes studies of the initiation of waves by an air flow (Cohen & Hanratty,
1965) in an enclosed horizontal channel with a height of 2.54cm and a width of 30.5 cm.
The height of the liquid layer was large enough for the waves to receive their energy
from induced gas-phase pressure variations in phase with the wave slope. The critical
gas-phase Reynolds number, defined with the height of the gas space and the spatially
averaged gas velocity, was measured as ReG ¼ 3600 for a liquid layer with a viscosity
of 3.9cP and a height of 0.0147ft flowing along the bottom of the channel. Observed
waves extended over the whole width of the channel. At the same liquid flow, these
two-dimensional waves were observed up to ReG ¼ 9600. The waves became three-
dimensional above this gas flow.
Figure 6.10 shows loci of cI ¼ 0 calculated by Cohen & Hanratty (1965) using the
quasi-laminar assumption in a Cartesian coordinate system that was explored by Miles
and by Benjamin (see Section 6.6.3). Stable conditions (cI 50 ) are predicted to exist
outside the curve labeled “Miles–Benjamin.” Note that the gas Reynolds number below
which unstable conditions do not exist is predicted to be about ReG ¼ 4100. The points
represent wave numbers of observed two-dimensional waves. The dashed curve is the
fastest-growing wave number predicted by linear theory.
400
Model D*
Wave number k (f t–1)
100
Model A
Miles–Benjamin
20
Observed wave number
4000 5000 6000

Gas Reynolds number ReG
Figure 6.10 Calculation of neutral stability curve and loci of maximum growth for water–glycerine
(H = 0.0147 ft, µ = 3.9 cP).
Frederick (1982) also calculated the behavior in a boundary-layer coordinate system

for two cases: (1) where wave-induced variations of the turbulence are ignored (Model A
in Figure 6.10); (2) where wave-induced variations of the turbulence are calculated
(Model D ) with the relaxation theory outlined in Section 6.7.2, using k 1 ¼ 55 and
k L ¼ 3000. The wave velocities observed in the experiments were small, cþ ¼ 1 3.
Therefore, the influence of c was ignored; that is, the equations developed for a wavy
solid boundary were used. (Frederick showed that P ~ I is not sensitive to changes in cR
when the wave velocity is small.)
As seen in Figure 6.10, the calculations are not highly dependent on the model
~ I . The chief limitation is that there are small differences in the
that is used to calculate P
prediction of the critical gas Reynolds number. The use of a relaxation model for the
wave-induced turbulence produces a value of the critical gas Reynolds number close to
observation, Re ¼ 3600.
6.10 Dissolution patterns
A consideration of the variation of mass transfer along a small-amplitude wavy surface is

an interesting extension of the theory outlined in Section 6.7. If a turbulent fluid flows
over a soluble planar surface, material need not be removed uniformly along the surface.
As outlined by Hanratty (1981) and by Thorsness & Hanratty (1979b), a growing
two-dimensional wave can emerge if the function describing the variation of the
convective mass transfer rate along a small-amplitude sinusoidal waveform has a
maximum somewhere in the trough region. Depending on whether the maximum is on
the downstream side of the trough, the upstream side of the trough or right at the bottom
of the trough, the wave will propagate in the direction of flow, will propagate in a
direction opposite to the flow, or will remain stationary.
The chief motivation for carrying out such a stability analysis is to explain the origin of
wavelike patterns that are caused by diffusional processes and to relate the wavelength
and wave velocity to flow conditions that prevailed during their formation. Hanratty
(1981) provides a discussion of the occurrence of this type of process in nature.
Wavelike patterns are often observed on limestone walls that are or have been exposed
to flowing water. It is generally accepted that they result because CO2 diffuses to the
surface and bicarbonate diffuses back to the main stream (Curl, 1966). The patterns
usually appear as closely packed hollows called “scallops.” However, they also appear as
two-dimensional structures characterized by sharp, parallel crests that lie transverse to
the flow direction. These have been identified by Curl as “flutes.” Similar patterns have
been observed on snow or ice that is ablating into an air stream. Goodchild & Ford (1971)
argue that these patterns are associated with convective heat transfer. Blumberg & Curl
(1966) have indicated that flutes observed in limestone–water systems and in ice–air
systems are characterized by λU B = ≈ 23000 or by λv = ¼ 2200, where v is the friction
velocity and U B is the bulk velocity.
Waveforms similar to flutes have been observed on the underside of ice covers on
rivers or canals (Ashton & Kennedy, 1972; Hsu et al., 1979). These are described as
6.10 Dissolution patterns 143
regular, long-crested, downstream-migrating waves which develop under conditions

in which ice is melting. Controlled experiments by Hsu & Kennedy (1971) give
λv = ¼ 3180.
To consider the stability of a dissolving surface, define a disturbance at the interface of
the following form:
hðt Þ ¼ hðt Þ þ aðt Þ cos ðkx kct Þ ð6:154Þ
Term hðt Þ is the average height of the surface at any time, and term aðt Þ is the amplitude
of the disturbance wave. A mass balance for a surface which is dissolving solely because
of convective transfer is
dh
ρS ¼N ð6:155Þ
dt
where ρS is the density of the solid and N is the local rate of convective transfer per unit
area from the solid to the flowing fluid.
The surface flux can be decomposed into mean and fluctuating components defined in
the following way:
N ¼ N aj~
njcos½ðkx þ θÞ kct ð6:156Þ
where the minus sign is to be used when convective transfer is occurring from the fluid to
the solid. The phase angle, θ, is the number of degrees by which the maximum in the
mass-transfer rate precedes the maximum of the wave profile. By using the trigonometric
relation for the sum of two angles, (6.156) can be written in the following form:
N ¼ N aj~
nj cos θ cosðkx ctÞ aj~nj sin θ sinðkx ct Þ ð6:157Þ
If (6.156) and (6.154) are substituted into (6.155) the following relations are obtained for
a dissolving surface:
dh
ρS ¼N ð6:158Þ
dt
da
ρS ¼ aj~nj cos θ ð6:159Þ
dt
j~nj sin θ
c¼ ð6:160Þ
kρS
Equation (6.159) can be solved to get

a ¼ a0 eϖt ð6:161Þ
with
cos θ
ϖ ¼ j~nj ð6:162Þ
ρS
1.4
1.2
N/N
1.0
0.8
0.6
Flow
height
Wave
(cm)
0.04
0.02
0
0 0.4 0.8 1.2 1.6 2.0
x/λ
Figure 6.11 Variation of the mass transfer rate along a wavy surface for Re = 20 900.
Thorsness & Hanratty, 1979.
If ϖ is positive the surface waves will grow and if ϖ is negative they will decay. For a
dissolution process, a positive ϖ requires cos θ to be negative so that π=25θ53π=2.
Thus, the understanding of the stability of a dissolving surface depends on predicting the
phase of the mass transfer profile on a wavy surface.
Thorsness & Hanratty (1979a) used electrochemical techniques to measure the spatial
variation of N along a small-amplitude solid wavy wall which is exchanging mass with a
turbulently flowing liquid. The Schmidt number characterizing the system was equal to
729. A rectangular channel with a flat top and a wavy bottom was used. A typical profile
of N is shown in Figure 6.11 for Re ¼ 20 900 where Re is based on the height of the
channel and the bulk velocity. The wave profile is the dashed curve in the figure. The
solid curve is the least-squares fit to the data. Note that the maximum in N is upstream of
the crest and in the trough region.
This, to a large extent, reflects the phase shift of the wall shear stress due to viscous
effects (see Figure 6.3). In order to predict instability it is necessary to predict the
additional shift due to the mass transfer process.
The shift of the profile due to mass transfer is associated with wave-induced variations
of the velocity field and of the turbulence. It is found that the first of these effects is
not sufficient to produce phase angles of π=25θ53π=2. One needs to consider wave-
induced variations of the turbulence. The papers by Thorsness & Hanratty (l979a, b)
assume that the turbulent diffusivity equals the turbulent kinematic viscosity, described
in Section 6.7.2. Instability of a dissolving surface (Schmidt number 729) is predicted
for 3 104 5αþ 53 103, in rough agreement with the measurements of Thorsness &
Hanratty (1979a).
Ashton & Kennedy (1972) and Hsu et al. (1979) measured the wavelength and
velocity of small-amplitude waves at a water–ice interface. These covered a range of
water velocities of 9.1cm/s to 83.2cm/s. Wavelengths were found to decrease with
increasing velocity from 106.4cm to 13.1cm. These are represented quite well by
λv = ¼ 3180. This corresponds to a value of αþ ¼ 0:002, in good agreement with a
calculation for maximum growth (Hanratty, 1981). Furthermore, the predicted wave
References 145
velocity also agrees with measurements. These ice waves are controlled by heat transfer
from the water to the ice. The process is characterized by a Prandtl number of 13.7.
References
Abrams, J. & Hanratty, T.J. 1985 Relaxation effects observed for turbulent flow over a wavy
surface. J. Fluid Mech. 151, 443–455.
Ashton, G.D. & Kennedy, J.F. 1972 Ripples on the underside of river ice covers. J. Hydraulics
Div., Proc. ASCE 96, 1603–1624.
Beebe, P.S. 1972 Turbulent flow over a wavy boundary. Ph.D. thesis, Department of Civil
Engineering, Colorado State University.
Benjamin, T.B. 1959 Shearing flow over a wavy boundary. J. Fluid Mech. 6, 161–205.
Blumberg, P.M. & Curl, R.L. 1974 Experimental and theoretical studies of dissolution roughness.
J. Fluid. Mech. 65, 735–751.
Buckles, J.J., Adrian, R.J. & Hanratty, T.J. 1984 Turbulent flow over large-amplitude wavy
surfaces. J. Fluid Mech. 140, 47.
Caponi, E.A., Fornberg, B., Knight, D.D., McLean, J.W., Saffman, P.G. & Yuen, Y. 1982
Calculation of laminar viscous flow over a moving wavy surface. J. Fluid Mech. 124, 347–363.
Charles, M.E. & Lilleleht, L.U. 1965 An experimental investigation of stability and interfacial
waves in co-current flow of two liquids. J. Fluid Mech. 22, 217–224.
Cohen, L.S. 1964 Interaction between turbulent air and a flowing liquid film. Ph.D. thesis,
Department of Chemical Engineering, University of Illinois.
Cohen, L.S. & Hanratty, T.J. 1965 Generation of waves in the concurrent flow of air and a liquid.
AIChE JL 11, 138–144.
Cook, G.W. 1970 Turbulent flow over solid wavy surfaces. Ph.D. thesis, Department of Chemical
Engineering, University of Illinois.
Craik, A.D.D. 1966 Wind-generated waves in thin liquid films. J. Fluid Mech. 26, 369–392.
Curl, R.L. 1966 Scallops and flutes. Trans. Cave Res. Group, Gt. Britain 7, 121–160.
Feldman, S. 1957 On the hydrodynamic stability of two viscous, incompressible flluids in parallel
uniform shearing motion. J. Fluid Mech. 2, 343–370.
Finnicum, D.S. & Hanratty, T.J. 1988 Influence of imposed flow oscillations on turbulence.
Phys-Chem. Hydrodynamics 10, 585–598.
Frederick, K.A. 1982 Wave generation at a gas–liquid interface. M.Sc. thesis, Department of
Chemical Engineering, University of Illinois.
Frederick, K.A. 1986 Velocity measurements for a turbulent non-separated flow over solid waves.
Ph.D. thesis, University of Illinois.
Frederick, K.A. & Hanratty, T.J. 1988 Velocity measurements for a turbulent non-separated flow
over solid waves. Exper. Fluids 6, 477–486.
Gaster, M. 1962 A note on the relation between temporally-increasing and spatially-increasing
disturbances in hydrodynamic stability. J. Fluid Mech., 14, 222–224.
Goodchild, M.F. & Ford, D.C. 1971 Analysis of scallop patterns by simulation under controlled
conditions. J. Geol. 79, 52–62.
Hanratty, T.J. 1981 Stability of surfaces that are dissolving or being formed by convective
diffusion. Ann. Rev. Fluid Mech. 13, 231–252.
Hanratty, T. J. & Engen, J. M. 1957 Interaction between a turbulent air stream and a moving water
Hinze, J.O. 1975 Turbulence. New York: McGraw-Hill.

Hsu, S. & Kennedy, J.F. 1971 Turbulent flow over wavy pipes. J. Fluid Mech. 47, 481–502.
Hsu, K.S., Locher, F.A. & Kennedy, J.F. 1979 Forced convection heat transfer from irregular
melting boundaries, Report for the Iowa Institute of Hydraulic Research, University of Iowa,
Iowa City, Iowa.
Hussain, A. K. M. F. & Reynolds, W.C. 1970 The mechanics of a perturbation wave in turbulent
shear flow. Thermosciences Division Report RA-6, Stanford University, California.
Jeffreys, H. 1924 On the formation of water waves by wind. Proc. R. Soc. Lond. A107, 189–206.
Jeffreys, H. 1925 On the formation of water waves by wind. Proc. R. Soc. Lond. A110, 341–347.
Kendall, J.M. 1970 The turbulent boundary-layer over a wall with progressive surface waves.
J. Fluid Mech. 41, 249–281.
Kim, Y.Y. 1968 Wave motion on viscoelastic fluids. M.Sc. thesis, Department of Chemical
Kuzan, J.D. 1983 Separated flow over a large amplitude wavy surface. M.Sc. thesis, Department of
Chemical Engineering, University of Illinois, Urbana.
Long, R.R. 1963 Engineering Science Mechanics. Englewood Cliffs, NJ: Prentice Hall.
Loyd, R.J., Moffat, R.J. & Kays, W.M. 1970 The turbulent boundary-layer on a porous plate: An
experimental study of the fluid dynamics with strong favorable pressure gradients and blowing.
Report HMT13, Thermosciences Division, Department of Mechanical Engineering, Stanford
University, California.
Mao, Z. & Hanratty, T.J. 1986 Studies of the wall shear stress in a turbulent pulsating pipe flow.
J. Fluid Mech. 170, 545–564.
McLean, J.W. 1983 Computation of turbulent flow over a moving wavy surface. Phys. Fluids 26,
2065–2073.
Miles, J.W. 1957 On the generation of surface waves by shear flow. J. Fluid Mech. 3, 185–204.
Miles, J.W. 1959a On the generation of surface waves by shear flow. Part 2. J. Fluid Mech. 6,
568–582.
Miles, J.W. 1959b On the generation of surface waves by shear flows. Part 3 Kelvin–Helmholtz
instability. J. Fluid Mech. 6, 583–598.
Miles, J.W. 1962a A note on the inviscid Orr–Sommerfeld equation, J. Fluid Mech. 13, 427–432.
Miles, J.W. 1962b On the generation of surface waves by shear flows. Part 4. J. Fluid Mech. 13,
433–448.
Mitchell, J.E. & Hanratty, T.J. 1966 A study of turbulence at a wall using an electrochemical wall
shear stress meter. J. Fluid Mech. 26, 199–221.
Phillips, G.M. 1957 On the generation of waves by turbulent wind. J. Fluid Mech. 2, 417–445.
Ramaprian, B.R. & Tu, S.W. 1983 Fully-developed periodic turbulent pipe flow. Part 2. The
detailed structure of the flow. J. Fluid Mech. 137, 59–81.
Reiss, L.P. & Hanratty, T.J. 1963 An experimental study of the unsteady nature of the viscous
sublayer. AIChE Jl 9, 154–160.
Schlichting, H. 1968 Boundary-layer Theory, 6th edn. New York: McGraw-Hill.
Sigal, A. 1970 An experimental investigation of the turbulent boundary- layer over a wavy wall.
Ph.D. thesis, Department of Aeronautical Engineering, California Institute of Technology.
Son, J.S. & Hanratty, T.J. 1969 Velocity gradients at the wall for flow around a cylinder at
Reynolds numbers 5×103 to 105. J. Fluid Mech. 35, 353–365.
Thorsness, C.B. & Hanratty, T.J. 1979a Mass transfer between a flowing fluid and a solid wavy
References 147
Thorsness, C.B. & Hanratty, T.J. 1979b Stability of dissolving and depositing surfaces. AIChE Jl
25, 697–701.
Thorsness, C.B., Morrisroe, P.E. & Hanratty, T.J. 1978 A comparison of linear theory
with measurements of the variation of shear stress along a solid wave. Chem. Eng. Sci. 33,
579–592.
Uchida, S. 1956 The pulsating viscous flow superposed on the steady laminar motion of incom-
pressible fluid in a circular pipe. Z. angew. Math. Phys. 7, 403–431.
Zilker, D.P. 1976 Flow over wavy surfaces. Ph.D. thesis, Department of Chemical Engineering,
7 Large-wavelength waves; integral
equations
7.1 Prologue
As shown in Section 4.4, a shallow-wave assumption can be made when the

height of the liquid is small compared with the wavelength, that is, khL is a small
number. Under this condition, velocity fluctuations associated with wave motion
extend throughout the liquid layer; shear stresses at the interface or at the wall are
directly felt over the entire liquid layer. This chapter considers such flows.
Integral equation (7.7), which describes the variation of the average velocity, is
developed.
The principal application of this equation is the development of (7.11) and (7.12),
which describe the velocity field associated with the propagation of large-wavelength
waves at the interface. These show that the wave behavior is closely related to the wave-
induced variation of the shear stress at the wall. A pseudo-steady state is assumed to
capture this relation, equation (7.25).
An examination of laminar free flow on inclined planes provides an opportunity to test
the accuracy of (7.11) and (7.12) against analytical solutions. Periodic roll waves are
considered for free turbulent flows down inclined planes. Equations (7.11) and (7.12) are
used to predict the velocity of these waves and the channel angle required for their
appearance. These periodic roll waves can be modeled by considering their fronts to be
hydraulic jumps.
Intermittent waves with steep fronts are also observed at the interface of gas–liquid
layers in horizontal channels that are moved along by interfacial stresses. These are
modeled by applying equations (7.11) and (7.12) to the gas phase. That is, the gas space is
small enough compared with the wavelength for a large-wavelength analysis to be
applied to the disturbed gas flow. The critical gas velocity is correctly predicted at
large liquid flows.
These flow surges in the liquid layer have been called roll waves because their
appearance is similar to what has been observed in flows down inclined planes.
However, they differ because their fronts are not hydraulic jumps.
Ripples with steep fronts are always observed on the wall layers associated with
annular flows. These are also analyzed by using equations (7.11) and (7.12). This is an
interesting case for which the film is so thin that the use of a large-wavelength analysis
requires a consideration of the effect of surface tension.
7.2 Integral equation used to analyze the behavior of liquid layers
7.2.1 Approach
Detailed calculations of the velocity field in the liquid are not given. Instead, the spatially
averaged velocity, ua, is considered. Equations for conservation of mass and conservation
of momentum, which define the spatial and temporal variation of ua, are developed in
(7.3), (7.5). These are used to derive (7.11) and (7.12), which define the variation of ua
associated with the appearance of sinusoidal waves at the interface. These two equations
define the real and imaginary parts of the complex wave velocity, c = cR + icI. The
conditions for which cI = 0 define the initiation of waves.
7.2.2 Dispersion relation

The two-dimensional gas–liquid flow, sketched in Figure 7.1, is considered. An average
velocity in the liquid layer is defined as
ð
1 h
ua ¼ u dy ð7:1Þ
h 0
where, for simplification, h is used instead of hL. Conservation of mass can be expressed
as
∂ ∂h
ðhua Þ þ ¼ 0 ð7:2Þ
∂x ∂t
or as
∂ua ∂h ∂h
h þ ua þ ¼ 0 ð7:3Þ
∂x ∂x ∂t
If the wavelengths of the interfacial disturbances are large compared with the height of
the liquid, the shallow-liquid assumption can be made whereby pressure changes in the
Gas
Liquid
TS
y
τW
B–
h y=B
θg x h
z
y=0
g
Figure 7.1 Sketch of flow system.

150 Large-wavelength waves; integral equations
y-direction vary only because of hydrostatic head. Therefore, the pressure in the layer is
given as
∂2 h
p ¼ PS þ ðh yÞρL g sin θg σ 2 ð7:4Þ
∂x
where θg is the angle between the flow direction and the direction of gravity, PS is the
pressure at the surface and the last term represents the change of pressure across the
interface because of surface tension effects. A formal derivation of this result can be
found in a paper by Alekseenko et al. (1985), which also presents a justification for using
integral methods to calculate waves on thin films.
The momentum balance equation can be written as
ð ð
∂ h ∂ h 2 TS τW h ∂PS ∂h σ ∂2 h
u dy þ u dy ¼ þ gh cos θg þ gh sin θg þ
∂t 0 ∂x 0 ρL ρL ρL ∂x ∂x ρL ∂x2
ð7:5Þ
Conservation of mass, equation (7.3), is added to (7.5) and a velocity profile shape
parameter is defined as
ð
1 h 2
Γ S ¼ 2 u dy ð7:6Þ
hua 0
The following equation is obtained:
∂ua ∂ua u2 ∂h ∂ΓS

þ ð2Γ S 1Þua þ ðΓ S 1Þ a þ u2a
∂t ∂x h ∂x ∂x
ð7:7Þ
TS τW 1 ∂PS ∂h σ ∂2 h
¼ þ g cos θg g sin θg þ
hρL hρL ρL ∂x ∂x ρL ∂x2
Represent all the terms in (7.7) by the sum of an average and a fluctuating component
introduced by waves at the interface. The fluctuations are small enough for quadratic
terms to be ignored. They are defined by a wave equation
u0 Γ 0 P S0 T S0 τ W
0
h0 ¼ ¼ ¼ ¼ ¼ ¼ a exp ik ðx ct Þ ð7:8Þ
ua Γ~
~ ~ S T~ S ~τ W
P
~S; P
where ~ua ; Γ ~ S ; T~ S ; ~τ W and the wave velocity can be complex. The following equation
is obtained:
T~ S ~τ W ~S
P
c2 þ u2a Γ S 2ua Γ S c hu2a Γ~ S ¼ i i þh
ρL k ρL k ρL
ð7:9Þ
k 2 σh i dPS ig cos θg
þ gh sin θg þ þ
ρL kρL dx k
In some cases, it is convenient to substitute into (7.7)

∂PS T S τW
þ ρL g cosθg ¼ ¼ g̑ ð7:10Þ
∂x h
where g̑ is defined in equation (3.20) (see Asali & Hanratty, 1993; Craik, 1968). The
quantities on the left side of (7.9) are the inertia terms.
For plug flow in the film, Γ S ¼ 1 and Γ~ S ¼ 0, so the inertia terms are equal to
ðua cÞ2 and, consequently, are always destabilizing. For ua ¼ c, the effects of inertia
vanish for a plug flow. If the velocity varies linearly with distance from the wall,
Γ S ¼ 4=3. Then inertia effects vanish if the wave velocity equals the liquid velocity at
the interface, 2ua .
The real and imaginary parts of (7.9) are, respectively,
T~ SI ~τ WI khP~ SR k 2 σh
c2I þ c2R 2Γ S cR ua þ Γ S u2a hu2a Γ~ SR ¼þ þ þ gh sin θg þ
ρL k ρL k ρL k ρL
ð7:11Þ
~
T SR ~τ WR 1 dPS g cos θg khPSI ~
hu2a Γ~ SI þ 2cI cR Γ S ua ¼ þ þ ð7:12Þ
ρL k ρL k ρL k dx k ρL k
The term kcI represents the growth rate of unstable waves and cI = 0 defines neutral
stability. Since ~τ WR is strongly related to cR, equation (7.12) may be viewed as defining cR
under neutral stability conditions. This velocity is the kinematic wave velocity defined by
Lighthill & Whitham (1955). In this context, (7.11) with cI = 0 defines the dynamic
conditions needed for neutral stability.
Note that in (7.11) and (7.12), as kh becomes smaller, for a fixed gas velocity, T~ SR
and T~ SI become more important relative to P ~ SR and P
~ SI . The reason for this, as pointed
out by Cohen & Hanratty (1965), is that pressure and shear stress fluctuations in the
gas, feed energy into the film through normal velocity fluctuations and tangential
velocity fluctuations in the liquid, respectively. As kh ! 0, the ratio of the normal to
the tangential velocity fluctuations in the liquid approaches zero. Waves appearing
under these circumstances are kinematic (see Section 7.4) and called slow waves by
Craik (1966).
Non-zero values of cI indicate growing or decaying waves. In order to calculate the
fastest-growing wave it is necessary to develop relations for T~ S, P ~ S , ~τ W , and Γ~ S . Terms
~τ W and Γ~ S are evaluated by using a pseudo-steady state approximation.
7.2.3 Relation between the wall shear stress and ua

Chapter 3 on film flows provides relations between the volumetric flow and the wall shear
stress for both laminar and turbulent flows. For laminar flow, a solution of the Navier–
Stokes equations gives
2μL ua hg̑
τW ¼ ð7:13Þ
h 3
4 1 g̑ 2 h4 1 h2 g̑
ΓS ¼ þ þ ð7:14Þ
3 270 u2a μ2L 18 μL ua
dPS T S τW
g̑ ¼ ρL g cos θg ¼ ð7:15Þ
dx h
From Chapter 3, the following relation can be used for a sheared turbulent flow:

u yvc
¼ f ð7:16Þ
vc L
with
1=2
τc
vc
¼ ð7:17Þ
ρL
2 1 1
τ c ¼ τ W þ τ S þ g̑ ð7:18Þ
3 3 3
The integration of (7.16) from 0 to h+ yields

hvc
¼ γðReL Þ ð7:19Þ
L
h 2:5 2:5 i0:4
γ ¼ 1:414Re0:5
L þ 0:132Re0:9
L ð7:20Þ
where ReL ¼ 4Γ ρL =μL and Γ is the volumetric flow per unit spanwise length. Hanratty &
Hershman (1961) show that equations (7.16–7.20) give
2μL ua 1
τW ¼ BðReL Þ hg̑ ð7:21Þ
h 3
γ2 ðReL Þ
BðReL Þ ¼ ð7:22Þ
2ðReL Þ
For a laminar flow, B(ReL) = 1. From (7.6), the following equation for ΓS(y+) is
obtained:
ð þ
hþ h 2 þ þ
ΓS ¼ 2 f ðy Þdy ð7:23Þ
ReL 0
The function f(y+) has been evaluated by Henstock & Hanratty (1976) by using the van
Driest mixing-length model with A = 26, κ = 0.4. The calculated dependence of h+, B and
ΓS on ReL is given by equations (7.19), (7.20) and Figures 7.2, 7.3, 7.4.
Equations for ~τ W and Γ~ are obtained by assuming that the relations of τW and ΓS to local
values of ua and h are the same as would exist for an undisturbed flow (the pseudo-steady-
state approximation). The sum of a time-average and a fluctuating component is sub-
stituted for the variables in (7.21), (7.22) and (7.23). The fluctuations are represented by
(7.8) and ~ua is eliminated with
h~ua ¼ c ua ð7:24Þ
obtained from conservation of mass, equation (7.3). Thus,

3μL ua BðReL Þ c c ReL dB 1
~τ WR ¼ 2
2þ ~τ SR ð7:25Þ
h u a u a B d Re L 2
102
A = 26
κ = 0.4
h+ 101
100 0
10 101 102 103
ReL = 4 ΓρL / μL
Figure 7.2 Calculation of film height. Henstock & Hanratty, 1976.
3.10
A = 26
κ = 0.4
2.48
B = γ2 (ReL) / 2 ReL
1.86
1.24
0.62
0
100 101 102 103
ReL = 4ΓρL / μL
Figure 7.3 Calculation of the wall shear parameter as a function of the liquid film Reynolds number. Henstock
& Hanratty, 1976.
1.35
A = 26
κ = 0.4
1.28
1.21
ΓS
1.14
1.07
1.00
100 101 102 103
ReL = 4ΓρL / μL
Figure 7.4 Calculation of the velocity profile shape parameter as a function of the liquid film Reynolds
number. Henstock & Hanratty, 1976.
where ~τ WI ¼ 0, T~ SI ¼ 0. Similarly, from (7.23),

dΓ S c
Γ~ S ¼ ReL ð7:26Þ
d ReL ua h
7.3 Laminar free flow
Laminar free flow provides a simple application of the integral equations (7.11) and
(7.12). The pressure gradient and TS are zero; that is, the film is pulled along only
by gravity. Since the flow is laminar, the velocity profile has the shape of a
parabola. From a force balance,
τ W ¼ ρL hg cos θg ð7:27Þ
and from (7.6)

6
ΓS ¼ ð7:28Þ
5
The goal is to predict the wave velocity and the conditions needed for the initiation of
waves. Since the shape of the velocity profile does not change, Γ~ SR ¼ 0. Also, since there
is no gas flow, T~ SI ¼ 0, T~ SR ¼ 0, P
~ SR ¼ 0, dPG =dx ¼ 0, P
~ SI ¼ 0. Equations (7.11) and
(7.12) can then be written as
12 6 ~τ WI k 2 σh
c2I þ c2R ua cR þ u2a ¼ þ gh sin θg þ ð7:29Þ
5 5 ρL k ρL

6 ~τ WR g cos θg
2cI c R ua ¼ þ ð7:30Þ
5 kρL k
For neutral stability, cI = 0. Equation (7.30) defines the wave velocity, cR. Equation (7.29)
relates cI to the imbalance between the destabilizing effect of inertia and the stabilizing
effects of gravity and surface tension.
From (7.13) and (7.15),
2μL ua h
τW ¼ þ ρL g cos θg ð7:31Þ
h 3
Substitute (7.27) into (7.31) to get

3μL ua
τW ¼ ð7:32Þ
h
Represent the variables as the sum of an average and a fluctuating quantity:
3μL ðua þ u0a Þ

τ W ¼ τ w þ τ 0W ¼ ð7:33Þ
h þ h0
7.3 Laminar free flow 155
Multiply the numerator and denominator by ðh h0 Þ and neglect second-order terms in

the fluctuations:
3μL u0a 3μL h0 ua
τ 0W ¼ 2
ð7:34Þ
h h
Represent the disturbance by (7.8), so that

3aμL ~ua 3μL aua
a~τ W ¼ 2
ð7:35Þ
h h
From conservation of mass, equation (7.3),

c ua
~
ua ¼ ð7:36Þ
h
This is substituted into (7.35), to give

3ðcR ua ÞμL 3μL ua
~τ WR ¼ 2
2
ð7:37aÞ
h h
Also, from (7.35) and (7.36),

~τ WI ¼ 0 ð7:37bÞ
since, at neutral stability, cI = 0. Thus, (7.30) and (7.37a) give
3ðcR ua ÞμL 3μL ua g cos θg

0¼ 2
þ 2
þ ð7:38Þ
h kρL h kρL k
at neutral stability. Eliminate g cos θg with (7.27). The following equation is obtained by
using (7.32) to eliminate τW:
cR
¼3 ð7:39Þ
ua
By substituting (7.39) and (7.37b) into (7.29), one obtains the following criterion for the
initiation of an instability ðcI ¼ 0Þ:
gh sin θg k 2 σh
¼3 ð7:40Þ
2
ua ρL u2a
Equation (7.40) defines the angle of the plane above which an instability will occur for a
given liquid velocity.
Benjamin (1957) presented a more rigorous analysis of free laminar flow down an
inclined plane by solving the Orr-Sommerfeld equation (6.19) with a series solution in y
(instead of using an integral approach):
ψ ¼ A0 þ A1 y þ A2 y2 þ A3 y3 þ … ð7:41Þ
where ψ is the stream function defined for a two-dimensional flow by using conservation
of mass: ux = −∂ψ/∂y, uy = ∂ψ/∂x. This solution is valid laminar flow
for for which the
velocity profile is given by a power series. For large kh hua ρL =μL , it is not useful
since it does not converge rapidly enough. Benjamin considered only four terms. The
constants Ai are determined using the kinematic relation at the interface, the shear stress
condition at the interface, and the conditions ψ = 0, ∂ψ/∂y = 0 at the lower wall (see
equation (6.20)). The solution is then substituted into the normal stress condition to
obtain c = f(k). The constants A0 to A3 are of order (k ReL). The constants A4 to A7 are of
order (k ReL)2. Therefore, if the series is truncated at y3, the solution ignores terms of
order (k ReL)3. For small kh, the Benjamin analysis gives
cR
¼3 ð7:42Þ
ua
gh sin θg 18 k 2 σh
¼ ð7:43Þ
u2a 5 ρL u2a
These compare favorably with the results from the integral analysis, equations (7.39) and
(7.40).
For a vertical wall, sin θg = 0 so (7.40) gives the following stability conditions:
k 2 σh
¼3 ð7:44Þ
ρL u2a
or
2 ρ u2 h
kh ¼ 3 L a ð7:45Þ
σ
An increase in ua is predicted to cause a decrease in the wavelength at neutral stability.

Thus, as pointed out by Benjamin (1957), a critical ua is not predicted for a very long flow
on a vertical wall. Binnie (1957) studied the initiation of waves on a water film flowing
down the outside of a vertical tube. He found that transition occurs at Re = 4.4 for waves
with cR =ua ¼ 3:5 and λ=h ¼ 102. For Re = 4.4 equation (7.45) predicts λ=h ¼ 74. Thus,
approximate agreement is realized between the integral analysis and laboratory observa-
tions for vertical films. (See Hanratty & Hershman (1961) for a discussion of these results.)
7.4 Kinematic waves
Lighthill & Whitham (1955) and Whitham (1974) have demonstrated that the concept of
a kinematic wave is useful in analyzing many phenomena. As indicated by the name,
kinematic waves are different from classical wave motions (discussed in Chapter 4)
encountered in dynamical systems, in that they are defined from the equation of con-
servation of mass. From (7.2)
∂q ∂h
þ ¼0 ð7:46Þ
∂x ∂t
7.4 Kinematic waves 157
where q ¼ hua is the volumetric flow per unit breadth. Suppose a pseudo-steady-state
approximation can be made where
q ¼ f ðhÞ ð7:47Þ
which is of the same form as exists under steady conditions

q¼f h ð7:48Þ
Then, from (7.46),

∂q ∂h ∂h
þ ¼0 ð7:49Þ
∂h ∂x ∂t
For a disturbance propagating at a velocity c

∂h ∂h
c þ ¼0 ð7:50Þ
∂x ∂t
Therefore the equating of (7.49) and (7.50) gives

∂q
cKW ¼ ð7:51Þ
∂h
For laminar flow down an inclined plane, the averaged volumetric flow and the average
velocity are
3
gh cos θg
q¼ ð7:52Þ
3ðμL =ρL Þ
2
ρ gh cos θg
ua ¼ L ð7:53Þ
3μL
where θg is defined in Figure 7.1. Make a pseudo-steady-state assumption. From (7.52),

gh3 cos θg
q¼ ð7:54Þ
3ðμL =ρL Þ
dq 3gh2 cos θg
cKW ¼ ¼ ¼ 3ua ð7:55Þ
dh 3ðμL =ρL Þ
where ua ¼ q=h is given by (7.53). The result of a stability analysis (7.39) can also be
obtained by assuming that cR is the kinematic wave velocity. Thus, Lighthill & Whitham
point out that neutral stability can be defined as the condition for which
cKW ¼ c ð7:56Þ
where c is the dynamic wave velocity. For free laminar flow this is given by (7.29) with
cI = 0 and ~τ WI ¼ 0.
7.5 Roll waves
7.5.1 Outline
Roll waves have been discussed in Chapter 2. Equations (7.11) and (7.12) are now used
to describe these waves for free turbulent flows down inclined planes and for gas–liquid
flows.
7.5.2 Free turbulent flows on an inclined plane

Roll waves are observed for turbulent flows down inclined planes, such as spillways
of dams and run-off channels. They are periodic and give the appearance of rollers
that are observed at a beach. (However, as pointed out by Dressler (1949, 1952), their
underlying mechanism is quite different, since they depend critically on the wall
resistance.) A famous photograph of roll waves presented by Cornish (1934) is
reproduced in Figure 7.5 and in Figure 11.33 of Water Waves by Stoker (1957).
These waves are believed to evolve from large-wavelength regular waves. Thus, a
prediction of their appearance can be obtained by considering the stability of a free
Figure 7.5 Roll waves looking downstream. Cornish, 1934.

7.5 Roll waves 159
turbulent flow on an inclined plane. The following equations are obtained from (7.11)
and (7.12), if ΓS = 1:
~τ WI
c2I þ ðcR ua Þ2 ¼ þ gh sin θg ð7:57Þ
ρL k
~τ WR g cos θg
2cI ðcR ua Þ ¼ þ ð7:58Þ
ρL k k
Here, the influence of surface tension is ignored because very large wavelengths are
considered. Also the influence of the gas flow on the liquid flow can be ignored, so
T~ SI ¼ T~ SR ¼ P
~ SR ¼ P
~ SI ¼ ∂P=∂x ¼ 0.
The wall shear stress for a smooth plane can be represented by
1
τ W ¼ ρL f W u2a ð7:59Þ
2
f W ¼ 0:046 Re0:2
L ð7:60Þ
However, for the applications cited above, the bottom boundary may be considered to be
completely rough. Thus, the friction factor varies as the ratio of the length scale, kS,
characterizing the roughness, to the height of the liquid
kS
fW / ð7:61Þ
h
Thus,
1 ρL u2a k S
τW ¼ ð7:62Þ
2 h
Substitute for τW and ua

τ W ¼ τ W þ τ 0W ua ¼ ua þ u0 a ; where τ 0W ¼ a~τ W ð7:63Þ
From conservation of mass, equation (7.3),

aðc ua Þ
u0a ¼ ~
ua a ¼ ð7:64Þ
h
Neglect second-order terms in the fluctuating quantities and assume neutral stability,
cI = 0. From (7.62), (7.63) and (7.64), the following equation is obtained
1 2ρL k S ua cR ρL k S u2a 1 k S 2u2a
~τ W ¼ 2 ρL ð7:65Þ
2 h h h 2 2h h
From (7.58), since cI = 0 for neutral stability

~τ WR g cos θg
0¼ þ ð7:66Þ
kρL k
Substitute (7.65) into (7.66). From a force balance, g cos θg =k ¼ τ W =kρL h. From (7.62),
τ W ¼ ð1=2ÞρL f W u2a . Equation (7.66) provides the result
cR
¼2 ð7:67Þ
ua
This agrees with measurements made by Cornish on the waves shown in Figure 7.5.
From (7.57), with ~τ WI ¼ 0 and cR given by (7.67), the following condition is obtained for
neutral stability:
gh
sinθg ¼ 1 ð7:68Þ
u2a
Equation (7.68) predicts the liquid flow-rate at which roll waves will appear on a declined
spillway. This defines the minimum angle at which roll waves will be observed.
7.5.3 Generation of roll waves by a gas flow

Figure 7.6 shows waves observed by Hanratty & Engen (1957) for air and water flowing
in a rectangular enclosed channel with a large aspect ratio. Because of the similarity with
the free waves described in Section 7.5.2, Engen & Hanratty called these large-
wavelength disturbances roll waves. They differ from what is observed for free flows
on inclined walls in that the liquid is moved by interfacial stresses associated with the air
flow rather than by gravity. The conditions for neutral stability have been calculated by
Hanratty & Hershman (1961), by Lin & Hanratty (1986), by Andreussi et al. (1985) and
by Hanratty (1983).
Neutral stability is defined by (7.11) and (7.12) with cI = 0. Since a fully developed
horizontal flow is considered, sin θg = 1, cos θg = 0 and dh=dx ¼ 0. Thus, if Γ ≈ 1
and Γ~ ≈ 0,
Figure 7.6 Roll observed by Hanratty and Engen for air–water flow in an enclosed rectangular channel.
Hanratty & Engen, 1957.
7.5 Roll waves 161
T~ SI ~τ WI kh P
~ SR
ðc ua Þ2 ¼ gh þ þ ð7:69Þ
ρL k ρL k ρL k
dPS
0 ¼ T~ SR ~τ WR ~ SI
þ khP ð7:70Þ
dx
where surface tension effects are not considered and (7.25) describes ~τ W . Equation
(7.70) defines the wave velocity and equation (7.69) defines the initiation of waves as
occurring when inertia and gas-phase pressure variation overcome the stabilizing effect
of gravity.
In the analyses by Hanratty & Hershman (1961) and by Lin & Hanratty (1986), an
~ S is obtained by considering a momentum balance for the gas. Because the
equation for P
wavelength is considered to be large compared with the height of the gas space, B − h,
integral equations of the type used for the liquid layer can be developed. Since ΓSG ≈ 1,
conservation of mass and momentum for the gas flow are written as
∂h ∂
þ ½ðB hÞuGa ¼ 0 ð7:71Þ
∂t ∂x
∂ ∂ ðB hÞ ∂PS ∂h
½ðB hÞuGa þ ðB hÞu2Ga ¼ þ g sin θg
∂t ∂x ρG ∂x ∂x
ð7:72Þ
1
ðT S þ τ B Þ þ g ðB hÞcos θg
ρG
where uGa is the average velocity of the gas.

The interaction of the gas with the liquid is represented by T~ SR, T~ SI , P
~ SR , P
~ SI . The
~ SR and P
pressures P ~ SI are obtained from (7.71) and (7.72) by using procedures outlined in
this chapter to analyze the flow in a liquid layer perturbed according to (7.8). The gas
phase is fully turbulent, so it is approximated by a plug flow with velocity uGa.
Gravitational effects can be ignored.

~ SR ¼ ρG ðuGa cR Þ2 1 T~ SI þ ~τ BI
P ð7:73Þ
Bh kρG
" #
1 T þ τ
P~ SI ¼ S
þ T~ SR þ ~τ BR
B
ð7:74Þ
k Bh Bh
Because of the assumption of a large-wavelength disturbance, the disturbed shear

stresses at the interface and the top wall can be calculated by using the pseudo-steady-
state approximation, whereby the instantaneous stress is related to flow variables by the
equation derived for undisturbed flow in the gas:
1
T S ¼ ρG f S ðuGa cR Þ2 ð7:75Þ
2
1
τ B ¼ ρG f B u2Ga ð7:76Þ
2
f B ¼ f S ¼ 0:042 Re0:2
G ð7:77Þ
Substitute T S ¼ T S þ T 0S , uGa ¼ uGa þ u0Ga , f S ¼ f S þ f 0S and neglect terms which are

second order in the fluctuations to obtain
" #
~f ~
u
T~ S ¼ T S
Ga
S
þ2 ð7:78Þ
fS ðuGa cÞ
From conservation of mass

ðuGa cÞ
~
uGa ¼ ð7:79Þ
ð B hÞ
So (7.78) can be written as

" #
~f 2
~
T SR ¼ T S S
þ ð7:80Þ
fS Bh
T~ SI ¼ 0 ð7:81Þ
Also, from (7.76) and (7.79),

" # " #
ðuGa cÞ 2
~τ BR ¼ τ B 2 ≈ τB ð7:82Þ
B h uGa Bh
~τ BI ¼ 0 ð7:83Þ
The fluctuations in the friction factor, f S0 , are approximated by Andreussi et al.

(1985) as

f S0 a~f 1 ∂f S ∂f S
¼ S¼ ReL0 þ ReG0 ð7:84Þ
fS fS f S ∂ReL ∂ReG
From conservation of mass, Re0G = 0. Also,

ReL0 c h0
¼ ð7:85Þ
ReL ua h
Thus
~f 1 ∂f S ReL c
S
¼ ð7:86Þ
f S f S ∂ReL ua h
So (7.80) gives, for cI = 0,

2 ReL ∂f S cR 1
T~ SR ¼ T S þ ð7:87Þ
ð B hÞ f S ∂ReL ua h
The second term on the right side of (7.87) recognizes that the liquid layer is on the
bottom of the channel and that the stress at the interface could depend on the flow of the
liquid.
7.5 Roll waves 163
Equation (7.25) can be used to represent ~τ WR . By assuming T~ SI ¼ ~τ WI ¼ ~τ BI ¼ 0,

equation (7.69), for neutral stability, becomes
~ SR
hP
ðcR ua Þ2 ¼ gh þ ð7:88Þ
ρL
and (7.73) gives

~ SR ¼ ρG
P ðuGa cR Þ2 ð7:89Þ
Bh
Equation (7.89) represents a Bernoulli effect whereby the squeezing of gas-phase stream-
lines at a wave crest causes an increase in velocity and, consequently, a decrease in gas-
phase pressure at the crest. This is a destabilizing effect.
The wave velocity is not given by classical theory developed in Chapter 4 for an
inviscid flow. As already mentioned, it is given by the equation for a kinematic wave,
whose behavior is, to first order, dependent on the viscous drag at the wall. It is calculated
by using (7.72), with T~ SR ; ~τ WR ; P
~ SI defined by (7.87), (7.25), (7.74). Equation (7.74)
requires the specification of τ B ; T S ; ~τ BR . These are estimated by using (7.75), (7.76),
(7.82).
By substituting (7.89) into (7.88), the following stability condition is obtained:

2 ρG h
ðcR ua Þ ¼ gh ðuGa cR Þ2 ð7:90Þ
Bh ρ L
The term on the left side represents the stabilizing of liquid inertia to an interfacial
disturbance. The first term on the right side represents the stabilizing effect of
gravity. The second term on the right side is the destabilizing effect of the wave
induced variation of the gas phase pressure. The numerical solution of (7.90) to
obtain the critical gas velocity was pursued by Lin & Hanratty (1986), Hanratty
(1983) and by Asali et al. (1985). The influence of liquid inertia is not large, so a
simplification of the calculation of (c − ua) is warranted. Thus, Hanratty (1983)
used cR/ua = 2 and compared (7.90) with measurements by Hanratty & Hershman
(1961), Woodmansee & Hanratty (1969), Miya et al. (1971). This comparison is
reproduced in Figure 7.7. Good agreement is noted at large liquid Reynolds
numbers. At very small liquid Reynolds numbers, transition is approximately
independent of gas velocity and a critical liquid Reynolds number can be defined.
Andreussi et al. (1985) suggest that it is associated with relaxation phenomena
whereby τi and τW do not respond to changes in film height as is predicted by
pseudo-steady-state relations. This explanation has not, however, been substanti-
ated. Another possibility is to assume that a necessary condition is that the wall
film needs to be able to sustain turbulence for a disturbance wave to appear. Thus,
the film Reynolds number must be larger than a critical value at large gas
velocities.
Roll wave transition data

25
Miya
Hershman
Woodmansee
20
uG (m/s)
15 14.76 m/s
10
0
0 2 4 6 8 10
Liquid Reynolds number, ReL(10–2)
Figure 7.7 Comparison of an observed roll wave transition with calculations using ^τ SI ¼ 0. Meyer, 1983.
7.6 Model for a fully developed roll wave
Dressler (1949) presented a model for fully developed roll waves, in a declined channel,
which uses non-linear shallow liquid theory. This work motivated a study for gas–liquid
flow by Miya et al. (1971).
The equations of conservation of mass (7.2) and momentum (7.5) are written by assuming
steady flow in a coordinate system moving with the velocity of the waves, c; that is,
X ¼ x ct ð7:91Þ
so that ∂/∂t = −c(d/dx). The integral form of conservation of mass is

hðc ua Þ ¼ q ð7:92Þ
where the integration constant is called the progressive discharge by Dressler (1949).
From conservation of momentum (7.7), one gets

dh 2 dΓ S
c 2Γ S þ ReL ua c þ S ua gh sin θg
2
dX dReL
ð7:93Þ
1 h dPG
¼ ðτ W T S Þ þ gh cos θg
ρL ρL dX
where ∂ua/∂x is eliminated by using conservation of mass. For the case analyzed by
Dressler, free flow down an inclined wall, TS = 0, dPG/dX= 0, ΓS = 1.
Equation (7.93) then simplifies to
dh τW
c u2a gh sin θg ¼ gh cos θg ð7:94Þ
dX ρL
7.6 Model for a fully developed roll wave 165
where τ W is given by (7.62). Thus (7.94) is of the form

dh
M1 ¼ M2 ð7:95Þ
dX
where M1 and M2 are functions of h. This is integrated, numerically, to give the variation
of h along the wave, from an initial value, h0, to a peak value of hP, where a hydraulic
jump returns it to h0. Thus, the solution is discontinued when
2* +1=2 3
h hP 1 4 8ð c u0 Þ 2
¼ ¼ 1þ 15 ð7:96Þ
h0 h0 2 gh0
Term M1 reaches a value of zero at
ðc ua Þ2 ¼ ghsinθg ð7:97Þ
Dressler arranged for M2 to equal 0 at this critical point by selecting parameters for the
solution. In this way, the predicted wave profile is continuous.
Gas–liquid flow in a channel is fundamentally different from flow down an inclined
plane in that the front of the roll wave is not a hydraulic jump. This is demonstrated in
Miya et al. (1971) by showing that the measured hP =h0 is smaller than predicted by
(7.96).
For this flow, sin θg = 1 and cos θg = 0; the pressure gradient in the gas is obtained from
(7.72) as
dPG ρG ðuG0 cÞ2 ðB h0 Þ2 dh ðT S þ τ B Þ

¼ ¼0 ð7:98Þ
dX ðB hÞ3 dX ð B hÞ
where the effect of gravity is ignored because of the low gas density. Again, taking ΓS = 1,
equations (7.93) and (7.98) give
dh h i 1 h
ðc ua Þ2 gh þ Gh ¼ ðτ W T S Þ ðT S þ τ B Þ ð7:99Þ
dX ρL ρ G ð B hÞ
ρG ðuG0 cÞ2 ðB h0 Þ2
G¼ ð7:100Þ
ρ ð B hÞ 3
The term G (dh/dX) represents the pressure gradient needed to balance inertial changes in
the gas (a Bernoulli effect). In the situation being considered, it has the same effect as a
negative gravitational force. For relatively high gas velocities (−g + G) is positive or has
a very small negative value. Therefore, the range of variables of interest is such that a
solution of (7.99) is sought for which M1 is always positive. The sign of the slope then
depends on the sign of M2. The term [h/ρG (B − h)](TS + τB) appearing in M2 is a force on
a unit length of the liquid layer due to the gas-phase pressure gradient and is usually
small compared with TS. Therefore, the sign of the slope depends largely on the sign of
(τW −TS). This term is positive in the back of the wave and it becomes negative at the front
due to relaxation of the wall shear stress. At h = h0 and at h = hP, dh/dX = 0; that is,
1 h
0¼ ðτ W T S Þ ðT S þ τ B Þ ð7:101Þ
ρL ρG ðB hÞ
At these locations there is a balance between forces at the interface and at the solid
boundary.
A comparison of the above analysis with experiment is given by Miya et al. (1971). A
critical issue is the modeling of the relaxation of the wall shear stress in the front of the
roll wave. Miya et al. speculate that the flow laminarizes in this region.
7.7 Ripples generated on a liquid film at high gas velocities
The wave structure at the interface for a high gas velocity (>20 m/s) has received
considerable attention because of its importance in understanding interfacial stresses in
annular flows (Hewitt & Hall-Taylor, 1970; Asali & Hanratty, 1993; Asali et al. 1985;
Hanratty, 1991). Waves appear at arbitrarily small liquid flows, both in vertical and
horizontal conduits.
The film is covered by long-crested slow-moving ripples, shown in Figure 7.8.
These ripples have a steep front and a low ratio of the amplitude (10–20 µm) to
wavelength (2–3 mm). Between ripples the surface is smooth and the flow appears
laminar. At sufficiently high liquid rates, disturbance or roll waves appear on the
film (see Figure 3.3). These have a much larger velocity than the ripples and very
large spacing between successive waves. (They might be described as a collection
of ripple waves.)
In this section, an analysis of ripple waves is presented. The goals are to identify the
physical processes responsible for their appearance and to predict the characteristic
length between them. This length scale is of importance in characterizing interfacial
stresses in vertical gas–liquid annular flows. For example, Andritsos & Hanratty (1987)
and Bontozoglou & Hanratty (1989) have argued that in separated flows the ratio of the
μL = 2.0 cP
ReG = 168 000 ρL = 1.056 g/cm3
ReL = 13.75 ReL = 24.33 ReL = 34.9

4 4 4
5.08 cm Flow
Figure 7.8 Ripple waves on a film with a viscosity of 2 cP. Asali & Hanratty, 1993.
7.7 Ripples generated on a liquid film at high gas velocities 167
interfacial drag to that for a smooth surface should scale with Δh/λ, where Δh is the wave
height and λ is the wavelength.
The approach is to solve the linear momentum equations to determine the growth of
small-amplitude two-dimensional wavelike disturbances at the interface. It is argued that
the fastest growing wave is the precursor of the ripples (Taylor, 1963).
As mentioned above, the thicknesses of the films on which these ripple waves occur
are quite small. For example, at a liquid Reynolds number of 36 and a gas Reynolds
number of 78 000, a water film on the wall of a 4.2 cm pipe would have a height of 147
µm. This dictates that the analysis considers wavelengths which are large enough for a
shallow-liquid assumption to be made, yet small enough for surface tension effects to be
important. Thus, equations (7.9) to (7.12) are applicable. The initial growth of unstable
waves is governed by an imbalance between the destabilizing effects of inertia, the
component of surface stress in phase with the wave slope, the component of pressure
180° out of phase with the wave height and the stabilizing effect of surface tension.
Calculations for the growth rate of ripples are presented by Asali & Hanratty (1993).
The analysis differs from that for roll waves in that surface tension is important, the flow
is laminar and the wavelength is small compared with the height of the gas space. The
large-wavelength analyses of the gas flow that have been used to calculate the wave-
induced variation of the pressure and interfacial stress (7.73, 7.74, 7.75) are not valid.
Thus, the analysis described in Sections 6.6 and 6.7 was used by Asali & Hanratty.
For gas flows over very thin films at high gas velocities, a number of simplifications in
the analysis can be made. Since the flow is laminar, the average velocity at the interface,
ūS, is equal to 2ūa, so τ S ≈ τ W ¼ 2uμL =h. This means that τ S =ρu2a ¼ 2=ReL . From (7.10),
∂P∂x þ ρL cos θ g ¼ 0, since T S ≈ τ W . Asali & Hanratty (1993) show that equation (7.9)
S
simplifies to
2 1
c 3 c huS c
1 þ 1 þ 3i kh 1
uS 4 uS vL uS
ð7:102Þ
k 2 σh hP ~ S gh 3 ~τ S
¼ þ þ sin θg þ i
ρL u2S ρL u2S u2S 2 ρL ku2S
where ūS is the liquid velocity at the interface. Craik (1966) solved the problem by using
the series approach developed by Benjamin (1957). For small kh, he obtained
2 1
6 c 7 c huS c
1 þ 1 þ 3i kh 1
5 uS 8 uS vL uS
ð7:103Þ
k 2 σh hP ~ S gh 3 T~ S
¼ þ þ sin θg þ i
ρL u2S ρL u2S u2S 2 ρL ku2S
2
where terms of order ðkhÞ are ignored. By comparing (7.102) and (7.103), it can be seen
that the integral approach and the series approach produce similar results.
The complex wave velocity, c = cR + ic1, is substituted into (7.102). The real and
imaginary parts of the resulting expression provide two equations that can be solved for
cR and cI. Asali & Hanratty (1993) provide results obtained from such a calculation.
Studies of the properties of ripple waves have been carried out by Würz (1977), Taylor
(1963), Shearer & Nedderman (1965). Asali & Hanratty compared their analysis with the
study by Würz in a 2.54 cm horizontal channel, as well as with their own measurements
in a 4.2 cm vertical pipe. Figure 7.9 shows calculations of the wave growth, kc1, for
different liquid Reynolds numbers where ReLF =4 ¼ hua ρL =μL . The maxima in the curves
are the maximum rates of growth. Figure 7.10 shows that the ratio of the wavelength to
7.00
6.13
5.25
51.35
(u)
4.38
a
cI
(kh) ReLF
3.50 35.6
4
2.63 ReLF
dt = 4.2 cm = 21.91
4
μL ρG 1/2
1.75 = 2.13
μG ρ L
ReG = 168 000
0.88
ReLF = 4 h ua ρL /μL
0
0 0.20 0.40 0.60 0.80 1.00 1.20
(kh)
Figure 7.9 Examples of calculated growth rates of disturbances. Asali & Hanratty, 1993.
0.6
μL ρG 1/2
μG ρL = 2.13
0.5
0.4
(kh)max
51.35
0.3 35.6
21.91
ReLF = 14.72
0.2 4
0.1
1000 1400 1800 2200 2600 3000 3400
σ
vG*μG
Figure 7.10 Calculated wave numbers for maximum growth. Asali & Hanratty, 1993.
7.7 Ripples generated on a liquid film at high gas velocities 169
2.6
μL ρG 1/2
= 2.13
2.5 μG ρL
14.72
2.4
max
2.3
21.91
ua
CR
2.2
2.1
35.60
2.0 ReLF
= 51.35
4
1.9
1000 1400 1800 2200 2600 3000 3400
σ
vG*μG
Figure 7.11 Calculated velocities of the fastest growing waves on a water film. Asali & Hanratty, 1993.
the film height for maximum growth is weakly dependent on the Reynolds number of the
liquid film. Figure 7.11 shows the calculated velocity of the fastest growing waves. The
dimensionless wave velocity, cR =ua is calculated to be approximately equal to 2. It is
noted in the discussion before equations (7.11) and (7.12) that one should expect liquid
inertia to have a small effect on the stability of the film. Instability occurs when the
destabilizing effects of wave induced variations of the gas-phase pressure and shear stress
at the interface overcome stabilizing effects of gravity and surface tension. Thus, the
modeling of T~ SI and P~ SR is critical.
The calculations presented in the paper by Asali & Hanratty (1993) employed
models for turbulent flow over wavy surfaces discussed in Sections 6.6 and 6.7 to
calculate T~ Si and P
~ SR . They used a slight modification of the Model D explored by
Thorsness et al. (1978). This approach is the relaxation analysis of Abrams & Hanratty
(1984) or Model D* in Abrams (1984). It uses a boundary-layer coordinate system
embedded in the wave surface and a modification of the van Driest model outlined by
Loyd et al. (1970).
The main result is that the observed wavelength is twice the value of the predicted
wavelength of the ripple waves at maximum growth. This is illustrated in Figure 7.12.
Theory predicts the observed trends in the measurements. However, there is not an
exact agreement between predicted and measured wavelengths. This should not be sur-
prising because the observed waves, although broad-crested, are not two-dimensional.
Asali & Hanratty (1993) provide the possible explanation that ripples evolve from a
two-dimensional wave with a wavelength equal to the fastest-growing wave predicted by
linear theory. As these waves grow in amplitude, they become unstable to a spanwise
disturbance, as shown in Figure 7.13. According to this picture the two-dimensional waves
are altered in such a way that the crests break into a group of isolated long-crested three-
dimensional waves. Such an instability has been explored by McLean et al. (1981) for
gravity waves on deep fluids.
100
μL (cP)
1.1 Würz (1977)
1.1
Asali & Hanratty (1993)
2.0
2λ predicted (cm)
10–1
10–2 –2
10 10–1 100
λobserved (cm)
Figure 7.12 Comparison of calculated spacing of the most rapidly growing waves with measurements. Asali &
Hanratty, 1993.
Ideal
wavelength
Observed
wavelength
(a) Ideal ripples (b) Observed ripples

Figure 7.13 Postulated growth of infinitesimal two-dimensional ripples into a finite-amplitude
three-dimensional wave pattern.
7.8 Roll waves in pipelines
The analysis presented in Section 7.5.2 for a rectangular channel was applied to a circular
pipeline by Lin & Hanratty (1986). The simplified configuration for a stratified flow,
presented in Chapter 5, was used. This was followed with analyses by Hurlburt &
Hanratty (2002) and by Soleimani and Hanratty (2003). These show that viscous large-
wavelength theory predicts the appearance of roll waves. At low gas velocities, these roll
References 171
waves grow into slugs because the height of the stratified layer is such that slugs would be
stable, if generated (see Chapter 2).
Soleimani & Hanratty (2003) present a detailed description of the disturbances in the
pseudo-slug region, which are identified as roll waves. A striking feature is the relative
insensitivity of liquid holdup (time-mean height) to changes in the liquid flow rate in the
pseudo-slug regime. For example, at a superficial gas velocity of 8m/s, h/dt varies from
0.17 to 0.18 as the superficial liquid velocity increases from 0.15 m/s to 0.25 m/s. This is
explained by an increase in the velocity and frequency of roll waves with increasing
superficial liquid velocity. The formation of slugs is initiated by the coalescence of roll
waves to form stable slugs. The transition is predicted to occur when the holdup is such
that propagating slugs would be stable.
References
Abrams, J. 1984 Turbulent flow over small amplitude solid waves. Ph.D. thesis, Univerity of
Illinois.
Abrams, J. & Hanratty, T.J. 1984 Relaxation effects observed for turbulent flow over a wavy
surface. J. Fluid Mech. 151, 443–455.
Alekseenko, S.V., Nakoryakov, V.E. & Pokusavev, B.G. 1985 Wave formation on vertical falling
liquid films. Int. J. Multiphase Flow 11, 607–627.
Jl 31, 119–126.
Andritsos, N. & Hanratty, T.J. 1987 Influence of interfacial waves in stratified gas–liquid flows.
AIChE Jl 33, 444–454.
Asali, J.C. & Hanratty, T.J. 1993 Ripples generated on a liquid film at high gas velocities. Int. J.
Multiphase Flow 19, 229–243.
Asali, J.C., Andreussi, P. & Hanratty, T.J. 1985 Interfacial drag and film height for vertical annular
flow. AIChE Jl 31, 895–902.
Benjamin, T.B. 1957 Wave formation in laminar flow down an inclined plane. J. Fluid Mech. 2,
554–574.
Binnie, A.M. 1957 Experiments on the onset of wave formation on a fluid flowing down a vertical
plane. J. Fluid Mech. 2, 551–553.
Bontozoglou, V.L. & Hanratty, T.J. 1989 Wave height estimation in stratified gas–liquid flows.
AIChE Jl 35, 1346–1350.
Cohen, L.S. & Hanratty, T.J. 1965 Generation of waves in concurrent flow of air and a liquid.
AIChE Jl 11, 138–143.
Cornish, V. 1934 Ocean Waves and Kindred Phenomena. Cambridge: Cambridge University Press.
Craik, A.D.D. 1966 Wind generated waves in thin films. J. Fluid Mech. 26, 369–392.
Craik, A.D.D. 1968 Wind generated waves in contaminated liquid films. J. Fluid Mech. 31, 141–162.
Dressler, R.F. 1949 Mathematical solution of the problem of roll waves in inclined open channels.
Comm. Pure Appl. Math. 2, 149–194.
Dressler, R.F. 1952 Stability of uniform flow and roll wave formation. US Natl Bur. Std. Circular
521, 237–241.
Hanratty, T.J. 1983 Interfacial instabilities caused by air flow over a thin layer. In Waves on Fluid
Interfaces, ed. R.E. Meyer. New York: Academic Press, pp. 221–259.
Hanratty, T.J. 1991 Separated flow modeling and interfacial transport phenomena. Appl. Scient.
Res. 48, 353–390.
Hanratty T.J. & Hershman, A. 1961 Initiation of roll waves. AIChE Jl 7, 489–497.
Henstock, W.H. & Hanratty, T.J. 1976 The interfacial drag and height of the wall layer in annular
flows. AIChE Jl 7, 489–497.
Hewitt, G.F. & Hall-Taylor, N.S. 1970 Annular Two-Phase Flow. London: Academic Press.
Hurlburt, E.T. & Hanratty, T.J. 2002 Prediction of transition from stratified to slug and plug flows
for long pipes. Int J. Multiphase Flow 28, 707–729.
Lighthill, M.J. & Whitham, G.B. 1955 Flood movement in long rivers. Proc. R. Soc. Lond. A229,
291.
Lin, P.Y. & Hanratty, T.J. 1986 Prediction of the initiation of slugs with linear stability theory. Int.
Loyd, R.J., Moffat, R.J. & Kays, W.M. 1970 The turbulent boundary on a plate: an experimental
study of the fluid dynamics with strong favorable pressure gradients and blowing. Report No.
HMT-13, Stanford University California.
McLean, J.W., Ma, Y.C., Martin, D.V. & Saffman, P.C. 1981 Three-dimensional instability of
finite amplitude water waves. Phys. Rev. Lett. 46, 817–820.
Meyer, R.E. (ed.) 1983 Waves on Fluid Interfaces. New York: Academic Press, pp. 221–259.
Miya, M., Woodmansee, D.E. & Hanratty, T.J. 1971 A model for roll waves in gas–liquid flow.
Chem. Eng. Sci., 26, 1915–1931.
Shearer, C.J. & Nedderman, R.M. 1965 Pressure gradient and liquid film thickness in concurrent
upwards flow of gas/liquid mixtures: application to film cooler design. Chem. Eng. Sci. 20,
679–683.
Soleimani, A. & Hanratty, T.J. 2003 Critical liquid flows for the transition from the pseudo-slug
and stratified patterns to slug flow. Int. J. Multiphase Flow 29, 707–729.
Stoker, J.J. 1957 Water Waves. New York: Interscience.
Taylor, G.I. 1963 Generation of ripples by wind blowing over a viscous liquid. In The Scientific
Papers of Sir Geoffrey Ingram Taylor. ed. G.K. Batchelor. Cambridge: Cambridge University
Press, pp. 244–254.
Thorsness, C.B., Morrisoe, P.H. & Hanratty, T.J. 1978 A comparison of linear theory with
measurements of the variation of shear stress along a solid wave. Chem. Eng. Sci. 33, 579–592.
Whitham, G.B. 1974 Linear and Non-linear Waves. New York: John Wiley.
Woodmansee, D.E. & Hanratty, T.J. 1969 Base film over which roll waves propagate. AIChE Jl 15,
712–715.
Würz, D. E. 1977 Flüssigkeits – Filmströmung unter einwirkung einer Uber-lufströmung. Thesis,
Univ. Karlsruhe, Inst. für thermische strömung maschinen.
8 Bubble dynamics
8.1 Prologue
A central issue to be addressed in analyzing the behavior of bubbles in a gas–liquid flow

is understanding the free-fall velocity of a spherical solid particle and the rise velocity of
a spherical bubble in an infinite stationary fluid. Analytical solutions for these systems
are available for very low particle Reynolds numbers (Stokes law and the Hadamard
equation). A derivation of Stokes law is presented in the first part of this chapter.
Experiments show that Stokes law is valid for particle Reynolds numbers less than
unity. For larger ReP, empirical correlations of the drag coefficient are used. The
description of the rise velocity of bubbles is complicated by possible contamination of
the interface. Measurements of the rise velocity of single bubbles are usually presented as
plots of US versus the bubble size for a given system. The structure of these plots reflects
changes in the shape and behavior of the bubbles. Very large bubbles take the shape of a
cap. A prediction of the rise velocity of these cap bubbles, developed by Batchelor, is
presented in Section 8.7.
Other very large bubbles encountered in gas–liquid flows are the Taylor bubble,
which is a part of slug flows in vertical pipes, and the Benjamin bubble used to model
the shedding of liquid from slugs in horizontal pipes. These are described in Sections
8.8 and 8.9.
Bubbly flows are situations for which a swarm of bubbles is considered. A feature of
these flows is that the rise velocities of bubbles are less than would be observed for single
bubbles. In this sense, the bubble motion is hindered. There is a kinship of bubbly flows
to sedimenting and fluidizing liquid–solid systems, where the critical issue is to relate the
sedimenting and the fluidizing velocities to the free-fall velocity and the volume fraction
of solids. These systems are described in Section 8.14.
8.2 Free-fall velocity of a solid sphere
A sphere released into a fluid would experience forces due to gravity, equal to +ρP VPg,
where V P ¼ πd 3P =6 is the volume of the sphere and the plus sign indicates a force in the
direction of gravity. The resisting force, FD, is related to the magnitude of the relative
velocity between the sphere and the fluid. The usual method for describing FD is to define
a drag coefficient, CD.
174 Bubble dynamics
1
F D ¼ C D ρF jU S j2 S ð8:1Þ
2
where jU S j2 is the square of the magnitude of the relative velocity between the particle
and the fluid and S is the projected area of the particle, S ¼ πd 2P =4.
The pressure in a stagnant fluid changes because of the variation of hydrostatic head.
This gives rise to a buoyancy force FB, which is related to the volume of fluid displaced
by the particle, VP,
F B ¼ ρF V P g ð8:2Þ
The acceleration of the sphere is given as mP dU P =dt where mP ¼ ρP V P and UP is the

particle velocity. A particle moving through a fluid drags a volume of fluid with it.
Thus, the force needed to accelerate the particle is given as ðρP V P þ ρF V P f ÞðdU P =dt Þ,
where VPρP is the mass of the particle, ρFVPf is the effective mass of the fluid dragged by
the sphere and f is usually taken as a constant. Thus, from Newton’s law of motion,
dU P 1
ðρP V P þ ρF f V P Þ ¼ þρP gV P ρF gV P C D ρF jU S j2 S ð8:3Þ
dt 2
Under free-fall conditions, the mean acceleration is zero, so that, from (8.3), the mean
free-fall velocity is
V P g ðρ P ρ F Þ
U 2S ¼ ð8:4Þ
SρF 12 C D
8.3 Stokes law for a solid sphere
The force FD of a fluid on a falling sphere can be calculated exactly for the limit of very
small particle Reynolds numbers, ReP ¼ d P U S ρF =μ51.
This is done by using the laws of conservation of mass and momentum. These are
given for a Cartesian coordinate system by equations (4.7), (6.1) and (6.3). Their
representation in a spherical polar coordinate system is given in Tables 3.4–1, 3.4–4,
3.4–7 of the book by Bird et al. (1960). The relations of fluid stresses to components of
the velocity gradient tensor are given in Bird et al. (1960), Table 3.4–7, for a Newtonian
fluid.
The following analysis leading to Stokes law is also taken from Bird et al. A sketch of a
sphere settling with a velocity US and the definition of θ are given in Figure 8.1a. For a
symmetric flow, uϕ = 0, so there is no variation of the velocity in the ϕ-direction. For an
incompressible fluid, the density is constant. Thus, the following equations define ur,
uθ and p:
1 ∂ð r 2 u r Þ 1 ∂ðuθ sin θÞ
þ ¼0 ð8:5Þ
r2 ∂r r sin θ ∂θ

∂ur ∂ur uθ ∂ur u2θ
ρ þ ur þ
∂t ∂r r ∂θ r
ð8:6Þ
∂p 1 ∂2 2 1 ∂ ∂ur
¼ þ μ 2 2 r ur þ 2 sin θ þρgr
∂r r ∂r r sin θ ∂θ ∂θ

∂uθ ∂uθ uθ ∂uθ ur uθ
ρ þ ur þ þ
∂t ∂r r ∂ϕ r
ð8:7Þ
1 ∂p 1 ∂ 2 ∂uθ 1 ∂ 1 ∂ 2 ∂ur
¼ þμ 2 r þ 2 ðuθ sin θÞ þ 2 þ ρg θ
r ∂θ r ∂r ∂r r ∂θ sin θ ∂θ r ∂θ
As seen in Figure 8.1, boundary conditions on the surface of the sphere are
ur ¼ US cos θ, uθ ¼ U S sin θ. The velocities are zero at large distances from the
sphere since the fluid surrounding the sphere is stagnant.
The solution is more easily obtained by using a coordinate system which is moving,
with the sphere, at a velocity US (see Figure 8.1b). An observer in the moving coordinate
system sees a steady flow around the sphere, so the transient terms ∂ur =∂t and ∂uθ =∂t are
set equal to zero. The boundary conditions are that
(a)
θ
ur = US cosθ
uθ = US sinθ
US
(b)
ur = uθ = 0
uθ = US sinθ
θ
ur = US cosθ
US US US
Figure 8.1 Solid sphere moving at a velocity through an infinite stagnant fluid: (a) as seen by a stationary
observer, (b) as seen by an observer moving with the sphere.
176 Bubble dynamics
ur ¼ uθ ¼ 0 ð8:8Þ
at the surface of the sphere and that

ur ¼ U S cos θ ð8:9Þ
uθ ¼ U S sin θ ð8:10Þ
at r ¼ ∞:
The three equations to be solved have three unknowns ur, uθ and p. A stream function,
ѱ(r, θ), can be defined from the equation of conservation of mass, (8.5).
1 ∂ψ
ur ¼ ð8:11Þ
sin θ ∂θ
r2
1 ∂ψ
uθ ¼ ð8:12Þ
r sin θ ∂r
uϕ ¼ 0 ð8:13Þ
Its validity can be tested by substituting (8.11) and (8.12) into (8.5). The function ψ
describes the current lines, that is, the amount of fluid that flows between two curves of
constant ψ.
The substitution of (8.11) and (8.12) into the momentum equations eliminates ur and
uθ. The pressure is eliminated by calculating ∂2 p=∂θ∂r from (8.6) and ∂2 p=∂r∂θ from
(8.7). These are equated to obtain a relation for ψ. The assumption is made that fluid
inertia can be neglected in a coordinate moving with velocity US. The following equation
is obtained for the stream function, ψ:
E4 ψ ¼ 0 ð8:14Þ
where the operator is defined as

2 2
∂ sin θ ∂ 1 ∂
E ¼
4
þ 2 ð8:15Þ
∂r2 r ∂θ sin θ ∂θ
The boundary conditions suggest a solution of the form

ψ ¼ f ðrÞsin2 θ ð8:16Þ
This satisfies (8.14) if

d2 2 d2 2
2
2 f ðr Þ ¼ 0 ð8:17Þ
dr r dr2 r2
The term rn is the solution of this equation. Its substitution into (8.17) produces a fourth-
order algebraic equation in n. Thus, there are four independent solutions: r‒1, r1, r2, r4.
The solution of (8.17) is given as
A
f ðrÞ ¼ þ Br þ Cr2 þ Dr4 ð8:18Þ
r
The constants A, B, C, D are specified so as to satisfy the boundary conditions:

A ¼ ð1=4ÞU S r2P B ¼ ð3=4ÞU S rP C ¼ ð1=2ÞU S D¼0
where rP is the radius of the sphere. The velocity components are found by substituting
(8.16) into (8.11) and (8.12) with f (r) given by (8.18):

3 rP 1 rP 3
ur ¼ U S 1 þ cos θ ð8:19Þ
2 r 2 r

3 rP 1 rP 3
uθ ¼ U S 1 sin θ ð8:20Þ
4 r 4 r
The fluid exerts a normal force due to the variation of the pressure around the sphere (the
form drag) and a frictional force due to shear stresses exerted on the sphere (the skin
friction). The pressure can be calculated with (8.7), (8.19) and (8.20). The local frictional
stress at the surface of the sphere, τ rθ , is calculated from the velocity field since

∂ uθ
τ rθ ¼ μ r at r ¼ rP
∂r r
The total force due to skin friction and form drag can be calculated by integrating p and
τ rθ over the surface of the sphere. Bird et al. (1960) show that the total force is
F D ¼ 2πμrP U S þ 4πμrP U S ð8:21Þ
where the first term is the skin friction and the second term is the form drag.
Stokes law is given as
F D ¼ 6πμrP U S ð8:22Þ
or as
24
CD ¼ ð8:23Þ
ReP
Where CD is the drag coefficient defined by (8.1). If (8.23) is substituted into (8.4) a
free-fall velocity of
g ðρP ρF Þd 2P
US ¼ ð8:24Þ
18μ
is predicted for Reynolds numbers at which Stokes law is applicable. Thus US varies as
d 2P at very low ReP. Typical inertial and viscous terms appearing in the Navier–Stokes
equations are ρur ð∂ur =∂rÞ and μ=r2 ∂2 =∂r2 ðr2 ur Þ . If ur scales with US and r scales with
dP, the ratio of the inertia forces to the viscous forces scales as
ρU 2S
dP
=μ dU
S
2
P
S
¼
dPU Sρ
μ
¼ ReP
Thus, one can expect that the neglect of inertia terms is valid for very small ReP.
178 Bubble dynamics
8.4 Rise velocity of bubbles at low ReP
The calculation of the rise velocity of a bubble in a stationary fluid at low ReP is similar to
that for the free-fall velocity of a solid particle. If ρP is taken as the density of the gas in the
bubble and ρF as the density of the surrounding fluid, the sum of the gravitational force
and the buoyancy force is given as ðρP ρF Þ gVP, where a positive sign indicates a
force or velocity opposite the direction of gravity.
Note that, in using (8.3), the mean inertia of the system is dominated by the mass
of the liquid dragged along by the bubble; that is, ρF V P f >> ρP V P in (8.3). As for a
solid sphere, an equation for CD can be developed at small Reynolds numbers if
inertia terms in the Navier–Stokes equations are ignored. A spherical bubble differs
from a spherical solid sphere in that the inside of the sphere can be in motion and
the fluid velocity at the boundary need not be the same as for a solid sphere. Hadamard
(1911) developed an equation for the velocity fields both inside and outside the
bubble. The following relation was obtained for the terminal velocity in a stationary
fluid:
d 2P g ðρP ρF Þ ð3μP þ 3μF Þ
US ¼ ð8:25Þ
18μF ð3μP þ 2μF Þ
where μP is the viscosity of the gas in the bubble. For μP ¼ ∞, (8.25) gives (8.24), the
equation for a solid sphere at low ReP. For μP = 0, (8.25) gives
d 2P gðρF ρP Þ
US ¼ ð8:26Þ
12μF
Thus, because of motion inside the bubble the terminal velocity is predicted to be 3/2 of
the value for a rigid sphere (if the interface is clean).
8.5 Measurements of CD
8.5.1 Solid spheres

Values of CD for flow around solid spheres have been obtained in a number of experi-
ments. These are presented in plots of CD versus ReP (Bird et al., 1960; Schlichting,
1955; Soo, 1967; Crowe et al., 1998). The agreement among different experiments is not
perfect. This is because they were done for different global flow conditions (Soo, 1967;
Crowe et al., 1998).
The plots by Bird et al. and by Schlichting agree with Stokes law at very low Reynolds
numbers. They show that (8.23) provides a good representation of the data for ReP < 1.
P for 2 < ReP 5 5 10 : A rough representation
Bird et al. suggest that C D ≈ 18:5=Re0:6 2
for the drag coefficient at large ReP is Newton’s law, CD = constant. Bird et al. give
CD ≈ 0.44 for 5 102 5ReP 52 105 : A sharp drop in CD is found at very large ReP.
This can be associated with a sudden transition to turbulence in the boundary layer, which
8.5 Measurements of CD 179
causes a decrease in the low-pressure wake region. The substitution of CD = 0.5 into (8.4)
gives the following relation for the free-fall velocity of a solid sphere:

2 d P ðρP ρF Þg 1=2
US ¼ ð8:27Þ
3 ρF
Equations (8.24) and (8.27) indicate that U S / d 2P at very low Reynolds numbers and
1=2
that U S / d P at large Reynolds numbers.
8.5.2 Gas bubbles

Studies of the rise velocity of single spherical bubbles in water at small Reynolds
numbers usually give values of US close to what is predicted for a solid sphere, rather
than (8.25). Levich (1962) explains this finding as due to contamination of the interface.
He points out that gradients of the concentration of contaminants along the interface
cause changes in surface tension which result in forces that can oppose the motion of the
interface.
The size of a bubble can be represented by its volume, VP, or by an equivalent radius,
rP, defined from the volume, that is, V P ¼ ð4=3Þπr3P . For ReP = 1, the equivalent radius
is approximately 0.2 mm for air bubbles in water. A summary of measurements of the
rise velocity of air bubbles in filtered or distilled water is given in Figure 8.2 (see also
Hetsroni, 1982, Figure 1.3.24; Soo, 1967, Figure 3.20). For bubble radii less than
0.2 mm, the rise velocity is found to increase with the square of the bubble radius, as
60
50
40
30
20
Terminal velocity (cm/s)
15
10
9
8
7 Cross-section of container
6 1
4 2ft × 25 ft
5
4 3 ft × 3 ft
15 cm × 15 cm
3 Diameter × 7 cm
Diameter × 5 cm
2
Diameter = 10.3 cm
1.5
1
0.01 0.02 0.04 0.06 0.080.1 0.2 0.4 0.6 0.8 1.0 2.0 4.0
Equivalent radius (cm)
Figure 8.2 Terminal velocity of single air bubbles in filtered or distilled water. Haberman & Morton, 1953.
180 Bubble dynamics
predicted by (8.24). The rise velocity shows a sharp local maximum of about 33 cm/s for
rP ≈ 0.7 mm and a broad local minimum of about 24 cm/s for rP ≈ 3 mm.
At low ReP, bubbles injected into the field keep a spherical shape up to rP ≈ 0.7 mm
because surface tension effects are much greater than inertia effects. The bubbles move
in straight lines. For equivalent radii of 0.7 to 3 mm, the bubbles behave differently.
Their shape is in the form of a flattened ellipsoid whose minor axis is in the direction of
mean motion; the bubbles vibrate and move in a spiral motion (Grace & Weber, 1982).
For radii greater than 3 mm, the bubbles assume a turbulent motion. For radii of
1 to 10 mm, the rise velocity in water is roughly constant (approximately equal to
30 cm/s). For very large radii, say greater than 10 mm, the bubble size increases with
r0:5
P . This is the cap bubble regime described in Section 8.7. Studies of the effect of
changes of the liquid properties do not produce definitive results. One of the reasons for
this is that contamination of the interface can have an important effect (Haberman &
Morton, 1953).
A dimensionless group characterizing the effect of surface tension is U S =ðgσ=ρF Þ0:25 .
The following equation, proposed by Harmathy (1960), roughly represents measure-
ments for large bubbles over a range of conditions:

gDρσ 0:25
U S ¼ 1:53 ð8:28Þ
ρ2F
8.6 Gaseous volume fraction for bubbly flow
The gaseous volume fraction in a one-dimensional system, α, is defined as the ratio of the
volumetric flow of gas to the total volumetric flow, given as ðuGS þ uLS ÞA. This param-
eter is considered below for the case where variations of the gas density can be ignored.
One way to characterize the slip between the phases is to define a slip ratio, S, as equal
to the ratio of the local gas velocity to the local liquid velocity (equation (1.67)). Equation
(1.70) then defines α. Another way, used in the separated flow model (described in
Section 1.4), is to consider the difference between the velocity of the gas and the velocity
of the mixture, ðuGS þ uLS Þ:
V Drif t ¼ uG C 0 ðuGS þ uLS Þ ð8:29Þ
This is called the drift velocity of the gas (see Nicklin et al., 1962; Zuber & Findlay,
1965; Wallis, 1969). The constant C0 was introduced by Zuber & Findlay. It has a
value of 1.2, which recognizes that the nose of the bubble moves relative to the
centerline of the velocity profile in turbulent liquid slugs. The following equation is
given by Wallis (1969):
QG
α¼ ð8:30Þ
QL þ QG þ AV Drif t
where A is the column area.

8.7 Cap bubbles 181
8.7 Cap bubbles
Very large cap-shaped entities are characteristic of bubbles that contain a volume of gas
larger than 5 cm3. Batchelor (1967) presents photographs and describes these bubbles as
“looking like an umbrella.” He depicts them as a slice off a sphere of radius R and provides
an analysis in a framework which is moving with the velocity of the bubble, US. The front
of the bubble is, then, a stagnation point. It has a radius of curvature R. The bubble is large
enough for effects of surface tension and viscosity to be neglected. A spherical polar
coordinate system is used. The idealized bubble is symmetric in the ϕ-direction. The front
face is steady, smooth and close to spherical in shape. The Bernoulli equation
p q2
þ þ gz ¼ const: ð8:31Þ
ρ 2
is used to describe the flow along a streamline, where z is the distance upward in the
vertical direction. The goal is to develop an equation for the rise velocity, US.
Consider a streamline embedded in the surface of the bubble. The velocity of the fluid
at the nose of the bubble is zero in a moving coordinate system, so the constant in (8.31) is
pS =ρ þ gzS . Thus
p q2 p
þ þ gz ¼ S þ gzS ð8:32Þ
ρ 2 ρ
where zS is the location of the stagnation point. Since the region near a stagnation point is
considered, the pressure is constant, that is, p = pS, so
q2 ¼ 2gðzS zÞ ð8:33Þ
Equation (8.33) describes the velocity variation at the surface of the bubble in a
coordinate system moving with the velocity of the bubble.
The problem is to relate q to US. This is done by solving for the velocity field. The fluid
approaching the bubble has a uniform velocity and may be considered inviscid.
Therefore, an irrotational flow is assumed. As shown in Chapter 4, the velocity vectors
are then described as the gradient of a potential function. Conservation of mass for an
incompressible fluid dictates that the potential function is defined by the Laplace
equation (4.9).
The front of the bubble has a spherical shape, so flow around a sphere is consid-
ered. The Laplace equation in spherical coordinates is given in Table A.7–3 of Bird
et al. (1960):

∂ 2 ∂ϕ 1 ∂ ∂ϕ
r þ sin θ ¼0 ð8:34Þ
∂r ∂r sin θ ∂θ ∂θ
for a symmetric flow, where

∂ϕ 1 ∂ϕ
ur ¼ uθ ¼ ð8:35Þ
∂r r ∂θ
182 Bubble dynamics
Equation (8.34) is solved in a frame of reference moving with velocity US using the
boundary conditions that ur is zero at the boundary of the sphere and that US is the
velocity at large distances from the sphere.
A solution of (8.34) is obtained by a separation of variables technique
ϕ ¼ f ðrÞ cos θ ð8:36Þ
This satisfies (8.34) if

d2f df
r2 þ 2r 2f ¼ 0 ð8:37Þ
dr2 dr
Substitution of rn into (8.37) gives two solutions with n = –2, n = 1 so

A
f ¼ þ Br ð8:38Þ
r2
To satisfy the condition that the velocity at a large distance from the surface is US
B ¼ US ð8:39Þ
For a viscous flow, both uθ and ur are zero at the surface. However, for an inviscid flow
only one of these conditions, ur = 0, can be satisfied, so
R3 U S
A¼ ð8:40Þ
2
Thus, uθ ≠ 0 at the bubble surface. The solution

U S R3
ϕ¼ cos θ þ U S r cos θ ð8:41Þ
2r2
is given in Streeter (1948, p. 67). The velocity at any point on the surface of the
sphere r = R is
1 ∂ϕ 3
¼ q ¼ U S sin θ ð8:42Þ
r ∂θ 2
The following equation for the rise velocity is obtained by substituting (8.42)
into (8.33):
9
2Rgð1 cos θÞ ¼ U 2S sin2 θ ð8:43Þ
4
U S 8 ð1 cos θÞ
2
¼ ð8:44Þ
gR 9 sin2 θ
An expansion of ð1 cos θÞ gives ð1 cos θÞ= sin 2 θ ≈ 1=2 for small θ, so that
2 pffiffiffiffiffiffi
US ¼ gR ð8:45Þ
3
Equation (8.45) is found to agree with measurements of Davies & Taylor (1950).
8.8 Taylor bubbles / vertical slug flow 183
The radius R can be related to the volume of the bubbles by representing the bubbles by
a spherical cap with an occluded angle of 100º and a relatively flat tail (Wallis, 1969,
p. 249). Thus (8.45) can be written as
1=2
1=3
U S ¼ 0:79 gV P ð8:46Þ
or as
pffiffiffiffiffiffiffi
U S ¼ 1:00 grP ð8:47Þ
where rP is the equivalent radius defined by the relation V P ¼ ð4=3Þπr3P . From (8.4),
the above result indicates that the drag coefficient is approximately equal to 1.0.
This is the same as found for solid discs at large Reynolds numbers (Lapple &
Shepherd, 1940).
8.8 Taylor bubbles / vertical slug flow
A description of vertical slug flow is given in Section 2.7.1 repeated here: “At large gas
velocities a slug flow can appear. It consists of a progression of bullet-shaped bubbles
(called Taylor bubbles) that fill most of the pipe cross-section. The fronts resemble cap
bubbles and the backs are approximately flat. The slugs of liquid between the Taylor
bubbles can contain small bubbles. Large pressure fluctuations are observed. The liquid
between the gas and the wall is moving downward. Thus, to an observer moving with the
gas bubble, the liquid moves around the bubble.” At the back of the bubble, the interaction
of the wall layer with the liquid slug resembles a plunging jet (Lin & Donnelly, 1966). This
produces the bubbles that aerate the liquid slug which follows the Taylor bubble (see Kaji
et al., 2009).
A discussion of the slug flow regime and of Taylor bubbles is given in Wallis (1969,
Chapter 10) and Batchelor (1967, pp. 475–479). If the effects of viscosity and surface
tension can be ignored, the velocity of a Taylor bubble in a stationary liquid is
calculated to be
1=2
U S ¼ k 1 ρL ½gd t ðρL ρG Þ1=2 ð8:48Þ
where dt is the pipe diameter. Wallis suggested that k1 = 0.345. If ρG is assumed to be

negligible compared with ρL then
pffiffiffi
U S ¼ 0:345 2ðgrt Þ1=2 ð8:49Þ
where rt is the pipe radius. This is to be compared with the velocity of cap bubbles,
equation (8.45). Wallis also develops equations that include the effects of viscosity and
surface tension.
The analysis of the slug regime for vertical flows can be simplified by ignoring the
influence of bubbles on liquid slugs. An equation for α is then obtained by substituting (8.29)
for VDrift into (8.30). A force balance gives the following equation for the pressure gradient.
184 Bubble dynamics
dp 2
¼ g ½ρL ð1 αÞ þ ρG α þ τ W ð1 αÞ ð8:50Þ
dz rt
The first term on the right side gives the contribution due to hydrostatic head. The second
term is the frictional pressure loss. The term (1 – α), multiplying τW, appears because it is
assumed that only the turbulent liquid in the slugs is making a contribution. The resisting
wall stress can be approximated by the friction factor relation for single-phase flow
f
τ W ¼ ρL u2L ð8:51Þ
2
where uL ¼ uLS þ uGS and f is a function of the slug Reynolds number, Re ¼ d t uL ρL =μL .
8.9 Benjamin bubble
Consider a horizontal rectangular channel or circular tube, filled with a liquid, that is
opened at one end. The liquid will drain out, as indicated in Figure 8.3. A cavity is formed
at the top of the channel. A layer of liquid of height hL2 moves out the back of the channel.
The cavity moves upstream at a velocity cB.
View the flow in a coordinate system moving with velocity cB. The cavity then appears
stationary. Location 0 is a stagnation point where the velocity is zero. The flow at Station
1 is assumed to be moving uniformly with a velocity cB. Thus, from this framework, fluid
appears to be moving into the channel at a rate Q = cBH, where Q is the volumetric flow
per unit breadth. From conservation of mass, the fluid appears to be moving out of the
channel at a rate Q = cBH and at a velocity of uL2 ¼ Q=hL2 , where H is the height of the
channel.
Benjamin (1968) provides an analysis of this flow in his paper “Gravity currents and
related phenomena.” Effects of surface tension are ignored. He applied the momentum
theorem to a volume between upstream (Station 1) and downstream (Station 2) locations,
which are in fully developed regions. Again, a coordinate system moving with a velocity
of cB is assumed. The momentum flowing into the volume is ρc2B H. The momentum
associated with gas flow outward can be neglected because of the small density. Thus,
Station 1 Station 2
cB
H
hL2 uL2
Figure 8.3 Sketch of liquid draining from a rectangular channel of height H: a Benjamin bubble moving with a
velocity cB.
8.9 Benjamin bubble 185
the
momentum flowing
out is ρuL2 uL2 hL2 . The average force on the upstream face is
p1 H þ ρgH 2 =2 , where the second term is the pressure
force due to the gravitational
head. The average force on the downstream face is p2 H þ ρgh2L1 =2 . Thus, the momen-
tum theorem gives

H2 ρgh2L2
ρu2L2 hL2 ρc2B H ¼ p1 H þ ρg p2 H þ ð8:52Þ
2 2
Benjamin used (8.52), conservation of mass

hL2
cB ¼ uL2 ð8:53Þ
H
and the following two equations developed by applying the Bernoulli equation to the top
wall from Station 1 and 0 and to the free streamline (0 to 2), along which the pressure is
constant and equal to p0
p1 c2B p0 p0 p u2
þ ¼ þ 0 þ gH ¼ 0 þ L2 þ ghL2 ð8:54Þ
ρ 2 ρ ρ ρ 2
He obtained

g H 2 h2L2 H
c2B ¼ ð8:55Þ
ð2H hL2 ÞhL2
H
hL2 ¼ ð8:56Þ
2
uL2 ¼ ð2ghL Þ1=2 ð8:57Þ
The shedding rate is given as
Q ¼ uL2 hL2 ¼ hL2 ð2ghL2 Þ1=2 ð8:58Þ
This analysis describes a flow caused by gravitational forces.

Benjamin also considered flow out of a circular tube of radius rt and obtained
uL2
¼ 1:322 ð8:59Þ
ðgrt Þ1=2
cB
¼ 0:767 ð8:60Þ
ðgrt Þ1=2
The shedding rate is
Q ¼ cB πrt 2 ¼ 0:767ðgrt Þ1=2 πrt 2 ð8:61Þ
where Q is the volumetric flow. These results have found application in analyses of slug
and plug flows, where the liquid shed by a moving slug or plug is modeled as a Benjamin
bubble.
186 Bubble dynamics
8.10 Gas lift pumps
Gas lift pumps, which have no moving parts, are widely used. See, for example, de
Cachard & Delhaye (1996), Clark & Dabolt (1986). The principle of operation is that gas
is introduced into an immersed vertical pipe. The gas–liquid mixture in the pipe is lighter
than the surrounding liquid, so it rises.
Consider the simple example shown in Figure 8.4 for an air–water system. A pipe is
submerged in water to a depth H, so the water level in the pipe is the same as in the tank.
An air injector is located close to the bottom of the submerged pipe at z0. The pressure at
this location is
ðp0 pa Þ ¼ ρL gH ð8:62Þ
where pa is the ambient pressure in the splash box.

When gas is admitted at a small rate, the gas–liquid flow in the pipe looks like a bubble
column (discussed in Section 8.11). As the air rate increases, the height of this column
(and the void fraction) increases. Eventually it attains a sufficient height to spill over into
the splash pot. The liquid carryover increases with increasing gas flow. The theoretical
challenge is to predict the relation between the liquid flow and the gas flow. This is done
Air
Water
Air
Lift height
Submergence H
Water
Figure 8.4 Sketch of a gas lift.
8.10 Gas lift pumps 187
by using a momentum balance. The simplification is made that the flow may be
considered to be uniform over the cross-section of the pipe. Then the one-dimensional
balance is given by equation (1.73):
2 2
G Ax GAx2
d þ ¼ Adp g ½ρG α þ ð1 αÞρL Adz ð8:63Þ
αρG ð1 αÞρL
Phτ W idz þ AGðuG uL Þdx
where α is the volume fraction of the gas and x is the quality, defined as the mass fraction
of the flowing mixture that is a gas or a vapour and the flow is considered to be horizontal.
The flow considered in this section does not involve a phase change so x is not varying
along the pipe. That is, the last term in (8.63) can be ignored. The term on the left side
represents changes of the momentum flux. The second term on the right side is the change
of pressure due to changes in hydrostatic head; the third term represents the frictional
losses.
The pressure drop causing the flow is given by (8.62). Thus, for a given gas flow, the
liquid flow increases with the submergence H/L, where L is the distance through which
the liquid is lifted. Equation (8.63) can be further simplified by ignoring the influence of
the momentum change. Then, a prediction of the liquid flow for a given gas flow and
submergence requires information on the influences of liquid flow on the void fraction
and on the frictional pressure loss.
The homogeneous model, described in Section 1.3, is not adequate. The separated flow
model (Section 1.4) is needed. However, the Lockhart–Martinelli analysis (Section 1.5)
is flawed since it does not, directly, recognize changes in the configuration of the phases.
The working curves for the operation of a gas lift are plots of the liquid flow as a
function of the gas flow for different values of the submergence ratio, H/L. These curves
show an increase in the liquid flow at small gas flows, where the pattern is a slug flow in a
small-diameter pipe. Eventually, a maximum output is realized. This can occur because
of a change in the flow pattern, which is associated with increases of frictional losses and
of the void fraction in the column.
Clark & Dabolt (1986) give the following relation for an isothermal operation which
has a 100% efficiency (no friction losses):
ρL gLV L ¼ V G ½pa lnðp0 =pa Þ ð8:64Þ
The term on the left side is the work to lift a volume of liquid, VL, through a height L. The
term on the right side is the work needed to compress an ideal gas with a volume, VG,
(isothermally) from a pressure pa to a pressure p0. The ratio VG/VL represents the ratio of
the volumetric gas flow to the volumetric liquid flow, if the operation has a 100%
efficiency.
The efficiency can be calculated along the operation curves by considering the ratio of
the volumetric flows of the gas and the liquid. This increases with increasing gas velocity
and reaches a maximum value before the operating curve reaches its maximum. Thus, the
location of the maximum efficiency in the operating curve is at a lower gas velocity than
required to reach a maximum liquid flow (see Clark & Dabolt, 1986).
188 Bubble dynamics
The void fraction is obtained from (8.30). The momentum balance is given by (8.63).
If phase changes and changes in the momentum flux are ignored, one obtains
ðL ðL
AρL gH ¼ g ½ρG α þ ð1 αÞρL ÞAdz þ Phτ W idz ð8:65Þ
0 0
where (8.62) has been substituted for ðp0 pa Þ. Slug flow is considered. The friction
term is important only when liquid slugs are present. Thus τW is given by (8.51) and
(8.53) and
hτ W i ¼ ð1 αÞτ W ð8:66Þ
The friction factor is obtained from relations for single-phase turbulent flows if aeration
of the liquid slugs is ignored. (Clark & Dabolt (1986) used the Martinelli equation to
calculate frictional pressure losses.)
Measurements by Garich & Besserman (1986) for an air–water system in a column
with a diameter of 1 inch are shown in Figure 8.5. Void fractions are plotted in Figure 8.6
Submergence ratios of 0.55, 0.68 and 0.75 were used. Slug flow is the dominant pattern at
low uGS. A maximum in the liquid flow occurs when a churn flow was observed.
However, the location for maximum efficiency is in the region where a slug flow exists.
Transition from a slug flow to a churn flow is not well defined. Garich & Besserman (and
0.7
H/L = 0.55
0.6 H/L= 0.68

H/L = 0.75
Transition to
Superficial liquid velocity, uLS (m/s)
churn flow
0.5
H/L = 0.75
0.4
H/L = 0.68
H/L = 0.55
0.3
0.2
0.1
0
0 0.5 1 1.5 2 2.5 3
Superficial gas velocity, uGS (m/s)
Figure 8.5 Operation curve for a gas lift in a 1.0 inch column containing water.
0.7
0.6
0.5
Void fraction, α
0.4
0.3
0.2
H/L = 0.55
H/L = 0.68
H/L = 0.75
0.1
Transition to
churn flow
0
0 0.5 1 1.5 2 2.5 3
Superficial air velocity, uGS (m/s)
Figure 8.6 Void fractions for a gas lift in a 1.0 inch column containing water.
others) have suggested transition at a volume fraction of α ≈ 0.52. (It should be noted that
Figure 2.11 indicates a transition at uGS ≈ 2 m=s.)
Gas lifts are used for pipes with a wide range of lengths and diameters. Information
about the development of slug patterns is given by Kaji et al. (2009). If the pipe is long
enough, the gas density and the void fraction can vary along the length. Then, the
simplification of ignoring the change of momentum in (8.63) could be in error. Also,
as pointed out in Section 2.4, slug flow is not observed in large-diameter pipes. Instead,
the flow pattern is dominated by cap bubbles. Since the rise velocities of Taylor bubbles
and cap bubbles are close, the equation used to predict α for slug flow might be
applicable.
8.11 Bubble columns
8.11.1 Volume fraction of gas in bubble columns

Bubble columns involve the passage of gas bubbles in a stationary liquid. They find
application as gas absorbers, gas strippers and reactors. All of these are mass transfer
operations which involve a transfer of gaseous or vapor components between a gas (or
190 Bubble dynamics
vapor) and a liquid. The rate of transfer is strongly related to the interfacial area per unit
volume, which increases with increases of volume fraction, α, and decreases in the
bubble diameter. The expansion of the column, α, depends on the velocity of the bubbles,
US, relative to the surrounding liquid.
There is a similarity between bubble columns and sedimenting or fluidizing beds of
solid particles, where the solids fraction can be related to the free-fall velocity in an
infinite fluid (discussed in Section 8.14). An important feature is the phenomenon of
hindered settling, whereby the presence of surrounding particles decreases the free-fall
velocity. Section 8.13 presents an interpretation (there are others), which explains
hindered settling as due to an effective increase in the viscosity of the suspension. It is
tempting to use this approach to describe bubble columns. However, bubble columns are
different in many ways. Changes in liquid viscosity cause changes in the shape and
orientation of the bubbles. Coalescence, associated with increased gas velocity, can
create larger bubbles. Contamination of the interface can greatly affect the behavior of
bubble columns (Anderson & Quinn, 1970). Hindrance of the bubble rise velocity can be
seen at very low gas velocities where the bubbles are spherical. See, for example,
measurements in tap water at small superficial gas velocities in Anderson & Quinn
(1970, Figure 3). The usual operation involves much more complicated configurations
of the gas and increases of US with increasing gas velocity.
8.11.2 Behavior of a bubble column

An established theoretical prediction of the behavior of bubble columns is not
available. However, it is useful to give a qualitative description by citing a few
examples. These rely on chemical engineering research projects performed at the
University of Illinois (Mueller & Wilger, 1979; Norman & Hummel, 1980; Allen &
Gillogly, 1979).
Mueller & Wilger used a sintered glass support plate of medium porosity in a glass
column with a diameter of 4.45 cm. Water (1 cP, 72 dynes/cm) and pentadecane
(2.81 cP, 30.2 dynes/cm) were studied. A swarm of tiny bubbles (less than 1 mm)
formed right at the support. These rapidly coalesced in the vicinity of the support for
low gas velocities and within 10 cm of the support at superficial gas velocities of
greater than 1 cm/s.
Two major flow regimes were observed both for water and for pentadecane. These are
analogous to the particulate and aggregative regimes defined for fluidized beds of solids.
For want of a better choice, this terminology will be used to describe bubble beds. At low
gas rates, the particulate regime exists. The bubbles remain distinct, even when in close
proximity. For superficial gas velocities of about 0.3 cm/s, small spherical entrance
bubbles (diameters of about 1 mm) were still present at one-third of the distance up the
column. However, most of these entrance bubbles coalesced to form bubbles which are
spherical for smaller bubbles and ellipsoidal for larger bubbles. They had sizes of 0.4 cm
and 0.3 cm by 0.8 cm, respectively. As the flow was increased to a superficial gas velocity
of about 0.97 cm/s, more collisions occurred and the entrance bubbles were present only
up to 10 cm from the bottom. The flow was dominated by ellipsoidal bubbles of
Figure 8.7 Bubbles of pentadecane near the top of the column. Superficial air velocity is 0.97 cm/s.
approximately uniform size (0.6 cm by 1.3 cm). A photograph for pentadecane is shown
in Figure 8.7.
As the gas velocity increases the bubbles take on a coherent large-scale singular motion.
Pockets which are devoid of bubbles, and larger bubbles which have a cap shape, appear.
Figure 8.8 shows this aggregative pattern in pentadecane for a superficial air velocity
of 7.7 cm/s.
The studies of Norman & Hummel (1980) in 5.15 cm and 10.45 cm columns and of
Allen & Gillogly (1979) in a 5.15 cm column used plates that were punctured with
0.37 mm holes to support the liquid column. The behavior can be represented by plots of
the volume fraction, α, as a function of the superficial velocity of the gas. Operation
curves were found to be roughly the same for sintered glass and for sieve tray supports.
Changes of the height of the liquid in the original column of liquid had only a small
effect because of the increased opportunity for coalescence. Increases in the diameter of
the column were accompanied by slightly smaller void fractions.
Examples of operating curves for different liquids are shown in Figure 8.9 where D is
the diameter of the tube and L0 is the height of the liquid before aeration. These are
constructed from measurements made by Mueller & Wilger (1979), Norman &
Hummel (1980) and Allen & Gillogly (1979). No consistent effect of liquid properties
is noted.
For example, runs with deionized water and with cyclohexane are situations for which
the viscosities are the same and the surface tension decreases from 72 dynes/cm to
192 Bubble dynamics
Figure 8.8 Aggregative pattern in pentadecane for a superficial air velocity of 7.7 cm/s.
25.5 dynes/cm. The decrease in surface tension is associated with an increase in void
fraction, as would be expected if the bubble diameter decreases. The results from runs
with deionized water and with benzene show small differences due to a large change in
surface tension for liquids with viscosities of 0.652 cP and 1 cP. The remaining liquids
Tap water (1 cP; 72 dynes/cm) Cyclohexane (1.020 cP; 25.5 dynes/cm)

Sintered glass support D = 4.45 cm; L0 = 46.8 cm 29 hole support D = 5.51 cm; L0 = 50 cm
Pentadecane (2.81 cP; 26.9 dynes/cm) Methyl alcohol (0.597 cP; 22.61 dynes/cm)
Sintered glass support D = 4.45 cm; L0 = 49.8 cm 29 hole suppor D = 5.15 cm; L0 = 65.2 cm
Oleic acid (32.5 cP; 32.5 dynes/cm) Isopropyl alcohol (1.77 cP; 21.7 dynes/cm)
29 hole support D = 5.15 cm; L0 = 62.9 cm 29 hole support D = 5.15 cm; L0 = 60 cm
n-butyl alcohol (2.948 cP; 24.6 dynes/cm) Benzene (0.652 cP; 28.86 dynes/cm)
29 hole support D = 5.15 cm; L0 = 60 cm 30 hole support D = 4.9 cm; L0 = 73 cm
Toluene (0.590 cP; 28.5 dynes/cm) Deionized water (1 cP; 72 dynes/cm)
29 hole support D = 5.15 cm; L0 = 50 cm Sintered glass support D = 4.45 cm; L0 = 40.8 cm
0.3
0.25
0.2
α 0.15
0.1
0.05
0
0.3
0.25
0.2
α 0.15
0.1
0.05
0
0 1 2 3 4 5 6 7 8 9 10
uGS (cm/s)
Figure 8.9 Operating curves for bubble columns.
considered in Figure 8.9 show the effect of liquid viscosity for surface tensions which are
varying over a small range, 21.7 to 28.9 dynes/cm.
A consideration of the data for oleic acid (with μ = 32.5 cP) would suggest that
increases in liquid viscosity are associated with large decreases in void fraction.
However, the operation curve for isopropyl alcohol (μ = 1.77 cP) is close to that for
oleic acid. Furthermore, a comparison of the measurements for methyl alcohol and
toluene shows a large difference even though the viscosity and surface tension are close.
These experiments seem to be consistent with the findings of Anderson & Quinn
(1970) that interfacial impurities are playing an important role in determining the
performance of bubble columns.
194 Bubble dynamics
8.11.3 Analysis of a bubble column

The rise velocity of a bubble, U S∞ , in stagnant water is discussed in Section 8.5.2, where
rP is the radius the bubble would have if it were spherical. As discussed in Section 2.2.2,
the rise velocity, U S ¼ uG uL , in a swarm of bubbles is inhibited so that U S is less than
U S∞ . Equation (2.20), developed by Zuber & Hench, gives
US
¼ ð1 αÞ1=2 ð8:67Þ
U S∞
The analysis in Section 2.7.2 can be applied to a bubble column, for which uL = 0 and
the gas velocity is low. Figure 8.7 presents a photograph of a column operating with an
air velocity of 0.97 cm/s. The liquid was pentadecane. At large gas velocities, this
“particulate” pattern breaks down. Thus, Figure 8.8, for a gas velocity of 7.7 cm/s,
represents a situation that is in the transition region between bubbly and slug flow.
From conservation of mass, the following equation can be used to define the relation
between the void fraction and the superficial velocity for a bubble column operating in
the particulate regime
uGS A ¼ U S αA ð8:68Þ
Substituting (8.67) into (8.68) gives
AuGS ¼ U S∞ ð1 αÞ1=2 αA ð8:69Þ
For very low volume fractions

uGS ¼ U S∞ α ð8:70Þ
Thus, the data in Figure 8.9 can be fitted with a linear relation between α and uGS at low
uGS with a slope equal to ðU S∞ Þ1 . The data show that α becomes less sensitive to
changes in uGS at larger uGS, as predicted by (8.69).
According to (8.69), the differences amongst the curves in Figure 8.9 indicate differ-
ences in U S∞ , the rise velocities of single bubbles in the liquid under consideration. (See
the discussion of measurements of rise velocities of single bubbles in Section 8.5.2.)
8.12 Bubbly gas–liquid flow in vertical pipes, distribution of bubbles
Figure 2.9 maps the bubbly regime for upward flow of air and water in a pipe. The results
for bubble columns would be applicable in the limit of a zero value of the superficial
liquid velocity. At finite liquid velocities, and small gas velocities, the bubble pattern is
similar to that shown in Figure 8.7. Section 2.7.2 considers the hypothetical case of a
uniform velocity field. Equation (2.18) gives a prediction of the dependency of the void
fraction on the superficial gas and liquid velocities, where U S is given by (2.20).
However, the liquid velocity profile is not uniform. This gives rise to a lift force in the
r-direction, FLr, equation (10.29). For upward flows, a drift velocity of bubbles toward
the wall, VDL , results from this force. Measurements of void fractions, for upward bubbly
8.13 Performance of downcomers in tray columns 195
flow of air and water in a 1.75 cm pipe, have been reported by Serizawa et al. (1975a, b).
Under fully developed conditions, these show a relatively uniform distribution of void
fraction over most of the pipe cross-section.
However, large maxima are observed at the wall. Turbulent mixing balances VDL. A
force balance in the radial direction gives an expression for VDL:
4 3 1
πrP F Lr ¼ C D πr2P ρf ður vPr Þj!
u !
v Pj ð8:71Þ
3 2
where !v P is the bubble velocity. The average liquid velocity in the r-direction, ur , is zero
under fully developed conditions, so (8.71) provides an equation for the drift velocity,
where V DL ¼ vPr . From (10.29)
d uz
F Lr ¼ C L ðuGz uLz ÞρL ð8:72Þ
dr
where CL is the lift coefficient. Substitute (8.72) into (8.71) and take an average. The
substitution of zero for ur provides a prediction of V DL ¼ vPr .
One approach to predicting the distribution of voids is to balance VDL with a diffusion
velocity
dαP
V DL ¼ εEP ð8:73Þ
dr
Measurements of the profiles of mean velocity in the liquid phase show an approximate
agreement with what would be observed for turbulent flow of a single phase. Similarly,
one could assume that εEP is proportional to the eddy viscosity (Serizawa et al., 1975a, b).
Integration of (8.73) provides an equation for the profile of αP.
The velocity difference, uGz uLz , changes sign in downflows so that the radial lift
force is toward the center of a pipe. The consequence of this is discussed in Section 2.9.
The procedure outlined above could be used to calculate the spatial distribution of
bubbles. However, possible coalescence could be an issue.
8.13 Performance of downcomers in tray columns
An account of the behavior of trays in a distillation column is given in Lockett (1986). A

typical stage in a sieve plate contactor consists of a perforated tray supporting a liquid. A
gas or vapor passing through the tray forms bubbles at the perforations. Mass transfer
occurs between these bubbles and the liquid on the tray. Liquid is transferred to the tray
from the tray above by a downcomer.
The level of the liquid on the tray is controlled by a weir which is part of a second
downcomer that transfers liquid to the tray below it. For low flows, the liquid moves over
the weir and rolls down the wall on to the liquid in the downcomer. As the liquid flow
increases, it shoots over the weir and strikes the liquid with a relatively large velocity. This
creates bubbles by a mechanism similar to that observed for plunging jets (Lin & Donnelly
1966; Burgess et al., 1972; Van De Sande & Smith, 1973; Koga, 1982; McKeogh &
196 Bubble dynamics
Ervine, 1981). The mixture in the downcomer then starts to resemble a bubble column.
Furthermore, the impingement of the jet can cause a large-scale circulation which carries
bubbles to the bottom of the downcomer and out on the lower tray. This contributes to a
decrease in the efficiency of mass transfer processes. For large enough liquid flows a
condition is reached for which the downcomer is completely filled with froth and foam.
This could lead to flooding. A knowledge of the behavior of the gas–liquid mixture in the
downcomer is thus central to understanding the behavior of the tray column.
This prompted the development of a two-dimensional transparent column to display
the hydraulics of a downcomer (deZutel & Mahoney, 1992; Lee, 1993; Remus, 1994;
Mclean, 1996). An example from these works is the effort of McLean to reduce bubble
creation in the downcomer. The column that he used had three trays. It was 46 in high,
24 in wide and 2 in deep. Air was admitted at the bottom and liquid at the top. The
downcomer on the middle tray was modified so that the impingement of a strong jet on
the surface of the liquid was avoided. Thus, aeration and large-scale circulation were
eliminated. A screen (50 wires per inch) at a 45º angle was located in the downcomer.
This prevented the jet formed by liquid flowing over the weir from striking the
surface of the liquid. Thus, a clear liquid formed under the screen. Figure 8.10
Figure 8.10 Prevention of flooding in a sieve tray column. Insertion of a screen in the downcomer prevents the
creation of bubbles/foam.
8.14 Sedimentation and fluidization 197
shows a picture of the flow when this screen was inserted. The conditions were
such that the downcomer would be completely filled with froth and foam if the insert
was not used.
8.14 Sedimentation and fluidization
8.14.1 Focus
Sections 8.2 and 8.3 focus on single solid particles in a stagnant fluid. However, most
applications involve a swarm of particles or bubbles for which gravity causes a relative
velocity which is smaller than would be realized for single particles. This hindered settling
increases with increasing volume fraction. Important parameters are the volume fraction,
αP, of the particles and the Reynolds number characterizing the flow, ReP ¼ d P U S ρF =μF ,
where US is the average slip velocity between the particles and the fluid, dP is the particle
diameter, ρF is the density of the continuous phase and μF is its viscosity.
Two systems are considered in this section: sedimentation and fluidization.
Sedimentation involves particles settling through a stagnant fluid. Fluidization involves
the flow of fluid through a bed of particles at a large enough velocity for the particles to be
suspended in the fluid. Hanratty & Bandukwala (1957) provide a discussion of these
processes.
8.14.2 Sedimentation
The studies of sedimentation and fluidization were carried out by Hanratty &
Bandukwala in a glass pipe that had a screen on which a bed of uniformly sized spherical
particles was supported. These particles were fluidized by a liquid flow. The height of the
bed, and therefore the volume fraction, increases with increasing flow rate after the
critical velocity needed to suspend the particles is reached.
The flow was turned off and the height of the bed decreased. Particles settled out on to
the support screen. The settling velocity, UC, was determined by measuring the rate at
which the top of the bed moves downward. The volume fraction remains constant during
this process. It can be changed by increasing the fluid velocity and, therefore, the height
of the original expanded bed.
For single spherical particles, the settling velocity is given by equation (8.4). For small
ReP, the drag coefficient, CD, is given by (8.23) and the settling velocity, by (8.24).
Hawksley (1950) adapted equation (8.24), for a single particle, to describe a settling
suspension by considering that the medium through which a particle is settling has
different properties from the fluid. Its density is that of the suspension
ρH ¼ αP ρS þ ð1 αP ÞρF ð8:74Þ
where αP is the volume fraction of the particles. View the bed from a frame of reference
moving with the settling velocity of the bed, UC. Because the presence of the particles
constricts the area, the velocity of the fluid in the bed is
198 Bubble dynamics
UC
US ¼ ð8:75Þ
ð1 α P Þ
Hawksley represented the suspension viscosity by using an equation developed by Vand

(1948), for the viscosity of a concentrated suspension of spherical particles.
!
2:5αP
μC ¼ μF exp ð8:76Þ
1 39
64 αP
The following equation was derived for the sedimentation velocity at a low Reynolds
number by substituting (8.74) for ρF, (8.76) for μF and (8.75) for US in (8.24):
!
UC 2:5αP
¼ ð1 αP Þ2 exp ð8:77Þ
US 1 39
64 αP
1.0
Equation Iron spheres

0.9 (8.77) ReP = 0.001 – 0.08
Fluidization
0.8 Sedimentation } ReP = 0.001 – 0.005
Fluidization
Sedimentation } ReP = 0.03 – 0.07
0.7
Fluidization
0.6 Fluidization
uC
0.5
uS
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Fraction solids
Figure 8.11 Fluidization and sedimentation data for iron spheres in glycerine–water solutions (ReP = 0.001–
0.08). Hanratty & Bandukwala, 1957.
8.14 Sedimentation and fluidization 199
Figure 8.11 presents a plot of the measured UC versus αP, where UC is the settling
velocity and US is given by (8.24). As seen in Figure 8.11, equation (8.77) provides an
excellent description of measurements at ReP< 0.07.
8.14.3 Fluidization
In studies of fluidization, the fluid is passed through a bed of particles resting on a porous
support. At a large enough flow rate, the particles are suspended in the fluid. Increases in the
flow cause the bed to expand, so that the volume fraction of the solids decreases with
increasing fluid velocity. This relationship was investigated by Hanratty & Bandukwala
(1957) in the same equipment in which sedimentation was studied. As seen in Figure 8.11,
the Vand–Hawksley equation also describes the fluidization data at low ReP. At high
particle Reynolds numbers one should use (8.4) rather than (8.24) to calculate US. The
relations presented in the literature for CD for a single sphere were used by Hanratty &
Bandukwala (1957). The density and viscosity of the fluid were used to calculate ReP
appearing in the relation for CD.
Figure 8.12 shows measurements for ReP > 0.07. Here, UC is the settling velocity of a
suspension or the fluidizing velocity and US is given by (8.4). The data for individual runs
are not described so well as indicated in Figure 8.11 for low ReP. However, rough
1.0
- ReP = 0.13 – 0.21 - Steel shot
Equation - ReP = 0.48 – 0.86 - Glass beads
0.9 (8.77) - ReP = 0.55 – 1.20 - Steel shot
- ReP = 0.48 – 1.01 - Glass beads
0.8 - ReP = 1.12 – 1.94 - Steel shot
- ReP = 0.9 – 0.2 - Glass beads
0.7
0.6 - ReP = 45 – 58 - Glass beads
uC ReP = 0.07 – 58.5

0.5
uS
0.4
0.3
0.2
0.1 Open points-sedimentation

Filled symbols-fluidization
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Fraction solids
Figure 8.12 Fluidization and sedimentation data at high Reynolds numbers. (ReP = 0.07‒58.5). Hanratty &
Bandukwala, 1957.
200 Bubble dynamics
agreement between fluidization and sedimentation measurements is noted. Both

Figure 8.11 and Figure 8.12 show the hindered settling phenomenon whereby the
presence of surrounding particles causes the settling velocity or the fluidization velocity
to decrease relative to US.
Some comments about the appearance of fluidized beds is appropriate. At particle
Reynolds numbers below 0.8, the bed had a uniform appearance, with no large move-
ments of the particles. The particles, however, were continually coming together in small
groupings and then dispersing. At particle Reynolds numbers greater than 2, coherent
movements of large groups of particles were observed. For the highest Reynolds
numbers studied by Hanratty & Bandukwala (for both iron shot and glass spheres
beads) the particle flow pattern consisted of a random eddying motion. There were
large variations in the solids concentration, and the motion of the solids was quite similar
to that observed in “aggregative” gas–solid systems (bubbly fluidized beds).
References
Allen, G.T. & Gillogly, D.H. 1979 The effect of viscosity on bubble coalescence. Chem. Eng.
Project Report, University of Illinois.
Anderson, J.L. & Quinn, J.A. 1970 Bubble columns: transitions in the presence of trace contam-
inants. Chem. Eng. Sci. 25, 373–380.
Batchelor, G.K. 1967 An Introduction to Fluid Dynamics. Cambridge: Cambridge University Press.
Benjamin, T.B. 1968 Gravity currents and related phenomena. J. Fluid Mech. 31, 209–248.
Bird, R.B., Stewart, W.E. & Lightfoot, E.N. 1960, Transport Phenomena. New York: Wiley.
Burgess, J.M., Molloy, N.A. & McCarthy, M.J. 1972 A note on the plunging liquid jet reactor.
Chem. Eng. Sci 12, 442.
de Cachard, F. & Delhaye, J.M. 1996 A slug-churn model for small-diameter airlift pumps. Int. J.
Clark, N.N. & Dabolt, R.J. 1986 A general design equation for air lift pumps operating in slug flow.
AIChE Jl 32, 56–64.
Crowe, C.C., Sommerfeld, M. & Tsuji, Y. 1998 Multiphase Flows With Droplets and Particles.
Boca Raton, FL: CRC Press.
Davies, R.M. & Taylor, G.I. 1950 Proc. R. Soc. Lond. A200, 375–390.
Garich, R. & Besserman, M. 1986 Design of an air lift pump. Chem. Eng. Project Report,
Grace, J.R. & Weber, M.E. 1982 Hydrodynamics of drops & bubbles, 1.204–1.246. In Handbook
of Multiphase Systems, ed. G. Hetsroni. New York: McGraw-Hill.
Haberman, W.L. & Morton, R.K. 1953 An experimental investigation of the drag and shape of air
bubbles arising in various liquids. David W. Taylor Model Basin Report 802.
Hadamard, J.S. 1911 Mouvement permanent lent d’une sphere liquide et visqueux. C. R. Acad. Sci.
Paris 152, 1735–1738.
Hanratty, T.J. & Bandukwala, A. 1957 Fluidization and sedimentation of spherical particles,
AIChE Jl 3, 293–296.
Harmathy, T.Z. 1960 Velocity of large drops and bubbles in media of infinite and restricted extent.
AIChE Jl 6, 281–288.
Hawksley, P.G.W. 1950 Some Aspects of Fluid Flow. New York: Arnold Press.
References 201
Hetsroni, G. 1982 Handbook of Multiphase Systems. New York: McGraw-Hill.

Kaji, R., Azzopardi, B.J. & Lucas, D. 2009 Investigation of flow development of co-current
gas–liquid slug flow. Int. J. Multiphase Flow 35, 335–348.
Koga, M. 1982 Bubble entrainment in breaking wind waves. Tellus 34, 481.
Lapple, C.E. & Shepherd, C.B. 1940 Calculation of bubble trajectories. Ind. Eng. Chem. 32,
605–617.
Lee, D.Y. 1993 Flow regimes in a sieve tray tower. B.Sc. thesis, Chemical Engineering, University
of Illinois.
Levich, V.G. 1962 Physicochemical Hydrodynamics. Englewood Cliffs, NJ: Prentice-Hall.
Lin, T.J. & Donnelly, H.G. 1966 Gas bubble entrainment by plunging laminar liquid jets, AIChE Jl
36, 1161.
Lockett, M.J. 1986 Distillation Tray Fundamentals. Cambridge: Cambridge University Press.
McKeogh, E.J. & Ervine, D.A. 1981 Air entrainment rate and diffusion Pattern of plunging liquid
jets. Chem. Eng. Sci. 36, 1161.
McLean, R. 1996 Downcomer hydraulics in a sieve tray column. B.Sc. thesis, University of Illiois.
Mueller, M.K. & Wilger, D.M. 1979 Operation of a bubble column. Chem. Eng. Project Report,
Nicklin, D.J., Wilkes, J.O. & Davidson, J.F. 1962 Two-phase flow in vertical tubes. Trans. IChem
E 40, 61.
Norman, C.D. & Hummel, D. 1980 Operation of a bubble column. Chem. Eng. Project Report,
Remus, M. 1994 Fluid mechanics of a sieve tray distillation column. B.Sc. thesis, Chemical
Schlichting, H. 1955 Boundary Layer Theory. New York: McGraw-Hill.
Serizawa, A., Kataoka, I. & Mishiyoshi, I. 1975a Turbulence structure of air–water bubbly flow.
Serizawa, A., Kataoka, I. & Mishioshi, I. 1975b Turbulence structure of air–water bubbly flow.
Soo, S.L. 1967 Fluid Dynamics of Multiphase Systems. Waltham, MA: Blaisdell.
Streeter, V.L. 1948 Fluid Dynamics. New York: McGraw-Hill.
Vand, V.J. 1948 J. Phys. Colloid Chem. 52, 271.
Van De Sande, E. & Smith, J.M. 1973 Surface entrainment of air by high velocity water jets. Chem.
Eng. Sci. 28, 1161.
Wallis, G.B. 1969 One Dimensional Two-Phase Flow. New York: McGraw-Hill.
Zuber, N. & Findlay, J.A. 1965 Average volumetric concentration in two-phase flow systems. J.
Heat Transfer, Trans. ASME Series C, 87, 453–468.
de Zutel, C. & Mahoney, A.W. 1992 Hydraulics of a sieve tray tower. Chemical Engineering
Projects Laboratory, University of Illinois.
9 Horizontal slug flow
9.1 Prologue
Chapter 2 gives considerable attention to slug flow because of its central role in under-
standing the configuration of the phases in horizontal and inclined pipes. Several criteria
have been identified to define the boundaries of this regime: (1) viscous large-wavelength
instability of a stratified flow; (2) Kelvin–Helmholtz instability of a stratified flow;
(3) stability of a slug; (4) coalescence of large-amplitude waves. Bontozoglou &
Hanratty (1990) suggested that a sub-critical non-linear Kelvin-Helmholtz instability
could be an effective mechanism in pipes with very large diameters, but this analysis
has not been tested. A consideration of the stability of a slug emerges as being
particularly important. It explains the initiation of slugs for very viscous liquids, for
high-density gases, for gas velocities where wave coalescence is important and for the
evolution of pseudo-slugs into slugs. Chapter 2 (Section 2.2.5) outlines an analysis of
slug stability which points out the importance of understanding the rate at which slugs
shed liquid. Section 9.2 continues this discussion by developing a relation for Qsh and
for the critical height of the liquid layer needed to support a stable slug. Section 9.3
develops a tentative model for horizontal slug flow. Section 9.4 considers the frequency
of slugging.
9.2 Slug stability
9.2.1 Necessary conditions for the existence of slugs

Figure 9.1 presents simplified sketches of the front and the tail of a slug in a pipeline. The
front has a velocity cF; the back has a velocity cB. The stratified liquid layer in front of
the slug has a velocity and area designated by uL1, AL1. The mean velocity of the liquid in
the slug is uL3. The slug is usually aerated; the mean volume fraction of gas in the slug is
designated by α. The gas at station 1 is moving from left to right at a velocity uG1. The
assumption is made that the velocity fields can be approximated as being uniform.
A first necessary condition for the existence of slugs, proposed by Ruder et al. (1989),
is that the front of the slug may be considered to be a hydraulic jump, which is an
irreversible flow with a large amount of dissipation. This model is more easily formulated
for a rectangular channel. Consider a control volume surrounding the front of the slug.
3 2
p = p1
B cF – uL3
1
cF – uL1 hL1
(a) Front end of a slug seen by an observer moving with a

velocity cF
5 4
7
hL6
(b) Steady-state inviscid tail of a slug

Figure 9.1 Simplified sketches of the front and tail of a slug in a pipeline.
The sides are the walls of the channel. Station 1 and station 3 are sufficient distances in
front of the slug and behind the slug for uniform flows to be considered. The influence of
gas entrainment in the slug is ignored. Mass and momentum balances, written in a
framework moving with velocity cF, give
ðcF uL1 ÞhL1 ¼ ðcF uL3 ÞB ð9:1Þ
ρL gh2L1 ρL gB
2
ρL ðcF uL3 Þ2 B ρL ðcF uL1 Þ2 hL1 ¼ ðp1 p3 ÞB þ ð9:2Þ
2 2
where ρL is the liquid density, g is the acceleration of gravity and p3 is the pressure at the
top of the conduit at station 3. The last two terms in (9.2) are the hydrostatic forces on
the front and the back of the jump; hL1/2 and B/2 are the centroids of these areas.
Equation (9.1) can be used to eliminate (cF – uL3) from (9.2). Since p2 = p1
$ 2 %
ðp3 p2 Þ ðcF uL1 Þ2 hL1 hL1 1 1 hL1 2
¼ þ ð9:3Þ
ρL gB gB B B 2 2 B
If the flow changes reversibly from 1 to 3, the Bernoulli equation is applicable, so

$ 2 %
ðp3 p2 Þrev 1 ðcF uL1 Þ2 hL1 hL1 hL1
¼ 1 þ 1 ð9:4Þ
ρL gB 2 gB B B B
By subtracting (9.3) from (9.4) one obtains

$ %
ðp1 p3 Þrev ðp3 p2 Þ ðcF uL1 Þ2 1 hL1 2 1 hL1 2
¼ 1 1 ð9:5Þ
ρL gB ρL gB gB 2 B 2 B
The left hand side of (9.5) must be positive since it represents the mechanical energy that
is dissipated. From (9.5), it is seen that this requires
ðcF uL1 Þ2
>1 ð9:6Þ
gB
Thus, for an unaerated slug to exist in a rectangular channel it is necessary that (9.6) is
satisfied (see equation (2.1)).
Ruder et al. (1989) also considered the front of a slug in a circular pipe and derived a
similar necessary condition
ðcF uL1 Þ
>N ð9:7Þ
ðgd t Þ1=2
where N varies with hL1/dt as indicated in Table 1 of their paper. It takes on values of
0.9459 to 1.2106 (see equation (2.2)).
The second necessary condition for a circular pipe (previously discussed in Chapter 2
depends on the behavior of the tail of the slug. An unaerated slug picks up liquid from the
stratified flow in front of it at a rate given by
Rate of pickup ¼ ðcF uL1 ÞðAL1 Þ ð9:8Þ
Liquid is shed out of the back of a slug at a rate Qsh. For a growing slug, it is necessary
that
ðcF uL1 ÞðAL1 Þ > Qsh ð9:9Þ
For a neutrally stable slug, the stratified layer over which the slug is moving has an area,
AL0, given by
ðcF uL0 ÞðAL0 Þ ¼ Qsh ð9:10Þ
Thus, a slug cannot develop on a stratified layer with an area less than AL0. The
implementation of this necessary condition requires an understanding of Qsh, that is, a
physical model for the tail of a slug. An appealing approach would be to use a pseudo-
steady-state solution, such as has been done for the tail of a roll wave (Sections 7.5, 7.6),
for which viscous drag at the wall is an important consideration. However, photographs
of the tails of slugs (Ruder et al., 1989) do not seem to support such an assumption. At
low throughputs, the height of the liquid behind the slug (at least, in the portion close to
the top wall) is observed to decrease rapidly enough for inertia effects to dominate effects
of viscous and turbulent stresses. One might, therefore, expect that, to first order, the flow
may be considered inviscid.
This led Ruder et al. (1989) to suggest that the back of an unaerated slug can
be approximated as a Benjamin bubble (see Section 8.9). Thus, for a rectangular
channel
hL6
¼ 0:5 ð9:11Þ
B
ðcB uL3 Þ ¼ ðgBÞ1=2 ð9:12Þ
1=2
Qsh ¼ 0:5BðgBÞ ð9:13Þ
For a circular pipe

hL6
¼ 0:563 ð9:14Þ
dt
ðcB uL3 Þ ¼ 0:542ðgd t Þ1=2 ð9:15Þ
2
πd t
Qsh ¼ 0:542 ðgd t Þ1=2 ð9:16Þ
4
Far enough behind a slug the liquid must return to the height of the liquid in front of the
slug. This suggests that the Benjamin bubble, if it exists, would need to be matched to a
region which is not inviscid.
If (9.16) is substituted into (9.10), the following equation is obtained:
2
πd t
ðcF uL0 ÞðAL0 Þ ¼ 0:542 ðgd t Þ1=2 ð9:17Þ
4
Height hL0 can be calculated from AL0 using the geometric relations given by (5.7) and
(5.11). AL0 or hL0 provides a critical condition for a slug to be stable.
Figure 9.2, from Ruder et al. (1989), shows height tracings, at different uGS for a
constant uLS = 0.95 m/s, downstream in a pipe. These were obtained by measuring the
conductance between two wires which were strung across the pipe diameter in the
vertical direction. Note that hL/dt does not assume a value of unity when a slug is
present. This type of measurement can be used to determine the void fraction in the
slugs. The solid line, in Figure 9.2, is the height needed for the formation of stable
slugs, hL0/dt. Thus, the necessary condition provides a theoretical prediction of the
base layer over which fully developed slugs move. The dashed line represents the hL/dt
predicted for the initiation of intermittent flow by a viscous large-wavelength hydro-
dynamic instability (Lin & Hanratty, 1986). The slugs were moving too fast for
the liquid height measurements to capture the detail shown in photographs. Thus,
they could not be used to see whether (9.14) approximates the liquid height just
behind a slug.
Figure 9.3 shows a photograph of a slug in a 2.54 cm pipe. Note that bubbles are
occluded. These are caught in the tumbling action at the front of the slug; some of
these escape and move out of the back of the slug. This photograph is at a low gas
velocity, for which aeration is not severe. The bubble at the tail resembles a Benjamin
bubble.
uSG = 1.0 m/s: uSL = 0.95 m/s

0.8
hL/dt 0.7
0.6
0.5
0.4
0.3
uSG = 1.90 m/s: uSL = 0.95 m/s

0.8
0.7
0.6
hL/dt
0.5
0.4
0.3
0.2
uSG = 3 m/s: uSL = 0.95 m/s

0.8
0.7
0.6
hL/dt
0.5
0.4
0.3
0.2
uSG = 5 m/s: uSL = 0.95 m/s

0.7
0.6
0.5
hL/dt
0.4
0.3
0.2
0.1
0.0
0 1000 2000 3000 4000 5000
Time (s) × 50
Figure 9.2 Liquid height tracings; uLS = 0.95 m/s. Ruder et al., 1989.
Figure 9.3 Photograph of a slug in a 2.54 cm pipe at a low gas velocity. Ruder et al., 1989.
9.2.2 Transition to plug flow

As pointed out in Chapter 2, researchers have considered an intermittent regime that has
two parts: slug flow and plug flow. The latter identifies a series of elongated bubbles
moving along the top of the pipe.
(a) Typical back of a gas cavity in the slug flow

(uGS = 1.45 m/s; uLS = 0.95 m/s, Fr = 2.53)
(b) Typical two-stage, staircase-like back of a gas cavity in the plug flow
(uGS = 0.6 m/s; uLS = 0.95 m/s, Fr = 1.8)
(c) The back of a symmetrical gas cavity in a plug flow (“a Benjamin bubble”)
(uGS = 0.1 m/s; uLS = 0.95 m/s, Fr = 1.2)
Figure 9.4 Photographs of the front of a slug, or the back of an air pocket. Ruder & Hanratty, 1990.
Figure 9.4 (from Ruder & Hanratty, 1990) shows changes in the front of a slug (the back
of the air cavity between slugs) for air–water flow in a horizontal 9.53 cm pipeline. The
superficial liquid velocity was 0.95 m/s. The superficial air velocity decreased from 1.45 m/s
to 0.1 m/s. A Froude number can be defined as Fr ¼ ðcF uL1 Þ=ðgd t Þ0:5 where uL1 is the
velocity of the liquid carpet in front of the slug. The analysis presented by Ruder et al.
(1989) suggests that Fr > ~ 1 is required for a slug to exist (see equations (9.6) and (9.7)).
Figure 9.4a is for Fr = 2.53. It shows a highly aerated hydraulic jump. Figure 9.4b, for
Fr = 1.8, shows the breakdown of the hydraulic jump into a staircase shape. Figure 9.4c,
for Fr = 1.2, shows an evolution to a Benjamin bubble. For all of the situations shown
in these images, the backs of the slugs/plugs are described as a Benjamin bubble. Thus,
for the conditions characterizing Figure 9.4c, a symmetric elongated bubble exists
between the slugs. These results are consistent with the observation by Barnea et al.
(1980) that the plug pattern is a limiting case of slug flow when the liquid is free of
entrained gas bubbles. Hurlburt & Hanratty (2002) used the above results to provide a
method to predict the transition from a slug flow to a plug flow. They selected Fr = 1.2 as a
criterion.
9.2.3 Measurements of the shedding rate

Measurements of the shedding rate under actual flow conditions were obtained by Woods
& Hanratty (1996) for air and water flowing in a horizontal pipe at atmospheric conditions.
Conductance profiles were used to determine the liquid holdup at several locations along
the pipeline. These conductance measurements established the profiles of the liquid carpet
and the tail of the slug, and the degree of aeration in the slug. A translating control volume
was attached to the back of the slug (station 5 in Figure 9.1). This volume extends out into
the stratified flow, where the area of the liquid is AL1. By using the analysis for a stratified
flow presented in Chapter 5, uL1 can be calculated from the measurements of hL1.
Conservation of mass is applied to the control volume moving at velocity cB to give the
following equation for Qsh:
dV
Qsh ¼ ðcB uL1 ÞðAL1 Þ ð9:18Þ
dt
where Qsh is the volumetric flow into the volume, the first term on the right side is the net
flow out and V is the volume of liquid in the control volume. The accumulation term,
dV/dt, was included in the conservation equation in order to take into account changes in
the liquid volume that occur in the control volume as it translates downstream. In this
way, Qsh was calculated from the measurements for stable, growing and decaying slugs.
9.2.4 Equations for the bubble velocity and the shedding rate
Apply conservation of mass between the back of the slug, station 5, and station 3 in the
body of the slug. Choose a frame of reference moving with velocity cB. Then, from this
mass balance,
Qsh ¼ ðcB uL3 Þð1 αÞA ð9:19Þ
where Qsh is the flow out of the volume. For small mixture velocities, the back of the slug
may be considered to be a Benjamin bubble and α may be assumed equal to zero. Thus
(cB − uL3) and Qsh are given by (9.15) and (9.16).
For large mixture velocities, the nose of the bubble behind the slug moves downward
in the pipe cross-section and the liquid in the slug is highly aerated. More general
relations are needed for bubble velocity and Qsh.
Common practice for vertical pipes is to relate the velocity of large bubbles to the
superficial gas and liquid velocities, uSG and uLS. Nicklin et al. (1962) assumed that the
bubble velocity for vertical slug flow is the sum of the centerline velocity in the liquid in
front of the bubble and the drift velocity of the bubble in a stagnant liquid.
cB ¼ K 0 uL3 þ V Drift ð9:20Þ
where K0 is a coefficient and VDrift is the drift velocity, defined by (8.29). Nicklin et al.
suggest that, for a turbulent liquid, K0 is approximately 1.2, that is, the ratio of the
centerline liquid velocity to the mean liquid velocity, uL3, in front of the bubble.
Assuming incompressible flow, a volume balance between the inlet of the pipe and
station 3 within the body of the slug gives
uGS þ uLS ¼ αuG3 þ ð1 αÞuL3 ð9:21Þ
Equation (9.21) is solved for uL3 and this is introduced into (9.20), using SuL3 = uG3. The
following equation for cB is obtained:
cB ¼ C 0 ðuGS þ uLS Þ þ V drift ð9:22Þ
where
(a) 16
14
12
10
cB (m/s)
8 uLS = 0.7 m/s

uLS = 0.9 m/s
6
uLS = 1.1 m/s
4 uLS = 1.2 m/s
uLS = 1.25 m/s
2
0
0 2 4 6 8 10 12 14 16
uGS (m/s)
(b) 16
14
12
10 cB = 1.10 umix
cB (m/s)
+ 0.52
8
6
4 cB = 1.20 umix
2 Benjamin bubble solution
0
0 2 4 6 8 10 12 14 16
uGS + uLS (m/s)
Figure 9.5 Measurements of the bubble velocity in the back of a slug. Woods & Hanratty, 1996.
K0
C0 ¼ ð9:23Þ
1 þ ðS 1Þα
and S is the slip ratio between the liquid and gas phases within the slug. For horizontal
pffiffiffiffiffiffiffi
flow, V Drif t ¼ 0:542 gd t (see equation (9.15) with uL3 = 0). Many authors neglect the
drift velocity,VDrift. Therefore, expressions such as cB = 1.35 (uLS + uGS) (Gregory & Scott,
1969) and cB = 1.25−1.28 (uLS + uGS) (Dukler & Hubbard, 1975) are commonly used.
A plot of measurements of cB by Woods & Hanratty (1996) for air and water flowing
in a horizontal 9.53 cm pipe is given in Figure 9.5 in units of m/s. The data at low gas
velocities are represented by
cB ¼ 1:10ðuGS þ uLS Þ þ 0:52 ð9:24Þ
where 0.52 m/s is the velocity of a Benjamin bubble. For larger mixture velocities the
influence of the drift velocity is small so that
cB ¼ 1:20ðuLS þ uGS Þ ð9:25Þ
Woods & Hanratty show that measurements in a 15 cm pipe by Kouba & Jepson (1990), in
a 5 cm pipe by Theron (1989) and in a 5 cm pipe by Nydal et al. (1992) agree approx-
imately with this equation.
2.5
Plug flow region
2
uLS = 0.7 m/s
Qsh/A (m/s)
1.5 uLS = 0.9 m/s

uLS = 1.2 m/s
uLS = 1.25 m/s
1 uLS = 0.68 m/s
from Ruder &
Hanratty (1990)
0.5 uLS = 0.95 m/s
from Ruder &
Benjamin bubble solution Hanratty (1990)
0
0 2 4 6 8 10 12 14 16
uGS + uLS (m/s)
Figure 9.6 Measurements of the mean value of the shedding rate. Woods & Hanratty, 1996.
0.8
0.6
(1 − α)
0.4
Experimental measurements by
Woods & Hanratty (1996)
0.2 Correlation from Gregory et al. (1978)
Semi-empirical equation from
Andreussi & Bendiksen (1989)
0
0 2 4 6 8 10 12 14 16
uGS + uLS (m/s)
Figure 9.7 Liquid holdup measurements in a slug. Woods & Hanratty, 1996.
Measurements of Qsh (Woods & Hanratty, 1996) are presented in Figure 9.6. At small
(uGS + uLS), QshA is given by the inviscid solution of Benjamin. The shedding rate
increases rapidly with increasing uGS + uLS and, at umix = 14.2 m/s, it is approximately
four times that given by the Benjamin solution. The use of (9.16) to represent Qsh is valid
only for very small values of (uLS + uGS).
Equation (9.19) provides a relation for Qsh. From (9.19), (9.21) and (9.22)

Qsh uGS þ uLS
¼ cB ð1 α Þ ð9:26Þ
A 1 þ ðS 1Þα
where SuL3 has been substituted for uG3.

Figure 9.7 presents measurements of the volume fraction of gas in a slug. Values
of S were calculated with (9.26) using measurements of Qsh (Figure 9.6) and of α
(Figure 9.7). A slip ratio equal to 1 is a rough approximation at low gas velocities.
At high mixture velocities ðuGS þ uLS Þ > 7 m=s, a slip ratio of 1.5 is indicated (see
Figure 9.8).
1.5
S = uG3/uL3
1
uLS = 0.7 m/s
uLS = 0.9 m/s
0.5 uLS = 1.2 m/s
uLS = 1.25 m/s
0
0 2 4 6 8 10 12 14
uGS + uLS (m/s)
Figure 9.8 Measurements of the velocity ratio in slugs. Woods & Hanratty, 1996.
Bendiksen (1984) suggested that Froude number

ðuGS þ uLS Þ
Frmix ¼ ð9:27Þ
ðgd t Þ1=2
is a measure of the relative roles of inertia and gravity on the behavior of the tail of the
bubble behind a slug. At small Froude numbers, gravitation effects are large enough for
the nose of the bubble to be at the top of the pipe. As the Froude number increases, inertia
becomes more important and, eventually, becomes dominant; the nose of the bubble
moves from the top of the pipe toward the centerline. This asymptotic behavior is shown
in Figures 9.6 and 9.8 at mixture velocities larger than 8 m/s. For mixture velocities
greater than 14 m/s, the slug flow changes to an annular flow.
Of interest is the finding from Figure 9.8 that (S − 1) is positive at large mixture
velocities. This suggests that the gas flow in the slug changes from a situation (at low
mixture velocities) for which gas is entrained in the front and rises to the top of the pipe
because of gravitational effects, to one for which gas enters at the back of the slug. One
could then speculate that a transition to annular flow occurs when the gas blows through
the back of the slug. More attention needs to be given to this suggestion.
9.2.5 Calculation of the stability of a slug

Slug stability is usually discussed by defining the critical height (or area) needed to
sustain slugs. This is defined by (9.10). Thus, (9.26) can be substituted into (9.10) to
produce the following expression:

1
C0 umix þ U S ð1 αÞ
AL0 1 þ ðS 1Þα
¼ ð9:28Þ
A C 0 umix þ U S uL1
For large mixture velocities, US and uL1 can be neglected, so

1
C0
AL0 1 þ ðS 1Þα
¼ ð1 αÞ ð9:29Þ
A critical C0
1
Initiation of slug flow
Initiation of KH waves
0.8
Prediction of transition by
Kelvin–Helmholtz theory
0.6
hL/dt
Slug flow
0.4
0.2
Stratified flow Stratified flow
(Smooth interface) (Large-amplitude waves)
0
0 5 10 15 20 25
uGS (m/s)
Figure 9.9 Initiation of slugs and Kelvin–Helmholtz waves for air and a 100 cP liquid flowing in a 0.0953 m
pipe. Woods & Hanratty, 1996.
From Figures 9.8 and 9.7, S ≈ 1:5 and ð1 αÞ ≈ 0:5 at very large mixture
velocities. From (9.22) and Figure 9.5, C 0 ≈ 1:2umix . Substitute into (9.29) to get
(AL0/A)critical ≈ 0.16. From Figure 5.1 and Figure 10.38 in Govier & Aziz (1972),
the critical height ratio is then given, approximately, as hL/dt = 0.2.
Figure 9.9 presents measurements made by Andritsos et al. (1989) for air and a 100 cP
liquid flowing in a 9.53 cm pipe. (Note that the transition data are plotted as hL/dt versus
uGS, rather than as uSL versus uSG. For given values of hL/dt and uSG, uSL is fixed.) The open
points represent observations of the conditions needed for the initiation of slugs. The solid
points designate the initiation of Kelvin–Helmholtz waves. The curve is the prediction of
instability based on KH theory. The above estimation of a critical hL =d t ≈ 0:2 at very large
superficial gas velocity is close enough to the data to support the notion that the initiation of
slugs is dictated by the stability of slugs (see Fan et al., 1992).
At low gas velocities, the interface is smooth. The initiation of waves is predicted by
KH theory. At uGS 5 5 m=s the dimensionless height is larger than the critical height so
slugs are predicted to be stable; KH waves rapidly evolve into slugs. At uGS > 5 m=s the
KH waves do not immediately evolve into slugs because the slugs would be unstable.
The initiation of slugging is then predicted from a consideration of the conditions needed
for stable slugs to exist.
9.3 Modeling of slug flow in a horizontal pipe
Dukler & Hubbard (1975) developed a model for slug flow that is characterized by the
intermittent appearance of slugs. A slug is accompanied by a sudden increase in both the
liquid level and the pressure. This implies that there would be a discontinuous increase in
pressure over each of the slugs in the pipeline. This section uses this picture to describe a
flow which is approximately fully developed.
9.3 Modeling of slug flow in a horizontal pipe 213
From discussions presented in this chapter, the stratified flow between slugs can
be pictured as having a height, hL0, which is defined from the necessary condition for
the existence of a slug. This portion of the flow could be described by the methods
used in Chapter 5, as mentioned earlier in this chapter. A critical issue is the descrip-
tion of the behavior of a slug and, in particular, the pressure drops over individual
slugs. A second issue is the prediction of the distribution of slug lengths. This is
addressed in Section 9.4.
Figure 9.1 presents an idealized drawing of a slug. The pressure drop is considered as
being composed of three parts: the change associated with the hydraulic jump in front of
the slug, from station 1 to 3, DP ~ h ; a frictional loss in the body of the slug, from station 3 to
4, DP~ f ; a pressure variation associated with the velocity change in the rear of the slug
from station 4 to 5, DP ~ r . Thus, for a horizontal pipe, the total pressure change over a slug
is given as
~ T ¼ DP
DP ~ h þ DP
~ f þ DP
~r: ð9:30Þ
by Fan et al. (1993), where P ~ is the pressure made dimensionless with ρLgdt. This is
similar to the treatment by Singh & Griffith (1970), who used only the first two terms on
the right side of (9.30), but did consider gravitational effects needed to analyze slugging
in inclined pipes.
The front is moving downstream with a velocity cF. The liquid and gas velocities at
locations 1, 3 and 4 are assumed to be uniform. The liquid layer in front of the slug has a
velocity uL1 in a laboratory framework. The flows of the liquid and gas in the slug body
are uniform after a short distance behind the slug front. For simplification, the gas and
liquid are assumed to have the same velocity in the slug (S = 1). The pressure drop over a
hydraulic jump in a rectangular channel is given by equation (9.3). For small ρG/ρL Fan
et al. (1993) give the following relation for a horizontal circular pipe:
" 2 #
~ h ¼ ρL AL1 A L1 1 AL1 C
DP ðc uL1 Þ2 þ ρL g hL1 ð1 α3 ÞhCL3
AL3 AL3 1 α3 AL3
ð9:31Þ
where the terms with a superscript of C are centroid heights for hydrostatic pressure in
the liquid layer and the slug body. They are length scales that give the average hydrostatic
pressure over the area being considered.
The pressure drop in the body of the slug may be pictured as resulting from the
frictional drag at the wall, so that
~f ¼ P
~4 P
~2 ¼ τ W πd t LS
DP ð9:32Þ
A
where τW is the average resisting stress at the wall and LS is the length of the slug. Because
of a lack of better information, the wall stress is approximated by the Blausius equation
0:20
dt ρL uL3 1:8
τ W ¼ 0:046 ð9:33Þ
L 2
1.0 3
hL
∼
P
0.8
2
∼ ∼
(ΔPr)th (ΔPr)exp
hL/dt
0.6
~
P
∼
(ΔPf)th ∼
(ΔPf)exp
1
∼ ∼
0.4 (ΔPh)th (ΔPh)exp
0.0 0
0 1 2 3
Time (seconds)
Figure 9.10 ~ is the pressure made
Pressure distribution for a stable slug at uGS = 1.10 m/s, uLS = 0.6 m/s, where P
dimensionless with ρL gd t . Fan et al., 1993.
where L is the kinematic viscosity of the liquid. Dukler & Hubbard (1975) used
ρL ¼ ρL ð1 αÞ þ ρG α ð9:34Þ
in (9.33). Fan et al. (1993b) simply use the density of the liquid. Their approach is
followed in the calculations presented in this section.
Fan et al. (1993b) made measurements of the pressure variation over individual slugs,
with a piezoresistive transducer, for air and water flowing in a 0.095 m pipe. The time-
varying liquid holdup was determined from measurements of the electric resistance
between two parallel wires strung across the pipe diameter. The results from one of
the experiments, at uGS = 1.10 m/s and uLS = 0.6 m/s, are plotted in Figure 9.10. Both
dimensionless holdup, hL/dt, and the dimensionless pressure P ~ ¼ P=ρL gd t , are plotted.
For hL/dt = 1 the liquid is not aerated so the value of hL/dt provides a measurement of
the void fraction. The experiment was conducted at a low gas velocity, for which the
resolution of the pressure profile is good.
Note that the pressure experiences
a sudden increase when a slug appears. This is
defined in Figure 9.10 by DP~ h . This is followed by a more gradual increase,
exp
~f
DP . Another sudden change ~r
occurs in the back of the slug, DP . Calculated
exp exp
values of DP ~ h and DP ~ f were obtained from (9.31) and (9.32). These show good
th th
agreement with values suggested from the measurements. The calculation of DP ~f
th
requires the specification of the length of the slug. This was obtained as the product of
the slug velocity and the time interval over which the slug was observed.
Fan et al. (1993b) calculated the pressure change in the rear by picturing it as a
Benjamin bubble. This requires that the top of the slug tail is a stagnation point in a
frame of reference moving with the velocity cF and the pressure change along the top
wall can be calculated with the Bernoulli equation. In this framework, the top of the tail,
station 5 in Figure 9.1, is a stagnation point. Furthermore, the pressure at 5 equals the
pressure at 6.
9.3 Modeling of slug flow in a horizontal pipe 215
Table 9.1 Pressure drop over stable slugs

uSG uSL cF ~h
DP ~f
DP ~r
DP ~T
DP ~T
DP
th th th exp
1.109 0.60 2.07 0.29 0.64 0.19 1.12 1.23

1.994 0.50 4.09 1.59 1.53 0.27 3.39 3.23
1.994 0.60 3.73 1.59 1.55 0.38 3.52 4.05
1.994 0.60 3.73 1.73 1.48 0.50 3.71 4.53
1.994 0.70 3.80 1.76 1.53 0.43 3.72 3.87
2.965 0.60 4.35 2.19 2.02 0.64 4.85 5.21
2.965 0.60 4.94 2.43 2.45 0.25 5.13 5.56
2.965 0.90 5.96 3.20 2.39 0.51 6.10 6.43
4.060 0.50 6.33 4.36 2.55 0.81 7.72 8.07
4.060 0.50 6.53 5.22 3.22 1.38 9.81 9.26
4.060 1.00 6.60 5.18 4.20 1.22 10.60 11.21
4.985 0.60 8.17 7.23 3.46 1.29 11.98 11.19
5.976 0.70 7.60 4.30 4.51 0.83 9.64 9.16
7.036 0.60 9.02 8.72 1.52 4.03 14.27 13.90
7.036 0.60 9.74 5.66 6.60 0.85 13.10 15.45
7.036 0.90 10.92 8.60 9.49 1.45 19.53 19.97
9.087 0.60 13.19 10.74 6.09 1.70 18.53 19.01
9.087 0.90 15.88 14.06 8.98 1.90 24.93 23.38
13.18 0.80 17.27 15.09 12.80 0.58 28.47 28.55
13.18 0.90 18.45 18.29 11.97 2.45 32.72 30.67
15.93 0.80 17.27 12.02 17.44 1.22 30.67 34.50
15.93 1.00 18.36 12.16 23.79 0.95 36.90 40.57
~ stands for DP=ρL gd t and velocities are given in units of m/s. Subscript th refers to theoretical
In this table DP
calculation; subscript exp refers to a measurement. Fan et al., 1993
DPr ¼ P6 P4 ¼ P5 P4 ð9:35Þ
Applying the Bernoulli equation between 4 and 5

1
DPr ¼ ρL ðcF uL4 Þ2 ð9:36Þ
2
Equation (9.36) was used to calculate ðDPr Þth in Figure 9.10. At higher gas velocities, the
delineation of the different contributions to the pressure drop becomes more difficult.
Therefore, Table 9.1, which summarizes all of the measurements with stable slugs by Fan
et al. (1993b), compares calculated and measured overall pressure drops, made dimen-
sionless with ρLgdt.
The calculated pressure drops over slugs roughly predict measurements. This agree-
ment is surprising considering that the following simplifying assumptions were made: In
analyzing the hydraulic jump, the bubbles in the slug body were assumed to be uniformly
distributed and to be moving with the liquid velocity. Friction factor relations for single-
phase flow are used to calculate the frictional pressure losses in the body of the slug. The
effect of voidage is taken into account only by considering its role in increasing the liquid
1. Slug forms discontinuity in liquid layer

uG LD
uL hB h0
2. Gravity wave translates upstream

uG
c
uL
3. Gravity wave reaches inlet of pipe

uG
c
uL
4. Liquid is replenished to original height

uG
uL
Figure 9.11 Schematic of a gravity wave translating upstream after the formation of a slug. Woods &
Hanratty, 1999.
velocity. The Benjamin bubble model, which was used to calculate the pressure drop in
the rear of the slug, could be too simple at high gas velocities. Clearly, more work on this
problem is needed.
9.4 Frequency of slugging
9.4.1 Shallow liquid inviscid theory

The formation of a slug on a stratified flow is accompanied by a depletion of the liquid.
Replenishment occurs rapidly enough for inertia to dominate. Therefore, the effects of
viscous shear stresses, turbulent shear stresses, and air flow can be ignored. At the instant
a slug appears, the liquid behind the slug is approximated as a discontinuity (see
Figure 9.11). Two means exist for rebuilding the liquid level: the flow of new liquid
into the pipe and the upstream propagation of a depression (gravity) wave from the
location at which the slug formed. This treatment is the same as the breaking dam
problem discussed by Stoker (1957). The relative importance of these two mechanisms
can be determined by considering the propagation velocity of a large-wavelength
inviscid gravity wave in a stationary shallow liquid (see Section 4.4):
pffiffiffiffiffiffiffiffi
c ¼ ghL ð9:37Þ
In a flowing liquid with a velocity uL, a gravity wave moves upstream with a velocity
c ¼ ghL þ uL ð9:38Þ
1
Slug flow
C
III
B II
uLS (m/s)
I
0.1 A
Wavy stratified flow
0.01
1 10
uGS (m/s)
Figure 9.12 Flow regime for air–water flow in a horizontal 0.0763 m pipe. Curve A indicates the transition
to slug flow. Between curves A and B, slugs form downstream of L=d t ≈ 40. Along curve C, Fr = 1
at the inlet. Woods & Hanratty, 1999.
A consideration of (9.38) shows it is not possible for a depression wave to move upstream
if the Froude number, Fr ¼ uL = ghL ; is greater than unity. In this case, the refilling will
occur by a bore (Stoker, 1957, p. 320) moving downstream.
The inviscid shallow liquid equations can be used to describe the refilling for Fr < 1:
∂hL ∂hL ∂uL
þ uL þ hL ¼0 ð9:39Þ
∂t ∂x ∂x
∂uL ∂uL ∂hL
þ uL þg ¼0 ð9:40Þ
∂t ∂x ∂x
where (9.39) is conservation of mass and (9.40) is conservation of momentum.

Figure 9.12 presents a flow regime map for air and water in a horizontal 7.63 cm pipe
(Woods & Hanratty, 1999). Curve A represents conditions for the initiation of slugs.
Slugs are observed to form far downstream for conditions close to the transition curve. As
uLS increases for a fixed uGS , slugs form closer to the inlet. For conditions below B, slugs
appear at L=d t 5 ~40 so that the formation could be dependent on the design of the inlet.
Curve C describes conditions for which Fr = 1 at the inlet. Thus refilling can occur with
a depression wave at conditions below this curve. Zone I in Figure 9.12 is the region in
which Fr < 1 and slugs form downstream of L=d t 5 ~40. The region of the flow map in
which Fr > 1 and slugs form at L=d t 5 ~40 is defined as zone II. For high uLS slugs form
upstream of L=d t 5~40. This is denoted as zone III.
9.4.2 Regular slugging

Three situations can be cited for which the slugging is regular, that is, periodic. These are
horizontal flows at low gas and low liquid velocities, inclined pipes, declined pipes.
The first of these involves situations for which Fr < 1. It is illustrated in Figure 9.13,
which describes measurements for air–water flow at Fr = 0.36, uL ¼ 25 m=s by Woods &
Hanratty (1999). The phases were combined in a tee section with water in the run and air
(a) L/dt = 12 (b) L/dt = 44

1 1
0.8 0.8
0.6 0.6
hL/dt
hL/dt
0.4 0.4
0.2 0.2
0 0
0 25 50 75 100 0 25 50 75 100
Time (s) Time (s)
(c) L/dt = 98 (d) L/dt = 131
1 1
0.8 0.8
0.6 0.6
hL/dt
0.4 hL/dt 0.4

0.2 0.2
0 0
0 25 50 75 100 0 25 50 75 100
Time (s) Time (s)
(e) L/dt = 148 (f) L/dt = 180
1 1
0.8 0.8
0.6 0.6
hL/dt
hL/dt
0.4 0.4
0.2 0.2
0 0
0 25 50 75 100 0 25 50 75 100
Time (s) Time (s)
Figure 9.13 Liquid holdup measurements for uGS =1.2 m/s, uLS=0.16 m/s. Woods & Hanratty, 1999.
entering from the top. The air flow was forced through an orifice located five diameters
upstream of the tee section. The velocity of the air through this orifice approaches the
local velocity of sound. Consequently, downstream variations in the gas-phase pressure
caused by the formation of slugs do not strongly affect the inlet air flow.
Measurements of liquid holdup, presented in Figure 9.13 were obtained with six
conductance probes located along the pipe. No slugs form for L/dt < 40.The presence
of slugs at L/dt > 98 is indicated by peaks in hL/dt.
At the transition from a stratified flow to slug flow, hL/dt was approximately 0.6.
Measurements at L/dt = 12 show that the stratified flow at the inlet is maintained at
hL/dt = 0.6, where hLS is the height at transition. If no slugs were present in the pipe, the
stratified flow would assume an equilibrium height, hLe, which is larger than hLS. Thus,
the slugs feed information to the inlet, through backward moving waves, to regulate
the inlet height. The height of the stratified flow required for a stable slug to exist was
hL0/dt = 0.3. One notes that this is a lower limit for the height of the layer behind the slug.
The time interval over which a slug is detected at L/dt = 180 is larger than at L/dt = 98.
This suggests that the slug is growing in length since its velocity is approximately
Flow direction
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
(i)
(j)
Figure 9.14 Video images illustrating the formation of a slug within zone I. Woods & Hanratty, 1999.
constant as it moves downstream. The height of the liquid in front of the slug is greater
than was needed to stabilize it, hL > hL0. The slugs grow by consuming liquid in the
stratified layer in front of them. In a much longer pipe than was used in this experiment,
one could presume that slugs stop growing (unless they coalesce) and that the stratified
layer between slugs would have a height slightly greater than hL0.
Figure 9.14 shows the formation of a slug in zone I. It gives 10 sequential frames
obtained with a high-speed video camera. The time between frames is 1/30 s. The flow
2.5
Experimental data
2 Model prediction
1.5
fS LD
uLS
0.5
0
0 0.2 0.4 0.6 0.8 1
Fr = uLS /(ghLS)0.5
Figure 9.15 Measurements of the frequency of slugging. Woods & Hanratty, 1999.
250
uLS = 0.08 m/s
200 uLS = 0.10 m/s
uLS = 0.12 m/s
150 uLS = 0.14 m/s
LD / dt
uLS = 0.16 m/s

100
uLS = 0.18 m/s
50
0
0 1 2 3 4
uSG (m/s)
Figure 9.16 Variation of LD/dt with uGS and uLS. Woods & Hanratty, 1999.
is from right to left and a pipe length of 0.5 m is shown. Three wave crests, 16–20 cm
apart are observed in Figure 9.14a. They correspond to a frequency 0.5 Hz. The wave
velocity is approximately 0.9 m/s. The third wave from the left becomes unstable as it
propagates downstream. It grows to form a slug in frame (h). The liquid behind the slug
decreases to hL0. The waves behind the slug decay because they are propagating over a
thinner layer.
Figure 9.11 gives a sketch of the refilling of the pipe by a backward moving wave in
the pipe. Woods & Hanratty (1999) used equations (9.39) and (9.40) to calculate the
time interval between slugs. Figure 9.15 gives the results of this calculation plotted as
uLS =f S LD versus Fr ¼ uLS =ðghLS Þ0:5 . Here, t S ¼ 1=f S is the time interval between slugs
and uLS ; hLS are the liquid velocity and the height corresponding to the instability
conditions. The length of pipe needed to develop an unstable wave train is designated
by LD. The bars in Figure 9.15 reflect the errors in estimating LD. Figure 9.16 shows the
variation of the mean values of LD with uGS and uLS. This parameter was estimated by
measuring the liquid holdup at various locations in the pipeline for several different flow
conditions and by estimating the probe at which a slug was first observed. Each mean
value in Figure 9.16 is based on an ensemble of at least 100 slugs.
The influence of the declination of a pipe on the initiation of slug flow is discussed in a
paper by Woods et al. (2000) and in Section 5.5.3. As seen in Figure 5.8, transition to
slugging occurs at larger uLS when the pipe is declined. The difference between hori-
zontal and declined pipes is that the liquid velocity is larger in declined pipes, when
comparison is made for the same height of the liquid layer. Both horizontal and declined
pipes show regular slugging (see Figure 5.9). A declined pipe, however, will be charac-
terized by Fr > 1 at the inlet. Thus, refilling by backward-moving gravity waves does not
occur.
In a horizontal pipe, slugs evolve as a result of a viscous large-wavelength instability,
so the frequency of slugging equals the frequency of these waves. (The actual transition
appears to be associated with a Kelvin–Helmholtz instability at the crest of a large-
wavelength wave, as shown in Figure 5.9.)
Section 5.5.2 describes the behavior in inclined pipes, for which reversed flows can
exist. In this regime an intermittent behavior is experienced. Slugs are formed at the inlet
by a Kelvin–Helmholtz instability. They move downstream and are followed by a
backward moving stratified flow. The liquid level at the inlet drops after the formation
of a slug. It rebuilds by liquid flowing into the pipe and by the backward-moving
stratified flow. When the height at the inlet reaches a critical value, a slug forms. Slugs
appear periodically with a frequency defined by the time required to rebuild the liquid
height to a critical value needed for a slug to form.
Thus, at low gas velocities regular slugging can occur for horizontal, inclined and declined
pipes. However, the mechanisms governing slugging for these three cases are different.
9.4.3 Stochastic slugging

Stochastic slugging is observed in cases for which the Froude number is greater than
unity (regions II and III in Figure 9.12). For these situations, a bore formed behind the
slug replenishes the liquid picked up by the slug. Since the slug moves downstream at a
faster velocity, the stratified layer between the bore and the slug lengthens. The stratified
flow has a height, hL0, which is equal to that required for a stable or growing slug to exist.
This process is illustrated in Figure 9.17, which displays holdup measurements
(Woods et al., 2006) obtained for air and water flowing through a horizontal 0.0763 m
pipe. The liquid and gas flows were uLS = 1.0 m/s, uGS = 1.8 m/s (zone III). A pressure
transducer at L/dt = 22 reflects the passage of a slug (see Figure 9.17c). A highly disturbed
interface is observed at L/dt = 4.0. The frequency of the disturbances decreases dramat-
ically downstream until at L/dt = 200 an approximately fully developed slug pattern is
observed.
A sharp drop in the level behind a disturbance, to a value of hL0, is the signature for the
presence of an incipient slug at L/dt = 4. At L/dt = 37, both slugs and large-amplitude
waves are present. The slower-moving large-amplitude waves are consumed by the slugs
so the number of peaks decreases between L/dt = 37 and L/dt = 200.
Measurements such as those presented in Figure 9.17 suggest the following picture of
the behavior of interfacial disturbances: Incipient slugs develop close to the entry. These
grow if hL > hL0 and decay if hL < hL0. For hL close to hL0, they will collapse if the slug is
uLS = 1.0 m/s uGS = 1.8 m/s

1
0.8
hL/dt
0.6
0.4
0.2 (a) L/dt = 4
0
0 2 4 6 8 10 12 14 16 18 20
Time (s)
1
0.8
hL/dt
0.6
0.4
0.2 (b) L/dt = 12
0
0 2 4 6 8 10 12 14 16 18 20
Time (s)
1.6
Pressure (psig)
1.2
0.8
0.4 (c) Pressure at L/dt = 22
0
0 2 4 6 8 10 12 14 16 18 20
Time (s)
1
0.8
hL/dt
0.6
0.4
0.2 (d) L/dt = 37
0
0 2 4 6 8 10 12 14 16 18 20
Time (s)
1
0.8
hL/dt
0.6
0.4
0.2 (e) L/dt = 200
0
0 2 4 6 8 10 12 14 16 18 20
Time (s)
Figure 9.17 Liquid holdup at four different locations and pressure pulsations at L/dt = 22, uGS = 1.8 m/s, dt =
0.0763 m. Woods et al., 2006.
not long enough. Collapsed slugs form large-amplitude roll waves which propagate
downstream. They are incorporated into faster-moving slugs.
Measurements of slug frequency at a pipe exit have been reported by a number of
investigators (Gregory & Scott,1969; Heywood & Richardson, 1979; Crowley et al.,
1986; Woods et al., 2006). These are summarized in Figure 9.18, where fSdt/uLS is plotted
against uLS =ðuLS þ uGS Þ. Note that this plot captures the effects of liquid flow and pipe
diameter and that fSdt/uLS = 0.05 is a rough approximation of the data.
dt = 0.019 m, Gregory & Scott (1969)

dt = 0.042 m, Heywood & Richardson (1979)
0.1 dt = 0.0763 m, Woods et al. (2006)

dt = 0.095 m, Fan et al. (1993a)
dt = 0.15 m, Crowley et al. (1986)
0.08
0.06
fS dt
uLS
0.04
0.02
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
uLS / (uGS + uLS)
Figure 9.18 Measurements of slug frequencies for different pipe diameters. Woods et al., 2006.
The slug frequency can be related to the distribution of slug lengths, IS, for a flow
consisting of slugs embedded in a stratified layer with height hL0 and velocity uL0. An
intermittency, I, is defined as the fraction of the time a slug is observed by a stationary
observer
f S LS
I¼ ð9:41Þ
c
where c is the slug velocity. It can be roughly approximated as

uLS
I≈ ð9:42Þ
uLS þ uGS
Furthermore (from Section 9.2.4 and equation (9.25).

c ¼ 1:2ðuGS þ uLS Þ ð9:43Þ
If (9.42) and (9.43) are substituted into (9.41), a first approximation for the relation
between fS and I ¼ f S LS =c under fully developed conditions is obtained:
1
f Sdt LS
¼ 1:2 ð9:44Þ
uLS dt
In addition to the use of a highly simplified model for slug flow, equation (9.44) ignores
the liquid carried in the stratified flow, and the gas carried by the slugs. A derivation that
takes these factors into account is given by Woods (1998).
Information on slug length was obtained by Woods et al. (2006) from measurements
of the holdup for air–water flow at the pipe outlet, such as shown in Figure 9.19. Values
of the void fraction in the slugs were calculated from peaks in hL/dt. The arrows in
Figure 9.19 that are pointing downward indicate large-amplitude waves, which are
(a) Measurements at L/dt = 190 for uSG = 1.05 m/s, uSL = 1.2 m/s
1
0.8
0.6
hL/dt
0.4
0.2 hL0
0
0 5 10 15 20
Time (s)
(b) Measurements at L/dt = 190 for uSG = 1.8 m/s, uSL = 1.2 m/s
1
0.8
0.6
hL/dt
0.4
0.2 hL0
0
0 10 20 30
Time (s)
(c) Measurements at L/dt = 201 for uSG = 3.0 m/s, uSL = 1.2 m/s
1
0.8
0.6
hL/dt
0.4
0.2
hL0
0
0 10 20 30
Time (s)
Figure 9.19 Holdup measurements at uLS = 1.2 m/s, dt =0.0763 m. Woods et al., 2006.
moving at a lower velocity than the slugs. These are overtaken by slugs, thus causing an
increase of the mass of liquid in the slugs.
Measurements of the average slug length were determined by setting an hL/dt at
which to measure the time interval for the slug passage. The triggers (the dashed lines
in Figure 9.19) were selected to give the correct slug frequency. For large gas velocities,
the measurements are less accurate because the slugs are highly aerated and the hL/dt are
smaller. This can make it difficult to differentiate slugs from large-amplitude waves. In
these cases, measurements of pressure pulses, characteristic of slugs, can be used to
differentiate fast-moving slugs from slow-moving large-amplitude waves at the outlet of
the pipe.
Average lengths measured close to the pipe outlet are not sensitive to changes in uLS
and weakly sensitive to uGS. A value of LS =d t ¼ 18 was obtained by Woods & Hanratty
(2006) for uGS = 2–4 m/s, L/dt = 200, where L is the length of the pipe. Grenier et al.
(1997) carried out a study in a 0.053 m pipe, which had a length of 90 m (L/dt = 1700),
and found LS =d t ¼ 20 for mixture velocities greater than 1.2 m/s. A similar result was
obtained by Ferre (1979) and by Bernicott & Drouffe (1991) in a pipe with L/dt = 9500. It
is of interest that this roughly equals the value of 18 obtained by Woods et al. (2006) for
L/dt = 200. The experiments of Nydal et al. (1992) are of particular interest since a large
range of gas (0.5–20 m/s) and liquid (0.6–3.5 m/s) superficial velocities were considered.
Two pipes with diameters of 0.053 m and 0.09 m, and lengths of 17 m were used. They
measured LS =d t ¼ 1520 in the 0.053 m pipe and LS =d t ¼ 1216 in the 0.09 m pipe.
Experiments by Saether et al. (1990) in a 0.032 m pipe with L/dt = 560 gave LS =d t ¼ 27
for ðuGS þ uLS Þ ¼ 7:58 m=s.
A striking feature of these measurements is that the reported average values of LS/dt
are roughly the same for a wide range of pipe lengths. Woods (1998) obtained measure-
ments of the standard deviation for the distribution of slug lengths of about 0.55 m for
uGS = 2–4 m/s and 0.5 m for uGS > 4 m/s. The ratio σ LS =LS was measured as 0.35
for uGS = 1 m/s and as 0.5 for uGS > 4 m/s. Nydal et al. (1992) obtained σ LS =LS ¼ 0:37
for ðuGS þ uLS Þ ≤ 5 m=s. They measured a drop of σ LS =LS at ðuGS þ uLS Þ ≤ 5 m=s. A
value of σ LS =LS ¼ 0:31 was obtained for ðuGS þ uLS Þ ¼ 10 m=s. The study of
Saether et al. (1990) at ðuGS þ uLS Þ ¼ 7:58 m=s gave σ LS =LS ¼ 0:26.
A number of researchers have suggested that there is a minimum length below which
slugs are unstable (Taitel et al. 1980). However, there are differences in its suggested
value (Barnea & Brauner, 1985; Dukler et al., 1985). Measurements by Woods et al.
(2006) of LS and dLS =dt suggest that ðLS =d t Þmin ≈ 5. The cumulative probability density
function of slug lengths obtained at uGS = 5 m/s by Nydal et al. (1992) gives
ðLS =d t Þmin ≈ 8. Measurements by Grenier et al. (1997) give a 100% chance for a slug
with LS =d t ¼ 9 to be stable.
The probability density function for slug length can be approximated by a log-normal
function (Brill et al., 1981; Nydal et al., 1992; Woods et al., 2006)
" #
1 ðln LS C 1 Þ2
pðLS Þ ¼ pffiffiffiffiffi exp ð9:45Þ
LS C 2 2π 2C 22
where constants C1 and C2 can be obtained from measurements of LS and σ 2LS .

Scott et al. (1987) have suggested two stages of slug growth from field measurements
in very long petroleum pipelines, at conditions close to those needed to initiate slugging.
The first involves an accumulation of waves by slugs which would culminate in the “fully
developed” pattern discussed above. The second involves the coalescence of slugs. This
second stage has been discussed by Barnea and Taitel (1993).
Several mechanisms for the growth of slugs in the second stage have been discussed.
The focus has been on differences in the velocities of the gas bubbles behind two
successive slugs with different lengths. (This approach presumes that the bubble velocity
behind a slug increases with decreasing slug length.)
Measurements of the bubble velocity at the back of a slug, cB, have been made by
Woods et al. (2006). These are shown in Figure 9.20. It is noted that cB can have a small,
but significant, range of velocities for given uLS and uGS. This indicates that adjacent
(a) uGS = 1.2 m/s, uLS = 1.2 m/s

3
2.9
2.8
cB (m/s)
2.7
2.6
2.5
2.4
0 0.5 1 1.5 2 2.5
Slug length LS (m)
(b) uGS = 2.4 m/s, uLS = 1.2 m/s

5
4.8
4.6
cB (m/s)
4.4
4.2
4
3.8
3.6
0 0.5 1 1.5 2 2.5 3
Slug length LS (m)
(c) uGS = 3.0 m/s, uLS = 1.2 m/s

6
5.5
5
cB (m/s)
4.5
4
3.5
3
0 0.5 1 1.5 2 2.5 3
Slug length LS (m)
Figure 9.20 Effect of slug length on the velocity of a bubble behind the slug. Woods et al., 2006.
slugs can overtake one another given sufficient time. However, note from Figure 9.20
that the average value of cB is independent of slug length. This seems to overrule the
mechanism involving coalescence outlined above.
These results, as well as the LS/dt measurements discussed earlier, suggest that the
occurrence of a “second stage” would require an extremely long pipe, such as used in
field operations. The coalescence of two adjacent slugs in this mode of operation should
be a stochastic event.
The holdup tracings shown in Figure 9.19 could be misleading in that spatial variations
in the flow direction are compressed. Slug velocities are given by (9.43). Thus, one
second in Figure 9.19b represents 36 m or 47 pipe diameters. Photographs taken with a
L/dt = 2
L/dt = 10 L/dt = 4 Air
(a) Water
(b)
(c)
(d)
(e)
uLS = 1.0 m/s uGS = 1.8 m/s
Figure 9.21 Video images at the entrance of a pipe for uGS = 1.8 m/s, uLS = 1 m/s, dt = 0.0763 m. Woods et al.,
2006.
high-speed video camera (Woods et al., 2006) give a better representation of what is
happening. Figure 9.21 is an example. The conditions are uGS = 1.8 m/s, uLS = 1 m/s,
dt = 0.073 m (the same as for Figure 9.17). The liquid was dyed to produce contrast. The
tee section in which air and water are mixed is at the right. The probes at L/dt = 2, 4 and 10
are shown in (a). The dimensionless length L/dt = 10 corresponds to a length of 0.76 m.
The velocity of a slug under these conditions is 3.4 m/s so the photos would correspond to
a time interval of 0.22 s. The formation of an incipient slug is shown in (a)–(c). It appears
as a growing slug in photo (d), where LS/dt = 5. A developing tail is shown in photos
(d) and (e). The height of the liquid in front of the slug is larger than hL0 so it is anticipated
that this slug reaches a large enough length to be stable. In photo (e), it can be seen that
the depleted liquid behind the slug is being replaced by an incoming bore.
9.4.4 Stochastic models

A critical issue in establishing a fundamental understanding of slugging at large liquid
flows is the formulation of a stochastic model for how the fully developed region evolves.
Bernicott & Drouffe (1991) proposed a model whose prerequisite is a probability
function for a slug to appear at a given location, x, at a given time. Woods & Hanratty
(1999) and Woods et al. (2006) used an approach similar to that of Bernicott & Drouffe
and developed computational schemes.
The 1999 paper focuses on low liquid velocities and low gas velocities for which slugs
appear at L/dt > 40. A stratified flow with a height hLS is assumed to exist for L/LD.
Beyond LD, an unstable irregular wave pattern appears over a length LU. It is assumed
that it is equally probable for roll waves to coalesce at all locations between LD and
LD+LU. After a slug forms, it translates downstream, rapidly picks up liquid at its front
and sheds liquid at its tail to form a stratified flow of height hL0, the critical height above
which stable slugs can exist. Since Fr > 1, a hydraulic jump appears at the location at
which the slug was formed; it moves downstream with velocity cjump to restore the height
of the liquid to hLS. The probability of forming a slug between and x and x + dx during a
time interval Δt is defined as NLU(t), which is taken to be independent of x. The
probability of forming a slug in the interval Δt is then given as
DtPðt Þ ¼ DtN LU ðt Þ ð9:46Þ
The length of the unstable wavy stratified flow increases with time at a rate given by
LU ¼ Lf þ cjump t ð9:47Þ
where Lf + LD is the location in the pipe at which a previous slug was formed. Thus

DtPðt Þ ¼ DtN Lf þ cjump t ð9:48Þ
The velocity cjump is determined analytically by applying conservation of mass in a

reference plane moving with the hydraulic jump

cjump uslug ALS ¼ cjump uL0 AL0 ð9:49Þ
where ALS is the area of the liquid in a slug and AL0 is the critical area needed to sustain a
stable or growing slug. The randomness of the slugging process was modeled through the
use of random number generators. At t + Δt, a random number, Ri, between 0 and 1 is
selected and compared with the value of ΔtP(t). If Ri >ΔtP(t), a slug does not form. If
Ri < ΔtP(t), a slug has formed somewhere in LU. A second random number generator was
then used to determine where, along LU, this occurred. This becomes a new value of Lf.
The algorithm is repeated to provide an ensemble of time intervals between slug
formation and an ensemble of slug lengths, LS.
In the second paper, Woods et al. (2006) used this computational scheme to represent
stochastic slugging for large superficial liquid velocities. For this case, slugs form close to
the inlet so LD was taken to be zero. Furthermore, the 2006 paper employed the concept
of minimum slug length for a slug to be stable. A value of (LS/dt)min = 4 was used. The
9.5 Triggering of slug formation 229
(a) uLS = 1.2 m/s, uGS = 1.2 m/s

0.3
0.25 Experimental
Fraction of total
Model
0.2
0.15
0.1
0.05
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Slug length (m)
(b) uLS = 1.2 m/s, uGS = 1.8 m/s

0.2
Experimental
Fraction of total
0.15 Model
0.1
0.05
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Slug length (m)
Figure 9.22 Comparison of calculated and measured distributions of slug lengths for air–water flow in a
0.0953 m pipe. Woods et al., 2006.
probability N was selected so that the measured mean LS/dt is the same as the calculated
mean LS/dt.
Calculated and measured distributions of LS for air–water are compared in Figure 9.22
(Woods et al., 2006) under conditions that the flow is fully developed or close to fully
developed. Rough agreement is noted. Both show skewed distributions. However, the
model predicts the existence of slug lengths larger than observed in experiments. Clearly,
there is room for improvement.
9.5 Triggering of slug formation
Two modes of operation can be considered. The data obtained by Woods & Hanratty
(1999) and by Woods et al. (2006) were obtained while the gas flow rate was kept
constant by using a critical orifice.
Another possibility is to keep the pressure drop constant. This type of operation is
discussed by Scott et al. (1987). A slug leaving the pipe gives rise to a sudden increase in
the flow rate. This, in turn, can trigger the formation of a new slug in the pipe.
References
Andreussi, P. & Bendikson, K. 1989 Investigation of void fraction in liquid slugs for horizontal and
inclined gas–liquid pipe flow. Int. J. Multiphase Flow 15, 937–946.
Andritsos, N., Williams, L. & Hanratty, T. J. 1989 Effect of liquid viscosity on stratified-slug
transition in horizontal pipes. Int. J. Multiphase Flow 18, 877–892.
Barnea, D. & Brauner, N. 1985 Holdup of the liquid in two-phase intermittent flow. Int.
Barnea, D. & Taitel, Y. 1993 A model for slug length distribution in gas–liquid slug flow. Int.
Barnea, D., Shoham, O., Taitel, Y. & Dukler, A.E. 1980 Flow pattern transition for gas–liquid flow
in horizontal and inclined pipes. Int. J. Multiphase Flow 6, 217–225.
Bendiksen, K.H. 1984 Experimental investigation of the motion of long bubbles in inclined tubes.
Bernicott, M.F. & Drouffe, J.M. 1991 A slug length distribution law for multiphase transportation
systems. SPE Prod. Eng. 19, 829–838.
Bontozoglou, V. & Hanratty, T.J. 1990 Capillary-gravity Kelvin–Helmholtz waves close to
resonance. J. Fluid Mech. 217, 71–91.
Brill, J.P., Schmidt, Z., Coberly, W.A. & Moore, D.W. 1981 Analysis of two-phase tests in large
diameter flow lines in Prudhoe Bay Field. Soc. Petrol. Eng. J. 363–377.
Crowley, C.J., Sam, R.G. & Rothe, P.H. 1986 Investigation of two-phase flow in horizontal and
inclined pipes at large pipe sizes and high gas density. Report prepared for the American Gas
Association by Creare Inc.
Dukler, A.E. & Hubbard, M.G. 1975 A model for gas–liquid slug flow in horizontal tubes. Ind.
Eng. Chem. Fund. 14, 337–347.
Dukler, A.E., Maron, D.M. & Brauner, N. 1985 A physical model for predicting minimum slug
length. Int. J. Multiphase Flow 40, 1379–1385.
Fan, Z., Jepson, W.P. & Hanratty, T.J. 1992 A model for stationary slugs. Int. J. Multiphase Flow
18, 477–494.
Fan, Z., Lusseyran, F. & Hanratty, T.J. 1993a Initiation of slugs in horizontal gas–liquid flows.
AIChE Jl 39, 1741–1753.
Fan, Z., Ruder, Z. & Hanratty, T.J. 1993b Pressure profiles for slugs in horizontal pipelines. Int.
Ferre, D. 1979 Ecoulements diphasiques aporches en conduite horizontale. Rev. Ind. Fr. Pet. 34,
113–142.
Govier, G.W. & Aziz, K. 1972 The Flow of Complex Mixtures in Pipes. New York: Van Nostrand
Rheinhold.
Gregory, G.A. & Scott, D.S. 1969 Correlation of liquid slug velocity and frequency in concurrent
gas–liquid slug flow. AIChE Jl 15, 833–835.
Gregory, G.A., Nicholson, M.K. & Aziz, K. 1978 Correlation for liquid volume fraction in the slug
for gas–liquid flow. Int. J. Multiphase Flow 4, 33–39.
Grenier, P., Fabre, J. & Fagundes Netto, J.R. 1997 Slug Flow in Pipelines: Recent Advances and
Future Developments. Bedford: BHR Group, pp. 107–121.
Heywood, N.I. & Richardson, J.F. 1979 Slug flow in air–water mixtures in a horizontal pipe;
determination of liquid holdup by gamma ray absorption. Chem. Eng. Sci. 28, 17–30.
References 231
Hurlburt, E.T. & Hanratty, T.J. 2002 Prediction of the transition from stratified to slug and plug
flow for long pipes. Int. J. Multiphase Flow 20, 707–729.
Kouba, G.K. & Jepson, W.P. 1990 The flow of slugs in horizontal two phase pipelines. Trans.
ASME 112, 20–25.
Lin, P.Y. & Hanratty, T.J. 1986 Prediction of the initiation of slugs with linear stability theory. Int.
Nicklin, D.J., Wilkes, J.O. & Davidson, J.F. 1962 Two phase flow in vertical pipes. Trans. Inst.
Chem. Engs. 102, 61–68.
Nydal, O.J., Pintus, S. & Andreussi, P. 1992 Statistical characterization of slug flow in horizontal
pipes. Int. J. Multiphase Flow 18, 439–452.
Ruder, Z. & Hanratty, T.J. 1990 A definition of gas–liquid plug flow in horizontal pipes. Int.
Ruder, Z., Hanratty, P.J. & Hanratty, T.J. 1989 Necessary conditions for the existence of stable
slugs. Int. J. Multiphase Flow 15, 209–226.
Saether, G., Bendiksen, K., Muller, J. & Froland, E. 1990 The fractal statistics of liquid slug
lengths. Int. J. Multiphase Flow, 16, 1117–1126.
Scott, S.L., Shoham, O. & Brill, J.P. 1987 Modeling slug growth in large diameter pipes. Paper
presented at the Third International Joint Conference on Multiphase Flow, The Hague,
Netherlands, May 1987, paper B2.
Singh, G. & Griffith, P. 1970 Determination of the pressure optimum pipe size for two-phase flow
in an inclined pipe. TASME J. Eng. Ind. 92, 717–726.
Stoker, J.J. 1957 Water Waves. New York: Interscience, pp. 313–333.
Taitel, Y., Barnea, D. & Dukler, A. E. 1980 Modelling of flow pattern transitions for steady upward
gas–liquid flow in vertical tubes. AIChE Jl 3, 345–354.
Theron, B. 1989 Ecoulements diphasiques instatationaires en conduite horizontale. Ph.D. thesis,
Institut National Polytechnique de Toulouse.
Woods, B.D. 1998 Slug formation and frequency of slugging in gas–liquid flows. Ph.D. thesis,
Woods, B.D. & Hanratty, T.J. 1996 Relation of slug stability to shedding rate. Int. J. Multiphase
Flow 22, 809–828.
Woods, B.D. & Hanratty, T.J. 1999 Influence of Froude number on physical processes determining
frequency of slugging in horizontal gas–liquid flows. Int. J. Multiphase Flow 25, 1195–1223.
Woods, B.D., Hurlburt, E.T. & Hanratty, T.J. 2000 Mechanism of slug formation in downwardly
Woods, B.D., Fan, Z. & Hanratty, T.J. 2006 Frequency and development of slugs in a horizontal
pipe at large liquid flows. Int. J. Multiphase Flow 32, 902–925.
10 Particle dispersion and deposition
10.1 Prologue
Particles entrained in a turbulent fluid are dispersed by velocity fluctuations; they assume a
motion that is related to the fluid turbulence. If the suspension flows through a conduit,
deposition on a wall depends on the particle turbulence. An understanding of these
processes is needed to describe the annular flow regime for which liquid flows along the
walls and as drops in the gas flow. The fraction of liquid that is entrained by the gas depends
on the rate at which the film is atomized and the rate at which drops deposit on the film.
Equations for trajectories of spherical drops and bubbles in a turbulent flow field are
developed. These are used to relate the turbulence properties and the dispersion of
particles to the turbulence properties of the fluid in which they are entrained. Of particular
interest is the development of relations for the influence of drop size on drop turbulence
and on drop dispersion.
The deposition of particles is strongly related to their size. A wide range of behaviors is
observed. Brownian motion is realized for very small particles. Particle motion becomes
independent of fluid turbulence for very large particles. It is hard to capture all of these
behaviors in a single laboratory experiment. Therefore, numerical studies have been
carried out in which a modified Langevin equation is used to provide a stochastic
representation of the fluid turbulence.
Results from laboratory and numerical experiments are used to develop correlations
for the rate of deposition and for the velocities with which particles strike a wall.
10.2 Dispersion of fluid particles in homogeneous isotropic turbulence
A starting point for this chapter is the description of the dispersion of marked fluid
particles from a point source, provided by Taylor (1921). The problem is simplified by
assuming a homogeneous, isotropic turbulence and a zero mean velocity (ūi = 0).
The velocity of a diffusing fluid particle is designated by three components in the x1,
x2, x3 directions, v1, v2, v3. The location is designated by y1, y2, y3. The magnitude of the
particle turbulence is represented by the mean-square values of the velocity fluctuations,
v21 ; v22 ; v23 . For a homogeneous field, the components v2i do not vary with location.
Consider a marked fluid particle located at yi = 0, at time zero. The distance traveled by
the particle in the yi-direction by time t is given as
10.2 Dispersion of fluid particles in homogeneous isotropic turbulence 233
ðt
yi ð t Þ ¼ vi ðt 0 Þdt 0 ð10:1Þ
0
If this measurement is repeated a large number of times, the ensemble average at time t
is given by yi ðt Þ ¼ 0 since vi is just as likely to be plus or minus. However, y2i ðt Þ≠0 ; it is a
measure of the average spread determined from a number of trials.
Einstein (1905) has shown that the molecular diffusion coefficient, D, can be related to
the change of mean-square dispersion over a time interval dt by the equation
1 dy2i
¼D ð10:2Þ
2 dt
From kinetic theory,

D / λL c ð10:3Þ
where c is the molecular velocity and λL is the molecular mean-free path.

Taylor (1921) derived an analogous equation for turbulent dispersion. He pointed out
that
ðt
dy2i dyi
¼ 2yi ¼ 2vi vi ðt 0 Þdt 0
dt dt 0
ðt
dy2i
¼2 vi ðt Þvi ðt 0 Þdt 0 ð10:4Þ
dt 0
Thus, a turbulent diffusivity can be defined as

ðt
1 dy2i
ε¼ ¼ vi ðt 0 þ sÞvi ðt 0 Þdt 0 ð10:5Þ
2 dt 0
where t = t 0 + s has been substituted into (10.4). The turbulence is steady in time so
vi ðt 0 þ sÞvi ðt 0 Þ ¼ vi ð0Þvi ðsÞ. Thus, the turbulent diffusivity, ε, is time-dependent
ðt
εðt Þ ¼ v2i RLi ðsÞds ð10:6Þ
0
where
vi ð0Þvi ðsÞ
RLi ðsÞ ¼ ð10:7Þ
v2i
is the Lagrangian correlation coefficient. For s → 0, vi(0) and vi(s) are perfectly correlated
so RLi ðt ! 0Þ ≈ 1 and
εðt ! 0Þ ¼ v2i t ð10:8Þ

As s increases, vi(0) and vi(s) eventually become completely uncorrelated; that is, vi(s)
is just as likely to be plus or minus if vi(0) is plus, so RLi ð∞Þ ¼ 0. Define a Lagrangian
time-scale as
ð∞
τL ¼ RLi ðsÞds ð10:9Þ
0
From (10.6)
εð∞Þ ¼ v2i τ L ð10:10Þ
Define a Lagrangian length scale as

1=2
ΛL ¼ v2i τL ð10:11Þ
Thus,
1=2
εð∞Þ ¼ v2i ΛL ð10:12Þ
where the two terms on the right side are analogous to the molecular velocity and the
molecular mean-free path.
An integration of (10.5) gives
ð t ð t0
yi ¼ 2vi dt 0 RLi ðsÞds
2 2 ð10:13Þ
0 0
As shown by Hinze (1959), the integration of (10.13) gives the following relation:
ðt
yi ¼ 2vi ðt sÞRLi ðsÞds
2 2 ð10:14Þ
0
For s close to zero, RLi ðsÞ ≈ 1. Equation (10.14) then gives
y2i ¼ v2i t 2 ð10:15Þ
for small t. For large t,

ðt ðt
y2i ðt Þ ¼ 2v2i t RLi ðsÞds 2v2i sRLi ðsÞds ð10:16Þ
0 0
In order to calculate y2i ðt Þ, one needs to specify RLi ðsÞ. Hinze defines a Lagrangian micro-
time-scale as

1 1 ∂2 RLi
T 2 ¼ ð10:17Þ
λ 2 ∂s2 s!0
Thus, as s → 0
s2
RLi ðsÞ ¼ 1 2 ð10:18Þ
λT
10.3 Equation of motion of a particle or a bubble 235
A commonly used functionality for RLi ðsÞ is

RLi ðsÞ ¼ exp s=τ L ð10:19Þ
Substitution of (10.19) into (10.16) yields

2
y2i ðt Þ ¼ 2v2i τ L t 2v2i τ L 1 exp t=τ L ð10:20Þ
Equation (10.19) is not accurate as t → 0 since it gives

s
RLi ðsÞ ¼ 1 ð10:21Þ
τL
in disagreement with (10.18). However, at large Reynolds numbers, this is not a serious
problem since (10.18) is valid only over a small range of diffusion times.
Measurements of the dispersion of marked fluid elements originating from the center
of a pipe have been made by Towle & Sherwood (1939), Boothroyd (1967), Groenhof
(1970) and Taylor & Middleman (1974). Concentration profiles downstream of the origin
are represented by Gaussian distributions with standard deviations of y2i . A turbulent
diffusivity is calculated from (10.5). Data from the above references for large times are
represented by the equation
ε
¼ 0:037 ð10:22Þ
v 2rt
where rt is the pipe radius and v* is the friction velocity, ðτ W =ρf Þ1=2 . The shear stress at
the wall, τW, is calculated from measurements of the pressure drop under fully developed
conditions.
rt DP
τW ¼ j j ð10:23Þ
2 Dz
where vertical bars indicate an absolute value.
10.3 Equation of motion of a particle or a bubble
The equation of motion of a spherical particle, entrained in turbulence, at relative

velocities in the range where the force of the fluid on the particle is small enough for
Stokes law to be applicable has been given by Maxey and Riley (1983) as
4 3 d! vP 4 3
πrP ¼ πrP ðρP ρf Þ!g þ 6πμf rP !u ! vP
3 dt 3
ð t ! !
d u vP dτ
þ 6πμf rP ð10:24Þ
0 dτ ½πf ðt τ Þ1=2
2 d ! ! 4 3 D! u
þ πr3P ρf u v P þ πrP ρf
3 dt 3 Dt

where ! u is the fluid velocity, ! v P is the particle velocity and ! u ! v P is the relative
velocity which dictates the force of the fluid on the particle. Lift forces have been ignored.
The term on the left side is the mass of the particle times its acceleration. The first term on
the right side is the force of gravity minus the buoyancy force on the particle. The second
term is Stokes drag force for a steady flow. The third term is the Basset force
(Basset, 1888), which is a correction to Stokes law because the flow is unsteady. The
fourth term is the added mass, which accounts for the fluid in the continuum which is
accelerated when the particle is accelerated. The last term is the force due to pressure
gradients in the fluid. This can be understood by recognizing that D! u =Dt is the
substantial derivative of the fluid. According to inviscid theory ρf D! u =Dt equals the
local pressure gradient accompanying the fluid acceleration.
Mei et al. (1991) show that the Basset force predicts too long a memory in the time
domain. Thus, its use in (10.24) is questionable. For the present, it is best to ignore this term.
In cases for which ρP >> ρf, such as a solid or liquid sphere in a gas, (10.24) can be
simplified to
4 3 d!vP 4 ! 4
πr ρ ¼ πr3P !
g ρP þ F L πr3P
3 P P dt 3 3
1 ! ! ! !
þ C D πrP ρf u v P j u v P j
3
ð10:25Þ
2
2 d ! ! 4 3 D! u
þ πr3P ρf u v P þ πrP ρf
3 dt 3 Dt
!
where a lift force per unit volume, F L , is included, !
u is the fluid velocity seen by the
particle and non-Stokesian behavior is considered. For cases in which ρf >> ρP, such as a
spherical gas bubble in a liquid, (10.25) can be written as
4 4 1
0 ¼ πr3P ρf !
g þ πr3P F L þ C D πr2P ρf !u !
v j!
u !
v Pj
3 ! 3 2
ð10:26Þ
2 d u vP 4 D!u
þ πr3P ρf þ πr3P ρf
3 dt 3 Dt
Equation (10.25) describes the time change of the velocity of a particle, !

v P , as it moves
around in a fluid. The path is given by solving the equation
d!y
¼!
vP ð10:27Þ
dt
Note that the solution of (10.25) requires a knowledge of the fluid velocity along the path
of the particle, which is not known a priori. This presents a major problem which has
been solved with an iterative method by Reeks (1977).
Equations (10.26) and (10.27) can be used to describe the change of bubble velocity
and location with time. Equation (10.26) differs from (10.25) in that the time change of
the momentum of the particle is not an important consideration for bubble motion,
because of its small density. Instead, the acceleration of the added mass introduces the
temporal change of the particle velocity.
10.4 Lift forces and lateral drift 237
10.4 Lift forces and lateral drift
10.4.1 Inviscid lift forces

Several lift forces have been defined. One is the classical inviscid result for a rotating
sphere (Magnus effect) described by Schlichting (1979). For this situation,
! !
F L ¼ C L ρL U S !
ωP ð10:28Þ
!
where U S is the slip velocity of the particle or bubble (velocity of the particle minus the
velocity of the surrounding fluid) and ! ω P is the angular velocity of the particle (which is
!
positive for a counterclockwise rotation). For a bubble in upward flow, U S is positive and
the lift force is toward the wall if !
ω P is positive.
As already discussed in Section 2.9 in the chapter on flow regimes, bubbles tend to
accumulate close to the wall for upward flows in a pipe and away from the wall for
!
downward flows. This supports (10.28) if ωP ¼ i r duz =dr; so that
duz
F Lr ¼ C L ðuG2 uL2 ÞρL ð10:29Þ
dr
However, this agreement does not prove that the bubbles are rotating. As shown by
Beyerlein et al. (1985), a bubble embedded in a flow which has gradients in the mean
velocity would experience a difference in fluid velocity on the two sides of the bubble.
Because of a Bernoulli effect this results in a pressure difference. This, in turn, provides a
lateral net force on the bubble.
Furthermore, the bubbles are not spherical in gas–liquid flows. The force on these
bubbles could have a component in the lateral direction which is related to their shape and
orientation. Thus, the theoretical justification for using (10.29) is not established.
These lift forces on bubbles, associated with a non-uniform mean velocity, lead to drift
velocities in the radial direction. A drift velocity can also be realized if the fluid
turbulence varies spatially. The particles will then tend to move from regions of high
turbulence to regions of low turbulence. Section 10.9 develops an analytical expression
for this effect.
10.4.2 Saffman viscous lift force

Saffman (1965, 1968) showed that, for low particle Reynolds numbers in a shear flow,
the lift force due to particle rotation can be neglected. He developed the following
relation for a particle in a steady laminar flow due to uniform simple shear
1=2
! 2 1 duz
F L ¼ 6:46μf rP ðuz vPz Þ ð10:30Þ
f dy
where y is the distance from the wall. If the particle leads the fluid, FL is negative, that is,
toward the wall. If the particle lags the fluid, FL is positive. Since the fluid drag is
described by Stokes law, the particle Reynolds number needs to be small. Furthermore,
the dimensionless shear rate needs to be large and the dimensionless particle diameter
needs to be small.
1=2
duz 1=2
f
dy duz ðd P Þ2 ð10:31Þ
>> 1 551
juz vPz j dy f
These restrictions suggest that Saffman’s equation should not be used for y+ > 70 in a
turbulent flow (see McLaughlin, 1991).
10.5 Characterization of fluid turbulence
Particles in the presence of a turbulent velocity field will assume a turbulent motion,
because of the fluctuating drag associated with the fluctuating fluid velocity. Because of
the inertia of the particles and the gravitational force, the particles do not follow
the velocity fluctuations exactly. The goal of theory is to relate the particle turbulence
and the particle diffusivity to the fluid turbulence and the fluid turbulent diffusivity.
Homogeneous, isotropic turbulence with zero mean velocity is assumed.
Fluid turbulence is characterized by the magnitudes of the three components of the
fluid velocity fluctuations, u2i , by correlation coefficients and by spectra. Eulerian spatial
and temporal correlation coefficients can be defined analogous to the Lagrangian
correlation (10.7). Thus, for a fixed location, the temporal correlation is defined as
ui ðt Þui ðt þ sÞ
REi ðx; sÞ ¼ ð10:32Þ
u2i
where ui(t) ui (t + s) is the product of ui at two different times and at a fixed location, xi.
A spatial Eulerian correlation can be defined for a fixed time as
ui ðxi Þuðxi þ ξ Þ
REi ðt; ξ Þ ¼ ð10:33Þ
u2i
where ui(xi)ui(xi+ξi) is the product of ui at two different locations at a fixed time. The
overbars are averages for a large number of realizations of the field. For ξ = 0 or s = 0,
these correlation coefficients are equal to unity. For ξ ¼ ∞; REi ¼ 0, since for a fixed time
the velocities at locations xi + ξ and xi are uncorrelated for large ξ. Thus, from (10.33),
ð∞
Λi ¼ REi dξ ð10:34Þ
0
is a measure of the extent of the turbulent velocity fluctuations. An Eulerian time-scale is

obtained from (10.32):
ð∞
τ Ei ¼ REi ðsÞds ð10:35Þ
0
10.5 Characterization of fluid turbulence 239
which is a measure of the duration of the turbulent fluctuations at a given location.

An alternative way to describe turbulence is through frequency spectra, whereby the
intensity u2i can be viewed as containing contributions from a number of frequencies.
Thus Ei(n)dn represents contributions to u2i from frequencies between n and n + dn and
ð∞
ui ¼
2 Ei ðnÞðdnÞ ð10:36Þ
0
Hinze (1959) derived the following relations between the spectral function Ei(n) and the
Eulerian temporal correlation:
ð
1 ∞
R i ðt Þ ¼
E
dnEi ðnÞcos 2πnt ð10:37Þ
u2i 0
ð∞
Ei ðnÞ ¼ 4ui dtREi ðt Þcos 2πnt
2 ð10:38Þ
0
Note that
ð∞
E i ð0Þ ¼ 4u2i dtREi ðt Þ ¼ 4u2i τ Ei ð10:39Þ
0
For a homogeneous isotropic field, u21 ¼ u22 ¼ u23 , two fundamental spatial correlations
can be defined. The longitudinal correlation considers two components in the same
direction, say x1, separated by a distance r. Thus,
f ðrÞ ¼ u1 ðx1 Þu1 ðx1 þ rÞ ð10:40Þ
The lateral correlation relates lateral components, u2 or u3, of the velocity
gðrÞ ¼ u2 ðx1 Þu2 ðx1 þ rÞ ð10:41Þ
For isotropic turbulence, conservation of mass gives

1 df
gðrÞ ¼ f ðrÞ þ r ð10:42Þ
2 dr
See Hinze (1959).

Consider the case
f ðrÞ ¼ er=Λ ð10:43Þ
The lateral correlation is given as

r r=Λ
g ðrÞ ¼ 1 e ð10:44Þ
2Λ
This shows interesting differences between f(r) and g(r). The longitudinal correlation f(r)
is positive for all values of r. However, g(r) takes on negative values. The positive values
of g(r) correspond to a region in which the lateral velocity has the same sign. The
correlation must have negative values at large r in order that conservation of mass is
satisfied.
If a homogeneous isotropic turbulence is being transported with a uniform velocity in
the x1-direction, U1 , the Taylor hypothesis can be used whereby the time variation of the
turbulence can be related to the spatial variation by assuming
∂ ∂
¼ U 1 ð10:45Þ
∂t ∂x1
Thus, the temporal correlation can be related to the spatial correlation as follows

f ðx1 Þ ¼ RE1 t ¼ x1 =U 1 ð10:46Þ
(See Hinze (1959, pp. 40–41) for a discussion of the Taylor hypothesis.)
From (10.37) and (10.38), the spectral function and the correlations can be defined as
follows:
ð
1 ∞ 2πnx1
f ðx1 Þ ¼ dnE1 ðnÞcos ð10:47Þ
2
u1 0 U1
ð∞
dx1 cos 2πnx1
E1 ðnÞ ¼ 4u1 2 f ð x1 Þ ð10:48Þ
0 U1 U1
Suppose f(x1) is approximated by an exponential function, f (x1) = exp (−x1/Λ1), except

for very small x1 (which is representative of high frequencies). Then from (10.48)
U 1 E 1 ð nÞ 1
¼ ð10:49Þ
4u2i Λ1 1 þ 4π 2 n2 =U 21 ðΛ1 Þ2
A plot of data, constructed by Favre and colleagues is presented by Hinze (1959, p. 61).
The measurements were made behind a square grid, located in a wind tunnel, so as to
approximate a homogeneous, isotropic turbulence convected with a velocity U1. The
solid line in this plot represents (10.49). The filled points are measurements. The open
points are calculated from measurements of f(x1). Hinze points out that (10.49) “appears
to approximate measured spectrum curves satisfactorily, except, of course, for that part of
the E1(n) -curve that pertains to high values of n. When n approaches zero, the value of
U1 E1 ðnÞ=u21 Λ1 ¼ 4 agrees very satisfactorily with experimental data.”
10.6 Relation of particle turbulence to fluid turbulence
An inertial time constant, 1/β, can be defined as

1 3C D ρf ! !
¼ j u v pj ð10:50Þ
β 4d p ρp
10.6 Relation of particle turbulence to fluid turbulence 241
where, for Stokes flow,

18μf
β¼ ð10:51Þ
d 2P ρP
Equation (10.25) for ρP >> ρf can then be written as follows if the effect of gravity and
the last two terms of (10.25) are ignored, and β is defined by (10.50)
d!
vP
¼β !u !
vP ð10:52Þ
dt
or
d!
vP
þ β!
v P ¼ β!
u ð10:53Þ
dt
Thus, 1/β is a time constant which characterizes the ability of a particle to follow the fluid
turbulence. It plays a central role in understanding the behavior of particles in a turbulent
field. (Consider a particle released in a stagnant fluid with a velocity ! v 0. The product of
!v 0 and 1/β represents the stopping distance.)
The integration of (10.53) yields
ðt
vPi ¼ vPi ð0Þexpðβt Þ þ βexpðβt Þ expðβT Þui dT ð10:54Þ
0
where vPi(0) is a component of the particle velocity vector and ui is a component of the
fluid velocity vector that the entrained particle sees at time T. The first term on the right
side of (10.54) is the solution to the homogeneous part of (10.53) and the second term is
the particular solution. Squaring (10.54) and, then, taking the mean value, over many
particles, of each of the terms, gives
ðt
v2Pi ¼ v2Pi ð0Þexpð2βt Þ þ 2βexpð2βt Þ expðβT ÞvP ð0ÞuðT ÞdT
ðt
0
ð10:55Þ
path
þui βexpð2βt Þ Ri ðθÞ½expðβθ þ 2βt Þ expðβθÞdθ
2
0
where
ui ðt Þui ðt þ θÞ
Rpath
i ðθ Þ ¼ ð10:56Þ
u2i
The correlation Rpath

i ðθÞ is not an Eulerian or a Lagrangian correlation in that it considers
velocities of the fluid seen by an entrained particle as it moves around the field. In the
limit of large times, (10.55) gives
ð∞
v2Pi ¼ u2i β expðβθÞRpath
i ðθÞdθ ð10:57Þ
0
Suppose

Rpath
i ðθÞ ¼ expðαθÞ ¼ exp θ=τ path ð10:58Þ
where τ path
i ¼ α1 is the time-scale for this correlation. Then (10.57) gives the limiting
value of v2Pi

β
vPi ¼ ui
2 2 ð10:59Þ
αþβ
Since β is the reciprocal of the inertial time constant, equation (10.59) predicts that
particles with large inertial time constants (small β) will have much smaller turbulent
velocity fluctuations than the fluid. This is because the entrained particles do not
respond to high frequencies and thus “see” only a fraction of the fluid velocity
fluctuations.
Equation (10.55) is plotted in Figure 10.1 for the case of v2Pi ð0Þ ¼ 0. The abscissa, βt, is
the ratio of the time to the inertial time constant of the particles. Note that there is a
transient region in which the particle becomes entrained in the fluid turbulence. Its
duration is given as βt = 2−3. Thus, particles with large inertial time constants require
a long time to come to a stationary state.
1.2
1.0
α /β
0.8
14
1.7
/β + 1)
0.66
0.6
(α
u i2
vP2
0.4
0.2
0
0 1 2 3 4 5
βt
Figure 10.1 Particle turbulence as a function of the dimensionless time for the case of v2P ð0Þ ¼ 0.
10.7 Dispersion of particles in a turbulent gas 243
10.7 Dispersion of particles in a turbulent gas which is homogeneous

and isotropic
10.7.1 No effect of gravity

Friedlander & Johnstone (1957) integrated (10.53) twice, using Stokes law, to obtain the
following equation for the dispersion of entrained particles, from a point source, at large
times:
ðt
X 2P
εP ¼ ¼ ui Rpath
2
i dθ as t ! ∞ ð10:60Þ
2t 0
where u2i represents the fluid turbulence and Rpath

i is the correlation coefficient of fluid
velocity fluctuations seen by the entrained particle. If the correlation coefficient is
represented by (10.58), equation (10.60) yields
u2i u2 τ L ε ð ∞ Þ
εP∞ ¼ ¼ i L ¼ L ð10:61Þ
α ατ ατ
where τL is the Lagrangian time-scale for the fluid turbulence, defined by (10.9). For
homogeneous turbulence, the Eulerian average of the fluid velocity fluctuations at a
given location, u2i , is usually taken to be equal to the Lagrangian average over particle
paths. Thus, in the above equation, the fluid diffusivity ε(∞) is defined by (10.10).
If βτL is large, the particle closely follows the fluid (ατL = 1). Equations (10.59) and
(10.61) yield v2Pi ¼ u2i and εP(∞) = ε(∞). If βτL is small, the particles will not follow the
turbulence. In this limit, the particles will be almost stationary; they are, effectively,
sampling the turbulence at a fixed location; that is, α−1 = τE, where τ E is the Eulerian
time-scale, defined by (10.35). Theoretical work by Reeks (1977) gives τ E ¼ ð1 1:3Þτ L
as βτL varies from 2 to 0. Reeks, therefore, suggests εP∞ ≈ ε∞ .
This is consistent with the analysis of Hinze (1959, p. 466, equation 5.207), which uses
spectral representations of RLP ðt Þ and of RL(t) to suggest that
εP∞ ¼ ε∞ ð10:62Þ
The above result is remarkable since (10.59) predicts that the intensity of the particle
turbulence, v2Pi , can be much smaller than the intensity of the fluid turbulence, u2i , if the
inertial time-scale of the particles is large.
From Section 10.2, one can describe the diffusivity of particles at large times as the
product of a velocity and a length scale.
1=2
εP ð∞Þ ¼ v2Pi ΛPi ð10:63Þ
where the length scale is defined as

1=2
ΛPi ¼ v2Pi τ LP ð10:64Þ
For large inertial time constants, the velocity scale decreases with increasing 1/β. Hinze’s
suggestion that εP ≈ ε implies that ΛP must increase with increasing 1/β just enough for
εP(∞) to remain roughly constant. This can be explained because inertia of entrained
particles inhibits their ability to respond to high-frequency velocity fluctuations. Thus,
the entrained particles have larger time constants and, therefore, larger length scales.
The prediction of v2Pi in equation (10.59) depends on the representation of Rpath i by
(10.58) and the definition of an empirical reciprocal time constant α. As mentioned
above, some success in relating this correlation to Eulerian properties of the fluid
turbulence has been realized by using the iterative approach of Reeks (1977).
[Equation (10.62) is strictly applicable to a homogeneous turbulence, which is
approximated in the central region of a pipe or channel. In fact, computational results
exist which suggest that εP ≠ εf close to a wall where the turbulence is non-homogeneous.
See, for example, Brooke et al. (1994).]
10.7.2 Effect of gravity; heavy particles

The preceding section points out that heavy particles do not follow the fluid turbulence
exactly. Their inertia causes the root-mean-square of the turbulent velocity fluctuations to
be smaller than the root-mean-square of the velocity fluctuations of the fluid.
Furthermore, high-frequency fluid velocity fluctuations become less effective in dispers-
ing particles, so that the length scale characterizing their dispersion increases.
Yudine (1959) pointed out an additional effect due to the influence of gravity,
which he called a “crossing of trajectories.” Turbulent dispersion of fluid particles is
strongly related to turbulent structures in which fluid motion is highly correlated.
Heavy particles entrained in the fluid possess a free-fall velocity. Yudine recognized
that this velocity could cause the particles to move out of eddies responsible for
the dispersion of fluid particles. Thus, the velocity correlation of heavy particles
decreases more rapidly with time than does the correlation for fluid particles. This
“crossing of trajectories” effect can have a strong influence on the dispersion of heavy
particles.
However, the expectation is that the “crossing of trajectories” should have little, if any,
effect on the measured intensity of turbulent velocity fluctuations of the particles since it
only affects the method of sampling.
For the case of very large ratios of the free-fall velocity to the fluid turbulence,
1=2
U S = u2 , the turbulence seen by a particle may be considered as frozen. Thus, the
fluid velocity fluctuations seen by a particle are characterized by the Eulerian spatial
correlation for the fluid turbulence:

RLii ðsÞ ¼ REii x ¼ U S s ð10:65Þ
where x is the coordinate in the direction of free fall. The particle diffusivity for large
diffusion times is then given by
10.8 Measurements of particle turbulence for suspension flow in a pipe 245
ð∞
εP ¼ v2Pi REii ðx ¼ U S sÞdx ð10:66Þ
0
1=2
For very small U S = u2i and large diffusion times, the turbulent diffusivity of the
1=2
particles is given by (10.60). For intermediate U S = u2i , interpolation formulas
(Csanady, 1963; Meek & Jones, 1973) can be used.
For example, Csanady (1963) suggests that the correlation coefficient can be given by
1=2
s 2
RðsÞ ¼ exp U S þ v2P ð10:67Þ
Λi
1=2
Thus, for large U S = v2P

sU S
RðsÞ ¼ exp ð10:68Þ
Λi
1=2
For small U S = v2P
0 1=2 1
s v2P
B C
RðsÞ ¼ exp@ A ð10:69Þ
Λi
The analysis of Meek & Jones (1973) uses the spectral function. They argue that the average
wave number is unchanged by the crossing of trajectories. However, the convection velocity
from c0 to c0 þ U S . Similarly, the frequency changes from ω0 = kc0 to
changes
ω ¼ k U S þ c0 . Meek & Jones suggest the following interpolation formula for ω:
1=2
US
ω ¼ ω0 1 þ ð10:70Þ
c0
Thus, the crossing of trajectories can have a much greater effect on the spatial scale of the
turbulence seen by the particles than does particle inertia. The calculations performed by
Reeks (1977) indicate that the effect of crossing of trajectories starts to become important
when the free-fall
velocity is roughly equal to the magnitude of the turbulent fluctuations
U S =q ≈ 1 :
10.8 Measurements of particle turbulence for suspension flow in a pipe
10.8.1 The experiments of Lee et al.

Photographic measurements of particle turbulence have been made by Lee et al. (1989) and
by Young & Hanratty (1991). These were carried out for downward flow in a 5.08 cm pipe.
The experiments of Lee et al. were done by injecting drops of water, with diameters of
50–150 µm, into an air stream at the center of a pipe. Light sheets were formed at three
1.0
Re z (cm) LAUFER (u 2r )1/2
0.8 36 000 77.5
52 000 77.5
(v2Pr)1/2/v*
36 000 212.1
0.6 52 000 212.1
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0
(rt – r)/rt
1.2
LAUFER (u 2θ )1/2
Re z (cm)
1.0 36 000 77.5
52 000 77.5
36 000 212.1
(v2Pθ)1/2/v*
0.8
52 000 212.1
0.6
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0
(rt – r)/rt
Figure 10.2 Measurements of particle turbulence for 50 μm drops in air. Lee et al., 1989.
locations along the pipe by electronic flash units. A camera, located at the bottom of
the pipe, obtained axial-view photographs of the particles at illuminated cross-sec-
tions. The timing between flashes provided three photographs of particles as they
moved down the pipe. The distances traveled by the particles and the time interval
between flashes can be used to calculate a velocity. In this way, the motion of the
droplets in a pipe cross-section could be measured. This technique has two interesting
aspects: (1) The use of axial-viewing photography filters out the much larger axial
flow component so that accurate measurements of turbulent motion in planes perpen-
dicular to the mean flow direction are possible. (2) Direct (rather than the usual
indirect) measurements were made of local mass fluxes in the radial direction, needed
to define Eulerian eddy diffusion coefficients.
Figure 10.2 shows measurements of the dimensionless root-mean-square of the
1=2
radial and tangential components of the velocity fluctuations of 50 μm drops, v2Pr
10.8 Measurements of particle turbulence for suspension flow in a pipe 247
100
Averaged across the pipe
Averaged for rt 1 cm
Equation (10.59), ατL = 0.7
1/2
v 2Pr u2r
/
10–1
10–2 10–1 100
βτL
Figure 10.3 Spatially averaged ratio of the particle to fluid turbulence as a function of βτ L
1=2
and v2Pθ over a pipe cross-section. The curved line represents measurements of the
1=2
fluid velocity fluctuations, u2i =v , obtained by Laufer (1954) in a pipe flow. It is
noted that the turbulence of entrained particles is significantly lower than the fluid
turbulence. Figure 10.3 shows a plot of the spatially averaged ratio of particle turbulence
to fluid turbulence as a function of βτL, where τL is the Lagrangian time-scale of the fluid.

The fluid time-scale is obtained from (10.22) since ε ¼ v2r τ L . The solid curve is
calculated with (10.59), using α = 0.7/τL. This fitting parameter is reasonable since
it indicates that the time-scale of the fluid velocity fluctuations seen by the particles
(in the absence of gravitational effects) is approximately 30 % larger than the Lagrangian
fluid time-scale. From Figure 10.3, it is seen that particle inertia affects the intensity of the
particle velocity fluctuations for βτL less
than 1. No evidence of an effect of a crossing of
trajectories on v2r was observed, since U S =q 51.
Values of the average velocity in the r-direction, V r , were also obtained. These give a
direct measurement of the flux of particles since
N ¼ V rC ð10:71Þ
The Eulerian diffusion coefficient could be calculated from measurements of V r and C ðrÞ
since
V rC
εEP ðrÞ ¼ ð10:72Þ
∂C=∂r
The experiments of Lee et al. show that concentration profiles of particles downstream of
the injector could be approximated by a Gaussian function. Then (10.72) gives
V r X 2P
εP ¼ ð10:73Þ
r
Table 10.1 Estimates of turbulent diffusion coefficients
dP (µm) Re zP (cm) X 2P (measured) εP (cm2/s) εf (cm2/s) εP/εf
90 36 000 77.5 1.4 8.4 9.5 .088

90 52 000 77.5 1.4 13.4 13.1 1.02
150 36 000 114.0 1.32 5.9 9.5 0.62
150 52 000 114.0 1.0 8.9 13.1 .068
Source: Lee et al. (1989)
Furthermore, the measurements could be fitted with the equation

V r ¼ kr ð10:74Þ
where r is the radial location and k is an empirical constant.

Values of εP calculated in this way are presented in Table 10.1, taken from Lee et al.
1=2
(1989). These measurements show that εP is close to εf provided U S = v2Pr is less
than 0.85. The values of εP/εf = 0.62, 0.68 for 150 μm particles can be interpreted as due
to a “crossing of trajectories” effect.
10.8.2 The experiments of Young & Hanratty

The experiments of Young & Hanratty (1991) used a water flow. Glass or steel spheres
with a diameter of 100 μm were injected at the center of a 5.08 cm pipe. In one set of
experiments the cross-section of the pipe was illuminated over 3–5 mm lengths by flash
units. The positions of the particles were photographed at different distances from the
injector by a camera located at the bottom of the pipe. These experiments provided
measurements of X 2P versus time ¼ z=vPz . The Lagrangian turbulent diffusivity at large

times was obtained from the relation εP ¼ ð1=2Þ dX 2P =dt . In a second set of experi-
ments the velocity components of the particles in a plane perpendicular to the flow were
determined by using five colored flashes located immediately downstream of one
another. This allowed measurements of the particle acceleration, as well as the particle
velocity.
The equation of motion for a solid particle in a liquid is a situation for which the
assumption that ρP >> ρf might not be satisfied. Based on the work of Tchen, discussed in
Hinze (1959, 1975), Young & Hanratty (1991) recommended the following approxima-
tion for (10.25) and (10.52) for liquid–particle flows:

!
d v P ~! ρ ρ
¼ β u vP þ
P f
!
g ð10:75Þ
dt ρP
where
3C D ρf j!
u ! v Pj
β~ ¼ ð10:76Þ
2d P ð2ρP þ ρf Þ
Note that (10.76) is the same as (10.50) for small ρf. The relative velocity appearing in
(10.76) is given as
X
i¼3 2
j!
u !
2
v P j ¼ ðuz vPz Þ2 þ uti vtPi ð10:77Þ
i¼1
The second term (due to particle turbulence) on the right side was ignored so that
j!
u ! v P j ¼ juz vPz j. Young & Hanratty (1991) used their measurements of particle
turbulence to justify this assumption.
The experiments are characterized by βτ ~ L much greater than unity so the inertial time
constant was small enough for the particles to follow the fluid velocity fluctuations closely.
Thus, v2Pr ¼ u2r . One of the experiments, using stainless steel particles, was done with
Uþ ~ L
S ¼ 1:77 and βτ ¼ 53:8. This produced a value for the Eulerian particle diffusivity of
εP/εf = 0.57, where U þS is the average slip velocity between the solid particles and the fluid
(see Young & Hanratty, 1991, Table 4). This result, for large U þ S , is consistent with the
“crossing of trajectories” theory of Csanady (1963). Young & Hanratty support the
1=2
suggestion that this effect becomes important for U S = v2Pr ≥ 0:85.
10.9 Turbophoresis
10.9.1 The concept of turbophoresis

The results presented above consider the turbulence to be homogeneous. However, for
flow in a pipe or channel, the turbulence is changing in space. This can lead to a force that
causes particles to move from regions of high turbulence to regions of low turbulence
(Caporaloni et al., 1975; Reeks, 1983). This process is called turbophoresis. Caporaloni
et al. and Reeks derived the following equation for the turbophoretic velocity:
1 dvt2
V tp ¼ Pr
ð10:78Þ
β~ dr
where vtPr is the r-component of the turbulent velocity of the particle.

A straightforward derivation of this relation is presented by Young & Hanratty (1991).
From (10.75), if gravitational effects are ignored,
APi ¼ β~ðui vPi Þ ð10:79Þ
where APi is the acceleration of the particle in the i-direction and β~ is given by (10.76). For
very large ρP/ρf (particles in a gas)
3C D ρf j!
u ! v Pj
β~ ¼ ð10:80Þ
4d P ρP
Take an average of (10.79) at a given location and let i = r,

1
APr ¼ ður vPr Þ ð10:81Þ
β~
Since ur ¼ 0
1 1
V tp ¼ vPr ¼ APr ≈ APr ð10:82Þ
~
β β~
Thus, if APr is not equal to zero, a turbophoretic velocity, V tP ¼ vPr , exists.
10.9.2 Formulation in cylindrical polar coordinates

Consider the Eulerian representation of APr in cylindrical polar coordinates
dvPr v2Pθ
APr ¼ ð10:83Þ
dt r
where v2Pθ =r is the centripetal acceleration and dvPr =dt is the average of the substantial
derivative, DvPr/Dt.
dvPr ∂vPr ∂vPr vPθ ∂vPr ∂vPr
¼ þ vPr þ þ vPz ð10:84Þ
dt ∂t ∂r r ∂θ ∂z
If changes of the particle concentration are ignored (or if the concentration is very small),
conservation of mass dictates that

∂vPr vPr 1 ∂vPθ ∂vPz
0 ¼ vPr þ þ þ ð10:85Þ
∂r r r ∂θ ∂z
Add (10.85) to (10.84) and take a time average

dvPr 1 ∂ rvPr
2
1 ∂ðvPr vPθ Þ ∂ðvPr vPz Þ
¼ þ þ ð10:86Þ
dt r ∂r r ∂θ ∂z
From (10.83) and (10.86)

0 1
∂ rv2Pr
APr
vt2 1
¼ Pθ þ @ A þ 1 ∂ðvPr vPθ Þ þ ∂ðvPr vPz Þ ð10:87Þ
r r ∂r r ∂θ ∂z
Substitute v2Pr ¼ vPr 2 þ vt2

Pr , where vP is a turbulent velocity fluctuation. If the flow is
t
fully developed, z-derivatives can be ignored. Because of symmetry, θ-derivatives can

also be ignored. Then (10.87) simplifies to

vt2 1∂ rv2Pr 1 ∂ rv t2
Pr
APr ¼ Pθ
þ þ ð10:88Þ
r r ∂r r ∂r
Since v2r 55vt2

r
vt2 vt2 ∂vt2

APr ¼ Pθ
þ Pr þ Pr ð10:89Þ
r r ∂r
From (10.82)
!
1 ∂vt2 vt2 vt2
V tp ¼ Pr þ Pθ Pr ð10:90Þ
β~ ∂r r r
Pθ ≈ vPr (see Figure 10.2). Then (10.78) is obtained.

Assume vt2 t2
The measurements of dvP =dt presented by Young & Hanratty (1991) are plotted in
Figure 10.4, where runs 1 & 2 represent glass spheres at Re = 70 800 and at Re = 16 400.
Average particle acceleration, Arrt/v*
(a) 20
10
–10
–20
0.0 0.2 0.4 0.6 0.8 1.0
Dimensionless distance from the wall, y/rt
Average particle acceleration, Arrt/v*
(b) 20
10
Run 1
–10 Run 2
Run 3
Run 4
–20
0 40 80 120 160 200
Dimensionless distance from the wall, y +
Figure 10.4 (a) The temporal change of VR measured by Young & Hanratty. The solid curve is calculated with
(10, 86). (b) Measurements close to the wall. Young & Hanratty, 1991.
Runs 3 and 4 are for steel spheres at Re = 72 900 and Re = 15 700. The curves in
Figure 10.4a and b represent theoretical calculations. The interesting feature of these
measurements is that the sign of dvP =dt changes. For low flow rates, this change occurs at
y/rt ≈ 0.2 and for high flow rates, at y/rt ≈ 0.05, where y is the distance from the wall. This
can be understood better when measurements are plotted against y, made dimensionless
with wall parameters, as shown in Figure 10.4b. Here, it is seen that the change in sign
occurs in the range y+ = 80−120. This corresponds to the location of a maximum in the
root-mean-square of the radial velocity fluctuations.
Thus, turbophoresis exists if dvP =dt has a finite value. Then a force given by the right
side of (10.75) will be acting on the particle. This requires that vP ≠ 0.
10.9.3 Formulation in Cartesian coordinates

Several researchers have formulated the turbophoretic velocity in Cartesian coordinates.
These works are summarized by Mito & Hanratty (2004b). From (10.82), The turbopho-
retic velocity is exactly defined as
1 dv2P
V tp ¼ ð10:91Þ
~β dt
where v2P is the x2-component of the particle velocity.Young & Leeming (1997) derived
the following relation for a compressible suspension (equation (10.85) is not valid):
dv2P ∂v2P ∂vt2 ∂lnC t

¼ v2P þ 2P þ v viP ð10:92Þ
dt ∂x2 ∂x2 ∂xi 2P
where the repeated i-index indicates a summation. The third term on the right side
accounts for compressibility of the suspension. Cerbelli et al. (2001) suggested that it
∂ðlnC Þ
can be approximated by vt2
2P ∂x2 .
10.9.4 Physical interpretation of turbophoresis; effect on concentration profiles

Several people have investigated the effect of turbophoresis on concentration profiles.
They have argued that the turbophoretic flux is balanced by turbulent mixing.
dC
0 ¼ V tp C εEP ð10:93Þ
dy
Measurements reveal that the use of (10.93) requires that the Eulerian turbulent diffusivity
is negative in parts of the concentration field. This is non-physical. Mito & Hanratty
(2004b) and Brooke et al. (1994) have suggested an explanation for this paradox.
Define a flux, F, given by
dC
F ¼ V tp C εEP ð10:94Þ
dy
This need not be zero, as suggested by (10.93). Mito & Hanratty (2004b) carried out
numerical experiments in a channel flow for which the fluid turbulence is represented
by the Langevin equation (see Section 10.11). One aspect of the investigation by Mito &
Hanratty was an extensive study of turbophoresis. This work and the study by
Brooke et al. (1994) speculate that turbophoretic motions resemble “free-flight” deposi-
tion described in Section 10.10.
Particles entrained in a turbulent field tend to follow the turbulence. However,
if the particles experience a sudden decrease in the fluid turbulence, they will
become detached from the turbulence and undergo a free-flight. Thus, there will
be two populations of particles. One is entrained in the fluid turbulence and a
second, which is smaller, is undergoing free-flights. Particles can end their free-
flights by again becoming entrained in the turbulence or by depositing in the wall
region. The velocity of these free-flight particles is associated with the turbophoretic
velocity.
At a given y, particles can be leaving or entering the entrained population at rates of S0
and Si (in the units of mass per unit time per unit area). Thus, the equation describing the
variation of the flux with y is given, for a stationary state, as
dF
¼ Si S0 ð10:95Þ
dy
where F is given by (10.94) and a unit length in the spanwise direction is assumed. Mito
& Hanratty (2004b) suggest that Si and S0 represent contributions from a large-scale
velocity field similar to a secondary flow.
10.10 Deposition of particles
The design equation that is used to describe the rate of deposition on a unit wall area, RD, is
RD ¼ k D C B ð10:96Þ
where CB is the bulk composition and kD is the deposition coefficient, with units of
velocity. The most important parameter characterizing the deposition is the inertial time
constant τ = β−1, defined by (10.50).
For small τ+ = τv*2/ particles closely follow the fluid velocity fluctuations. Under
these circumstances, deposition occurs by Brownian motion, which is characterized by a
very large Schmidt number, Sc = /D. The equation for the deposition coefficient is
kD
¼ cScn ð10:97Þ
v
where c is a constant. The classical approach used to explain the effect of Schmidt
number on the average mass transfer rate is to assume an analogy between momentum
and mass transfer. This leads to the assumption that the eddy diffusivity varies with the
cube of the distance from the wall throughout the concentration boundary layer and a
Dimensionless deposition constant, kD/v*

1
10–1 kD/v* = 0.086 Sc–0.7
10–2 kD/v* = 0.17
10–3
kD/v* = 3.25 × 10–4τ+2

10–4
10–5
10–2 10–1 1 101 102 103 104 105 106
Dimensionless particle relaxation time, τP+
Figure 10.5 Summary of deposition data in vertical flows by McCoy & Hanratty. McCoy & Hanratty, 1977.
prediction that n = −2/3. However, experiments by Shaw & Hanratty (1977) and by Na
et al. (1999) show that n = −0.704. For Sc = 631−37 200,
kD
¼ 0:0889Sc0:704 ð10:98Þ
v
The argument that the eddy diffusion coefficient varies with the cube of the distance from
the wall is correct since uty / y2 and utx / y, where uty is the turbulent velocity component
perpendicular to the wall and utx is the turbulent velocity component parallel to the wall.
However, the assumption that this holds throughout the concentration boundary layer is
not correct. This prompted theoretical work by Vasiliadou & Hanratty (1988), which
shows, for extremely large Sc, that n = −3/4. Since Brownian motion is characterized by a
much larger Sc than studied by Shaw & Hanratty (1977), it is expected that n = −3/4
would be more appropriate for very small aerosol particles.
When τ þ P > ~0:2, particles do not follow the fluid turbulence exactly over the whole
field; they impinge on the wall by an inertial mechanism, called a “free-flight.” For
0:25τ þP 520, particles moving toward the wall disengage from the turbulence in the
viscous wall region where the magnitudes of the turbulent fluid velocities are decreasing
rapidly as the distance from the wall decreases. Thus, the rate of deposition increases
rapidly with increasing τ þP.
Figure 10.5 was created by McCoy & Hanratty (1977). It summarizes measurements
of kD made in vertical flows by Farmer (1969), Forney & Spielman (1974), Friedlander &
Johnstone (1957), Llori (1971), Liu & Agrawal (1974), Schwendiman & Postma (1961),
Sehmel (1973), Wells & Chamberlain (1967), Jagota et al. (1973), Cousins & Hewitt
(1968). At very small τ þ *
P , the dimensionless deposition coefficient, kD/v , is seen to be
þ þ
independent of τ P . (The particles mix by Brownian motion.) For 0:25τ P 522:9; k D =v
increases with τ þ2
P , as suggested by Kneen & Straus (1969) and by Liu & Agrawal
(1974). It is noted that, at a given τ þ
P , there can be an order of magnitude variation of the
measurements made in different laboratories. McCoy & Hanratty (1977) suggested that
the following equation, very roughly, captures the trend:
kD
¼ 3:25 104 τ þ2 ð10:99Þ
v P
In the range where (10.99) is applicable, there is a remarkably strong increase of kD with
gas velocity and with particle diameter, kD ∝ v*5 and k D / d 4P .
For a given particle size, kD/v* becomes independent of τ þ þ
P at very small τ P . The data
shown in Figure 10.5 are in this range for particles with diameters of 0.65 to 1.1 µm,
which correspond to Sc of 3.3 ×105 to 1.2 ×106.
At τ þ > ~ 22:9, the value of kD/v* appears to be relatively insensitive to particle diameter
or fluid velocity. McCoy & Hanratty suggest that an average fit to the data at large τ þ P is
kD
¼ 0:17 ð10:100Þ
v
Friedlander & Johnstone (1957) were the first to model deposition in the range
0:25τ þ P 520 as due to a free-flight mechanism. They pictured the particles as arriving at
a certain distance from the wall, yff, by turbulent diffusion and then moving to the wall by a
free-flight. Here, yff, the location at which the particles start a free-flight to the wall, is
defined as yff = vffτP. Friedlander & Johnstone assumed that vff is the root-mean-square of the
fluid velocity fluctuations at y+ = 30. Theoretical work had mainly involved modifications of
this theory. For example, Davies (1966) argued that vff is representative of the fluid velocity
fluctuations at yff, rather than y+ = 30, but predicted deposition rates which are too small.
Studies of particle motion in direct numerical simulations (DNS) of the Navier–
Stokes equation for turbulent flow in a channel have provided an opportunity to study
free-flights (Chen & McLaughlin, 1995; McLaughlin, 1991; Brooke et al., 1992, 1994).
These studies involved the seeding of the flow with spherical particles of uniform size,
and removing them from the field when they are located at one particle radius from the
wall. Their paths were calculated with (10.25) and (10.27).
Over the outer flow field the particles follow, but lag, the fluid velocity fluctuations.
Eventually, they move toward the wall where the fluid velocity decreases rapidly and the
particles disengage from the fluid turbulence as they move toward the wall with a higher
velocity than the surrounding fluid. The change from a situation where the particles lag
the fluid velocity fluctuations to one in which they are disengaged from the fluid velocity
fluctuations defines the initiation of a “free-flight.”
These free-flight particles can move directly to the wall or they can stop in a layer
closer to the wall where velocity fluctuations in a direction perpendicular to the wall are
very small. The trapping of particles at the wall has been noted by several investigators
(Kallio & Reeks, 1989; McLaughlin, 1989; Sun & Lin, 1986; Brooke et al., 1992).
Observations by Young & Hanratty (1991) are displayed in Figure 10.6. These authors
showed that the particles are trapped in a necklace formation at a distance of less than
one particle diameter from the wall. The location is dictated by a balance between
the Saffman lift force and a wall-induced force (Brenner, 1961; Maude, 1961;
Goldman et al., 1967).
Figure 10.6 Particles trapped at a wall. Young & Hanratty, 1991.
Several new or revised concepts arose from computer experiments (Brooke et al.,
1992, 1994):
(1) At a given distance from the wall, a small fraction of the particles, with the highest
velocities perpendicular to the wall, start a free-flight. Thus, the probability distri-
bution of the fluctuations of the velocity, rather than the root-mean-square of the
velocity fluctuations, is needed to interpret the phenomenon. This explains why early
theories on inertial impaction predicted that particles strike the wall at much smaller
velocities than is observed.
(2) Particles start their free-flight to the wall from a number of locations.
(3) Particles can stop a free-flight before they hit the wall. This can lead to their being
trapped in a region close to the wall where the fluid turbulence is small.
(4) Particles are defined as striking a smooth wall when their centers are at a distance of
one radius from the wall, where the fluid turbulence is very small, but not zero.
Although turbulent deposition is a possible fate of trapped particles (as suggested by
several researchers), the most likely scenario is that trapped particles are caught in a
large outward flow to regions away from the wall, where they have another chance to
experience a free-flight to the wall.
(5) Kallio & Reeks (1989) and Brooke et al. (1992) have suggested that the buildup of
particles in the viscous sub-layer ( y+ < 5) could be due to turbophoresis. However,
computer studies show that turbophoresis aids the motion of particles toward the
wall, but that accumulation is mainly due to free-flights.
(6) The optimum distance at which free-flight deposition starts is roughly y+ ≈ 9.
10.11 Stochastic representation of fluid turbulence 257
10.11 Stochastic representation of fluid turbulence
10.11.1 Prologue
Studies of particle motion in a direct numerical solution for the flow field are limited in
that the complexity of the calculations requires a consideration of small Reynolds
numbers, small diffusion times and a range of variables which is insufficient to capture
the physics completely. This has led to the use of stochastic models to represent
turbulence seen by the particles. By pursuing this approach, it has been possible to
study a wide range of conditions, to identify several mechanisms for deposition and to
explore methods to describe the concentration field. In the order of increasing
inertial time constant, particles strike the wall by (1) Brownian motion, (2) turbulent
diffusion, (3) free-flight, (4) free-flight from regions outside the viscous wall layer,
(5) unidirectional motion from wall to wall.
10.11.2 The use of a modified Langevin equation in channel flow

The Langevin equation has been explored in studies by Perkins (1992), Sommerfeld et al.
(1993), Pozorski & Minier (1988), Iliopoulos et al. (2003), Mito & Hanratty (2003).
In its simplest form, it consists of a damping force and a Wiener process which
calculates velocities of passive tracers by assuming they are approximated by a
Markov sequence. Lin & Reid (1963) and Obukhov (1959) applied the Langevin
equation to homogeneous, isotropic turbulence. The change of the velocity of a tracer
over a time interval dt is given by
ui
dui ¼ dt þ dμi ð10:101Þ
τi
where τi is a time constant. The forcing function, dμi, has a zero mean, and fluctuations
are given by a Gaussian function. This equation provides the same result as obtained by
Taylor (1921) in his analysis of dispersion from a point source if the Lagrangian
correlation is represented by expðt=τ L Þ and τi = τL, the Lagrangian time-scale.
The Langevin equation has been adapted to describe dispersion of fluid particles in
non-homogeneous fields by Durbin (1983, 1984), Hall (1975), Iliopoulos & Hanratty
(1999), Legg & Raupach (1982), Reid (1979), Reynolds (1997), Thomson (1984, 1986,
1987), van Dop et al. (1985) and Wilson et al. (1981). The analysis by Mito & Hanratty
(2004b) used the approach of Wilson et al. (1981) and of Thomson (1984),

ui ui
d ¼ þ dμi þ dμ0i ð10:102Þ
σi σiτi
where the forcing function, the time-scale, τi, and the root-mean-square of the
velocity fluctuations, σi, are functions of x2. For a non-homogeneous flow, the forcing
function, dμi, consists of a mean component, dμi , and a fluctuating component, dμ0i .
The fluctuating component is assumed to be Gaussian. A number of investigators have
shown that dμi must be non-zero in order to avoid spatial accumulations which are not
physical. Equations for the mean drift, dμi , and the covariance, dμ0i dμ0j , are derived from
equation (10.102) by neglecting terms of higher order than dt (Iliopoulos & Hanratty,
1999; Mito & Hanratty, 2002).

u2 ui
∂
σi
dμi ¼ dt ð10:103Þ
∂x2

ui uj 1 1
dμ0i dμ0j ¼ þ dt ð10:104Þ
σiσj τi τj
A method for evaluating the unknowns in (10.102), (10.103) and (10.104) is outlined in
Mito & Hanratty (2002, 2003, 2004a), who considered flow in a rectangular channel of
infinite extent in the spanwise direction. The focus was vertical annular flow.
Woodmansee & Hanratty (1969) have shown that atomization of wall films occurs by
rapid growth and removal of capillary waves which create drops that are entrained by the
turbulence in a region outside the viscous wall layer. This process was represented by
introducing drops at a short distance from the wall with a velocity characteristic of
the0þvelocity fluctuations
outside the viscous wall layer. Injection
0þ velocities of
V 1 ; V 2 ; V 3 ¼ ð15; 1; 0Þ were used at the bottom wall and of V 1 ; V 2 ; V 3 ¼
0þ 0þ 0þ 0þ
ð15; 1; 0Þ at the top wall. Equations (10.27), (10.52) were used to calculate the
subsequent velocities and locations of these particles. The fluid velocity fluctuations
encountered by the particles were represented by (10.102). The influence of gravity was
ignored since vertical flow was considered. The influences of collisions and of feedback
particle forces on the turbulence were not considered. For calculations in which
Brownian motion is important, the approach adopted by Ounis et al. (1991) and by
Chen & McLaughlin (1995) was used.
The drag coefficient, CD, was represented by
24
CD ¼ 1 þ 0:15 Re0:687
P ð10:105Þ
ReP
where the particle Reynolds number, ReP, is defined with dP and the magnitude of the
relative velocity j!
u !v P j. The dimensionless inertial time constant of a particle is
defined as
4d þ
P ðρP =ρf Þ
τþ
P ¼ ð10:106Þ
3C j! u !
þ þ
D v j P
where !
u is the fluid velocity and !
v P is the particle velocity. For a Stokes law resistance
d þ2
P ðρP =ρf Þ
τþ
PS ¼ ð10:107Þ
18
In the non-Stokes region, τ þP is a function of x2. Thus, the volume-averaged inertial time
constant, τ þ
P , is a more appropriate parameter to describe particle turbulence:
ð 2H
1
τþ
PB ¼ τþ
P C ðx2 Þdx2 ð10:108Þ
2HC B 0
where CB is defined as
ð 2H
1
CB ¼ C ðx2 Þ dx2 ð10:109Þ
2H 0
The rates of injection of particles at the two walls are calculated as NB/AΔt, where A is the
area of the wall over which particles are discharged and Δt is the time interval over which
the NB particles are admitted from a wall source. Particles are deposited (removed from
the field) when their centers are located one particle radius from the wall. The run is
terminated when a stationary state is reached. (There is no further change in the concen-
tration field and the rate of injection, RA, equals the rate of deposition, RD.)
10.11.3 Identification of mechanisms for deposition

Stationary states were studied for two sets of calculations: In one of these, the dimension-
less particle diameter was kept constant at d þ
P ¼ 0:368 and the particle time constant was
varied by changing ρP/ρf (see Table 10.2). Results were obtained for τ þ PS ¼ 1 25 000.
This led to a consideration of unreasonably large ρP/ρf. In the second set of calculations,
Table 10.2 Stokesian and bulk-mean inertial time constants, bulk-mean particle Reynolds numbers for
constant inertial time constant and for constant density ratio
dþ
P ¼ 0:368 ρP/ρf = 1000
τþ
PS τþ
PB RePB ρP/ρf τþ
PS τþ
PB RePB dþ
P
1 0.977 0.0748 1.33 × 102 1 0.988 0.0274 0.134

3 2.91 0.118 4.00 × 102 3 2.92 0.0740 0.232
5 4.82 0.147 6.65 × 102 5 4.84 0.118 0.300
10 9.55 0.181 1.33 × 103 10 9.52 0.214 0.424
20 18.9 0.242 2.65 × 103 20 18.5 0.441 0.600
40 37.2 0.367 5.30 × 103 40 35.2 0.901 0.848
100 91.4 0.536 1.33 × 104 100 81.9 1.87 1.34
250 226 0.643 3.32 × 104 250 187 3.45 2.12
500 449 0.712 6.65 × 104 500 344 5.27 3.00
1000 888 0.810 1.33 × 105 1000 624 7.93 4.24
2500 2120 1.34 3.32 × 105 2500 1310 14.2 6.70
5000 4150 1.62 6.65 × 105 5000 2120 25.8 9.48
10,000 8260 1.71 1.33 × 106 10000 3180 52.1 13.4
25,000 20500 1.76 3.32 × 106 25000 6070 89.8 21.2
50000 9960 131 30.0
100000 16200 189 42.4
Source: Mito & Hanratty (2004b)

(a) 100
10–1 σP+/2
10–2
k +DB, VD+
10–3 dP+ = 0.368

No Brownian motion
10–4 k+DB VD+
Brownian motion
10–5 0.0889Sc–0.704
k+DB VD+
10–6
100 101 102 103 104
τ+PB
(b) 100
10–1 σP+/2
10–2
k +DB, VD+
ρP / ρf = 1000
10–3
No Brownian motion
10–4 k+DB VD+
0.0889Sc–0.704 Brownian motion
10–5
k+DB VD+
10–6
100 101 102 103 104
τ+PB
Figure 10.7 (a) Deposition constants for dþ

P = 0.368 (b) Deposition constants for ρP =ρf ¼ 1000. The fluid
turbulence is represented by a Langevin equation. Mito & Hanratty, 2004b.
ρP/ρf was kept constant and the dimensional constant was varied by changing d þ P . This
led to a consideration of unreasonably large particle sizes (compared with the mesh size
used in the calculation). The influence of Brownian motion was considered only for
τþ
PS ¼ 1; 3; 5.
Figure 10.7a, for d þ
P ¼ 0:368, presents calculations of mean values of the velocity with
which particles are depositing, VD . For τ þ PS ¼ 1; 3; 5, particles are striking the wall with
þ
very small velocities, 104 5V D 5103 . These are of the order of the wall-normal fluid
velocity fluctuations at d þ P =2. For this range of variables, particles trapped close to the
wall appear to be depositing by turbulent diffusion. For τ þ PS >> 5, the main contributors
to deposition are particles which have velocities which are characteristic of the
fluid outside the viscous sub-layer. Particles with these velocities are pictured to move
in “free-flight” through the layer of trapped particles close to the wall.
In the diffusion regime, the mean deposition velocity of the particles is not terribly
sensitive to changes in τ þ þ
PB . The slight decrease with increasing τ PB reflects a decrease in
the ability of the particles to follow fluid velocity fluctuations. The increase in V þ D in the
region 55τ þ PB 5 250 occurs because, on average, “free-flight” to the wall can be initiated
at larger distances from the wall as τ þ PS increases.
Values of k þD ¼ k D =v
are also plotted in Figure 10.7. A comparison of the calculated
þ
k D in Figure 10.7 with the plot of experimental results in Figure 10.5 shows agreement
within the spread of the experimental data.
6
4
0.1
kDB / V 20
8 Reτ = 590
6 Reτ = 300
Reτ = 150
4
V20+ = 0.5
2 V20+ = 1
V20+ = 2
0.01
8
0.01 0.1 1 10 100
τPB+V 20/2H
Figure 10.8 Effect of the inertial time constant τ PB V 02 =2H on the dimensionless deposition coefficient, for
dþ
P ¼ 0:368. The fluid turbulence is represented by a Langevin equation. Mito & Hanratty, 2007.
Hay et al. (1996) and Hanratty et al. (2000) show that, for particles starting free-flights
outside the viscous wall layer,
σP
k D ¼ pffiffiffiffiffi ð10:110Þ
2π
where σP is the root-mean-square of the wall-normal velocity fluctuations of the particles.

Equation (10.110) provides the average velocity of particles moving toward the wall at
y+ = 40. It decreases with increases in the inertial time constant since the particles
experience increasing difficulties in following the fluid velocity fluctuations. Equation
(10.110) is compared with the calculations in Figures 10.7 a, b. Good agreement is noted
for a range of conditions characteristic of annular flow.
Studies of the behavior of single wall sources by Mito & Hanratty (2003) show that
injected particles are transported to a distance slightly less than the stopping distance in a
stationary fluid
xþ 0þ þ
2stop ≈ 0:65V 2 τ PS ð10:111Þ
where they mix with the turbulence.

þ
Equation (10.111) suggests that the sudden change in the dependencies of V þ D and k D on
the inertial time constant at τ þ
PB ≈ 1800, shown in Figure 10.7, suggests a change in the
behavior of the particles in that their motion is not influenced by fluid turbulence. Under
these circumstances, the injection velocity is playing a strong role. At very large τ þ PB , the
deposition velocity equals the velocity at which the particles are injected into the field.
Deposition then occurs by the trajectory mechanism discussed by Anderson & Russel
(1970), Andreussi & Azzopardi (1983) and by Chang (1973). The system discussed
above was used by Mito & Hanratty (2007) to provide quantitative criteria for determining
when a trajectory mechanism will define the rate of deposition at very low concentrations.
They suggested that, in order to capture the trajectory regime, it is appropriate to
use V 02 and 2H as the velocity and length scales, rather than v* and /v*. A plot of k DB =V 02
versus τ PB V 02 =2H is presented in Figure 10.8. Note that the time it would take for a
particle with a velocity V 02 to move from wall to wall is 2H=V 02 , if fluid drag is ignored.
The calculations shown in Figure 10.8 for three particle Reynolds numbers and three
injection velocities fall on a single curve if τ PB V 02 =2H is greater than a number slightly
larger than 1. At τ PB V 2 =2H greater than about 10, k DB =V 02 is a constant equal to 0.5.
This follows since, in this limit, RD ¼ 0:5V 02 C B and k DB ¼ RD =C B ¼ 0:5V 02 . Thus, in
the region τ PB V 02 =2H > 1:2, the trajectories and the deposition are not affected by fluid
turbulence. The decrease of k DB =V 02 with decreasing τ PB V 02 =2H is associated with mean
resistance of the fluid to the motion of the particles.
The influence of Brownian motion is treated the same as molecular motion in that the
flux in the i-direction, due to a concentration gradient, is represented by Fick’s law
∂C
Fluxx ¼ D ð10:112Þ
∂x
As indicated in Table 10.2, the series of calculations carried out by Mito & Hanratty
(2004a) for constant ρP/ρf offer the opportunity to study the effect of Brownian motion.
Chen & McLaughlin (1995) show that the Schmidt number is given as
3π3 ρf d þ
Sc ¼ ¼ P
ð10:113Þ
D C c kT v
where Cc is the Cunningham slip factor, k is the Boltzmann constant, T is the absolute
temperature, D is the diffusion coefficient for Brownian motion and the gas is assumed
to be air at atmospheric conditions. This indicates increases in D with decreasing
particle diameter. Because of this, Brownian motion could show an important effect at
τþ þ
PS ¼ 1; 3; 5 in Figure 10.7b. At small enough τ PB Brownian motion could be controlling
deposition. The rate should then be similar to that found for turbulent mass transfer of
molecular species at large Schmidt numbers. A comparison of (10.98) with calculations
of the type summarized in Table 10.2 is awkward because D need not be a constant.
Thus, from (10.113), it is seen that Sc varies with d þ *
P and v . Mito & Hanratty (2004b)
followed the lead of Chen & McLaughlin(1995) by considering a constant friction
velocity of 0.6 m/s. The Schmidt number is considered to be a constant in the calculations
for d þ
P ¼ 0:368 given in Figure 10.7a.
Chen & McLaughlin (1995) studied particle deposition in a DNS of turbulent flow in a
channel at a very low Reynolds number, Reτ = 125. The analysis included the effects of
Brownian motion. The friction velocity was assumed to be 0.6 m/s. Calculations of
kþD are comparable with calculations of Mito & Hanratty (2004b) for ρP/ρf = 1000,
v* = 0.6 m/s. (This corresponds to a Schmidt number variation of 1.6 × 106 to 5 × 106
for τPS = 1, 3, 5.) The dashed curve on the bottom of Figure 10.7a represents (10.98).
The calculations of Chen & McLaughlin extrapolate to (10.98) at τ þ P ¼ 1. The Mito &
Hanratty calculations suggest that a smaller τ þ P is needed for Brownian motion to
control.
This could reflect the very small Reynolds number used by Chen & McLaughlin.
Indirect evidence for this is that their predicted k þ þ
D at large τ PB are smaller than the
calculations by Mito & Hanratty and available laboratory measurements.
References 263
(a) 105
τPS+ = 1, Brownian τPS+ = 1
104 τPS+ = 3
τPS+ = 3, Brownian
τPS+ = 5
Cv*/RAb
103 τPS+ = 5, Brownian

τPS+ = 10
102 τPS+ = 20
τPS+ = 40
101 τPS+ = 100
τPS+ = 250
100
0 20 40 60 80
x2+
(b) 105
τPS+ = 1, Brownian
Cv*/RAb
102
101
100
1 10 100 600
dP+/2
x2+
Figure 10.9 Concentration profiles for ρP =ρf ¼ 1000. The fluid turbulence is represented by a modified
Langevin equation. Mito & Hanratty, 2004b.
10.11.4 Concentration profiles

Calculated concentration profiles are presented in Figure 10.9a,b for ρP/ρf = 1000.
A logarithmic abscissa is used in Figure 10.9b. A range of τ þ PS ¼ 1 250 is covered.
The injection process causes small bumps at x = 13, 26, 65 for τ þ
+
PS ¼ 20; 40; 100. No
evidence of this influence is seen for τ þ
PS ≤ 10. A striking feature of these profiles is the
accumulation of particles at xþ
2 less than 1.
References
Anderson, R.J. & Russel, T. W. F. 1970 Circumferential variation of interchange in horizontal

annular two-phase flow. Ind. Eng. Chem. Fund. 9, 340–344.
Andreussi, P. & Azzopardi, B. J. 1983 Droplet deposition and interchange in annular two-phase
flow. Int. J. Multiphase Flow 9, 681–695.
Basset, A.B. 1888 Treatise on Hydrodynamics, Vol. 2. New York: Dover.
Beyerlein, S.W., Cossman, R.K. & Richter, H.J. 1985 Prediction of bubble concentration profiles
in vertical turbulent two-phase flow. Int. J. Multiphase Flow 11, 629–641.
Boothroyd, R. G. 1967 Turbulence characteristics of the gaseous phase in duct flow of a suspension
of fine particles. Trans. IChemE 45, T297.
Brenner, H. 1961 Slow motion of a sphere through viscous fluid toward a plane surface. Chem.
Eng. Sci. 16, 242.
Brooke, J.W., Kontomaris, K., Hanratty, T. J. & McLaughlin, J. B. 1992 Turbulent deposition and
trapping of aerosols at a wall. Phys. Fluids A 4, 825–834.
Brooke, J. W., Hanratty, T. J. & McLaughlin, J. B. 1994 Free-flight mixing and deposition of
aerosols. Phys. Fluids 6 (10), 3404–3415.
Cerbelli, S., Giusti, A. & Soldati, A. 2001 ADE approach to predict dispersion of heavy particles in
wall-bounded turbulence. Int. J. Multiphase Flow 27, 1861–1879.
Caporaloni, M., Tampieri, R., Trombetti, F. & Vittori, O. 1975 Transfer of particles in nonisotropic
air turbulence. J. Atmos. Sci. 32, 565–568.
Chang, D. R. 1973 The generation and deposition of droplets in annular two-phase flow. Ph.D.
thesis, University of Delaware.
Chen, M. & McLaughlin, J. 1995 A new correlation for aerosol deposition in vertical ducts.
J. Colloid Interface Sci. 169, 437–455.
Cousins, L.B. & Hewitt, G.F. 1968 Liquid mass transfer in annular two-phase flow to obtain
interchange and entrainment: droplet deposition and liquid entrainment. UKAEA Report No.
AERE-R5657.
Csanady, G.T. 1963 Turbulent diffusion of heavy particles in the atmosphere. J. Atmos. Sci. 20,
201–208.
Davies, C.N. 1966 Deposition of aerosols from turbulent flow through pipes. Proc. R. Soc. Lon. A
239, 235.
Durban, P. A. 1983 Stochastic differential equations on turbulent dispersion. NASWA Reference
Publication 1103.
Durban, P. A. 1984 Comment on papers by Wilson, et al. 1981 and by Legg and Raupach 1982.
Boundary-Layer Meteor. 29, 409–411.
Einstein, A. 1905 On the movement of small particles suspended in stationary liquids required by
the molecular-kinetic theory of heat. Ann. Physik (Ser. 5) 17, 549–560.
Farmer, R.A. 1969 Liquid droplet trajectories in two-phase flow. Ph.D. thesis, Massachusetts
Institute of Technology.
Forney, L.F. & Spielman, L.A. 1974 Deposition of coarse aerosols from turbulent flow. J. Aerosol
Sci. 5, 257–271.
Friedlander, S.K. & Johnstone, H.F. 1957 Deposition of suspended particles from turbulent gas
streams. Ind. Eng. Chem. 49, 1151–1156.
Goldman, A.I., Cox, N.G. & Brenner, H.B. 1967 Slow viscous motion of a sphere parallel to a
plane wall: II Couette flow. Chem. Eng. Sci. 22, 653–660.
Groenhof, H.C. 1970 Eddy diffusion in the central region of turbulent pipe flows in pipes and
between parallel plates. Chem. Eng. Sci. 25, 1005.
Hall, C.D. 1975 The simulation of particle motion in the atmosphere by a numerical random walk
model. Q. J. R. Met. Soc. 101, 235–244.
Hanratty, T.J., Woods, B.D., Iliopoulos, I. & Pan, L. 2000 The roles of interfacial stability and particle
dynamics in multiphase flows: a personal viewpoint. Int. J. Multiphase Flow 26, 169–190.
Hay, K.J., Liu, Z.C. & Hanratty, T.J. 1996 Relation of deposition to drop size when the rate law is
non-linear. Int. J. Multiphase Flow 22, 829–848.
Hinze, J.O. 1959 Turbulence. New York: McGraw-Hill.
Hinze, J.O. 1975 Turbulence, 2nd edn. NewYork: McGraw-Hill.
Iliopoulos, I. & Hanratty, T.J. 1999 Turbulent dispersion in a non-homogeneous field. J. Fluid
Mech. 392, 45–71.
Iliopoulos, I., Mito, Y. & Hanratty, T.J. 2003 A stochastic model for solid particle dispersion in a
non-homogeneous turbulent field. Int. J. Multiphase Flow 29, 375–394.
References 265
Jagota, A.K., Rhodes, E. & Scott, D.S. 1973 Tracer measurements in two-phase annular flow to
obtain interchange and entrainment. Can. J. Chem. Engng 51, 139–148.
Kallio, G.A. & Reeks, M.W. 1989 A numerical simulation of particle deposition in boundary
layers. Int. J. Multiphase Flow 15, 433–446.
Kneen, T. & Straus, W. 1969 Deposition of dust from turbulent gas streams. Atmos. Env. 3, 55–67.
Laufer, J. 1954 The structure of turbulent fully developed flow. NACA Report No. 1174.
Lee, M.M., Hanratty, T.J. & Adrian, R.J. 1989 An axial viewing photographic technique to study
turbulence characteristics of particles. Int. J. Multiphase Flow 15, 787–802.
Legg, B.J. & Raupach, M.R. 1982 Markov-chain simulation of particle dispersion in homogeneous
flows: The mean drift velocity induced by a gradient in Eulerian velocity variance. Boundary-
Layer Met. 24, 3–13.
Lin, C.C. & Reid, W.H. 1963. Turbulent flow, theoretical aspects. In Encyclopedia of Physics, Vol.
8/2. Berlin: Springer-Verlag, pp. 438–542.
Liu, B.Y. & Agarwal, J.K. 1974 Experimental observations of aerosol deposition in turbulent flow.
J. Aerosol Sci. 5, 145–155.
Llori, T.A. 1971 Turbulent deposition of aerosol particles inside pipes. Ph.D. thesis, University of
Minnesota.
Maude, A.D. 1961 End effects in a falling film viscometer. Brit. J. Appl. Phys. 12, 293–295.
Maxey, M.R. & Riley, J.J. 1983 Equation of motion for a small rigid sphere in a non-uniform flow.
Phys. Fluids 26, 883–889.
McCoy, D.D. & Hanratty, T.J. 1977 Rate of deposition of droplets in annular two-phase flow. Int. J.
McLaughlin, J.B. 1989 Aerosol particle deposition in numerically simulated channel flow. Phys.
Fluids A1, 1211.
McLaughlin, J.B. 1991 Inertial migration of a small sphere in linear shear flows. J. Fluid Mech. 22,
261–274.
Meek, C.C. & Jones, B.G. 1973 Studies of the behavior of heavy particles in a turbulent fluid flow.
J. Atmos. Sci. 30, 239–244.
Mei, R., Adrian, R.J. & Hanratty, T.J. 1991 Particle dispersion in isotropic turbulence under Stokes
drag and Basset force with gravitational settling. J. Fluid Mech. 225, 481–495.
Mito, Y. & Hanratty, T.J. 2002 Use of a modified Langevin equation to describe turbulent
dispersion of fluid particles in a channel flow. Flow, Turbulence, Combustion 68, 1–26.
Mito, Y. & Hanratty, T.J. 2003 A stochastic description of wall sources in a turbulent field: Part 1.
Verification. Int. J. Multiphase Flow 29, 1373–1394.
Mito, Y. & Hanratty, T.J. 2004a A stochastic description of wall sources in a turbulent field: Part 2.
Calculation for a simplified model of horizontal annular flow. Int. J. Multiphase Flow 30, 803–825.
Mito, Y. & Hanratty, T.J. 2004b Concentration profiles in a turbulent suspension when gravity is
not affecting deposition. Int. J. Multiphase Flow 30, 1311–1336.
Mito, Y. & Hanratty, T.J. 2007 Trajectory mechanism for particle deposition in turbulent flows. Int.
Na, Y., Papavassiliou, D.V. & Hanratty, T.J. 1999 Use of direct numerical simulation to study the
effect of Prandtl number on temperature fields. Int. J. Heat Fluid Flow 20, 187–195.
Obukhov, A.M. 1959 Description of turbulence in terms of Lagrangian variables. Adv. Geophys. 6,
113–116.
Ounis, H., Ahmadi, G. & McLaughlin, J.B. 1991 Dispersion and deposition of Brownian
particles from point sources in a simulated turbulent channel flow. J. Colloid Interface Sci.
147, 233–250.
Perkins, R.J. 1992 The entrainment of heavy particles into a plane turbulent jet. Proceedings of the
6th Workshop on Two-phase Flow Predictions, ed. M. Sommerfeld. Erlangen: Forschungszentrum
Julich, pp. 18–33.
Pozorski, J. & Minier, J.P. 1998 On the Lagrangian turbulent dispersion models based on the
Langevin equation. Int. J. Multiphase Flow 24, 913–945.
Reeks, M.W. 1977 On the dispersion of small particles suspended in an isotropic turbulent fluid. J.
Fluid Mech., 83, 529–546.
Reeks, M.W. 1983 The transport of discrete particles in inhomogeneous Turbulence. J. Aerosol
Science 14, 729–739.
Reid, J.D. 1979 Markov chain simulations of vertical dispersion in a neutral surface layer for
surface and elevated releases. Boundary-layer Meteorol. 16, 3–22.
Reynolds, A.M. 1997 On the application of Thomson’s random flight model to the prediction of
particle dispersion within a ventilated air pace. J. Wind. Eng. Ind. Aerodyn. 67–68, 627–638.
Saffman, P.G. 1965 The lift of a small sphere in a slow shear. J. Fluid Mech. 2, 340–385.
Saffman, P.G. 1968 Corrigendum to the lift on a small sphere in a slow shear flow. J. Fluid Mech.
31, 624.
Schlichting, H. 1979 Boundary-layer Theory. New York: McGraw-Hill.
Schwendiman, L.C. & Postma, A.K. 1961 USAEC Report. HW-65309.
Sehmel, G.A. 1973 Particle eddy diffusivities and deposition velocities for isothermal flow and
smooth surfaces. J. Aerosol Sci. 4, 145–155.
Shaw, D.A, & Hanratty, T.J. 1977 Turbulent mass transfer rates to a wall for large Schmidt
numbers. AIChE Jl 23, 28–37.
Sommerfeld, M., Kohnen, G. & Ruger, M. 1993 Some open questions And inconsistencies of
Lagrangian particle dispersion models. Proceedings of the Ninth Symposium on Turbulent Shear
Flows, Kyoto, Japan, Paper No. 15–1.
Sun, Y.F. & Lin, S.P. 1986 Aerosol concentration in a turbulent boundary-layer flow. J. Colloid
Interface Sci. 113, 315–320.
Taylor, A.R. & Middelman, S. 1974 Turbulent dispersion in drag reducing fluids. AIChE Jl 20,
454–461.
Taylor, G.I. 1921 Diffusion by continuous movements. Proc. Lond. Math. Soc. 151, 196–211.
Thomson, D.J. 1984 Random walk modeling in inhomogeneous turbulence. Q. J. R. Met. Soc.110,
1107–1129.
Thomson, D.J. 1986 A random walk model of dispersion in turbulent flows and its application to
dispersion in a valley. Q. J. R. Met. Soc. 112, 511–530.
Thomson, D.J. 1987 Criteria for the selection of stochastic models of particle trajectories in
turbulent flows. J. Fluid Mech. 180, 529–556.
Towle, W.L. & Sherwood, T.K. 1939 Mass transfer in the central portion of a turbulent air stream.
Ind. Eng. Chem. 31, 457.
van Dop, H., Nieustadt, F.T.M. & Hunt, J.C. R. 1985 Random walk models for particle displace-
ments in inhomogeneous turbulent flows. Phys. Fluids 28, 1639–1653.
Vasiliadou, E. & Hanratty, T.J. 1988 Turbulent transfer to a wall at large Schmidt numbers. In
Transport Phenomena in Turbulent Flows Washington, DC: Hemisphere.
Wells, A.C. & Chamberlain, A.C. 1967 Transport of small particles to vertical surfaces. Brit. J.
Appl. Phys. 18, l793–1799.
Wilson, J.D., Thurtell, G.W. & Kidd, G.E. 1981 Numerical simulation of particle trajectories in
inhomogeneous turbulence II: Systems with variable turbulent velocity scale. Boundary-layer
Meteorol. 21, 423–441.
References 267
Woodmansee, D.E. & Hanratty, T.J. 1969 Mechanism for the removal of a droplet from a liquid
Young, J.B. & Hanratty, T.J. 1991 Transport of solid particles at a wall in a turbulent flow. AIChE Jl
37, 1529–1536.
Young, J. & Leeming, A. 1997 A theory of particle deposition in turbulent pipe flow. J. Fluid
Mech., 340, 129–159.
Yudine, M.I. 1959 Physical considerations on heavy-particle diffusion. Adv. Geophys. 6, 185–191.
11 Vertical annular flow
11.1 Prologue
In the annular pattern, part of the liquid flows as a film along the wall and part flows as
drops entrained in the gas. The interfacial stress varies with the flow rate of the film. Thus,
the pressure gradient depends on the fraction of the liquid flow, E, entrained as drops in the
gas. A predictive approach is to view E as resulting from a balance between the rate of
atomization of the liquid film, RA, and the rate of deposition of drops, RD. Thus, measure-
ments of RA and RD are a priority. A knowledge of drop size is of importance since it is
needed to predict drop turbulence and the influence of gravity on the motion of drops.
Discussions of the initiation of annular flow, the initiation of atomization, the proper-
ties of waves at the interface, the prediction of interfacial stress and the prediction of film
height have been presented in previous chapters. It is useful to give a brief review.
Annular flow is observed for air and water flowing upward in a vertical pipe for
V GS ¼ uGS ρG ½gd t ðρL ρG Þ1=2 > ~1

1=2
ð11:1Þ
Thus, for air–water flow at atmospheric conditions, transition occurs at uGS of the order of
10 m/s. As discussed in Sections 2.7.4 and 3.2.3, this criterion can be explained by
suggesting that the transition from vertical annular flow to churn flow is associated with a
flooding mechanism whereby stresses associated with gas flow over the film cannot
balance the force of gravity on the wall film.
For very small liquid flows, the thin liquid film on the wall is covered with capillary
waves, as described in Section 7.7. These waves received their energy from wave-
induced variations of the shear stress over the interface, which cause the waves to tumble.
Thus, this instability does not lead to the removal of liquid from the film. Since this
laminar film is disturbed by waves, the average film height is given by (3.46), rather than
the classical relation for an undisturbed rectilinear shear flow. The waves create a
roughened interface; the interfacial friction factor is given by (3.44) or (3.47).
Highly disturbed ringlike patches appear when the film Reynolds number reaches a
value of approximately 280 (see Figure 3.3). These disturbances look like collections of
capillary waves. Their thicknesses are much larger than the base film over which they
propagate. Because of this, the waves within a disturbance induce pressure variations
along the wave surface (in addition to shear stress variations), which can cause them to
grow to form drops by a Kelvin–Helmholtz mechanism.
11.2 Distribution function for drop size
Thus, atomization of the wall film is initiated with the appearance of disturbance waves.
The mechanism for the creation of drops was studied in a transparent horizontal rectangular
channel by obtaining high-speed photographs (3500–6000 frames per second) with a
camera located beneath the bottom wall (Woodmansee & Hanratty, 1969). Woodmansee
gives the following description of the atomization process, shown in Figure 11.1.
One of these ripples suddenly accelerates to the front of the disturbance or roll wave. The central
section of the ripple is lifted by the air stream having one or both ends connected to the flowing
liquid film. As the detached region of the tube of liquid is lifted, it is blown into an arc which
narrows as it stretches downstream until it ruptures into a number of pieces. Though difficult to
Ripples being lifted

and shattered by 0.0 ms
60 ft/s airstream
Roll
wave
front
Looking up at
3.3 × 2.4 in
field of horizontal 5.75
liquid film
ms Event
Ripples being 6.50

0.00
accelerated
Lifting into
5.75
airstream
Stretching
6.50 7.25
downstream
Center
7.25
rupturing
8.00
Ends falling back

8.00
on film; droplets
transported in air
Figure 11.1 Photographs from Woodmansee & Hanratty (1969) showing atomization of a wall film.
count, as many as twenty drops can be sent streaming off into the flowing air after the filament
rupture. Finally the end attached to the water film lies down ahead of the roll wave and is passed
under the passing disturbance.
The entrained liquid in annular flows is carried along as droplets

Ð d with a large range of
sizes. A number distribution function, fn(dP), is defined so that 0 P fn dd p is the fraction of
the drops with diameters less than dP. The number mean diameter is given as
ð∞
d 10 ¼ d P fn ðd P Þd ðd P Þ ð11:2Þ
0
A normalized volume distribution is also used. It is defined by

dv
¼ f v ðd P Þ ð11:3Þ
dd P
ð∞
fv ðd P Þdd P ¼ 1 ð11:4Þ
0
It gives the fraction of the volume made up of drops between dP and dP+ddP. Since the
volume of a drop varies as d 3P , larger drops are contributing more to the total volume than
smaller drops. The volume distribution function is skewed more in the direction of large
dP than is the number distribution function. Thus, the volume median diameter, dvμ, is
considerably larger than the number mean diameter, d10.
Another way to characterize the droplet size would be one of the average diameters,
dqm, defined by Mugele & Evans (1951), where
Ð∞ q
d f dd P
d qm ¼ Ð 0∞ mP n ð11:5Þ
0 d P f n dd P
For the number mean diameter, q = 1 and m = 0. The Sauter mean diameter, d32, is
particularly useful in mass transfer operations, since a drop with this diameter has the
same surface-to-volume ratio as the whole spray (Wallis, 1969).
A widely used distribution function is the log-normal, given as
$ %
1 1 ln d P ln d vμ 2 dd P
f v ðd P Þ ¼ pffiffiffiffiffi exp ð11:6Þ
σ 2π 2 σ dP
Two parameters need to be determined from experiment: the standard deviation, σ, and
the volume mean diameter, dvμ. The Sauter mean diameter can then be calculated as
d vμ
d 32 ¼ ð11:7Þ
expðσ 2 =2Þ
The Rosin & Rammler (1933) distribution has been applied to annular flows by
Azzopardi et al. (1991), Jepson et al. (1989), Hay et al. (1996), and others. It is given as
$ %
dP N
f v ðd P Þ ¼ 1 exp ð11:8Þ
X
101.000
100.000
dP/dvμ
10–1.000 30 m/s, 2.54 cm

36 m/s, 2.54 cm, Log-normal fit
36 m/s, 2.54 cm
43 m/s, 2.54 cm
50 m/s, 2.54 cm
10–2.000 30 m/s, 9.53 cm
36 m/s, 9.53 cm
43 m/s, 9.53 cm
50 m/s, 9.53 cm
10–3.000
3%
9%
7%
4%
7%
7%
2%
%
29
33
33
.9
0.
95
97
88
68
73
90
99
23
70
95
.9
.7
.4
.9
.2
.3
5
5.
77
97
15
33
56
90
99
1.
Cumulative volume
Figure 11.2 Log-probability plot of drop sizes in horizontal 2.54 cm and 9.53 cm pipes, uLS = 0.041 m/s. Al-
Sarkhi & Hanratty, 2002.
where N and X are obtained from experiment. The log-normal distribution is the main
focus of this chapter.
The parameter σ in equation (11.6) characterizes the spread of drop sizes around dvμ.
It does not vary greatly with flow conditions. This is not the case for dvμ, which
represents the size of larger drops that carry most of the mass. Figure 11.2 shows
measurements of dP/dvμ for air and water flowing in 2.54 cm and 9.53 cm horizontal
pipes by Simmons & Hanratty (2001) and by Al-Sarkhi & Hanratty (2002). The plot
uses log-normal paper (a log-probability plot) for which equation (11.6) produces
straight lines that give a value of dP/dvμ = 1 at a cumulative volume fraction of 50%.
The parameter σ determines the slope. Note that a value of σ = 0.82 produces a good fit
for a wide range of conditions. The measured dvμ varied from 81.8µm to 181µm and
the liquid flow varied from 30 g/s to 170 g/s. This includes changes of the superficial
liquid velocity of 0.041–0.125 m/s (not indicated in Figure 11.2). Equation (11.6)
fits the data, except for small and large drop diameters (about 4.5% of the drop volume
for the 2.54 cm pipe). The departure from the log-normal distribution at small and
large drop sizes is more severe in the larger pipe. A comparison of the log-normal and
Rosin–Rammler equations with the data of Simmons & Hanratty (2001) indicates that
the Rosin–Rammler equation overpredicts the volume concentration of small drops
and underpredicts the volume contribution of large drops. The log-normal provides a
better fit of the measurements that they examined.
Wicks & Dukler (1966), and others, have used the upper limit log-normal distribution
(proposed by Mugele & Evans, 1951) for annular flows
$ 2 %
dv δd m dP
¼ pffiffiffi exp δ ln
2
ln a ð11:9Þ
dd P π d P ðd m d P Þ d m dP
where
d vμ
a¼ ð11:10Þ
d m d vμ
A third empirical parameter, the maximum drop diameter, dm, is introduced. This
parameter appears to vary with the size of the sample (Azzopardi, 1997), so it is not
easily defined.
11.3 Drop size
The critical issue in using the log-normal distribution is the specification of a character-
istic diameter, dvμ or d32, of the droplets carrying most of the volume. Taylor (1940) made
two proposals regarding the mechanism of atomization.
One is the suggestion that the liquid is sheared from a viscous boundary layer at the
interface by the shear stress imposed by a high-velocity gas stream. The drop size is
related to the thickness of this boundary layer. Experimental studies of annular flow are
inconsistent with this interpretation in that measurements of drop size and the rate of
atomization are not strongly affected by liquid viscosity.
The second proposal is that atomization occurs by the removal of waves and that drop
size scales with the wavelength of the unstable waves. Taylor used the inviscid theory
outlined in Chapter 4 (see Lane & Green, 1956) to calculate the wave which is growing
the fastest:
2πσ
λm ¼ f ðθ Þ ð11:11Þ
ρG uG 2
where uG is the relative velocity of the gas to the liquid and

ρL σ 2
θ¼ ð11:12Þ
ρG μ2L uG 2
Here μL is the liquid viscosity, σ is the surface tension and f(θ) is a function which corrects
for effects of liquid viscosity. Values of f(θ) are given by Lane & Green (1956). For
liquids with small viscosity, f(θ) = 1.5. Thus, the assumption that the drop diameter scales
with the wavelength yields the following relation
ρ G uG 2 d P
¼C ð11:13Þ
σ
where C is a constant. This equation is attractive since it also represents the gas velocity at
which a drop with a diameter dP will break up. See Lane & Green (1956). However,
(11.13) is not consistent with measurements, which indicate that dvμ ∼ u−1:1 G . This
disagreement led Tatterson et al. (1977) to consider another scaling.
The wavelets observed to be atomizing in annular flows are too small, relative to the
thickness of the wall layer, to be described by a deep liquid analysis. Consequently, the
11.3 Drop size 273
Pb
dL
Pa
σ σ
σ σ σ σ
(a) (b) (c)
Figure 11.3 Kelvin–Helmholtz mechanism for the formation of drops. Tatterson et al., 1977.
above approximation for the scaling of λm is not appropriate. Tatterson et al. suggest that
the unstable wavelengths scale with the height of the wall layer.
The unstable wave shown in Figure 11.3a is pictured to grow until the crest adopts the
configuration shown in Figure 11.3b. Here, the crest resembles a two-dimensional tube
with a diameter dL, which is removed from the wave because the suction pressure of the
gas, ΔP = Pb − Pa, is just balanced by surface tension (see Section 4.3).
DP ≈ σ =d2L ð11:14Þ
By using Kelvin–Helmholtz theory for the flow of an inviscid fluid over a small-
amplitude wave Tatterson et al. (1977) suggested that the pressure gradients in a gas
flowing over a wavy surface scale as ρG uG 2 k and that ΔP in (11.14) scales as ρG uG 2 kd L ,
where k = 2π/λ is the wave number of the wave producing the ligament. If the drop
diameter, dP, is assumed to be proportional to dL (rather than the wavelength) and if the
above scaling for ΔP is substituted into (11.14), one obtains
1=2
ρ uG 2 k
dP G ¼ constant ð11:15Þ
σ
It is expected that k would be the wave number of the fastest-growing wave, km. For
wavelets on thick layers k m ~ ρG uG 2 =σ. If this is substituted into (11.15), equation
(11.13) is obtained. As mentioned above, this is not consistent with experiments.
Tatterson et al. (1977) argue that the deep liquid assumption is not appropriate since
the wall film is too thin. They assumed that the wavelength of the unstable wave
scales with the film height, m. If this is used in (11.15) the following relation for dP is
obtained:
1=2
d P ρG muG 2
¼ C1 ð11:16Þ
m σ
where C1 is a constant. The use of (11.16) requires the specification of an appropriate

value for uG since the analysis was carried out on the assumption that uG is uniform. The
distance over which a surface disturbance penetrates a flowing gas stream is proportional
to the wavelength. The velocity profile over a wavy surface can be represented as
uG ð y Þ 1 y
¼ ln þ B ð11:17Þ
vG κ kS
where kS is a length scale characterizing the roughness of the interface. From this
observation and (11.17), Tatterson et al. suggested that uG in equation (11.16) should
be replaced with the friction velocity, vG ¼ ðτ i =ρG Þ1=2 . Thus, (11.16) changes to
1=2
d P ρG v2 Gm
¼ C2 ð11:18Þ
m σ
where
1
τ i ¼ f i ρG uG2 ð11:19Þ
2
and
1=2 1=2 1=2
fi f f
vG ¼ uG ¼ S uG i ð11:20Þ
2 2 fS
The term fS is the friction factor for a smooth surface,

0:046
fS ¼ ð11:21Þ
Re0:2
G
The work of Henstock & Hanratty (1976), described in Chapter 3, gives an approxima-
tion for m:
m 6:59
¼ ð11:22Þ
d t ð1 þ 1400F Þ1=2
where
rffiffiffiffiffi
γðReLF Þ L ρL
F¼ ð11:23Þ
Re0:9
G G ρG
Henstock & Hanratty also give

fi
¼ 1 þ 1400F ð11:24Þ
fS
Tatterson et al. used these equations to eliminate m and fi from (11.18). They rewrote
(11.18) as
1=2 1=2
d P ρG u2G ðf S =2Þd t m fi
¼ C2 ð11:25Þ
dt σ dt fS
11.3 Drop size 275
where (m/dt) is given by (11.22) and (fi/fS) is given by (11.24). The correlations of
Henstock & Hanratty suggest that the right side of (11.25) is a function of F. (Somewhat
different expressions are obtained if equations (3.47) and (3.49) in Chapter 3 are used to
calculate m/dt.)
Pogson et al. (1970) used still photography to measure drop size in an upward flow of
steam and water in a 0.32 cm pipe. Namie & Ueda (1972) measured drop size for air–
water flow in a horizontal 1 × 6 cm rectangular channel by collecting samples of the spray
in a silicone oil. The sample sizes, in both studies, were not large enough to measure the
volume median drop size, so Tatterson et al. (1977) calculated the average drop size, d10,
from their data on drop size distribution. These are plotted in Figure 11.4, as suggested by
(11.25). The ordinate is the left side of (11.25). The abscissa is the flow parameter. Terms
m/dt and fi/fS are calculated from (11.22) and (11.24). The curve, calculated from (11.25)
with C2 = 0.58, provides a good fit. Note that the data represent studies with surface
tensions of 40 dynes/cm and 70 dynes/cm.
Thus, the paper of Tatterson et al. tentatively suggests that d10 can be estimated by
1=2
ρG uG2 d t ðf S =2Þ d 10
≈ 2:4 103 ð11:26Þ
σ dt
The above analysis is speculative. Clearly, more measurements need to be considered.

Simmons & Hanratty (2001) give the following account of drop size measurements:
Several techniques have been employed: Tatterson et al. (1977) used charge removal from an
insulated probe. Wicks & Dukler (1966) used a needle bridging method, where two needles,
placed a small distance apart, were connected to a resistance battery. Drops larger than the gap
10–2
70 dynes/cm
Namie & Ueda (1972) 40 dynes/cm
Pogson et al. (1970) 70 dynes/cm
d10
dt
12
ρGu 2dt fS
2
σ
G
Theory
10–3 –3
10 10–2
F
Figure 11.4 Comparison of measurements of the length-averaged diameter with the theory of Tatterson et al.
(1977). Tatterson et al., 1977.
completed the circuit and caused an electric pulse. By varying the needle gap, the cumulative size
distribution was obtained. Semiat & Dukler (1981) and Lopez & Dukler (1985) used a laser
gradient technique, which produced a local measurement of drop size and velocity. All these
techniques were unable to detect very small drops, below 100–200 μm. Subsequent work has
shown that these techniques lead to significant error in measuring the volume median size since a
significant portion of the volume is carried by drops smaller than 100 μm. Further refinements
were made to the laser-grating technique by Fore & Dukler (1995). These allowed measurements
of droplets down to a size of 10 μm.
Several workers have used photographic techniques (Cousins & Hewitt, 1968; Pogson et al.,
1970; Andreussi et al., 1978). A problem with these techniques is that they, too, can favour large
drops, since it is more probable that part of a large droplet may be within the field of the camera
lens. Furthermore, smaller droplets may not be detected since they could be out of focus,
particularly if the resolution of the photographs is poor. Hay et al. (1996) addressed this problem
by using a sheet of laser light to illuminate the flow field. The laser sheet was thinner than the focal
depth of the camera lens, so only droplets illuminated by the laser sheet appeared in focus. Droplets
out of focus were eliminated from analysis by examining changes in the light gradient at the edges
of the drops.
A laser diffraction technique, invented by Switherbank et al. (1976) has been used by several
workers. This work has been reviewed by Azzopardi (1997). Instruments based on this approach
are marketed by Malvern instruments. Early versions required the assumption of a distribution
function for drop size, either the equation of Rosin and Rammler (1933) or the upper limit log-
normal distribution of Mugele & Evans (1951). Current versions employ a fifteen parameter
“model independent” algorithm which does not impose any unimodal function. The detection
range of the instrument is governed by the focal length of the lens used in the detector.
Combellack and Matthews (1981) and Azzopardi (1985) indicated the importance of using the
optimal focal length.
The use of the diffraction technique has been limited to situations in which the concentration of
droplets is low. The algorithms used to analyze the measurements assume that the light is scattered
only by one drop. At higher concentrations, multiple scattering of laser light by many drops can
exist. This will cause the scattered light to enter the detector at a larger angle; the distributions
obtained will overpredict the number of smaller drops.
A multiple scattering algorithm to correct for this effect is employed in the Malvern
Spraytec R5000 series.
An immersion technique was used by Namie & Ueda (1972) and by Okada et al.
(1995) to collect in situ samples in a viscous oil and photograph them after removal.
Advantages of this technique are that the droplets are spherical when photographed and
that the measurement does not require removal of the film from the wall to provide optical
access. Hurlburt & Hanratty (2002) further developed this technique so that it might be
employed in field operations.
Hay et al. (1996) studied upward flow of air and water in a 4.2 cm pipe. The experi-
ments were performed at a gas velocity of 36 m/s and liquid flow rates 30 to 170 g/s. The
volume median drop diameter increased from 91.2μm to 192μm over this range of liquid
flow rates. Changes in the gas velocity have a greater effect than changes in the liquid
flow rate. Many studies have correlated this increase with the increase in entrained liquid.
This seems to be motivated by an expectation that drop collisions would increase with
increasing drop concentration. However, Hay et al. showed, for a gas velocity of 36 m/s,
that drop size increases with the thickness of the wall layer. That is, they found that dμ/m
is a constant, as shown in Figure 11.5.
11.3 Drop size 277
0.8
0.6
dP/m
0.4
0.2 Sauter mean, d32

Volume median, dvμ
Volume mean, d30
0
0 0.05 0.1 0.15 0.2
Total liquid flow rate (kg/s)
Figure 11.5 Scaling of drop size with the average thickness of the wall layer. Measurements of Hay et al. in a
4.2 cm pipe. Hay et al., 1996.
It is useful, initially, to ignore the influence of liquid flow rate. Values of

1=2
ρG u2G d t ðf S =2Þ d vμ
¼ 2:76:1 103 ð11:27Þ
σ dt
are calculated for dvμ varying from 91.2 to 192μm.

Equation (11.27) seems consistent with measurements reported in the literature on the
influence of uG , ρG and σ on drop size. The prediction that dvμ increases with pipe
diameter is more contentious, since many calculations (Azzopardi, 1985; Pan &
Hanratty, 2002) presume that drop size is independent of pipe diameter. This is an
important issue because it is desirable to use measurements of drop size in small-diameter
pipes to predict the behavior of large commercial pipes, such as used to transport natural
gas/condensate. This prompted two studies (Simmons & Hanratty, 2001; Al-Sarkhi &
Hanratty, 2002) in horizontal 2.54 cm and 9.53 cm pipes, which use the Malvern Spraytec
to measure drop size. This has the advantage that comparison is made in the same flow
system with the same measurement technique.
One of the difficulties is that drop size decreases with height in a horizontal pipe
because of the gravitational force on the drops and because the height of the liquid film
varies around the circumference. The drop size was found to be constant along any
horizontal chord in the pipe. Figure 11.6 shows measurements made in a 9.53 cm pipe for
uGS = 43 m/s at the pipe center, 1.9 cm above the center and 1.9 cm below the center. Note
that the influence of gravity on drop size measurements is not great below the pipe center.
However, differences of about 10% were found at gas velocities of 30 m/s and 36 m/s.
Figure 11.7 compares plots of the data obtained at the center of the 2.54 cm pipe and at
1.9 cm below the center in the 9.53 cm pipe. These show that the drop size roughly
1=2
increases with d t and decreases with u1:1 GS .
140
130
120
110
100
90
d32 (μm)
80
70
60
50
40 uGS = 43 m/s
30 Above the center
20 Center
Below the center
10
0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
uLS (m/s)
Figure 11.6 Comparison of measurements of the Sauter mean diameter at different positions in a 9.53 cm pipe at
uGS. Simmons & Hanratty, 2001.
250
uGS
dt = 9.53 cm
200 30 m/s
36 m/s
dvμ(μm)
150 43 m/s
50 m/s
100
dt = 2.54 cm
50
0
0 0.05 0.1 0.15
uLS (m/s)
Figure 11.7 Comparison of volume median drop diameters obtained at the center of a 2.54 cm pipe (open
symbols) and at 1.9 cm below the center of a 9.53 cm pipe. Al-Sarkhi & Hanratty, 2002.
The following equation is given by Al-Sarkhi & Hanratty if the influence of liquid flow
is ignored:
1=2
d t ρG uG2 d vμ
¼ 0:1072 ð11:28Þ
σ dt
It is of interest to compare this result with that obtained by Hay et al. (1996) for upward
flow of air and water at 36 m/s in a 4.2 cm pipe, for which (fS/2)1/2 = 4.74 × 10−2. Hay
et al. obtained
!1=2
ρG u2G d t d vμ
¼ 0:0570:129 ð11:29Þ
σ dt
11.4 Rate of deposition 279
The average of the right side is 0.093. Thus, (11.28) agrees, approximately, with the
measurements of Hay et al.
Equation (11.29) offers a first approximation of data. It does not exactly match
measurements since the predicted dependency of drop diameter on uG 1 does not
agree with the dependency of uG 1:1 found in measurements. Furthermore, the role of
liquid flow is not defined.
Two interpretations of the influence of liquid flow have been proposed: the increase of
coalescence with increasing flow of entrained drops, GLE, and changes of the drops
generated by atomization of the wall layer. (A mechanism which has not been considered
is the decrease of gas turbulence with increasing particle concentration.)
11.4 Rate of deposition
For small drop concentrations, the rate of deposition of drops in vertical annular flow can
be represented by the linear rate equation
W LE
RD ¼ k D C B ¼ k D ð11:30Þ
SQG
where CB is the average drop concentration over the pipe cross-section, RD is the rate of
deposition per unit area, QG is the volumetric gas flow and S is the ratio of the drop
velocity and the axial velocity of the gas flow. The drops, in annular flow, are large
enough that τ þ
P > 20. Thus, the dominant mechanism involves drops starting a free-flight
to the wall from a region outside the viscous wall layer (see Chapter 10).
The rate of deposition can be described as the average velocity with which droplets
strike the wall, VW, so that
CW
RD ¼ V W C W ¼ V W C B ð11:31Þ
CB
where CW is the concentration at the wall. Measurements of C show that it is constant

over the pipe cross-section for a vertical annular flow (Hay et al., 1996). Thus CW ≈ CB
and kD = VpWffiffiffiffiffi
. If the turbulent velocities are Gaussian, the average velocity is given as
V W ¼ σ P = 2π , so
pffiffiffiffiffi
k D ¼ σ P = 2π ð11:32Þ
1=2
where σ P ¼ ðv2Pr Þ is the root-mean-square of the radial component of the particle
velocity fluctuations outside the viscous wall layer. Equation (11.32) has been verified for
dilute concentrations (Dykhno & Hanratty, 1996) and by Lee et al. (1989).
Measurements of ðv2Pr Þ are discussed in Section 10.8. The particle turbulence is related
to the fluid turbulence by equation (10.59). Thus, v2Pr ¼ u2r ½β=ðα þ βÞ with α ¼ 0:7=τ path
i
where vPr and ur are the r-components of the particle velocity fluctuations and the fluid
velocity fluctuations, respectively (see Figure 10.3). The term β ¼ 1=τ P , where τP is the
inertial time constant of the drops.
0.25
0.2
Rate of deposition (kg/m2 s)
0.15
0.1 Govan et al. (1988) dt = 3.18 cm

uGS = 26−74 m/s, WL = 20−79 g/s
Schadel et al. (1990) dt = 4.2 cm
uGS = 20, 53, 72 m/s
0.05 Schadel et al. (1990) dt = 4.2 cm
uGS = 36 m/s
Estimated from Schadel et al. data
Extrapolated from Schadel et al. data
0
0 0.5 1 1.5 2
CB = WLE/QG (kg/m3)
Figure 11.8 Measurements of rates of deposition. Hay & Hanratty, 1996.
Measurements of RD by Govan et al. (1988) and by Schadel et al. (1990) are shown in
Figure 11.8 as a function of CB. Equation (11.30) describes the limiting slope at small
WLE/QG. An interesting feature of these measurements is the breakdown of the limiting
law; that is, kD depends on concentration. At large CB (volume fractions roughly equal to
0.6 ×10−3) a limiting behavior is suggested such that k D / C 1
B , so that RD is independent
of CB. This behavior has also been observed for downward flow in a vertical pipe by
Andreussi et al. (1983). See the review article by Dykhno & Hanratty (1996). The data
shown in Figure 11.8 are roughly approximated by fitting them with two straight lines. At
small drop concentrations, the rate of deposition is given by (11.30). For WLE/Q > 0.75
(which corresponds to a volume fraction of drops of 0.75 × 10−3), k D / C 1 B and the rate
of deposition is constant at 0.17 kg/m2 s.
If kD is represented by (11.32) the measurements of RD shown in Figure 11.8
1=2
suggest the dependency on v2Pr displayed in Figure 11.9. Several explanations
for the decrease of particle turbulence with increasing drop concentration have been
proposed: (1) Drop size increases with liquid flow. This leads to a more sluggish
behavior of the drops and, therefore, a decrease in v2Pr . However, this effect is not
large enough to explain Figure 11.9 (see Hay et al., 1996). (2) The number of
encounters per unit time increases with the number of drops per unit volume.
Therefore, it is plausible to explore the notion that particle–particle encounters result
in a decrease in particle turbulence through inelastic interactions. Calculations (Hay
et al., 1996) show that this could lead to a significant decrease in the particle
turbulence and to the rate of deposition. (3) Namie & Ueda (1972) suggest that the
decrease in particle turbulence with increasing concentration is due to a decrease in
the fluid turbulence. The next section describes numerical experiments in a turbulent
velocity field, which support this suggestion.
1
(v P2)1/2 (m/s)
0.1
0.1 1 10
CB (kg/m3)
Figure 11.9 Root-mean-square particle velocity needed to explain deposition data. Hay et al., 1996.
11.5 Experiments in turbulence generated by a DNS
Direct numerical solutions (DNS) of the Navier–Stokes equations have provided results
on the behavior of a turbulent flow. Particles embedded in the turbulence experience a
drag force which is related to the difference in velocity between the particle and the fluid
3ρf C D !
f Pi ¼ jv !
uG jðvPi uGi Þ ð11:33Þ
4d P ρP P
The force of the particle on the fluid is −fPi. This feedback effect of particles on fluid flow has
been modeled with a point force method using the assumption that the particles are small
enough for the eddy shedding process not to occur. Squire & Eaton (1990) and Elgobashi &
Truesdell (1993) used this approach in a DNS of an isotropic field. Pan & Banerjee (1996),
Li et al. (2001), Mito & Hanratty (2006), and Hanratty & Mito (2009) used it in a DNS of
turbulent flow in a channel. The approach involves the introduction of forces of the particles
on the fluid into the Navier–Stokes equations, which are solved numerically.
The computational domain is divided into a number imaginary cells which contain one
grid point. The force of particles on the fluid motion at a grid point is the sum of the
reaction forces exerted by particles whose centers exist in the computational cell
surrounding the grid point (Li et al., 2001).
ρP V P X N cell
C
Fi ¼ f Pik ¼ h f Pi icell ð11:34Þ
ρf V cell k¼1 ρf
where VP is the volume of the particle, Vcell is the volume of the computational cell, Ncell is
the number of particles in the cell, fPik is the fPi for the kth particle, C = ρPVPNcell/Vcell is the
concentration of the particles and hf Pi icell is the average point force of the Ncell particles.
The equation for fluid motion is then represented as

∂uGi ∂uGi ∂P ∂2 uGi
ρf þ uGj ¼ þμ þ ρf F i ð11:35Þ
∂t ∂xj ∂xi ∂xj ∂xj
The method for introducing particles into the flow field differs from author to author.
Mito & Hanratty considered the annular configuration for upward flow in a vertical
channel. The ratio of the particle density to the fluid density was the same as for an
air–water flow, 1000 : 1. The atomization process is mimicked by injecting particles from
x2 = dP/2 with a velocity of 15v0 ; v0 ; 0 and a rate of atomization per unit area of Rib from
the bottom Particles were also injected from x2 = 2H − dP/2 with a velocity of
wall.
15v0 ; v0 ; 0 and a rate per unit area of Rit, where v0 is the friction velocity in the
absence of particles. The particles were removed from the field when they hit a wall. If the
rate of introduction of particles is less than a critical value, a stationary state can be
reached for which the rate of injection equals the rate of deposition. If it is above this
critical value, the concentration continues to increase with time since the rate of depo-
sition is not sufficient to balance the rate of injection.
The calculations of Mito & Hanratty (2006) are of interest because they mimic the
atomization process. However, they have limitations in that the walls are smooth. (The
direct influence of wall roughness was not considered.) The stress on the wall, τW, is
obtained from the calculated velocity gradient at the wall, τ W ¼ μðdu=dyÞW . The friction
pffiffiffiffiffiffiffiffiffiffiffiffi
velocity is defined as v ¼ τ W =ρf . The calculations were done for a fixed volumetric
flow of fluid. The friction is found to decrease with an increasing concentration of
particles. The Reynolds number, defined with the bulk mean velocity and the half-height
of the channel, H, was 2260. The Reynolds number, Reτ, defined with the friction
velocity in the absence of particles, v0 , and the half-height of the channel, H, was 150.
Figure 11.10 presents their calculations of the rate of deposition for cases in which
stationary states were reached. The left ordinate is the calculated rate of deposition made
dimensionless with v0 . The abscissa is the volume fraction of spheres, αP . The right
2 0.1
k D+ (1-way) 9
10–5 8
6 7
4 6
2 5
R D+ k D+B
10–6 R D+ (1-way) 4
6
4 k D+ ~ αP–1 3
2
10–7 2
2 3 4 5 67 2 3 4 5 67
10–5 10–4 10–3
αP
Figure 11.10 Effect of feedback, as demonstrated by the effect of αP on the rate of deposition, Rþ
D , and on the
deposition constant, kþ
D . Mito & Hanratty, 2006.
ordinate is the deposition coefficient made dimensionless with the friction velocity that
existed when no particles were present, k þ þ þ
D ¼ k D =v0 . The values of RD and k DB are also
shown for the one-way coupling situation, for which the effect of αP on the fluid
turbulence is ignored. The results for two-way coupling capture the trend of measure-
ments in annular flows.
Large changes in the normal velocity fluctuations at very small volume fractions of the
particles were observed. Instantaneous velocity fluctuations calculated at the center plane
are shown in Figure 11.11a for single-phase flow, in Figure 11.11b for a stationary case
with αP = 4.9 × 10−4 and in Figure 11.11c for a non-stationary case with αP = 3.0 × 10−3,
where αP is the volume fraction of particles embedded in the fluid. Dots show the
locations of particles. (They do not represent the actual size of the particles.)
Significant attenuation of the fluid turbulence is noted in Figure 11.11b. The particles
are seen to be almost uniformly distributed and to be affecting all turbulence structures.
The small-scale turbulence is seen almost to vanish for the non-stationary case shown in
Figure 11.11c. Instantaneous velocity fluctuations in a cross-section perpendicular to the
950
x +3
0 x +1 1900
(a) Single phase
(b) αP = 4.9 × 10−4 (stationary state)
(c) αP = 3.0 × 10−3 (non-stationary state)
Figure 11.11 Instantaneous fluid velocity fields in the center plane. Dots represent locations of particles.
(a) Single-phase flow. (b) A statistical stationary case at αP = 4.9 ×10–4. (c) A non-stationary case
at αP = 3.0 × 10–3. Mito & Hanratty, 2006.
300
x +2
0 x +3 950
(a) Single phase
(b) αP = 4.9 × 10−4 (stationary state)
(c) αP = 3.0 × 10−3 (non-stationary state)
Figure 11.12 Instantaneous fluid velocity fields in a cross-section perpendicular to the direction of mean flow.
Dots represent locations of the particles. Their sizes are magnified five times. (a) Single-phase flow.
(b) A stationary case at αP = 4.9 × 10–4. (c) A non-stationary case at αP = 3.0 × 10–3. Mito &
Hanratty, 2006.
direction of mean flow are shown in Figure 11.12. The vertical structures observed close
to the wall for single-phase flows are seen to disappear.
Under fully developed conditions the average of equation (11.35) gives
ðH
∂P ∂uG1
ðH x2 Þ ¼ ρf uG1 uG2 þ μ
t t ρf F 1 ðx2 Þdx2 ð11:36Þ
∂x1 ∂x2
x2
For single-phase flow, the force due to the pressure gradient is balanced by the Reynolds

shear stress ρf ut1 ut2 and the viscous shear stress ðμ∂uG1 =∂x2 Þ. In the presence of
particles the pressure gradient is balanced by particle forces, as well as the viscous and
turbulent stresses. This can explain the calculated decrease of Reynolds stress shown in
Figure 11.13 and the damping of fluid turbulence shown in Figures 11.11 and 11.12, since the
production of fluid turbulence is related to Reynolds shear stress (see Hanratty & Mito, 2009).
These calculations support the notion that the reduction of kD at large concentrations is
associated with the damping of fluid turbulence.
11.6 Rate of atomization 285
0.8
(a) Single-phase αP = 1.0 × 10–5
αP = 1.2 × 10–4
0.6
αP = 1.9 × 10–4
− u +1 u +2
αP = 3.0 × 10–4
0.4 αP = 4.9 × 10–4
0.2
0.0
0 20 40 60 80 100 120 140
x +2
0.8
(b) Single-phase
0.6
αP = 6.0 × 10–4 αP = 9.3 × 10–4
− u +1 u +2
0.4 αP = 2.2 × 10–3

αP = 3.0 × 10–3
0.2
0.0
0 20 40 60 80 100 120 140
x +2
Figure 11.13 Effect of feedback on the fluid Reynolds shear stress. (a) Statistically stationary cases. (b) Non-
stationary cases when 2.7 particles are injected per time step. Mito & Hanratty, 2006.
11.6 Rate of atomization
An important phenomenon in considering the rate of atomization and the entrainment is

the existence of a critical film flow rate, WLFC, below which atomization does not occur
(Hewitt & Hall-Taylor, 1970; Dallman et al., 1979; Asali et al., 1985a, b; Andreussi
et al., 1983; Schadel et al., 1990). The air–water data in small- and large-diameter pipes
give average values of the critical film Reynolds number in the range 241–280.
The rate of atomization has been represented by the empirical equation
0
k u2 ðρ ρ Þ0:5 ðW LF W LFC Þ
RA ¼ A G L G ð11:37Þ
σ P
where k 0A is a dimensionless constant. Equation (11.37) shows that the rate of

atomization varies linearly with the excess film flow rate at small WLF (Lopez de
Bertodano et al., 1997; Assad et al., 1998; Pan & Hanratty, 2002). For WLF < WLFC,
RA = 0. The linear relation in (11.37) breaks down at large WLF in that dimensionless
k 0A decreases with increasing film flow, as has been found for kD (see Figure 11.10).
At large WLF, this is particularly evident in the measurements of Andreussi & Zanelli
(1976, 1979) for downward annular flow. In contrast to upward flow, gravity is
aiding, so that thicker wall flows can be realized. A maximum RA at large WLF is
suggested.
Dykhno & Hanratty (1996) provide a speculative explanation for this non-linear
behavior: At low liquid flows, atomization occurs from disturbance waves which appear
intermittently in the wall layer. Deposition occurs, mainly, on the base film. This
deposited liquid is picked up by the disturbance waves, which move rapidly over the
base film. Thus, the base film does not change its height. However, at large deposition
rates this pickup is not large enough to keep the height of the base film constant.
A form of equation (11.37) that recognizes, directly, the existence of a critical gas
velocity for the initiation of entrainment is
k 0A ðuG uGC Þ2 ðρL ρG Þ0:5 ðW LF W LFC Þ

RA ¼ ð11:38Þ
σ P
A discussion of this modification of (11.37) is provided in the next section.
11.7 Entrainment in vertical flows
Entrainment is defined as
W L W LF
E¼ ð11:39Þ
WL
This section develops expressions for entrainment, except for cases for which the inertial
time constant is so large that the drops have unidirectional paths.
Plots by Willetts (1987) of WLF versus uG for a fixed liquid flow, WL, provide easily
identified critical gas velocities, uGC, above which WL > WLF. His results are fitted
approximately with the relation
0:25
t uGC ðρL ρG Þ
d 0:5
≈ 40 ð11:40Þ
σ 0:5
(see Pan & Hanratty, 2002).

By equating (11.38) to (11.30), the following equation is obtained
E=E M k 0 d t uG ðuG uGC Þ2 S ðρL ρG Þ0:5

¼ A ð11:41Þ
1 E=E M 4k D σ
where kD has the units of velocity and

W LFC
EM ¼ 1 ð11:42Þ
WL
Equation (11.41) requires that WLF and uG are above the critical values of WLFC and uGC
that are needed for drops to appear, so E = 0 if uG < uGC or if WLF < WLFC. Note that the
liquid flow does not appear in (11.41).
Strictly, (11.41) is valid only for small entrainments where linear rate equations are
observed. Non-linearities could be captured by allowing k 0A and kD to decrease with
increasing WL. The inclusion of these effects makes (11.41) clumsy. Pan & Hanratty
11.7 Entrainment in vertical flows 287
(2002) used these equations outside the range of liquid flows where they should be valid
by making the assumption that k 0A =k D is approximately independent of WL. They
substituted uG for kD/S and obtained the following rough fit to data of Schadel et al.
(1990), Binder (1991), Lopez de Bertodano & Jan (1998), Lopez de Bertodano et al.
(1997,1998), Willetts (1987), Andreussi & Zanelli (1976,1979):
E=E M A1 ðuG uGC Þ2 ðρG ρL Þ1=2 d t

¼ ð11:43Þ
1 E=E M σ
with A1 = 6×10−5.
Figure 11.14 shows measurements of E/EM for air and water flowing up a vertical
4.2 cm pipe. These indicate that E/EM is more sensitive to changes in the gas velocity than
to changes in the liquid flow. These data are plotted as E versus WL in Figure 11.15. For a
constant gas velocity, an influence of liquid velocity, in (11.41), is included in the term
(E/EM)/(1 − E/EM). The solid curves in Figure 11.15 were calculated using (11.43).
1.2
1
uG (m/s)
0.8 72
53
E/EM
0.6
36.5
0.4
19.5
0.2
0
0 0.03 0.06 0.09 0.12
WL (kg/s)
Figure 11.14 Plots of E / EM for air and water flowing up a vertical 4.2 cm pipe. Pan & Hanratty, 2002.
0.8
uG (m/s) WLFC /(πdt) = 0.085 (kg/s.m)
19.5
0.6 37
53
72
0.4
E
0.2
0
0 0.02 0.04 0.06 0.08 0.1
WLFC = 0.0112 (kg/s)
WL (kg/s)
Figure 11.15 Plots of entrainment, E, for air and water flowing up a vertical 4.2 cm pipe. Pan & Hanratty, 2002.
At WLF = WLFC = 0.0112 kg/s, the atomization rate is zero, so E = 0. Thus, there is a sharp
drop in E close to WL = WLFC, as indicated by the dashed lines. From (11.43), it would be
expected that entrainment reaches an asymptotic value at large WL. This type of behavior is
only approximately realized. The entrainment actually tends to increase at large WL. This is
not surprising since (11.43) was derived by using linear relations for RA and RD. It is
expected that k 0A =k D should change with WL at large values of this quantity. The increase of
E at large WL suggests that kD is decreasing more with increasing WL than is k 0A =k D .
It is of interest to see how well (11.43) describes observed effects of surface tension
and gas density:
Figure 11.16 shows results obtained by Willetts (1987) for air–water (σ = 0.073 N/m)
and for air–genkelene (σ = 0.01 N/m) flowing up a 1.026 cm pipe. The critical gas
velocity, calculated with (11.40), is included in the figure. Entrainments are seen to be
larger for the liquid with lower surface tension. The slight differences in the plot reflect
the observation that the constant in (11.40) is slightly different for the two systems (see
Pan & Hanratty, 2002, Table 2).
All measurements show that entrainment increases with increasing gas density. This is
illustrated in Figure 11.17 where data of Lopez de Bertodano & Jan (Private communi-
cation, 1998) for air–water flow for three gas densities and by Willetts (1987) for air–
water & helium–water are compared by plotting E/EM versus uG ρ0:25 G , as suggested by
(11.43). The measurements extrapolate to the critical gas velocity, uGC, and E/EM varies
linearly with uG uGC at small gas velocities. A first impulse would be to assume
entrainment increases linearly with the gas-phase kinetic energy or that E/EM varies
with ρ0:5
G uG . The results in Figure 11.17 and the results in Figure 11.16 for a 1 cm pipe
show that this is not the case.
Pan & Hanratty (2002) show that somewhat better agreement of theory and experi-
ments can be realized if the assumption that kD/S in (11.41) can be replaced by uG is
abandoned. Instead, kD is related to particle turbulence by equation (11.32). Particle
turbulence is related to fluid turbulence by the equations outlined in Section 11.4. A value
of k 0A ¼ 1:4 106 was used to correlate data for air and water flowing up pipes with
1
dt = 1.026 cm
atm. pressure
0.8
0.6
E/EM
Critical
0.4
σ (N/m)
0.2 0.073
0.026
0
0 100 200 300 400 500
uGρG0.25/σ0.5 (m/s)(kg/m3)0.25(N/m)−0.5
Figure 11.16 The effect of surface tension on E/EM. The line intersecting the abscissa represents uGC. Pan &
Hanratty, 2002.
References 289
1
dt = 1 cm
0.8 Critical
0.6
ρG (kg/m3)
E/EM
0.4 Air−water 1.83

Helium−water 0.27
Air−water 1.6
0.2 Air−water 2.8
Air−water 4.4
0
0 50 100 150 200
ρ
uG G0.25 (m/s)(kg/m3)0.25
Figure 11.17 An examination of the influence of gas density on E / EM. The line intersecting the abscissa
represents uGC. Pan & Hanratty, 2002.
diameters of 0.0254 m, 0.042 m, 0.00953 m (Lopez de Bertodano & Jan, 1998; Lopez de
Bertodano et al., 1997, 1998; Schadel et al., 1990) and for air and water flowing up and
down in a 0.024 m pipe (Andreussi & Zanelli, 1976, 1979). This calculation is somewhat
more complicated than (11.43) because drop diameter needs to be known.
References
Al-Sarkhi, A. & Hanratty, T.J. 2002 Effect of pipe diameter on drop size in a horizontal annular
gas–liquid flow. Int. J. Multiphase Flow 28, 1617–1629.
Andreussi, P. & Zanelli, S. 1976 Liquid phase mass transfer in annular two-phase flow. Ing. Chim.
12, 132–136.
Andreussi, P. & Zanelli, S. 1979 Downward annular-mist flow of air–water mixtures. In Two-phase
Flow, Momentum, Heat and Mass Transfer, Vol. 2, ed. F. Durst, G.V. Tsiklauri & N.H. Afgan.
Washington, DC: Hemisphere.
Andreussi, P., Romano, P. & Zanelli, S. 1978 Drop size distribution in annular-mist flow. In
Proceedings of the First Conference on Liquid Atomization in Spray Systems, Tokyo, August
27–31.
Andreussi, P., Asali, J.C. & Hanratty, T.J. 1983 Initiation of small waves in gas–liquid flows.
AIChE Jl 31, 119–126.
Asali, J.C., Hanratty, T.J. & Andreussi, P. 1985a Interfacial drag and film height for vertical annular
flow. AIChE Jl 31, 895–902.
Asali, J.C., Leman, G.W. & Hanratty, T.J. 1985b Entrainment measurements and their use in
design equations. Phys-Chem. Hydrodyn. 6, 207–221.
Assad, A., Jan, C.S., Lopez de Bertodano, M. & Beuss, S. 1998 Nucl. Eng. Des. 184, 437–447.
Azzopardi, B.J. 1985 Drop sizes in annular two-phase flow. Exp. Fluids 3, 53–59.
Azzopardi, B.J. 1997 Drops in annular two-phase flow. Int. J. Multiphase Flow 23, 1–53.
Azzopardi, B.J., Piercey, A. & Jepson, D.M. 1991 Drop size measurements for annular two-phase
flow in a vertical tube. Exp. Fluids 11, 191–192.
Binder, J.L. 1991 Use of Lagrangian methods to describe particle deposition and distribution in
dispersed flows. Ph.D. thesis, University of Illinois.
Combellack, J.H. & Matthews, G.A. 1981 Droplet spectra measurements of fan and cone atom-
izers using a laser-diffraction technique. J. Aerosol Sci. 12, 529–540.
Cousins, L.B. & Hewitt, G.F. 1968 Liquid mass transfer in annular two-phase flow; droplet
deposition and liquid entrainment. UKAEA Report AERE-R5657.
Dallman, J.C., Jones, B.J. & Hanratty, T.J. 1979 Interpretation of entrainment measurements in
annular gas–liquid flow. In Two-phase Flow, Momentum, Heat and Mass Transfer, Vol. 2, ed. F.
Durst, G.V. Tsiklauri & N.H Afgan. Washington, DC: Hemisphere, pp. 681–693.
Dykhno, L.A. & Hanratty, T.J. 1996 Use of the interchange model to predict entrainment in vertical
annular flow. Chem. Eng. Comm. 141–142, 207–235.
Elgobashi, S. & Truesdell, G.C. 1993 On the two-way interaction between homogeneous turbu-
ence and dispersed solid particles I: Turbulence modification. Phys. Fluids A5, 1101–1203.
Fore, L.B. & Dukler, A.E. 1995 The distribution of drop size and velocity in gas–liquid annular
flow. Int. J. Multiphase Flow, 21, 137–149.
Govan, A.H., Hewitt, G.F., Owen, D.G. & Bott, T.R. 1988 An improved CHD modeling code.
Paper presented at the Second UK National Conference, Strathclyde University, Glasgow,
pp. 33–52.
Hanratty, T.J. & Mito, Y. 2009 A unifying explanation for the damping of turbulence by additives
and external forces. Flow, Turbulence Combustion 83, 293–303.
Hay, K.J., Liu, Z.C. & Hanratty, T.J. 1996 Relation of deposition rate to drop size when the rate law
is non-linear. Int. J. Multiphase Flow 22, 829–848.
Henstock, W.H. & Hanratty, T.J. 1976 Interfacial drag and film height in annular flows. AIChE Jl
22, 990–1000.
Hewitt, G.F. & Hall-Taylor, N.S. 1970 Annular Two-Phase Flow. Oxford: Pergamon Press.
Hurlburt, E.T. & Hanratty, T.J. 2002 Measurement of drop size in horizontal annular flow with the
immersion technique. Exp. Fluids 32, 692–699.
Jepson, D.M., Azzopardi, B.J. & Whalley, P.B. 1989 The effects of gas properties on drops in
annular flow. Int. J. Multiphase Flow 15, 327–339.
Lane, W.R. & Green, H.L. 1956 The mechanics of drops and bubbles. In Surveys in Mechanics, ed.
G.K. Batchelor, Cambridge: Cambridge University Press, pp. 162–215.
Lee, M.M., Hanratty, T.J. & Adrian, R.J. 1989 The interpretation of droplet measurements with a
diffusion model. Int. J. Multiphase Flow 15, 459–469.
Li, Y., McLaughlin, J.B., Kontomaris, K. & Portelo, L. 2001 Numerical simulation of particle-
laden turbulent laden flow. Phys. Fluids 13, 2957–2967.
Lopez, J.C.B. & Dukler, A.E. 1985 Droplet sizes, dynamics and deposition in vertical
annular flow. U.S. Nuclear Regulatory Commission, Washington DC, Report NUREG/
CR-4424.
Lopez de Bertodano, M.A., Jan, C.S. & Beus, S.G. 1997 Annular flow entrainment rate experiment
in a small, vertical pipe. Nucl. Eng. Des. 178, 61–70.
Lopez de Bertodano, M.A., Jan, C.S. Assad, A. & Beus, S. 1998 Entrainment rate of droplets in
ripple-annular regime for small diameter vertical ducts. Paper presented at the Third
International Conference on Multiphase Flow, Lyon, France.
Mito, Y. & Hanratty, T.J. 2006 Effect of feedback and inter-particle collisions in an idealized gas–
liquid annular flow. Int. J. Multiphase Flow, 32, 692–717.
Mugele, R.A. & Evans, H.D. 1951 Droplet size distribution in sprays. Ind. Eng. Chem. 43,
1915–1931.
Namie, S. & Ueda, T. 1972 Droplet transfer in two-phase annular-mist flow. Bull. JSME 15,
1568–1580.
References 291
Okada, O., Fujimatsu, T., Fujita, H. & Nakajima, Y. 1995 Measurement of droplet size distribution
in an annular-mist flow in a vertical pipe by immersion liquid method. In Proceedings of the
2nd International Conference on Multiphase Flow, Kyoto, April, 3–7, IP2–11.
Pan, L. & Hanratty, T.J. 2002 Correlation of entrainment for annular flow in vertical pipes. Int.
Pan, Y. & Banerjee, S. 1996 Numerical simulation of particle interactions with wall turbulence.
Phys. Fluids 8, 2733–2755.
Pogson, J.T., Roberts, J.H. & Waibler, P.J. 1970 An investigation of the liquid distribution in
annular-mist flow. J. Heat Mass Transfer 92, 651–658.
Rosin, P. & Rammler, E. 1933 Laws governing the fineness of powdered coal. J. Inst. Fuel 7,
29–36.
Schadel, S.A., Leman, G.W., Binder, J.L. & Hanratty, T.J. 1990 Rates of atomization and
deposition in vertical annular flow. Int. J. Multiphase Flow 16, 363–374.
Semiat, R. & Dukler, A.E. 1981 Simultaneous measurements of size and velocity of bubbles and
drops, a new optical technique. AIChE Jl 27, 148–159.
Simmons, M.J.H. & Hanratty, T.J. 2001 Droplet size measurement in horizontal gas–liquid flow.
Squire, K.D. & Eaton, J.K. 1990 Particle response and turbulent modification in isotropic
turbulence. Phys. Fluids A2, 1191–1203.
Switherbank, J., Beer, J.M., Taylor, D.S., Abbot, D. & McCreath, G.C. 1976. A laser diagnostic
for the measurement of droplet and particle size distributions. Prog. Astronaut. Aeronaut. 1,
421–427.
Tatterson, D.F., Dallman, J.C. & Hanratty, T.J. 1977 Drop sizes in gas–liquid flows. AIChE Jl 23,
68–76.
Taylor, G.I. 1940 Generation of ripples by wind flowing over a viscous fluid. Reprinted in The
Scientific Papers of Sir Geoffrey Ingram Taylor, Vol. 3, ed. G.K. Batchelor. Cambridge:
Cambridge University Press, 1963, p. 244.
Wallis, G.B. 1969. One-dimensional Two-Phase Flow. New York: McGraw-Hill.
Wicks, M. & Dukler, A.E. 1966 In-situ measurements of drop size distribution in two-phase flow: a
new method for electrically conducting liquids. Paper presented at the Third International Heat
Transfer Conference, Chicago.
Willetts, I. 1987 Non-aqueous annular two-phase flow. Ph.D. thesis, University of Oxford.
Woodmansee, D.E. & Hanratty, T.J. 1969 Mechanism for the removal of droplets from a liquid
12 Horizontal annular flow
12.1 Prologue
Horizontal annular flows differ from vertical annular flows in that gravity causes
asymmetric distributions of the liquid in the wall layer and of droplets in the gas flow.
The understanding of this behavior is a central problem in describing this system.
Because of these asymmetries, entrainment can increase much more strongly with
increasing gas velocity than is found for vertical flows.
Theoretical analyses of the influence of gravity on the distribution of liquid in the
wall film and on the distribution of droplets in the gas phase are reviewed. As with
vertical annular flows, entrainment is considered to be a balance between the rate of
atomization of the wall film and rate of deposition of droplets. Because of the asym-
metric film distribution, the local rate of atomization varies around the pipe circum-
ference. This is treated theoretically by assuming that the local rate is the same as would
be observed for vertical annular flow. Gravitational settling contributes directly to
deposition so that the rate of deposition is enhanced. Thus, at low gas velocities,
entrainment can be much smaller for horizontal annular flows than for vertical annular
flows.
Equating the rates of atomization and deposition produces a prediction of the
entrainment for fully developed flow, equation (12.82). This equation is used in a
few calculations but, because of its complexity, simplified versions are used. Thus, in
correlating measurements of entrainment, Pan & Hanratty (2002b) use it to guide the
development of an empirical method for correlating data (equation (12.84)). A more
attractive approach is to match relations for low gas velocities, where gravity is having
a dominant role, and for high gas velocities, where equations developed for vertical
annular flow can be used.
The prediction of droplet distribution is an important consideration in predicting film
thickness at the top of the pipe. This can be used to predict the transition to annular flow.
It is also of importance in understanding the operation of large-diameter pipelines, where
it might be desirable to keep the top of the pipe wetted.
An important application of the theory, developed in this chapter, is the understanding
of the behavior of petroleum gas–condensate pipelines. These offer a challenge in
applying experimental results, obtained mainly for air–water flows, to conditions
which are quite different: pipes with diameters of 2–3 feet, high pressures, low surface
tensions.
12.2 Drop distribution and local gas velocity
12.2.1 Measurements
Impact tubes have been used by a number of researchers to measure the local drop flux,
the local gas velocity, and the slip ratio in vertical flows (Hewitt & Hall-Taylor, 1970;
Asali et al., 1985; Williams, 1990).
In the experiments of Asali et al., the tube was constructed with 1/8 inch stainless steel
tubing, which entered through the wall and faced upstream. The end of the tube was
beveled; its inside diameter was 1.457 mm. Local mass fluxes, GLE, were determined by
withdrawing liquid through the tube. The flux was calculated as the mass flow of the
liquid divided by the inside area of the tube. Williams (1990) and Asali et al. (1985)
tested the accuracy of the method by comparing fluxes obtained with tubes that have
different diameters.
The total entrainment, E, was calculated from measured droplet fluxes at different
locations over traverses from the pipe center to the time-averaged location of the wall
layer. Measurements could not be made too close to the layer, so an extrapolation of GLE
to the average location of the interface had to be made. This was the chief source of error
in determining E (about 6%).
Measurements of the difference in pressure at the opening of the impact tube and a
pressure tap on the wall of the pipe were used to calculate the local gas velocity. The
following equation, derived by Anderson & Mantzouranis (1960), was used:
1
Dp ¼ αρG u2G þ ec SuG GLE ð12:1Þ
2
where uG is the local time-mean gas velocity, α is the void fraction and ec is the capture
efficiency. For experiments in annular flow, all drops moving toward the impact tube
opening were captured, so ec = 1. The slip ratio, S, in equation (12.1) is the ratio of the
drop velocity to the gas velocity. It was assumed to be constant over the entire cross
section of the pipe; it was adjusted until the integrated gas-phase velocity profile agreed
with the measured gas flow rate into the system. Values of S determined in this way for
air–water flow in a 9.53 cm horizontal pipe (Williams, 1990) are presented in Figure 12.1.
(Figure 9.8 provides slip ratios of gas entrained in slugs, that are obtained from measure-
ments of void fractions in slugs.)
Concentrations were calculated by dividing the local drop flux by the local drop
velocity
GLE
C¼ ð12:2Þ
SuG
12.2.2 Maps of gas velocity and droplet flux

Impact tubes have been used to map the spatial variation of the gas velocity and droplet
flux in horizontal pipes. The impact tube was moved to obtain a traverse along a diameter.
0.9
0.8
Slip ratio
0.7
uLS (m/s)
0.6 0.03
0.06
0.5 0.09
0.12
0.4
25 30 35 40 45 50 55
uGS (m/s)
Figure 12.1 Slip ratios for air and water flowing in a vertical annular flow, estimated from (12.1) and the
measured gas velocity. Williams, 1990.
Since the distributions of gas velocities and droplet fluxes are asymmetric in a horizontal
annular flow, a device was needed to rotate the test section. This provided measurements
along diameters which were oriented at different angles to the vertical.
Maps of mean gas velocity for stratified air–water flows with and without atomization
were made by Dykhno et al. (1994). Measured velocity profiles along the vertical
diameter, for conditions at which little or no entrainment was observed, are presented
in Figure 12.2. The symbol y represents the distance from the crests of waves at the
bottom of the traverse. The lines in Figure 12.2a represent the average location of the
interface, hW. Figure 12.2b normalizes y with (dt – hW), rather than with dt. The maximum
velocities for the three runs were located in the lower half of the gas space. If the
maximum is the locale of zero shear stress, a force balance indicates that this maximum
would be located above the center of the gas space for a simple flow, because of the
roughness of the interface. However, the gas flow is not simple.
The layer on the bottom of the pipe is rougher than the layer on the top. This gives
rise to a variation of the drag of the gas on the liquid around the circumference of the
pipe. Hinze (1967) has demonstrated that this can cause a secondary flow in the gas,
which has associated with it an upward-oriented circumferential component of the
interfacial stress and a downward flow in the center of the pipe. Darling & McManus
(1969) developed an expression for the variation of the circumferential component of
the shear stress and argued for its importance in redistributing the wall layer. The results
presented in Figure 12.2 have been used to support the proposal by Darling &
McManus.
However, when a significant amount of entrainment is present, a different picture
emerges. Figure 12.3 shows velocity profiles measured under these circumstances.
The maximum of the mean velocity is located above the center of the gas space.
Dykhno et al. suggest, for these cases, that a secondary flow exists in the gas whereby
the flow is downward at the wall and upward at the center. They propose that this occurs
because the concentration of drops is higher close to the wall. The force of gravity on
(a) Top
0
Run 1
0.2 Run 2
Run 3
0.4
y/dt
0.6
3
0.8
2
1
1
5 10 15 20 25 30 35
uG (m/s)
(b) Top
0
Run 1
0.2 Run 2
Run 3
y/(dt − hW)
0.4
0.6
0.8
1
0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
uG/uGM
Figure 12.2 Regular and normalized velocity profiles for conditions with no and small amounts of atomization.
Run 1, uGS = 14.4 m/s, uLS = 1 cm/s. Run 2, uGS = 23.2 m/s, uLS = 4 cm/s. Run 3, uGS = 17 m/s, uLS
= 8 cm/s. Dykhno et al., 1994.
these drops is associated with a downward pull on the gas, which is larger close to
the wall.
Figure 12.4 shows contours of the streamwise component of the velocity of the
gas (isotachs), determined by Dykhno et al. (1994), where the velocities are normalized
with the maximum. Run 1 (uGS = 14.4 m/s, uLS = 0.01 m/s) is a stratified flow for
which atomization is not occurring. Run 2 (uGS = 23.2 m/s, uLS = 0.04 m/s) and Run 3
(uGS = 17 m/s, uLS = 0.08 m/s) are for stratified flows for which only a small amount of
atomization is occurring. These represent situations where the concentration of drops is
zero or close to zero.
The use of the method of Prandtl (1927) indicates a secondary pattern in the top of the
pipe with downward flow at the center plane and upward flow at the wall. The upward
bulges in the bottom part of the gas space are not understood. Dykhno et al. (1994)
discuss possible interpretations.
Figure 12.5 presents measurements of isotachs for situations in which a significant
amount of liquid is entrained in the gas. The dotted lines in Figures 12.4 and 12.5
represent the locations of the tops of the waves.
(a) Top
0
0.2 Run 4
Run 5
Run 6
0.4 Run 7
y/dt
Run 8
0.6
0.8
4
5 6 7 8
1
15 25 35 45 55 65
uG (m/s)
(b) Top
0
0.2
y/(dt−hW)
0.4
0.6
Run 4
0.8 Run 5
Run 6
Run 7
1
0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
uG/uGM
Figure 12.3 Regular and normalized velocity profiles for conditions with a large amount of atomization. Run 4,
uGS = 25.3 m/s, uLS = 9 cm/s. Run 5, uGS = 37.2 m/s, uLS = 6 cm/s. Run 6, uGS = 44.5 m/s, uLS =
6 cm/s. Run 7, uGS = 44.5 m/s, uLS = 9 cm/s. Run 8, uGS = 43.2 m/s, uLS = 3 cm/s. Dykhno et al., 1994.
Run 1 Run 2 Run 3
0.7 0.7
0.8 0.8
0.7 0.85 0.85
0.8 0.9 0.9
0.85 0.95
0.95
0.9 0.98
0.95 0.98 1.0
0.98 1.0
1.0
a b c
Figure 12.4 Velocity contours (isotachs) for conditions with no or small amounts of atomization, normalized
with the maximum velocity. Dykhno et al., 1994.
Run 4 Run 7 Run 8
1.0
1.0
0.98
1.0
0.98 0.95 0.98
0.95
0.95
0.9 0.9 0.9
0.85 0.85
0.8 0.85
0.8 0.8
a b c
Figure 12.5 Velocity contours (isotachs) for conditions with a large amount of atomization, normalized with the
maximum velocity. Dykhno et al., 1994.
Lin & Hanratty (1987) defined annular flow as a situation for which the top of the
pipe is wetted with a turbulent film (which need not contain disturbance waves). Run 4
(uGS = 25.3 m/s, uLS = 0.09 m/s) represents a situation close to the transition from
stratified to annular flow. Run 7 (uGS = 44.5 m/s, uLS = 0.09 m/s) represents an annular
flow for which the wall is wetted with disturbance waves over the bottom two-thirds of
the pipe wall. The bulges of the isotachs in the bottom of the pipe for runs 4 and 7 are
consistent with the existence of an upward secondary flow in the center plane of the
pipe.
Run 8 was performed at the same gas velocity as run 7 but at a much lower liquid flow
(uGS = 43.2 m/s, uLS = 0.03 m/s). It represents conditions close to the transition to annular
flow. It is a situation for which the maximum is located approximately at the center of the
pipe. This seems to be a condition for which entrainment is small, but significant.
Secondary flows due to density gradients could be just balancing a secondary flow due
to the circumferential variation of the wall roughness.
Measurements of droplet fluxes by Williams (1990) in a 9.53 cm horizontal pipe and
by Paras & Karabelas (1991) in a 5.08 cm horizontal pipe show much larger variations in
the vertical direction than in the horizontal direction. This prompted the development of a
diffusion model, by Paras & Karabelas, for which the concentration is constant in planes
perpendicular to the direction of gravity. The determination of C(r, θ) then requires only a
calculation of C(y), where y is the vertical axis.
12.2.3 Stochastic calculations for a 2-D rectangular channel

A simple system to begin a discussion of the diffusion model is the flow of a dilute
suspension of spheres with uniform size in an infinitely wide rectangular channel that
is horizontal. The particle density is considered to be much larger than the fluid density.
The coordinates perpendicular to the wall and in the direction of mean flow are x2 and x1.
For this field, mechanisms for promoting secondary flow are absent. The concentration
field is assumed to be fully developed, so that the concentration varies only with x2. The
development of a stochastic model for fluid turbulence, described in Section 10.11,
provided an opportunity to study the flow of a gas and suspended particles in a horizontal
channel over a wide range of conditions (Mito & Hanratty, 2005). Annular flow was
simulated by injecting particles with much larger densities than the continuum at the top
and bottom walls. By adjusting the relative rates of admission of particles at the two
walls, a fully developed condition can be realized. The concentration field is calculated
by using a Lagrangian, rather than an Eulerian, approach whereby the field is described as
resulting from distributions of point sources at the two walls. Particles are removed from
the field when they are a distance of dP/2 from a wall.
The location and velocity of the particles are described by the equations
dxi
¼ vPi ð12:3Þ
dt
dvPi 3ρ C D ! !
¼ f j u v P jðvPi ui Þ þ gi ð12:4Þ
dt 4d P ρP
where ! u is the instantaneous fluid velocity and ! v P is the instantaneous particle velocity.
Coordinate x2, perpendicular to the walls, has a value of zero at the bottom wall. Unlike
the approach outlined in Chapter 10, the force of gravity is included in the equation of
motion of a particle.
Calculated dimensionless concentration fields are presented for τ þ P ¼ 3 in Figure 12.6,
and for τ þ
P ¼ 20 in Figure 12.7, where τ þ
P ¼ τ P v
= f is the dimensionless time constant of
a particle
4d þ
P ðρP =ρf Þ
τþ
P ¼ ð12:5Þ
3C D jvþ
P u j
þ
105 V T+ = 0
104 V T+ = 1.7 × 10−4
103
V T+ = 1.7 × 10−3
102
Cv*/RAb
101 V T+ = 7.1 × 10−3

0
10 V T+ = 0.014
−1
10
V T+ = 0.028
10−2
10−3 V T+ = 0.058
0 200 400 600 800 1000
C0exp(−VTx2/εP)
x +2
Figure 12.6 Concentration profiles in a channel for τ þ

P ¼ 3, calculated by Mito & Hanratty (2005), where x2 is
the distance from the bottom wall. Mito & Hanratty, 2005.
V T+ = 0
102
V T+ = 1.7 × 10−4
1
10
V T+ = 1.7 × 10−3
C v*/RAb
100 V T+ = 0.014
V T+ = 0.012
10−1
V T+ = 0.021
10−2
V T+ = 0.044
10−3 V T+ = 0.086
0 200 400 600 800 1000
C0exp(−VTx2/εP)
x +2
Figure 12.7 Concentration profiles in a channel for τ þ

P ¼ 20, calculated by Mito & Hanratty (2005), where x2 is
the distance from the bottom wall. Mito & Hanratty, 2005.
For a Stokes law resistance

d þ2
P ðρP =ρf Þ
τþ
PS ¼ ð12:6Þ
18
The dimensionless acceleration of gravity and free-fall velocity are defined as gþ ¼

g=v and V þ þ þ þ
T ¼ V T =v , where V T ¼ τ P g .
Since the concentration field is fully developed, the rates of atomization and deposition
at the top and bottom walls satisfy the condition that
RAb ¼ RDb and RAt ¼ RDt ð12:7Þ
Since the net fluxes are zero at all x2 , the representation of the concentration field in terms
of fundamental mechanisms responsible for particle transport and mixing is simplified.
Thus, the flux of particles is not dependent on x2 , so
d
ðCvP2 Þ ¼ 0 ð12:8Þ
dx2
The concentration and the velocity are represented as the sum of an average and a
turbulent fluctuation. Thus
CvP2 ¼ CvP2 þ C t vt P2 ð12:9Þ
Turbophoresis (see Section 10.9) is ignored. The mean velocity in the x2-direction is then
given as
vP2 ¼ gτ P ¼ V T ð12:10Þ
If a diffusion model is used

dC
C t vtP2 ¼ εEP ð12:11Þ
dx2
Substitute (12.10) and (12.11) into (12.9) to get

dC
εEP V TC ¼ 0 ð12:12Þ
dx2
In the central regions of the channel, a reasonable assumption is that εEP is constant so that

VT
C ¼ C 0 exp E x2 ð12:13Þ
εP
where x2 is the distance from the bottom wall and C0 is the concentration at the bottom
wall. This diffusion model was proposed by O’Brien (1930) and Rouse (1937) to
describe sediment distribution in open channels and rivers.
From equation (8.4), the free-fall velocity is given as
4d P g ðρL ρG Þ
V 2T ¼ U 2S ¼ ð12:14Þ
3C D ρG
where CD is the drag coefficient. For Stokes law,

24
CD ¼ ð12:15Þ
ReP
For 1.92 < ReP < 500, the drag coefficient can be approximated as
18:5
CD ¼ ð12:16Þ
Re0:6
P
where ReP ¼ d P V T ρf =μf .

The dispersion of fluid and solid particles downstream of a point source in a pipe at
large times can be represented with a turbulent diffusivity. Section 10.7 shows that the
diffusivities of fluid and solid particles in such a situation are approximately the same.
Experiments give equation (10.22) for pipe flows:
εP ¼ 0:074rt v ð12:17Þ
For channel flows, H, the half-height of the channel, would be substituted for rt
εP ¼ ξHv ð12:18Þ
with ξ = 0.074. Since the use of an eddy diffusivity model in an Eulerian formulation is
not fundamentally correct (see Hanratty, 1956), ξ must be looked upon as an empirical
constant. It need not be the same for different systems.
If (12.18) is substituted into (12.13),

C V T x2
¼ exp ð12:19Þ
C0 ξHv
The dotted lines below the curves in Figures 12.6 and 12.7 represent (12.19) with ξ = 0.094.
(This is close to the value of 0.074, estimated from studies of point source diffusion.)
Note that particles injected at the bottom wall do not reach the top wall for
Vþ þ
T ≥ 0:11; 0:16; 0:31; 0:5 at τ P ¼ 5; 10; 20; 40. These results indicate that a limit for
the existence of annular flow can be roughly defined as g+ ≤ 0.012 for this range of τ þ P.
+
In the saltation regime, g > ~ 0.04, the trajectories of the particles are not affected by
fluid turbulence and the behavior depends strongly on the velocity with which the
particles enter their field (see Mito & Hanratty, 2005).
12.2.4 Formulation of the diffusion equation for particles in a pipe

The formulation of the proposal of Paras & Karabelas for a circular pipeline produces a
more complicated result than presented in the previous subsection. Concentrations are
assumed to be constant in horizontal planes. The location of these planes is given by the
distance above the bottom of the pipe. A mass balance over a differential volume
considers that the net flux due to free-fall and turbulent mixing equals the net contribution
of sources and sinks of drops that result from atomization and deposition at the boundary.
Thus

dC
d V T Cw þ wεP ¼ ðRA RD Þrt dθ ð12:20Þ
dy
where rt is the pipe radius, w is the width of the control volume and θ is measured from
the bottom of the pipe. The average of (RA – RD) around the pipe circumference is zero
since the field is fully developed. However, atomization is larger at the bottom of the pipe
because the wall film is thicker at that location. Thus, the wall is a source of drops at the
bottom and a sink at the top. The term on the right side of (12.20) causes a net transport of
liquid from the bottom of the pipe to the top.
The concentration C can be considered as an average over w for a given y. In order
to develop a solution of (12.20) either y or θ needs to be eliminated. The width w is
given as
w ¼ 2rt sinθ ð12:21Þ
Furthermore,
rt y ¼ rt cosθ ð12:22Þ
The following relation is obtained from (12.22):

δy
dθ ¼ ð12:23Þ
rt sinθ
so
dC dC 1
¼ ð12:24Þ
dy dθ rt sinθ
Thus, y or θ can be eliminated from (12.20) by using (12.24) and (12.21).

The implementation of (12.20) introduces complications not present in (12.12). One

needs to develop relations for local rates of atomization and deposition around the pipe
circumference. The width of the control volume, w, varies with y; this could have an
impact on the spatial variation of C.
The integration of (12.20) gives
ðy
dC ðRA RD Þ dy
V T C þ εP ¼ þB ð12:25Þ
dy 0 w sinθ
where B is the integration constant and RD, RA, w are functions of y. The right side can be
considered as a flux of droplets due to atomization and deposition. Paras & Karabelas
(1991a, b) represented the right side by parameter a, so that
dC
V T C þ εP ¼a ð12:26Þ
dy
They assumed, for a given set of conditions, that a may be considered independent of y.
They chose εP = 0.1rtv* and developed an empirical correlation for a.
12.2.5 Concentration distributions measured in a pipe

Baik & Hanratty (2003) presented measurements of Paras & Karabelas (1991a, b) for
air–water flow in a 5.08 cm pipe and of Williams et al. (1996) in a 9.53 cm pipe. They
used equation (12.19) to correlate the data by ignoring the effects described by the
right side of (12.25) so that C(y) is given by the same equation as for a rectangular
channel,

C V Ty
¼ exp ð12:27Þ
C0 ξrt v
The terminal velocity is given by (12.14). Thus, the application of (12.27) to annular flow
requires information on the drop size, dP, in order to evaluate VT.
For dilute concentrations droplet–droplet interactions can be neglected so (12.27) can
be combined with a relation for the probability distribution function for the drop size to
predict C ðyÞ. Baik & Hanratty (2003) considered the suspension as having a single drop
size, the volume median diameter. The equation chosen to predict dvμ was (11.28), that is,
1=2
d t ρG u2G d vμ
¼ 0:1072. An important aspect of this choice is that it predicts that
σ dt
dvμ depends on the pipe size.
Concentrations were calculated from measured droplet fluxes by using (12.2),
C ¼ GLE =SuG , where uG is the local gas velocity. “To simplify the analysis,” Baik &
Hanratty substituted uGS for uG. From Figure 12.1, a value of S = 0.7 was used. A bulk
mean concentration is calculated from the entrainment, E, as
EW L
hCi ¼ ð12:28Þ
QG S
where QG is the volume flow of gas and WL is the mass flow of liquid. The concentration,
C0, that appears in (12.27) is related to hCi by integrating (12.27) over the cross-section
and assuming a plug flow
ð
1 V Ty
hCi ¼ C 0 w exp dy ð12:29Þ
A ξrt v
The measurements by Williams et al. and by Paras & Karabelas are presented in Figures
12.8 and 12.9. The numbers in parentheses are values of V T =v . The lines were calculated
using (12.27), where C0 is given by (12.29) and ξ = 0.165. This result is to be compared
with ξ = 0.094, obtained from a stochastic analysis of flow in a two-dimensional channel
presented in Section 12.2.3, and 0.074, obtained from measurements of dispersion from
point sources located in the center of a pipe. (See equation (10.22), recognizing that it
uses 2rt as a length scale.)
The behavior defined by (12.27) and (12.29) indicates a strong influence of drop size
because of its influence on VT (see equation (12.14)). Previous papers by Pan & Hanratty
(2002a, b) used an equation for drop size which indicates no effect of pipe diameter. They
needed to use different values of ξ (0.04 and 0.08) to fit the data of Williams in a 9.53 cm
pipe and of Paras & Karabelas in a 5.08 cm pipe. The results presented in Figures 12.8
and 12.9 provide support for the finding that drop size depends on pipe diameter.
The rough fits displayed in Figures 12.8 and 12.9 suggest that there is room for
improvement both in theory and in measurements. A correlation that includes, directly,
the influences of RA and RD could introduce new dimensionless groups. A possible
improvement can also be obtained by abandoning the assumption of plug flow, so that C
is defined by using the local velocity, rather than uGS. Some justification for using the
simplified approach can be obtained from the study in a 9.53 cm pipe. Williams et al.
(0.064)
(0.086)
(0.162)
C/C0
0.1
(0.243)
uLS = 0.12 m/s uGS = 30.1 m/s

uLS = 0.12 m/s uGS = 45.4 m/s
uLS = 0.02 m/s uGS = 32.2 m/s
uLS = 0.02 m/s uGS = 49.6 m/s
0.01
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
y/dt
Figure 12.8 Concentration measurements for air–water flow by Williams et al. (1996). The solid lines are
(12.27), with ξ = 0.165. The values of VT/v* are shown in parentheses. Williams et al., 1996.
1
(0.035)
(0.072)
(0.127)
(0.185)
C/C0
0.1
(0.283)
uLS = 0.06 m/s uGS = 31 m/s
uLS = 0.06 m/s uGS = 37 m/s
uLS = 0.06 m/s uGS = 44 m/s
uLS = 0.06 m/s uGS = 68 m/s
uLS = 0.06 m/s uGS = 90 m/s
0.01
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
y/dt
Figure 12.9 Comparison of concentration measurements for air–water flow in a 5.08 cm pipe, by Paras &
Karabelas (1991a,b), where y is the distance from the bottom wall. The solid lines are equation
(12.27). Paras & Karabelas, 1991b.
(1996) measured the velocity profiles for some of the runs in which droplet fluxes were
measured, so that concentration profiles could be calculated with local gas velocities.
Baik & Hanratty (2003) reveal that results obtained in this way “were not appreciably
different from those obtained by using the plug flow assumption.”
The finding that drop size in a horizontal pipe depends on location in large-diameter
pipes (see Chapter 11) encourages an adaptation of the analysis so as to consider drop size
distribution.
12.3 Distribution of liquid in the wall layer of a horizontal pipe
12.3.1 Description of the wall film

Dykhno (1996) gives the following description of the flow of air and water in a 9.53 cm
pipe in an internal communication within the Hanratty research group.
At low superficial gas velocities (uGS = 5–9 m/s), a stratified flow is observed. At 12 m/s, waves
increase in amplitude with increasing gas velocity and have steep spilling fronts. They are the roll
waves described by a number of authors. The spreading of waves up the pipe wall is observed with
increasing gas velocity. At approximately 17 m/s, the gas flow starts to atomize the waves on the
liquid interface. Drops deposit on the wall and, in this way, greatly enhance the wetting of the wall
at locations above the pool of liquid on the bottom of the pipe. At uGS = 24 m/s, uLS = 0.09 m/s, a
thick layer with roll waves exists at the bottom of the pipe and a continuous film covered the upper
part of the pipe. The formation of a continuous film is slightly dependent on the liquid velocity.
Thus, at a superficial liquid velocity of 0.02 m/s it occurred at a greater superficial gas velocity,
30 m/s. As the height of the wall layer increased, disturbance waves of the type seen in vertical
annular flows exist over part of the film. These are haphazard packets of capillary waves which
could be considered packets of turbulence. Atomization occurs over these disturbance waves. This
enhanced the amount of liquid entrained in the gas. Eventually, at high enough gas velocities,
disturbance waves covered the whole circumference.
Dye injected into the film was observed to move downward (due to gravitational pull) when no
disturbance waves were present. However, when the dye encountered a disturbance wave, it was
observed to disperse explosively, in the circumferential direction, as would be expected in a
turbulent patch.
Pan & Hanratty (2002) and Williams et al. (1996) developed a framework from
which to describe entrainment in horizontal pipes. Let θ = 0 be the bottom of the pipe.
The local rate of atomization, RA, is assumed to be the same as for a symmetric flow
(equation (11.40))
k 0 Au2G ðρG ρL Þ1=2

RA ¼ ðΓ Γ c Þ ð12:30Þ
σ
where Γ is the local mass flow rate per unit perimeter length and Γc is the critical film flow
below which atomization does not occur.
The asymmetry of the wall layer is usually characterized by the ratio of the kinetic
energy of the gas, represented by ρGu2GS , to the work against gravity which is needed to
lift liquid from the bottom of the pipe to the top, ρLgdt, that is, the dimensionless group
1=2 1=2
ρG uGS =ðρL gd t Þ . Thus, the asymmetry can be characterized by a modified Froude
number, (ρG/ρL)1/2 Fr, where Fr = uGS/(gdt)1/2.
Conductance techniques have been used to measure the variation of the height of the
wall layer around the pipe circumference. (See the thesis by Williams (1990) for a
detailed description of the technique.) Hurlburt & Newell (2000) suggest that the
asymmetry of the wall layer can be characterized by the ratio of the height of the layer
at the bottom of the pipe to the average height around the circumference, hLB =hhL i.
Plots of hLB =hhL i for studies of air–water flow in horizontal pipes with diameters of (a)
dt = 9.53 cm, (b) dt = 5.08 cm, (c) dt = 2.31 cm have been constructed by Pan & Hanratty
(2002b). These are presented in Figure 12.10. Note that, for large Froude numbers, the
measurements are close to what is expected for a symmetric flow, hLB =hhL i ≈ 1. For small
Fr; hLB =hhL i approaches a value close to that suggested by the idealized model of a
stratified flow used in Chapter 5. The transition between these two extremes is seen to be
sharp for a 9.53 cm pipe. For a ρG/ρL characteristic of an air–water flow, it occurs at
Fr ≈ 47.
12.3.2 Asymmetric wall layer model

Henstock & Hanratty (l976, Section 3.3) used the following relation for the height of the
wall layer for vertical annular flow (equation (3.46)), where m signifies the average of the
height of the wall layer around the circumference:
m 6:59
¼ ð12:31Þ
d t ð1 þ 1400F Þ1=2
(a)
5
3
hLB / hL
uG/(gdt)0.5
uG (m/s)
2 32, 31
38, 37
47, 45
1 69, 67
dt = 9.53 cm 91, 88
0
0 0.3 0.6 0.9 1.2
WL (kg/s)
(b)
7
6 dt = 5.08 cm uG/(gdt)
0.5
uG (m/s)
5 16, 11
20, 14
hLB / hL
4 25, 18
34, 24
3 48, 34
62, 44
81, 57
2 99, 70
146, 103
1 186, 131
0
0 0.5 1 1.5 2
WL (kg/s)
(c)
4
3
0.5
uG/(gdt)
uG (m/s)
hLB / hL
2
32, 15
46, 22
69, 33
1 92, 44
116, 55
dt = 2.31 cm 185, 88
0
0 0.5 1 1.5 2
WL (kg/s)
Figure 12.10 Measurements of the ratio of the liquid height at the bottom of the pipe to the average around the
circumference: (a) Air–water in a horizontal pipe with dt = 9.53 cm. (b) Air–water with dt = 5.08 cm.
(c) Air–water with dt = 2.31 cm. Pan & Hanratty, 2002b.
The flow factor, F, is defined as

γðReLF Þ L ρL 1=2
F¼ ð12:32Þ
Re0:9
G G ρG
and
h 2:5 2:5 i0:4
mþ ¼ γðReLF Þ ¼ 0:707Re0:5
LF þ 0:0379Re0:9
LF ð12:33Þ
(equation (3.17)). The friction factor, defined in the equation below

1
τ i ¼ ρG u2G f i ð12:34Þ
2
is given as
fi
¼ 1 þ 1400F ð12:35Þ
fS
(equations (3.41) and (3.44)), where fS is the friction for a smooth wall
f S ¼ 0:046Re0:2
G ð12:36Þ
These expressions were modified by Dallman (1978) so as to describe horizontal annular

flow for large Fr (see Laurinat et al., 1984).
For horizontal flow, it is desired to calculate the average height around the circum-
ference, m ¼ hhL i. The argument is made that, locally, the relation of the film height to
the volumetric flow per unit pipe perimeter, Γ ðθÞ, is the same as for vertical annular flow.
Equation (12.33), for vertical annular flow, is an interpolation formula where the first
and second terms on the right side represent laminar and turbulent behavior, respectively.
For laminar flow (equation (3.7))
Γ hþ2
¼ L ð12:37Þ
L 2
where
hL ðτ i =ρL Þ1=2
hþ
L ¼ ð12:38Þ
L
Integrate (12.37) around the circumference to obtain

hΓ i 1 þ2
¼ hhL i ð12:39Þ
L 2
where
hΓ L i 1
¼ ReLF ð12:40Þ
L 4
The dimensionless spatially averaged film thickness is defined as
hhL ihτ i =ρL i1=2

mþ ¼ ð12:41Þ
L
Thus, from (12.39) and (12.40),

hhþ2
L i
ReLF ¼ 2mþ2 ð12:42Þ
mþ2
for laminar flow since, for an asymmetric wall layer, the average of the square, hh2L i, is not
equal to the square of the average, m2.
For high ReLF, the local film height is defined by the second term on the right side of
(12.33)

þ 4Γ L 0:9
hL ¼ 0:0379 ð12:43Þ
L
or

þ 1:1 4Γ L
hL ¼ ð0:0379Þ1:1 ð12:44Þ
L
The integration of (12.44) around the pipe circumference gives
0:0379Re0:9
mþ ¼ LF
0:9
ð12:45Þ
hhþ1:1
L i =mþ
For a horizontal pipe, the asymptotic behaviors for small and large ReLF are (12.42)
and (12.45). Measurements show that an interpolation formula of the same form
as (12.33) can be used to describe m+. Dallman (1978) approximated hhþ2 L i=m
þ2
þ1:1 0:9 þ 2 1=2 1:1 0:9
and hhL i =m , respectively, as hhL i =m and hh i =m. Experiments in a
1=2
2.54 cm pipe indicate that hh2L i =m approaches 1.4 at small ReLF. Thus, for
ReLF ! 0,
mþ ¼ 0:50Re0:5
LF ð12:46Þ
in a horizontal pipe. Dallman (1978) also presented measurements which suggest that a
0:9
value of hhþ1:1
L i =mþ of 1.38 can be used for the high Reynolds number correction so
that for ReLF ! ∞
mþ ¼ 0:028Re0:9
LF ð12:47Þ
Laurinat et al. (1984) suggested the following interpolation formula to represent meas-
urements of m+ (shown in Figure 12.11), which provides asymptotic behaviors close to
(12.46) and (12.47)
h 2:5 2:5 i0:4
mþ ¼ γH ðReÞ ¼ 0:566 Re0:5
LF þ 0:0303 Re0:9
LF ð12:48Þ
Note that 0.566 replaces 0.50 in (12.46) and that 0.0303 replaces 0.028 in (12.47). Good
agreement between (12.48) and measurements is noted. For comparison, equation
(12.33), representing data in vertical pipes, is also presented.
103
Equation (12.33)
102
m+
WGkg/s
101 0.0174
0.0252
0.0378
0.0500
0.0630
Equation (12.48)
100 1
10 102 103 104
ReLF
Figure 12.11 Correlation of film height measurements taken in a 2.54 cm pipe. Open points are for ρG =
1.34 kg/m3. Filled points are for ρG = 2.73 kg/m3. Laurinat et al., 1984.
A flow factor analogous to F can be defined for horizontal flows

rffiffiffiffiffi
γ ðReLF Þ L ρL
F H ¼ H 0:9 ð12:49Þ
ReG G ρG
A plot of m/dt against FH is presented in Figure 12.12. The data are represented quite
well by
m 6:59F H
¼h i0:2 ð12:50Þ
dt
ð2:3Þ0:5 þ ð90F H Þ0:9
(See Laurinat et al. (1984) for the development of equation (12.50).)

A correlation for the frictional pressure loss

dp
¼ 2f i ρ u2 ð12:51Þ
dx d t G GS
f
is also given by Laurinat. Measurements of fi / fS are plotted against m/dt in Figure 12.13,
where fS is the friction factor for a smooth wall, for air–water flow in a 2.54 cm horizontal
pipe. For a given m/dt, gas velocity increases with increasing WL. Such a plot has been
successful for vertical annular flows. At high gas velocities (greater than 30 m/s), where
the wall layer is described as an asymmetric film, the friction factor approaches an
asymptotic behavior. However, it fails to do this at small gas velocities, where the wall
layer is highly asymmetric, resembling a stratified flow with an agitated surface. The
friction factor relation at low gas velocities is a strong function of liquid Reynolds
number. This region is best treated as a stratified-annular flow (see Section 12.5).
10–1
10–2
Equation
(12.50)
m/dt
10–3 Butterworth & Pulling (1973)

Swanson (1966) 2.54 cm pipe
Swanson (1966) 5.08 cm pipe
Laurinat et al. (1984) 2.54 cm pipe
Laurinat et al. (1984) 5.08 cm pipe
10–4 –4
10 10–3 10–2 10–1
FH
Figure 12.12 Film holdup correlation. Laurinat et al., 1984.
10–1
10–2
fi/fS
WLkg/s WLkg/s
.0032 .076
10–3 .0063 .113
.0126 .151
.0252 .183
.032 .208
.050 .252
10–4 –3
10 10–2 10–1
m
dt
Figure 12.13 Comparisons of the friction factor ratio taken in a 2.54 cm pipe at 500 diameters from the entry.
Laurinat et al., 1984.
12.3.3 Distribution of the liquid in an asymmetric wall layer

A number of investigators have considered the modification of the behavior of the
wall layer by gravitational draining (McManus, 1961; Dallman, 1978; Laurinat,
1982; Pletcher & McManus, 1965; Lin et al., 1985; Fukano & Ousaka, 1989;
Butterworth, 1972; Butterworth & Pulling, 1973; Anderson & Russell, 1970;
Hutchinson et al., 1974; Fisher & Pearce, 1979; James & Burns, 1979; Wilkes
et al., 1980; Swanson, 1966). Several mechanisms for redistributing the wall layer
have been proposed. These are incorporated into a single equation by Laurinat et al.
(1985).
In formulating the problem, it is assumed that the wall layer is thin enough compared
with the pipe radius for the momentum and mass balance equations to be written in
Cartesian, rather than polar, coordinates. The x-coordinate is in the circumferential
direction; the y-coordinate is the distance from the wall; the z-coordinate is in the
flow direction. The goal is to predict hL ðθÞ from mass and momentum balances in the
x-direction, where x = rtθ and rt is the pipe radius.
The instantaneous volumetric flow in the x-direction, per unit circumferential length, is
given as Γ x ¼ huLx ihL .
The analytic framework used by Laurinat et al. (1985) is analogous to the formulation
for droplet concentration. Since uLx ¼ uLx þ utLx and hL ¼ hL þ htL
Γ x ¼ huLx ihL þ hutL x htL i ð12:52Þ
The inclusion of the turbulence term is motivated by visual observation of the dispersion
of dye by the disturbance waves (Section 12.3.1). This term can be represented by a
diffusivity model so that
dhL
utLx htL ¼ εh ð12:53Þ
dx
Deposition increases the volumetric flow in the film. Atomization decreases the volu-
metric flow, so that atomization and deposition contribute to a local change of the flow in
the film.
dΓ x RD RA
¼ ð12:54Þ
dx ρL ρL
ðx
RD RA
Γx ¼ dx ð12:55Þ
0 ρL
At the top of the pipe, x = πrt. Since a fully developed flow is assumed
ð πrt
RD RA
dx ¼ 0 ð12:56Þ
0 ρL
Thus, (12.55) describes a redistribution of the liquid flowing in the wall layer.
Conservation of mass gives
ðx
RA RD dhL
dx ¼ huLx ihL εh ð12:57Þ
0 ρL dx
where huLx i is the spatially averaged value of uLx at a given x.

A momentum balance in the x-direction is used to define huLx i. Surface tension effects
are ignored. From equation 12 in Laurinat et al. (1985)
∂τ xy ∂τ xx dhL
þ ρL g sin θ ρL g cos θ ¼0 ð12:58Þ
∂y ∂x dx
where τxy and τxx are local shear and normal stresses in the liquid, and
τ xx ¼ ρL ut2
Lx ð12:59Þ
Integrate (12.58) between y and hL . The variation of ut2 Lx with y is ignored. (A constant
value is used.) Then

dτ xx dhL
hL y þ τ ix τ xy ðyÞ hL y ρL g sin θ þ cos θ ¼0 ð12:60Þ
dx dx
where τ ix is the stress imposed at the interface in the x-direction. Take

τ xy μL t duLx
¼ þ ð12:61Þ
ρL ρL dy
where t is the turbulent eddy viscosity. Substitute (12.61) into (12.60). Integrate once to
obtain uLx ðyÞ. Integrate again to obtain huLx i.

τ ix I 2 dτ xx dhL
huLx ihL ¼ I 1 þ þ I 2 g sin θ þ cos θ ð12:62Þ
ρL ρL dx dx
ð h 1 ð y2
ðN Þ N
I1 ¼ dy1 dy2 ð12:63Þ
0 0 ½ ð μ L LÞ þ
=ρ t
ð h 1 ð y2
ðN Þ N ðhL yÞ
I2 ¼ dy1 dy2 ð12:64Þ
0 0 ½ðμL =ρL Þ þ
t
Substitute (12.62) into (12.57) to obtain

ðx
dhL ðRA RD Þ τ ix I 2 dτ xx dhL
ε þ ¼ I1 þ þ I 2 g sin θ þ cos θ
dx 0 ρL ρL ρL dx dx
ð1Þ ð2Þ ð3Þ ð4Þ ð5Þ
ð12:65Þ
This equation defines hL ðθÞ. The left side represents mixing due to turbulence and to
atomization/deposition. The right side represents contributions to huLx i due to forces
acting on the film. Term (3) considers forces due to gas-phase drag on the interface in the
x-direction. These could result from secondary flow in the gas and/or an orientation of the
interfacial waves to the direction of mean flow. Term (4) represents the influence
of gradients of turbulent normal stresses in the circumferential direction. Term (5)
represents the contribution of the force of gravity on the layer. The cosine term in (5)
is important only in the neighborhood of the bottom of the pipe where sin θ = 0.
Equation (12.65) has not been fully exploited.
A number of investigators have used only (2) and (5) (Anderson & Russell, 1970;
Hutchinson et al., 1974; Fisher & Pearce, 1979; James & Burns, 1979; Wilkes et al.,
1980). These and other papers show that a deposition-atomization mechanism, alone,
cannot account for the variation of film height. Darling & McManus (1969) used (3) and
(5) to calculate hL ðθÞ. Laurinat et al. (1985) used (2), (3), (4) and (5). Their motivation
was to use term (4) to describe the influence of turbulent velocity fluctuations. They
concluded that this approach was not adequate.
This led Pan & Hanratty to introduce term (1) to represent turbulent mixing in the wall
layer. For cases where disturbance waves cover the whole circumference, they used only
terms (1) and (5) in equation (12.65):

dhL dhL
εh ¼ I 2 g sin θ þ cos θ ð12:66Þ
dx dx
The turbulent mixing coefficient, εh, can be pictured as the product of a velocity and a
length. It is reasonable to assume that the velocity scales with the friction velocity. The
mixing in the disturbance waves can be pictured as a large-scale phenomenon. Pan &
Hanratty, therefore, modeled the turbulent diffusivity as
ε h ¼ ξ h r t v ð12:67Þ
where vis the friction velocity, rt is the pipe radius and ξh is a dimensionless constant.
Equation (12.66) can be made dimensionless using
r t v þ v3 þ hL v þ
rþ
t ¼ I2 ¼ I2 h ¼ g ¼ g 3 ð12:68Þ
2 L v
The following relation is obtained from (12.66)

dhþ þ þ dhþ
ξh L
¼ g I 2 sin θ þ cos θ þ L
ð12:69Þ
dθ dx
Pan & Dykhno (2001, private communication) compared (12.69) with measurements (in
unpublished calculations). The results are presented in Figure 12.14 and in Figure 12.15.
From (12.64), it is seen that I2 varies with hL(θ) for given flow conditions. One expects that
hL / rt or that hþ þ þ þ þ
L / rt . Thus, I 2 / r t . Plots of I 2 calculated from measurements by
Dallman in a 2.54 cm pipe (Run 67, uGS = 78.18 m/s, uLS = 0.0572 m/s, rþ t ¼ 1626, ρG =
1.538 kg/m3 ) and by Laurinat in a 5.08 cm pipe (Run 46, uGS = 56.8 m/s, uLS = 0.0752 m/s,
rþ
t ¼ 2775, ρG = 2.039 kg/m are shown in Figure 12.14. Here, θ is measured from the top
3

of the pipe. It is seen that I þ þ
2 ðθ ¼ 180 Þ minus I 2 ðθ ¼ 0 Þ are, respectively, 2600 and 1200
þ
for the 2.54 cm and 5.08 cm pipes. The change in I 2 scales linearly with rþ t .
Thus, (12.69) can be written as
dhþ Cg þ I þ
2 ð sin θ Þ
L
¼ ð12:70Þ
dθ rþ
t
with C = constant, if the term containing cos θ is ignored. This solution should be
matched to a solution valid at the top part of the channel, θ = 00. For Run 67 of Dallman
(a) (b)
3500 1600
Dallman (1978) Run 67 Laurinat (1982) Run 46
3000 uGS = 78.18 m/s uGS = 56.8 m/s
uLS = 0.0572 m/s uLS = 0.0752 m/s
rt+ = 1626 1200 rt+= 2775
2500
ρG = 1.538 kg/m3 ρG = 2.039 kg/m3
2000
I +2
I +2
800
1500
1000
400
500
0 0
0 45 90 135 180 0 45 90 135 180
Angle (degrees) Angle (degrees)
Pipe Bottom Pipe Bottom
top top
Figure 12.14 Angle θ is measured from the top of the pipe. (a) Calculated I þ2 (equation (12.64)) versus θ for
Dallman Run 67 in a 2.54 cm pipe. (b) Calculated I þ
2 versus θ for Laurinat Run 46 in a 5.08 cm pipe.
L. Pan & L.A. Dykhno, Private communication.
(a) (b)
30 20
Dallman Run 67 Laurinat Run 46
18
+
25 dhL CI +2 g+ sin θ
= 16
dθ rt+
14
20
12
h+
h+
15 10 C = 2523
8
C = 2780
10
6
4 +
5 dhL CI +2 g+ sin θ
=
2 dθ rt+
0 0
0 45 90 135 180 0 45 90 135 180
Angle (degrees) Angle (degrees)
Pipe Bottom Pipe Bottom
top top
Figure 12.15 (a) Comparison of equation (12.70) with measurements by Dallman (1978) Run 67, C = 2780.
(b) Comparison of (12.70) with measurements by Laurinat (1982) Run 46, C = 2523. Private
communication.
and Run 46 of Laurinat, this can be accomplished by using the measurements at θ = 0° as

the boundary condition for (12.70). A comparison of (12.70) with measurements is
shown in Figure 12.15 with C = 2523 for Laurinat Run 46 and C = 2780 for Dallman
Run 67.
The agreement of equation (12.66) with data supports the use of term (1) in equation
(12.65), to represent the turbulent mixing in the film when disturbance waves are present.
However, this is a limited test. One cannot conclude that only (1) and (5) are needed
in general calculations. For example, it might be necessary also to consider using either or
both of terms (2) and (3) in equation (12.65) to capture the behavior at the top of the pipe.
12.4 Entrainment in horizontal pipes
12.4.1 Entrainment theory

Entrainment for fully developed flow in vertical pipes is discussed in Section 11.7 by
considering it to result from a balance between the rates of atomization of the wall layer
and the rate of deposition. The enhancement of deposition in horizontal pipes because of
gravitational settling can result in much smaller entrainment than would be experienced
in vertical flows. Thus, the approach of using vertical flow equations to describe entrain-
ment in horizontal pipes is not correct (see, for example, Dallman et al., 1984).
The assumption is made that the local rate of atomization is given by (11.38),
developed for symmetric flow:
k 0A u2G ðρL ρG Þ1=2

RA ¼ ðΓ Γ C Þ ð12:71Þ
σ
where Γ is the local mass flow rate in the wall layer per unit length and Γc is the critical
film flow below which atomization does not occur. Take θ = 0 as the bottom of the pipe.
The integration of (12.71) around the perimeter gives
ð
1 π
hRA i ¼ d t RA dθ ð12:72Þ
πd t 0

k 0 u2 ðρ ρ Þ W LF
hRA i ¼ A G L G Γ C ð12:73Þ
σ P
ðπ
1
Γc ¼ Γc ðΓ c Γ Þdθ ð12:74Þ
π
θc
where θc is the angular location at which Γ = Γc. Thus hRA i is given by (12.71) with Γc
replaced by Γ c . If Γ > Γc around the whole circumference, θc = π and Γ c ¼ Γ c . If Γ < Γc
over a portion of the perimeter, from θc to π; Γ c 5Γ c .
The local rate of deposition can be represented by (11.31):
CW
RD ¼ V W C W ¼ V W C B ð12:75Þ
CB
where VW is the average velocity of the particles striking the wall. For vertical flows, VW
is related to the component of the turbulent velocity fluctuations of the particles normal to
the wall. If these are given by a Gaussian distribution with a mean-square value of v2P ,
" #
1 ð x μÞ 2
pð x Þ ¼ exp ð12:76Þ
ð2π Þ1=2 σ P 2σ 2P
1=2
where μ is the mean velocity, x = μ + vP and σ P ¼ V 2P . For a vertical flow, μ = 0; for a
horizontal flow μ = VT cos θ, where VT is the free-fall velocity. Thus, at the top of the pipe
cos θ = −1 and μ = −VT.
The velocity, VW, varies around the circumference, so
ð∞
VW ¼ xpðxÞdx ð12:77Þ
0
By substituting (12.76) into (12.77), the following expression was obtained by Pan &
Hanratty (2002a, b):
ð∞ " # ð∞ " #
ðx μÞ ðx μÞ2 μ ð x μÞ 2
VW ¼ pffiffiffiffiffi exp dx þ pffiffiffiffiffi exp dx ð12:78Þ
0 2π σ P 2σ 2P 0 2π σ P 2σ 2P
The integral in (12.78) considers only depositing particles, that is, positive values of x = μ
+ vP. (An integration from −∞ to +∞ would give VW = μ.) When (VT / σP) → 0, equation
(12.78) gives
2 0:5
σP
VW ¼ ð12:79Þ
2π
so that only particle turbulence is responsible for deposition. For (VT + σP) → ∞,
V W ! V T cosθ ð12:80Þ
so that gravitational settling is controlling deposition. Pan & Hanratty (2002a, b) present
the integration of (12.78) shown in Figure 12.16. It can be seen that deposition is
enhanced over what is predicted by (12.79) at the bottom of the pipe, where VT cos θ
is positive, and diminished below what is predicted by (12.79) at the top, where VT cos θ
is negative.
The averaging of RD (defined by equation (12.75)) around the circumference gives

C W W LE
hRD i ¼ V W ð12:81Þ
C B QG S
Under fully developed conditions, hRD i ¼ hRA i. Equations (12.73) and (12.81) give
ðE=EM Þ k 0 d t u3G S ðρL ρG Þ1=2

¼ A ð12:82Þ
1 ðE=EM Þ 4hk D ðC W =C B Þiσ
100
10
kD(θ)/(σP2/2π)0.5
0.1
0.01
–10 –5 0 5 10
VTcosθ/σP
Figure 12.16 Calculated variation of kD with θ, equation (12.78). The angle θ is zero at the bottom of the pipe.
Pan & Hanratty, 2002b.
where
πd t Γ c
EM ¼ 1 ð12:83Þ
WL
The critical flow, Γc, is given by Andreussi et al. (1985). One needs to develop a relation
for hk D ðC W =C B Þi to use (12.82).
12.4.2 Entrainment measurements

Measurements of entrainment by Williams (1986, 1990), Laurinat (1982), Dallman (1978)
and by Paras & Karabelas (1991a, b) are discussed by Pan & Hanratty (2002b). They are
for the air–water system and cover pipe diameters of 2.31–9.53 cm, gas velocities of
11–131 m/s, gas densities of 1.26–2.75 kg/m3. These experiments were carried out at low
enough gas velocities for gravitational settling to play a major role. A modified version of
(12.82), which substitutes the terminal velocity for hk D ðC W =C B Þi and takes S as a constant,
was used by Pan & Hanratty (2002a, b) to correlate measurements at low gas velocities:
!
1=2 1=2
ðE=EM Þ d t u3G ρL ρG
¼ A2 ð12:84Þ
ð1 E=EM Þ σV T
where A2 is a dimensionless constant. At very high gas velocities, where particle

turbulence would be the dominant mechanism for particle deposition, the equations for
vertical flows can be used.
Measurements of entrainment for a horizontal 9.53 cm pipe are presented in Figure 12.17.
As already noted for vertical annular flow, E/EM is much more sensitive to changes in gas
flow than to changes in liquid flow. In calculating EM, the term Γ c was set equal to Γc, which
was calculated with the equation proposed by Andreussi et al. (1985) for vertical annular
1.2
dt = 9.53 cm uG (m/s)
1 26
31
0.8 37
45
E/EM
0.6 67
88
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2
WL (kg/s)
Figure 12.17 Measurements of E/EM obtained by Williams (1986, 1990) for air and water flowing in a 9.53 cm
pipe. Pan & Hanratty, 2002b.
flows. The extrapolation of these results to E = 0 gives a rough estimate of a critical gas
velocity of UG ≈ 21 m=s. The measurements of E/EM in Figure 12.17 show a slight decrease
with increasing WL at large liquid flows. This reflects inaccuracies associated with the use of
linear relations for RA and RD. (See discussion by Pan & Hanratty (2002a, b).)
Results obtained in 9.53 cm and 2.31 cm pipes are presented in Figure 12.18 a, b. These
are plots of E, rather than E/EM. A rough estimate of a critical gas velocity of 15 m/s is
obtained for the 2.31 cm pipe. Lines representing WLFC are calculated with the equation
proposed for vertical annular flows. Justification for the substitution of ΓC for Γ C is
inconclusive.
Equation (12.84) is tested in Figure 12.19. A value of A2 = 9 ×10−8 was selected to fit
the data. Each point represents an average of measurements for different WL at a fixed gas
velocity. The terminal velocity is given by
4d P gρL
V 2T ¼ ð12:85Þ
3C D ρG
with CD = 24/ReP for ReP < 1 and C D ¼ 18:5=Re0:6 P for intermediate ReP (see Section 8.4,
that is, Bird et al., 1960). If Stokes law is used to calculate CD, theory predicts that
ðE=EM Þ=1 ðE=E M Þ varies as u5G . An intermediate behavior for CD gives a dependency
of u4:2
G . The change of the slope of the lines calculated using (12.84) represents a change
from Stokes law at large uG to the intermediate relation for CD at small uG . (This is
because there is a decrease of drop diameter with increasing gas velocity.)
The three lines at the top part of Figure 12.19 represent the entrainment that would be
observed in vertical pipes (see equation (11.43)). An important point to be made in
comparing these lines with data is that correlations based on results in vertical pipes will
overpredict at small gas velocities, and not capture the strong influence of gas velocity
observed in horizontal pipes.
(a)
1
dt = 9.53 cm
0.8 WLFC
uG (m/s)
0.6 26
31
E
0.4 37
45
67
0.2 88
0
0 0.2 0.4 0.6 0.8 1 1.2
WL (kg/s)
(b)
1
WLFC dt = 2.31 cm
0.8
uG (m/s)
0.6 15
22
E
0.4 33
44
55
0.2 88
0
0 0.1 0.2 0.3
WL (kg/s)
Figure 12.18 Measurements of entrainment in (a) a 9.53 cm horizontal pipe and (b) a 2.31 cm pipe. Pan &
Hanratty, 2002b.
10000
1000
100
E/EM(1 −E/EM)
dt, ρG
10 (cm), (kg/m3),
9.53, 1.6
5.08, 2.05
1
2.31, 2.74
9.53, 1.6
0.1 5.08, 2.05
2.31, 2.74
5.08, 1.7
0.01
10 100 1000
ρG0.5uG (kg/m3)0.5(m/s)
Figure 12.19 Comparison of measurements with empirical equation (12.84). The particle diameter, d32 is
calculated with equation (11.31). Pan & Hanratty, 2002b.
1000
Straight lines generated by eq. (11.43)
with A1 = 0.000088
100
kÁ= 3 × 10–6
E/EM(1 – E/EM)
10
dt (cm)
9.53
1 5.08
2.31
9.53
0.1 5.08
2.31
5.08
0.01
10 100 1000
ρG
0.5
uG (kg/m3)0.5(m/s)
Figure 12.20 Entrainment calculated with equation (12.82) using k 0A ¼ 3 106 and d32 given by (11.31). Pan
& Hanratty, 2002b.
The favorable comparison of results with (12.84) shows that the influence of gravita-
tional settling needs to be considered in order to capture the effect of gas velocity. For
very large gas velocities, the drops would be uniformly distributed and E/EM would be
close to unity.
Pan & Hanratty (2002a, b) explored the use of equation (12.82), instead of (12.84).
The variation of kD and CW/CB around the pipe circumference was calculated by methods
outlined earlier in this chapter. This approach automatically describes the transition from
a situation where gravity is dominating to one in which gravitational effects are
not important. The results of this calculation are presented in Figure 12.20. A value of
0
k A ¼ 3 106 was used. These theoretical calculations show a smooth transition from
a low to a high velocity behavior for which (E/EM)/1 − (E/EM) varies roughly as u5G and
u2G . One is characteristic of a system in which gravitational settling is controlling
deposition. The other is characteristic of a vertical flow.
12.4.3 Calculations for a natural gas pipeline

It is of interest to use the interpretation for air–water flows to describe the behavior of
a natural gas–condensate pipeline. The system used in the Bacton experiments done
by Shell Oil Company (Wu et al., 1987) is considered. The conditions are as
follows: d t ¼ 0:2032 m; ρG ¼ 65 kg=m3 ; ρL ¼ 720 kg=m3 ; σ ¼ 0:0118 N=m, pressure =
75 bars, μL ¼ 0:00055 mPa s; μG ¼ 1 105 mPa s. The range of V T =σ P for this natural
gas pipeline in the annular flow regime is 0.02–0.5 (where σP is the intensity of the
turbulent velocity fluctuations of the particles). This is to be compared with VT/σP =
0.25−2.5 for an air–water system in a 9.53 cm pipe.
Figure 12.21 compares calculations with (12.82), using k A0 ¼ 3 106 and an eddy
diffusivity for droplets of εP ¼ 0:04rt v , for both natural gas and the air–water system. The
10000
Natural gas pipeline
empirical methods
1000
100
E/EM(1−E/EM)
Natural gas
pipeline
10 Air−water
dt = 9.53
′ = 3 × 10−6
kA
0.1 ζ = 0.04
0.01
10 100 1000
0.5
ρG uG (kg/m3)0.5(m/s)
Figure 12.21. The solid curves use equations (12.82) and (11.41) to calculate entrainment for an air–water system
and for a natural gas pipeline. The dashed lines are calculated for a natural gas pipeline using
equations (12.84) and (11.31). Pan & Hanratty, 2002b.
1.2
uG (m/s)
1 34
11
0.8
9
E 0.6
7
0.4
6
0.2 Solid lines: theoretical

Dashed lines: empirical
0
0 1 2 3 4
WL (kg/s)
Figure 12.22. Plots of calculated E for a natural gas pipeline. The solid curves use equation (12.82) with
k 0A ¼ 3 106 and ζ ¼ 0:04. The dashed curves were calculated with equation (12.84).
Pan & Hanratty 2002b.
dashed lines represent calculations with (12.84) using A2 = 9 × 10−8 to describe the behavior
of the natural gas pipeline at low gas velocities. Equation (11.43) for vertical pipes, with A1 =
8.8 × 10−5, was used to represent the behavior of natural gas pipelines at large gas velocities.
Figure 12.22 compares calculations of E versus WL for different gas velocities. The
solid curves use (12.82); the dashed lines use (12.84) and empirical equation (11.43) for
vertical flows. Note that large differences in predicted values of ðE=E M Þ=1 ðE=E M Þ at
high gas velocities (shown in Figure 12.21) do not translate into large differences in the
predicted entrainment E.
12.5 Pool model
Studies of the distribution of liquid around the circumferences of horizontal pipes that are
transporting gas and liquid reveal a pool configuration at small gas flows. The criterion is
small (ρG/ρL)1/2 Fr, where Fr = uSG/(gdt)1/2. For air and water at atmospheric conditions,
this yields Fr < 50. For these conditions, most of the liquid is located at the bottom of the
pipe, so the asymmetric film model might not be applicable. Measurements of fi /fS in
Figure 12.13 show smaller values for small Fr when comparisons are made at the same
liquid holdup. In order to capture this behavior, Williams (1990) developed a pool model
for a horizontal annular flow.
The liquid at the bottom of the pipe is assumed to resemble the idealized configuration
discussed in Chapter 5 on stratified flow. The only difference is the inclusion of a thin
asymmetric film on the portion of the circumference indicated by PG and the inclusion of
droplets in the gas space.
Chapter 5 describes analyses by Andritsos & Hanratty (1987) of stratified flows which
provide relations for the height of the stratified layer, hL0, and for the friction factor, fipool,
for a given flow rate in the pool, WLP. If hL0 / dt is known, geometric parameters Si and PG
can be specified. For a fully developed flow, the pressure drop is calculated by applying a
force balance to the gas phase:
DPðA AL Þ ¼ τ ipool S þ τ iwall PG ð12:86Þ
where τipool is obtained from the analysis for stratified flow and τiwall is estimated by
assuming the wall film is smooth.
The use of the results on stratified flow to specify AL and τipool requires a prediction of
the liquid flow in the pool which is given by
W LP ¼ W L W LE W LF ð12:87Þ
Williams neglects WLF and develops an equation for the entrainment, WLE, by equating
relations developed for the rate of atomization and the rate of deposition.
Figure 12.23 contains measurements by Williams (1990), which show that almost all
of the liquid in the wall layer is contained in the “pool” at low enough gas velocities. The
“pool model” pictures atomization as occurring only from an idealized layer at the
bottom of the pipe. Some of the drops deposit on the walls of the pipe to be part of
the film which drains downward back to the pool. Studies of the rate of atomization in a
rectangular channel by Alonso (1977), by Tatterson (1975) and later work in vertical
pipes suggest the following equation for small (WLP − WLFC)

u2G ðρG ρL Þ1=2 W LP W LFC
RA ¼ k″A ð12:88Þ
σ Si
12.5 Pool model 323
1.0
0.8 uLS (m/s)

0.0305
0.0610
0.6 0.0914
0.1220
WPool
WWall
0.4
0.2
0.0
0 20 40 60 80 100
uGS (m/s)
Figure 12.23 Fraction of non-entrained liquid contained in the pool for a 0.0953 m pipe. Figure 8.14 of Williams,
1990.
where RA is the rate of atomization per unit length and WLP is the weight flow in the pool.
This is similar to equation (11.37), but the dimensionless constants appearing in the two
equations need not be equal. Williams used AL/h0, rather than Si (see notation in Chapter 5).
The rate of deposition can be given as
ðS i þ PG ÞhRA i ¼ C 0 S i V W0 þ hV W C W iPG ð12:89Þ
where VW0 and VW are the velocities of droplets striking the stratified layer and the wall.
Equation (12.88) can be written as

C0 CW
ðS i þ PG ÞhRA i ¼ C B S i V W0 þ V W PG ð12:90Þ
CB CB
where CB is the bulk concentration of drops and PG is the length of the wall above the
pool (see Chapter 5). The first term on the right side is the rate of deposition on the pool.
The second term is the rate of deposition on the wall layer, where hCCWB V W i is the average
around the length PG. The term hRD i is the average rate of atomization over Si and PG.
Equating the rate of atomization to the rate of deposition provides an equation for the
entrainment, E, analogous to (12.82).

u2 ðρ ρ Þ1=2 C0 Si CW PG
k″A G G L ðW LP W LPC Þ ¼ C B V W0 þ VW
σ C B ðS i þ P G Þ CB ðS i þ P G Þ
ð12:91Þ
where VW0 may be assumed equal to the settling velocity, VT. Since E ¼ W LE =W L and
C B ¼ W LE =QG , the solution of (12.91) for CB provides information on the entrainment
of drops in the gas phase.
10−1
Superficial
liquid velocity
0.0305 m/s
0.0610 m/s
0.0914 m/s
0.122 m/s
h0
dt 10−2
meas
10−3 −3
10 10−2 10−1
h0
dt calculated
Figure 12.24 Comparison of the liquid height measured at the bottom of a pipe with results calculated with the
analysis of Andritsos & Hanratty, (1987). Andritsos & Hanratty, 1987.
A knowledge of E allows a calculation of WLP if WLF is known since

W L W LE ¼ W LF W LP ð12:92Þ
The flow in the film, WLF, may be estimated with methods outlined in previous sections.
However, a good first approximation is to neglect WLF in (12.92).
The approach outlined above has not been exploited so its validity and usefulness have
not been tested. However, Williams (1990) compared measurements of the layer height at
the bottom of a 9.53 cm pipe carrying air and water with the analysis of Andritsos &
Hanratty (1987) for a stratified flow. To simplify the calculation, the assumption was
made that all of the non-entrained liquid is flowing in a pool at the bottom of the pipe. The
results for Froude numbers below 50, shown in Figure 12.24, are encouraging.
12.6 Transition to the annular regime
12.6.1 Air–water flow

Two mechanisms have been presented for the transition from stratified flow to annular
flow in horizontal pipes: One suggests that wetting of the top of the pipe is initiated by
large-amplitude waves (or flow surges) which wrap around the circumference. The other
suggests that annular flow is initiated by an entrainment–deposition mechanism. Lin &
Hanratty (1987) carried out a study on flow patterns for the air–water system and
concluded that the transition in a horizontal 9.53 cm pipe occurs primarily through the
deposition of droplets. This mechanism is dominant in a 2.54 cm pipe only for superficial
liquid velocities smaller than 0.015 m/s. The initial supposition for systems in which
droplet wetting is creating an annular flow was that transition to annular flow is
associated with the initiation of atomization. However, observations show that transition
occurs at higher gas velocities than that needed to create droplets.
Baik & Hanratty (2003) point out that droplets must migrate in sufficient quantity to
form a stable continuous film at the top of the pipe. They postulated that, for a smooth
pipe, the film at the top of the pipe would be laminar and covered with capillary waves.
Consider a coordinate system for which y is the distance from the wall and x is the
distance from the top of the pipe in the circumferential direction. A force balance on the
film gives
dτ yx
þ ρL g sinθ ¼ 0 ð12:93Þ
dy
where τ yx ¼ μyx ðdux =dyÞ and θ is the angular distance from the top of the pipe.
Equation (12.93) can be integrated to obtain ux by assuming that ux ¼ 0 at y = 0 and
that τ yx ¼ 0 at y ¼ hL , where τ yx is a shear stress in the circumferential direction on a
plane perpendicular to the y-axis. This implies that the influence of secondary flows in the
gas can be neglected and that ripple waves are perpendicular to the x-axis. The mass flow
per unit length is given as
ð hL
Γx ¼ ρL ux dy ð12:94Þ
0
The change of Γx in the circumferential direction is obtained from the mass balance
1 dΓ x
¼ RA RD ð12:95Þ
rt dθ
where rt is the pipe radius, RD is the rate of deposition of drops per unit area, RA is
the rate of atomization per unit area and θ = 0 is taken as the top of the pipe. The rate
of atomization at the top of the pipe would be zero at the initiation of annular flow by
droplet deposition. The height of the film at the top of the pipe is then calculated as

3RD d t L
hL ¼ ð12:96Þ
2gρL
where L is the kinematic viscosity of the liquid and dt is the pipe diameter.
The rate of deposition at the top of the pipe, RD, is given as
RD ¼ C W V W ð12:97Þ
where the velocity of the particles striking the wall, VW, is given by equation (12.78). The
concentration of drops at the top wall, CW, was assumed by Baik & Hanratty to be given
by equation (12.27).
Figure 12.25 compares calculations by Baik & Hanratty (2003) of the thickness of the
layer at the top of the pipe with measurements by Williams et al. (1996), which were
limited to heights above 60 µm. The system being considered is the flow of air and water
in a 9.53 cm pipe. The numbers in parentheses are calculated drop sizes. Note that, as
160
uLS = 0.06 m/s
140 Calculation
(80 μm)
Measurements
120
(109 μm)
Thickness (μm)
100
Transition to turbulence
80
(143 μm)
60
40 (178 μm)
Transition to
annular flow
20 (229 μm)
(288 μm) (266 μm)
(361 μm)
0
0 10 20 30 40 50 60 70 80 90 100
uGS (m/s)
Figure 12.25 Comparison of measurements of the liquid height at the top wall with equations (12.96) and
(12.97), for air–water flows in a 9.53 cm pipe. Drop sizes are shown in parentheses. Baik &
Hanratty, 2003.
drop size decreases with increasing gas velocity, the concentration of drops at the top of
the pipe increases, so RD increases.
Lin & Hanratty (1987) have found that the transition to annular flow (for uLS =
0.06 m/s) occurs at a superficial gas velocity slightly above 30 m/s. This is indicated
by one of the arrows in Figure 12.25. Note that the film height at the top of the pipe is
20 µm. Andreussi et al. (1985) observed that disturbance waves (or turbulence) occur for
a film Reynolds number of about 380. For the system considered in Figure 12.25, this
corresponds to a thickness of about 86 µm.
Thus, according the work of Baik & Hanratty, the transition to annular flow can be
related to the formation of a stable layer on the top of the pipe. Therefore, the solution of
this problem would involve the development of a theoretical construct that defines the
stability condition. Figure 12.26 shows transitions to annular flow defined by Lin &
Hanratty, for conditions where the change is controlled by droplets wetting the top wall.
The transition for a 9.53 cm pipe is indicated by the solid curve. The transition for a
2.54 cm pipe is indicated by the dotted curve at the bottom of the figure. The numbers
indicate the height of the film at the top of the pipe (in µm). It is noted in Figure 12.26 that
the transition defined by Lin & Hanratty, in 9.53 cm and 2.54 cm pipes, occurs when the
thickness of the liquid film at the top of the pipe is about 20 µm.
Baik & Hanratty explored the possibility that the critical condition is given by
dimensionless height, hþ L . They used methods developed by Asali et al. (1985) (see
Chapter 3) to show that the critical dimensionless height for the two air–water systems
considered in Figure 12.26 is hþ L ¼ 2.
Observations of the transition indicate that the droplets impinge on the top wall
where they coalesce to form larger drops. These coalesce to form rivulets which, at
large enough flows, spread out to create a continuous film. There is a similarity between
Natural gas data 2.54 cm air−water transition (Lin)

1 9.53 cm air−water transition (Lin)
Natural gas transition to annular

2.3 56 185 298 555 714 2.54 cm air−water
9.53 cm air−water
20.32 cm natural gas
0.8 30 117 187 394 524
0.3 18 81 132 312 436

Air−water
annular
uLS /(m/s)
0.38 23 52 88 124 152

0.1
0.04 4.3 34 70 191 283
0.2 15 43 72 106 130
0.01 2.2 18 50 148 220

0.07 9.4 28 61 81 100
hL ≅ 20 microns
Transition to
turbulence
0.01 4.8 17 38 46 47
4.3 16 35 43 44
3.7 14 31 39 40
0
0 10 100
uGS (m/s)
Figure 12.26 Transition from stratified flow to annular flow. Film thicknesses at the top of the pipe, calculated
with equation (12.96), are given in the units of micrometers. Baik & Hanratty, 2003.
this process and the process of dryout, which has received consideration in the heat
transfer community.
Therefore, it is of interest to discuss two studies carried out by Hewitt & Lacey (1965)
for air flowing over a water film that was moving upward over an acrylic wall under
conditions that the water film forms dry spots for hþ L of 1.5–2.7. In the second, a dry spot
was created by blowing air at an upward-flowing annular film. If the flow of the film was
large enough, the dry spot disappeared. A critical hþ L between 4.62 and 6.87 was
determined from this second study. These results are to be compared with the critical
hþL / 2 that is estimated from the experiments of Lin & Hanratty (1987) on the transition
to annular flow. It also should be noted that Asali et al. (1985) observed hþ L as low as
about 2 for water–glycerine solutions flowing in vertical annular flows under conditions
that atomization was not occurring.
12.6.2 Natural gas pipelines

Calculations of the film height at the top of a pipe for the Bacton studies, at 75 bar,
described in Section 12.4.3 are presented in Figure 12.26. The properties of the fluids are
quite different from air–water. The interfacial tension (σ = 0.012 N/m), the gas density
(ρG = 65 kg/m3), the liquid density (ρL = 0.72 g/cm3) and the liquid viscosity (μL = 0.55
cP) are to be compared with typical properties of the air–water system operating at
atmospheric pressure (σ = 0.072 N/m, μL = 1 cP, ρG = 1.2 kg/m3, ρL = 1 g/cm3). Also, the
pipe diameter, dt = 20.32 cm, is larger than the values of dt = 9.53 cm and dt = 2.54 cm
considered in Figure 12.26 for air–water flows.
From Section 11.3 it is seen that drop sizes are much smaller for the natural gas system
than for the air–water system, primarily because of the smaller surface tension. Thus,
mixing of drops is faster and the calculated thicknesses of the wall layer at the top of the
pipe are larger than would be experienced for air–water systems considered in Figure
12.26, even though the pipe diameter is much larger.
Data for the transition to annular flow (Zabaras, 2002, private communication) are also
presented in Figure 12.26. The calculated height of the layer at the top of the pipe at
transition is found to be an order of magnitude larger than for the air–water system. This
difference cannot be explained by assuming that the transition is associated with an
instability of the liquid layer at the top of the pipe, which is characterized by a critical
hþL (for a smooth wall). The pipe had a roughness size which was of the order of
100–250 μm. If a criterion for transition is that the liquid must flood these roughnesses,
then film thicknesses much larger than 20 μm would be needed. Clearly, the last word has
not been written on this problem.
References
Alonso, G. F. 1977 The effect of liquid viscosity on the rate of atomization in concurrent gas–liquid
flow. M.Sc. thesis, University of Illinois.
Anderson, G. H. & Mantzouranis, B. G. 1960 Two-phase (gas–liquid) flow phenomena I. Chem.
Eng. Sci. 12, 109–126.
Anderson, R. J. & Russell, T. W. F. 1970 Film formation in two-phase annular flow. AIChE Jl 16,
626–632.
Andreussi, P., Asali, J. C. & Hanratty, T. J. 1985 Initiation of small waves in gas–liquid flows.
AIChE Jl 31, 119–126.
Andritsos, N. & Hanratty, T. J. 1987 Influence of interfacial waves on hold-up and frictional
pressure drop in stratified gas–liquid flows. AIChE Jl 33, 444–454.
Asali, J. C., Hanratty, T. J., Andreussi, P. 1985 Interfacial drag and film height for vertical annular
flow. AIChE Jl 31, 895–902.
Baik, S. & Hanratty, T. J. 2003 Concentration profiles of droplets and prediction of the transition
from stratified to annular flow in horizontal pipes. Int. J. Multiphase Flow 20, 329–338.
Bird, R. B., Stewart, W. E. & Lightfoot, E. N. 1960 Transport Phenomena. New York: Wiley.
Butterworth, D. 1972 Air– water annular flow in a horizontal tube. Prog. Heat Mass 6, 239.
Butterworth, D. & Pulling, D. J. 1973 Film flow and film thickness measurements for horizontal
annular air–water flow. AERE Report R7576.
Dallman, J. C., 1978 Investigation of separated flow model in annular gas–liquid two-phase flows.
Ph.D. thesis, University of Illinois.
Dallman, J. C., Laurinat, J. E. & Hanratty, T. J. 1984 Entrainment for horizontal gas–liquid flow.
Darling, R. S. & McManus, H. N. 1969 Flow patterns in circular ducts with circumferential
variation of roughness: a two-phase analog. In Developments in Mechanics, Vol. 5,
Proceedings of the 11th Midwestern Mechanics Conference, ed. H. J. Weiss, D.F. Young, W.F.
Riley & T.R. Rogge. Ames: Iowa State University Press, pp. 153–163.
Dykhno, L. A., 1996 Waves generated on thin liquid films during transition to annular flow in a
large diameter pipe: visual observation and measurements. Internal document.
Dykhno, L. A., Williams, L. R. & Hanratty, T. J. 1994 Maps of mean gas velocity for stratified flow
with and without atomization. Int. J. Multiphase Flow 20, 691–702.
References 329
Fisher, S. A. & Pearce, A. 1979 A theoretical model for describing horizontal annular flow. In Two-
Phase Momentum, Heat and Mass Transfer in Chemical Process, and Energy Engineering
Systems, ed. F. Durst, G.V. Tsiklauri & N. Afgan. Washington, DC: Hemisphere.
Fukano, T. & Ousaka, A. 1989 Prediction of the circumferential distribution of film thickness in
horizontal and near-horizontal gas–liquid annular flows. Int. J. Multiphase Flow 15, 403–419.
Hanratty, T. J. 1956 Heat transfer through a homogeneous isotropic turbulent field. AIChE Jl 2,
42–45.
Hewitt, G. F. & Hall-Taylor, N. S. 1970 Annular Two-phase Flow. Oxford: Pergamon Press.
Hewitt, G. F. & Lacey, P. M. C. 1965 The breakdown of the liquid film in annular two-phase flow.
Int. J. Heat Mass Transfer 8, 781–791.
Hinze, J. O. 1967 Secondary currents in wall turbulence. Phys. Fluids Suppl. 10, 112.
Hurlburt, E. T. & Newell, T. A. 2000 Prediction of the circumferential film thickness distribution in
horizontal annular gas–liquid flow Trans. ASME 122, 396–402.
Hutchinson, P., Butterworth, D. & Owen, R. G. 1974 Development of a model for horizontal
annular flow. AERE Report. R7789.
James, P.W. & Burns, A. 1979 Further developments in the modeling of horizontal annular flow.
AERE Report R9373.
Laurinat, J. E. 1982 Studies of the effect of pipe size on horizontal two-phase flows. Ph.D. thesis,
Laurinat, J. E., Hanratty, T. J. & Dallman, J. C. 1984 Pressure drop and film height measurements
for annular gas–liquid flow. Int. J. Multiphase Flow 10, 341–356.
Laurinat, J. E., Hanratty, T. J. & Jepson, W. P. 1985 Film thickness distribution for gas–liquid
annular flow in a horizontal pipe. Phys.-Chem. Hydrodynamics 6, 179–195.
Lin, P. G. & Hanratty, T. J. 1987 Effect of pipe diameter on flow patterns for air–water flow in
Lin, T. F., Jones, O. C., Lahey, R. T., Block, R. S. & Mirase, M. 1985 Film thickness measurements
and modelling in horizontal annular flows. Phys.-Chem. Hydrodynamics 6, 197–205.
McManus, H. N. 1961 Local liquid distribution and pressure drops in annular two-phase flow.
Paper presented at the ASME-EIC Hydraulic Conference, Montreal, paper 61-HYD-20.
Mito, Y. & Hanratty, T. J. 2005 A stochastic description of wall sources in a turbulent field. Part 3:
Effect of gravitational settling on the concentration profiles. Int. J. Multiphase Flow 31, 155–178.
O’Brien, M. P. 1930 Review of the theory turbulent flow and its relation to sediment transport.
Trans. Am. Geophysical Union 14, 487–491.
Pan, L. & Hanratty, T. J. 2002a Correlation of entrainment for annular flow in vertical pipes. Int.
Pan, L. & Hanrattty, T. J. 2002b Correlation of entrainment in horizontal pipes. Int. J. Multiphase
Flow 28, 385–408.
Paras, S. V. & Karabelas, A. J. 1991a Properties of the liquid layer in horizontal annular flow. Int.
Paras, S. V. & Karabelas, A. J. 1991b Droplet entrainment and deposition in annular flow. Int.
Pletcher, R. H. & McManus, H. N. 1965 The fluid dynamics of three-dimensional liquid films with
free-surface shear: a finite difference approach. In Developments in Mechanics, Vol. 3:
Proceedings of the 9th Midwestern Mechanics Conference, ed. T. C. Huang & M. W Johnson.
New York: John Wiley.
Prandtl, L. 1927 Turbulent flow. NASA TM 435.

Rouse, H. 1937 Modern conceptions of the mechanics of fluid turbulence. Trans. Am. Soc. Civil
Engrs 102, 463–505; Discussion 506–543.
Swanson, R. W. 1966 Characteristics of the gas–liquid interface in two-phase annular flow. Ph.D.
thesis, University of Delaware.
Tatterson, D. F. 1975 Rates of atomization and drop size in annular two-phase flow. Ph.D. thesis,
Wilkes, N. S., Conkie, W. & James, P. W. 1980 A model for the droplet deposition rate for annular
two-phase flow. AERE Report R9691.
Williams, L.R. 1986 Entrainment measurements in a 4-inch horizontal tube. M.Sc. thesis,
Williams, L. R. 1990 Effect of pipe diameter on annular two-phase flow. Ph.D. thesis, University of
Illinois.
Williams, L. R., Dykhno, L. A. & Hanratty, T. J. 1996 Droplet flux distribution and entrainment in
horizontal gas–liquid flows. Int. J. Multiphase Flow 22, 1–18.
Wu, H. L., Pots, B. F. M., Hollenberg, J. F. & Meerhof, R. 1987 Flow pattern transitions in two-
phase gas/condensation flow at high pressures in a 8-inch horizontal pipe. Paper presented at the
3rd International Conference on Multiphase Flow, The Hague, The Netherlands.
Index
annular flow, wall layer trajectory regime 261

critical liquid Reynolds number 65 trapping of particles at the wall 255
disturbance waves 65, 269 direct numerical solutions (DNS) 281
effective stress 67 calculations of Mito and Hanratty 282
growth rate of ripples 167 damping of fluid turbulence 284
interfacial stress 67 feedback effect 281
slow moving ripples 166 point force method 281
atomization of the wall film 269 rate of deposition 282
intermittency factor 91 drag force on a particle
rate of atomization 87 drag coefficient 173, 178
form drag 177
balance theorems free-fall velocity 173, 174
energy theorem 3, 4 skin friction 177
momentum theorem 3, 53 Stokes law 177
Benjamin bubble 185, 205, 208 drag-reducing polymers 190
bubbles drop size
cap bubbles 173 number mean diameter 270
Hadamard 178 Sauter mean diameter 270
rise velocity 173, 177 Tatterson et al. 272
spatial distribution 195 volume median diameter 270
Taylor bubbles 183 drop size distribution
bubble columns 186 log-normal 270
operating curves 191 Rosin and Rammler 270
particulate and aggregative 190 upper limit log-normal 271
bubbly flow drop size measurements 275
coring 204 drop spatial distribution
downcomer 195 diffusion model 297
downflows in vertical pipes 49 for a circular pipeline 301
drift velocity 180
slip ratio 180, 209 entrainment in horizontal pipes 315
volume fraction 180 entrainment in vertical flows 282
critical gas velocities 286
capillary tubes 52 rough fit to data 287
characteristic stress 59, 63 equation of motion, particles
choking 1, 5, 21 Maxey and Riley 235
velocity of sound 5, 6 solid or liquid sphere in a gas 236
spherical gas bubble in a liquid 236
deposition evaporating and condensing flows 47
deposition coefficient 253 critical heat flux 48
free-flight 254
Langevin equation 173 film flow in a vertical pip
measurements of kD 254 Colburn–Carpenter hypothesis 61
mechanisms for deposition 259 flooding 65
rate of deposition 253 flow reversal 64
332 Index
flow in a nozzle 19 deep liquid assumption 80

two-fluid model 24 dispersion relation 77
two-phase flow 22 gravity waves 80
flow regimes in horizontal pipes 28 group velocity 85
annular flow 29 irrotational flow 71
bubble-foam regime 33 shallow liquid waves 80
effect of gas density 40 two-dimensional wave 74
effect of pipe size 34 wave packet 87
Mandhane 28 wave resistance 89
plug flow 30
slugs 30 kelvin–Helmholtz waves 212
stability of a slug 30 kelvin–Helmholtz instability 81
flow regimes in vertical pipes maximum growth 83
annular flow 53
bubble pattern 42 lift forces
dispersed bubble flow 40 bubble in upward flow 237
flow pattern map 43 Saffmann 237
slug flow 42
free-falling layer mass transfer
characteristic stress 59 stability of a dissolving surface 144
interpolation relation 60 waves at a water–ice interface 144
laminar 58 microchannels 53
turbulent 59 microgravity flows 50
flow regimes 50
gas lift pumps 186 models for gas–liquid flows
homogeneous model 1
horizontal annular flow Lockhart and Martinelli 15
asymmetry of the wall layer 305 separated flow model 1, 10, 12
film height at the top of a pipe 326
local rate deposition 315 one-dimensional single-phase flow
maps of mean gas velocity 293 conservation of mass 3
natural gas pipeline 320 energy 4
pool model 322 mechanical energy 9
secondary flow in the gas 294 momentum 3
wetting of the top of the pipe 324
particle turbulence
ideal gas 6 crossing of trajectories 244
adiabatic flow 6 experiments of Lee et al. 245
isothermal flow 8 experiments of Young and Hanratty 247
imposed temporal oscillations 136 inertial time constant 240
quasi-steady behavior 138 saltation regime 301
quasi-laminar behavior 138
relaxation theory 136 roll waves
inclination of a pipeline inclined planes 158
upflows 105 initiation of roll waves 37
declined flows 106 model 164
integral methods 150 pseudo-slug region
laminar free flow 154
momentum balance 150 sedimentation and fluidization 197
neutral stability 154 aggregative 200
pseudo-steady state 152 hindered settling 197
shallow liquid assumption 149 slugs, horizontal
shape parameter 150 base layer 205
wave velocity 150 first necessary condition 202
inviscid waves pressure drops 213
Bernoulli equation 74 regular slugging 217
capillary waves 80 second necessary condition 204
Index 333
shedding rate 207 Taylor 232

slug velocity 223 turbulent diffusivity 233
stability of slugs 212 turbulent liquid flow over solid waves
transition from a slug flow 207 Boussinesq approximation 131
stochastic slugging 221 effective pressure gradient 134
probability density function 225 Lloyd, Moffat, and Kays 133
slug frequency 223 relaxation theory 136
stochastic model 228 shear stress variation 134
stratified pattern wave-induced turbulence 128
design relations 103
flow in a pipe 95 viscous wavelength theory 185
interfacial friction factor 100 initiation of roll waves 37, 158
simplified model 94 instability of a stratified flow 202
steepness of the waves 101
substantial derivative 72 waves
Gaster 89
turbophoresis Jeffreys 29
physical interpretation 252 kinematic wave 156
turbophoretic velocity 249 Orr–Sommerfeld equation 155
turbulence, fluid viscous effects 111
Eulerian time scale 238 wave generation by a gas flow
frequency spectra 239 crticial gas phase Reynolds number 141
spatial Eulerian correlation 238 forces at the interface 125
temporal correlation 238 gas flow needed to generate waves
turbulent dispersion 126, 158
Lagrangian correlation coefficient 233 Miles and Benjamin 126
Lagrangian length scale 261 theory of sheltering 125
Lagrangian time scale 260 wave-induced turbulence 128

Physics of Gas-Liquid Flow

Uploaded by

Copyright:

Available Formats

Physics of Gas-Liquid Flow

Uploaded by

Document Information

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

Physics of Gas-Liquid Flow

Uploaded by

Copyright:

Available Formats

What are some of the important flow regimes discussed in the text?

What are some of the important flow regimes discussed in the text?

What are some of the design equations developed to predict important characteristics of gas-liquid flows?

What are some of the design equations developed to predict important characteristics of gas-liquid flows?

Physics of Gas–Liquid Flows

Thomas J. Hanratty is Professor Emeritus at the University of Illinois at Urbana-

Preface page xiii

Chapter 1 One-dimensional analysis 1

Chapter 2 Flow regimes 27

2.2.3 Annular ﬂow 29

Chapter 3 Film ﬂows 58

Chapter 4 Inviscid waves 71

Chapter 5 Stratiﬁed ﬂow 94

Chapter 6 Inﬂuence of viscosity on large Reynolds number interfacial waves; effect

6.4.1 General solution 116

Chapter 7 Large-wavelength waves; integral equations 148

Chapter 8 Bubble dynamics 173

8.4 Rise velocity of bubbles at low ReP 178

9 Horizontal slug ﬂow 202

10 Particle dispersion and deposition 232

10.5 Characterization of ﬂuid turbulence 238

11 Vertical annular ﬂow 268

12 Horizontal annular ﬂow 292

12.4 Entrainment in horizontal pipes 315

dh Hydraulic diameter = 4A/wetted perimeter dvμ

hL In a pipe, the length of the bisector of the liquid layer

pGR Pressure component which has a minimum at the wave crest

tC Time constant reﬂecting the rate of growth of waves

Vtp Turbophoretic velocity

α Volume fraction of gas in a ﬂowing mixture

β~ Reciprocal time constant for a particle in a liquid (10.50)

τ Magnitude of a shear stress

ReP Particle Reynolds number = d P ρU S =μ

h i Spatial average, ensemble average or phase average

1.2 Single-phase flow

1.2.1 Flow variables

1.2.2 Momentum and energy theorem

Figure 1.1 An arbitrary volume in space.

Figure 1.2 Control volume in a pipe.

where ρe is the energy per unit volume.

1.2.3 Steady flow in a duct of constant area

where term G is called the mass velocity.

d ðAGuÞ ¼ ðdpÞA τ W Pdz ρAg cos θg dz ð1:5Þ

The wall shear stress is usually represented as

From conservation of mass, if A is constant, du = G d(1=ρ) so that

The enthalpy per unit mass is given as

Therefore, a table or a diagram representing the thermodynamic variables provides

where the velocity of sound, c, is given by

This is a thermodynamic quantity which is the classical deﬁnition of the velocity of

1.2.5 Flow of an ideal gas in a pipe of constant area

where γ = CP /CV and

The application of the equations developed in Section 1.2.3 will be illustrated by

where subscript 1 indicates inlet conditions. From (1.21),

If (1.19) and (1.24) are used in (1.28)

The temperature is obtained from (1.19).

If an isothermal ﬂow is assumed

This can be integrated, using (1.19), to yield

1.2.6 Mechanical energy balance

Thus (1.41) can be rewritten as

A comparison of (1.43) with the momentum balance equation gives