THESIS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

Fluidized bed combustion

– modeling and mixing

David Pallarès

Department of Energy and Environment

Division of Energy Technology
CHALMERS UNIVERSITY OF TECHNOLOGY

Göteborg (Sweden), 2008

Fluidized bed combustion – modeling and mixing
David Pallarès
ISBN 978-91-7385-142-8

© David Pallarès, 2008

Doktorsavhandlingar vid Chalmers Tekniska Högskola

Ny serie nr. 2823
ISSN 0346-718X

Department of Energy and Environment

Division of Energy Technology
Chalmers University of Technology
SE-41296 Göteborg (Sweden)
+46 31 7721000

Chalmers Reproservice
Göteborg (Sweden), 2008
 Fluidized bed combustion – modeling and mixing

David Pallarès

Department of Energy and Environment

Division of Energy Technology
Chalmers University of Technology

Abstract

First demonstrated on industrial scale in 1976, fluidized bed combustion (FBC)

rapidly became an established technology for solid fuel combustion with a current
installed capacity of about 30 GWe and power plants ranging from 10 to 460 MWe.
The widespread use of the FBC technology both under so-called circulating and
bubbling conditions (i.e. with and without external recirculation of solids) is generally
considered to be due to two main advantages with respect to other competing
combustion technologies: the high fuel flexibility and possibilities of cost efficient
reduction of SOx and NOx emissions. Yet, the relative lack of knowledge within some
of the phenomena governing the process represents a major obstacle for the further
development of the FBC technology.
Efficient scale-up and development of the FBC technology depends on the
possibility to increase the knowledge on the main processes involved in FBC boilers.
A combination of experimental work and formulation of mathematical models is
important to increase the knowledge of the process. In this work, macroscopic
submodels for key processes in FBC (fluid dynamics, combustion and heat transfer)
are formulated based on experimental data. These submodels are then linked to form
a comprehensive model for large-scale fluidized bed combustion which is compared
to experimental data not used in the validation of the submodels. The submodels
include treatment of important phenomena such as fluctuations in the gas phase,
separate convective and radiative heat transfer accounting for absorption in the
particle suspension, influence of the pressure drop across the gas distributor and
corner effects in the solid phase. These phenomena were not included in previous
comprehensive FBC models and are the focus of this thesis.
The comparison between the model and experimental data from large-scale
fluidized bed boilers have been carried out for a wide range of data categories
covering the solid phases (inert bed material and fuel), the gas phase and the
temperature field. The modeled data generally shows good agreement with
experimental data.
Although the model developed is primarily aimed at circulating fluidized bed boilers,
its principles and formulation can also be applied to bubbling beds and gasifiers as
well as new FBC technologies such as oxy-fuel or chemical looping combustion.

Keywords: Fluidized bed combustion; Modeling; Mixing

List of publications
This thesis is based on the work contained in the following papers, referred to by
Roman numerals in the text.

I Pallarès, D., Johnsson, F. 2006. ”Macroscopic modelling of fluid dynamics

in large-scale circulating fluidized beds”. Progress in Energy and
Combustion Science, 32 (5-6), pp. 539-569.

II Pallarès, D., Johnsson, F. 2008. ”Modeling of fuel mixing in fluidized bed

combustors”. Accepted por publication in Chemical Engineering Science.

III Pallarès, D., Johnsson, F. 2008. ”Macroscopic modeling of the gas phase in
fluidized bed combustion”. Submitted por publication.

IV Pallarès, D., Johnsson, F. 2006. ”A novel technique for particle tracking in

cold 2-dimensional fluidized beds - simulating fuel dispersion”. Chemical
Engineering Science, 61 (8), pp. 2710-2720.

V Pallarès, D., Díez, P., Johnsson, F. 2007. ”Experimental analysis of fuel

mixing patterns in a fluidized bed”. Proc. of the 12th Int. Conf. on
Fluidization (Harrison, Canada).

Appendixes

A Macroscopic modeling of heat transfer in circulating fluidized bed units.

David Pallarès is the principal author and investigator in all papers. Filip Johnsson
has supervised the work in all papers and contributed to their writing. Pedro Díez
carried out the experiments in Paper V.
The following papers have also been published during the course of the work but are
not included in this thesis since their content either is outside the scope of the
present work or partly overlaps with the papers included in it.

• Pallarès, D., Johnsson, F. 2008. ”A comprehensive model of CFB

combustion”. Proc. of the 9th Int. Conf. on Circulating Fluidized Beds
(Hamburg, Germany).

• Pallarès, D., Johnsson, F. 2002. ”Fluid dynamic modeling of CFB units”. Proc.
of the 7th Int. Conf. on Circulating Fluidized Beds (Niagara Falls, Canada),
pp. 387-394.

• Pallarès, D., Johansson, A., Johnsson, F. 2006. ”Interpretation of dynamics of

pressure measurements”. Proc. of the 19th Int. Conf. on Fluidized Bed
Combustion (Vienna, Austria).

• Pallarès, D., Johnsson, F. 2007. ”Modeling fuel mixing in a fluidized bed

combustor”. Proc. of the 12th Int. Conf. on Fluidization (Harrison, Canada).

• Niklasson, F., Pallarès, D., Johnsson, F. 2006. ”Biomas co-firing in a CFB

boiler - the influence of fuel and bed properties on in-furnace gas-
concentration profiles”. Proc. of the 19th Int. Conf. on Fluidized Bed
Combustion (Vienna, Austria).

• Almendros-Ibáñez, J.-A., Pallarès, D., Johnsson, F., Santana, D. 2008. ”A

novel approach to characterize fluidized bed dynamics combining PIV and
FEM”. Submitted for publication in AIChE Journal.

• Almendros-Ibáñez, J.-A., Pallarès, D., Johnsson, F., Santana, D. 2008.

”Porosity distribution around bubbles in a fluidized bed”. Submitted for
publication in Powder Technology.
Acknowledgements
Foremost, I wish to express my most sincere gratitude to my supervisor Professor
Filip Johnsson for his skillful guidance throughout this and other works. His deep
knowledge of the field of FBC combustion, together with his confidence on this work,
provides a privileged ground which I deeply appreciate.
My gratitude must be extended to Associate Professor Henrik Thunman for the time
and effort he has spent on valuable discussions which saved me by far more time
and effort than otherwise would have been the case. Associate Professor Lars-Erik
Åmand is acknowledged for his passion for the experimental field, which burns like
no one else’s and grants him the ability to provide almost any experimental data one
can think of (and explanations for it).
Besides for wise advice, I am grateful to Professor Emeritus Bo Leckner specially for
giving me the opportunity to regularly attend the highly rewarding workshops
arranged by the Fluidized Bed Conversion group of the International Energy Agency,
where this work has been presented and given further qualified feedback.
Very special thanks must go to Adjunct Professor Bengt-Åke Andersson for his
strong trust in the embryo of what you hold in your hands and for giving it the chance
to develop within a successful framework of collaboration with the manufacturing
industry. It has been and will continue to be a pleasure to work with METSO
Power Oy; their flexibility and understanding for the nature of academic research is
highly appreciated and Mr. Marko Palonen is specially acknowledged in this sense.
Lennart Darell is thanked for always having a solution and a smile at hand for any
difficulty encountered in the experimental field.
Coworkers at the Thermal and Fluid Engineering Department of University Carlos III
(Madrid, Spain) and José Antonio Almendros-Ibáñez in special as well as Pedro-
Antonio Díez are thanked for fruitful cooperation.
Financial support from the European Commission (under contract RFC-CR-03001),
the Swedish Energy Agency and METSO Power Oy is acknowledged.
Highly-skilled surrounding, long field experience and funding have been a necessary
but not sufficient condition for the implementation of this work. Thus, last but not least
come the following acknowledgements:
The friendly and inspiring atmosphere at the division of Energy Technology has
played an important role, and not only in pure academic terms: chats about weekend
activities, movies, soccer, politics or life itself are a must to preserve some
reasonable level of mental health during a work like this.
A time ago Göteborg became my new home, and some people managed to make it
feel home too. These friends know who they are and how much I enjoy them.
My (increasing!) family at home in Barcelona has given me unconditional
understanding and support for my move to Sweden, which I deeply value: moltes
gràcies de debò. Friends I left there are also missed, even more now when distance
confirmed them to be real ones (both day- and night-time).
It is nothing but a blessing to share life with my love and best friend Emma and her
everlasting warmth and loyalty.
 “All models are wrong.
Some are useful”.

- George Box, chemist and statistician

Table of contents

Chapter 1 - Introduction ………………………………………..1

1.1 Scope: CFB boilers 1

1.2 Aim: Modeling and experimental work 3

1.3 Background: Macroscopic models 4

Chapter 2 - Submodels …………………………………………6

2.1 Governing phenomena in CFB combustion 6

2.2 Theory 9

Chapter 3 - Comprehensive Model ………………………….24

3.1 Coupling of submodels 24

3.2 Computational flow scheme 30

3.3 Furnace mesh 32

Chapter 4 - Results ……………………………………………. 34

Chapter 5 - Further work …………………………………….. 43

Chapter 6 - Conclusions ….…………………………………… 45

Notation ….……………………………………………………….46

References ……………………………………………………….48
Chapter 1 - Introduction

1.1 Scope: CFB boilers

The focus of this thesis is on Circulating Fluidized Bed Combustion. The principle of
a fluidized bed is that a bed of solid particles adopts a fluid-like behavior when a gas
injected from below flows through it. A fluidized bed is known to provide a good gas-
solids contact together with relatively high rates of solids and gas mixing. These
properties have made fluidized beds a competitive technology in a wide variety of
industrial sectors such as pharmaceutics, metallurgy, material science,
petrochemistry and, in the context of the present work, the energy sector.
For some applications, it is desirable
to inject a relatively high gas flow to the
fluidized bed reactor (generally called
riser), which can lead to significant
entrainment of bed material out of the
riser through the exit duct. In such
case, the entrained solids flow is
separated from the gas flow by means
of a cyclone and fed back into the riser
through a particle seal system
containing an auxiliary fluidized bed
and a return duct. Depending on the
magnitude of this externally recirculated
solids flow, FB units can be divided into
bubbling (BFB) and circulating (CFB)
fluidized beds. Although a clear-cut
value does not exist, FB units are
considered to be operated under Figure 1: The 177 MWth biomass-fired CFB boiler
in Pori (Finland). Courtesy of Metso
circulating conditions when a significant Power, manufacturer. The parts forming
flux of externally recirculated solids is the circulating loop are identified.
present (roughly above 0.5 kg/m2).
Otherwise, the bed is considered a BFB, also called stationary fluidized bed (SFB).
The scope of this work is CFB boilers, i.e. CFB units used for combustion of solid
fuels. Figure 1 illustrates a typical CFB boiler.
While the basic principle of fluidization is common to all FB units, there are
significant differences between large-scale and laboratory fluidized beds. Units of
both these types can be found operating at bubbling as well as circulating conditions.

1
 Introduction – Scope: CFB boilers

Large-scale FB units applied in combustion, which are in focus in this work, are
generally characterized by:

• An aspect ratio of the riser (Hriser/Deq, riser) lower than 10

• An aspect ratio of the settled bed (Hb, settled/ Deq, riser) lower than 1
• Bed material which belongs to group B in the Geldart classification

These characteristics give a flow and mixing pattern which significantly differs from
that obtained in typical laboratory units (see e.g. Werther, 1993). Thus, application of
data from laboratory units to large-scale units must be done with care and may not be
appropriate at all.
Large-scale CFB units are used for the thermal conversion of solid
fuels, i.e. combustion and gasification of coal, biomass, peat or solid waste. Fluidized
bed combustion is widely used whereas FB gasification has not yet reached
commercial status when it comes to difficult fuels such as biomass and waste. The
focus of this work is on the use of large-scale CFB units in combustion.
One of the main advantages of FB combustion is its fuel flexibility and the ability to
burn difficult fuels such as various biomass waste and other waste fuels. Thus, FB
plants are not limited to one fuel type but can operate on different fuel types or co-
firing a mix of fuels, depending on the availability and price of the fuels.
In addition, with the increasingly demanding regulations on emissions, FB
combustion gives low emissions of both SOx and NOx. The high gas-solids contact in
FB units facilitates in-bed SOx capture by addition of a solid sorbent (limestone or
dolomite) to the bed material, i.e. there is normally no need for scrubbers. In addition,
the low and homogenous in-furnace temperature, around 850 ºC, yields low
emissions of NOx (low thermal NOx).
A problem with FB combustion is that there is significant lack of knowledge and
understanding of the in-furnace processes which govern the combustion and, thus,
are crucial for design and scale-up of the process, optimal operation of existing units
and for preventing erosion of internals.
The first FB boilers were BFB units, with all bed material located in the furnace and
combustion to a large extent taking place in the bed and in the lower part of the
freeboard, especially when burning a low volatile fuel such as bituminous coal.
Further development of the FB combustion technology resulted in CFB boilers which
give an increased volumetric heat load as a result of the higher gas velocities. The
external recirculation of solids for typical CFB boiler designs is between 0.5 and
20 kg/m2·s. In CFB boilers (see Figs 2a and 2b) combustion is distributed more
homogeneously along the height of the furnace than in bubbling beds and heat
extraction by means of internals in the return leg is also possible and often used in
large boilers. At present, both BFB and CFB boilers are built, with BFB boilers
applied for small- and medium-sized (10 to 50 MWth) combined heat and power
plants (including waste combustors) whereas CFB boilers are also applied as power
boilers up to several hundreds of MW electric power (largest boiler is 465 MWe).
Hupa (2005) estimates that the currently installed capacities of CFB and BFB
boilers are 50 GWth and 15 GWth, respectively with the largest plants (single boiler) of
460 MWe and 142 MWe. There are ongoing feasibility studies for CFB capacities
around 800 MWe.

2
 Introduction – Scope: CFB boilers

a) b)
Figure 2: The world's largest (550 MWth) biomass-fired CFB boiler in Alholmen (Finland).
a. Photo of plant (courtesy of Alholmens Kraft, plant owner)
b. Illustration of circulating loop (courtesy of Metso Power, manufacturer)

In future, new approaches to CFB combustion directed towards CO2 capture such
as oxy-fuel and chemical looping combustion are believed to play an increasingly
important role in the development of CFB combustion technology.

1.2 Aim: Modeling and experimental work

Further development of the FBC technology requires an increased level of

knowledge and understanding of key phenomena in the process, i.e. gas and solids
mixing, fuel conversion and heat transfer. One way to increase understanding is a
thorough combination of modeling and experimental work where modeling is
validated against specific experimental data and thereafter applied to model a
particular FB unit for comparison with corresponding experimental data. Successful
development of models will then help to facilitate reliable design and scale-up of FB
units.
The aim of this work is to develop such a modeling procedure combined with
experimental work, with focus on key phenomena which are crucial for the process.
The phenomenon in focus for the experimental part of this work is the fuel mixing,
described in Papers IV and V. This is investigated based on a novel method for
particle tracking developed within the framework of this thesis. Fuel mixing is known
to be a critical issue in the optimization of the performance of FBC units. However,
there is a lack of knowledge on the phenomena governing fuel mixing, both in
qualitative and quantitative terms.
Regarding the modeling part of this work, described in the following chapters and in
Papers I to III, a model is developed with the aim to give a detailed 3-dimensional
description of the steady-state combustion process in the circulating loop of a CFB
unit at an affordable computational cost. This requires that fluid dynamics, heat
transfer and combustion are all taken into account in the modeling.

3
 Introduction - Aim: Modeling and experimental work

Two main modeling approaches exist at present: macroscopic modeling and

computational fluid dynamics (CFD).
Macroscopic modeling of multiphase flows has a semi-empirical character and
therefore needs experimental data for its formulation and validation. Experimental
data represents a critical point in the modeling in various ways: the measurements
must be reliable and not too scarce and cover a wide range of operational conditions
and geometries in order to make the model general at the same time that some data
must be available for an independent validation of the model. Macroscopic modeling
presents reasonably short computational times (from a few minutes to an hour or so).
CFD modeling is based on modeling from first principles, i.e. solving the transport
equations for mass, momentum and energy in microscopic or mesoscopic scales. As
far as FB units are concerned, CFD modeling can be divided into two sub-types
depending on the how the particulate phase is modeled: multifluid and discrete
element modeling. In multifluid modeling (also known as Eulerian modeling) the
particulate phase is handled as a continuum while in discrete element modeling (also
known as Lagrangian modeling) each individual particle is considered in the equation
system. CFD modeling may be an important modeling tool for FB units in the future,
but at present the long computational time (in the order of days to months), together
with uncertain assumptions in modeling and inaccuracies in the results, make it still a
tool under development, especially when aiming at including both fluiddynamics and
reactions/combustion.
With the above concerns, and considering the aim set, the macroscopic approach
is adopted for the modeling part of this thesis. It must be noted that even though the
modeling is focused on CFB combustion, its theoretical ground makes it a valid basis
also for the modeling of other large-scale FB processes, i.e. BFB combustion, BFB
and CFB gasification and new FBC techniques with inherent CO2 separation (such as
oxy-fuel combustion and chemical looping combustion).

1.3 Background: Macroscopic models

A number of macroscopic models are available in literature for describing certain

regions or phenomena in CFB boilers. However, so-called comprehensive models
covering all the major phenomena (fluid dynamics, heat transfer and combustion) in
the entire combustion space are scarce. Comprehensive models are built up by
linking several validated local submodels into an overall model by means of proper
boundary conditions and mass and energy balances. The main comprehensive
macroscopic models for large-scale CFB units given in literature are listed in Table 1.
All boiler manufacturers are known to have confidential in-house models which are
more or less comprehensive but to a very large extent based on experimental data,
becoming more of a correlation bank than a model.
While all models listed in Table 1 solve the coupling between their respective
submodels by means of similar energy and population balances and sets of
boundary conditions, the main differences are found in the dimensionality of the
model and the choice of submodels.
So-called 1.5-dimensional models discretize the furnace in the axial direction,
dividing each vertical layer in the freeboard into one core region and one wall region
cell in order to account for the core-annulus flow pattern observed and explained

4
 Introduction - Background: Macroscopic models

later on in connection to Fig. 3. This type of furnace discretization provides a more

proper picture of the process than 1-dimensional models. Yet, a 1.5-dimensional
model is not able to describe some of the crucial phenomena in CFB boilers, such as
the horizontal fuel mixing and all events deriving from it.

Table 1: Macroscopic comprehensive models for CFB combustion.

Years Developer References Dimensionality

International
- 1996 Energy Hannes (1996) 1.5-dimensional
Agency

Hamburg
1989 - present University of Lücke (2003)* 3-dimensional
Technology

Participants in Zhang and

1998 - 2001 1.5-dimensional
EU Joule project Leckner (2001)

Chalmers
2005 - present University of This work 3-dimensional
Technology

* latest published full version

The present work contains submodels which give some phenomena a new
approach or account for phenomena which were not included in previous models.
Examples of these phenomena included in the modeling of this work are dynamic gas
phase mixing, the coupling between fuel conversion and fuel mixing, the modeling of
radiative heat flux in suspensions accounting for optical shadowing and consideration
of corner effects in the solids flow. In addition, a wide range of experimental data has
been used for validation of the submodels developed.

5
Chapter 2 - Submodels

2.1 Governing phenomena in CFB combustion

When operating a CFB boiler, primary air is injected from the bottom of the furnace
where bed material (mostly sand-like particles) is contained. With this, the bed
material is fluidized and a turbulent-like gas-solid flow with both dense solid emulsion
and gas pockets is established in the bottom furnace region, providing good solids
and gas mixing rates and gas-solids contact. These conditions constitute a favorable
environment for combustion of solid fuels: fuel particles fed to the unit will spread
over the cross section of the furnace (solids mixing) at the same time as the char is
consumed in contact with oxygen in the primary air (gas-solid contact) and the
volatiles released are combusted as they meet oxygen from the primary air (gas
mixing).
Above the dense region formed in the bottom of the furnace, a splash zone is
formed, provided a high enough furnace pressure drop. The splash zone is
characterized by a strong back-mixing of solids in the form of solid clusters following
a ballistic movement. These clusters are thrown up from the bed due to eruption of
bubbles at the top of the bed.
Above the splash zone, solids are present in a more dispersed form and follow a
core-annulus structure with upflow in the core region and downflow in the annular
region at the furnace walls (so-called wall layers) with a net flow from the core to the
wall layers. Secondary air injection (which represents an enhancement of the gas
mixing) is generally introduced a few meters above the air distributor in order to
reduce NOx emissions and combust unburned volatile matter.
All gas and part of the solids reaching the top of the furnace flow into the cyclone
(there might be more than one cyclone) through the exit duct, see Fig. 1. In the
cyclone, high gas mixing makes any unburned volatiles and oxygen react. Also, gas
and fine particles are removed from the circulating loop and directed towards the
convection pass for heat extraction while most particles are separated from the gas
flow and remain within the circulating loop. The solids captured by the cyclone, which
contain a certain percentage of fuel particles, then flow down to a fluidized seal
system in most CFB designs, as seen in Fig. 1. Thereafter the solids are fed back to
the furnace through a return duct. Despite the fines leaving the cyclone with the gas
flow, the solids inventory in the circulating loop can be maintained more or less
constant due to the ashes fed with the fuel and controlled ash withdrawal.
Since the solid phase carries a large part of the total enthalpy flow in the circulating
loop of a CFB unit, a high solids mixing will act as thermal flywheel and ensure a high

6
 Submodels: Governing phenomena in CFB combustion

degree of homogeneity in the temperature field. Typically, about the same amount of
heat is removed from the circulating loop as from the flue gas in the convection pass.
With respect to the circulating loop, most of the heat extraction is through the
waterwalls in the furnace, although in some designs there is a contribution from
internals or waterwalls in the cyclone.
In the modeling of any process including several phenomena, it is a key task to
identify the governing phenomena in order to establish corresponding submodels. To
what extent this can be done depends on the focus and aim of the model and the
complexity of the overall process. For the present work, phenomena considered to
govern the process are listed and shortly described below, divided into three
categories: fluid dynamics, combustion and heat transfer. Superscripts after each
underlined item indicate the paper in this thesis where the phenomenon is addressed
in detail.

Fluid dynamics
- The dense bed expansion I, defined as the dependency of the time-averaged
voidage of a dense bed (in the furnace and in particle seals, when applicable) with
fluidization velocity, is a crucial phenomenon to account for in pressure or
population balances since most of the solids in the unit are contained in the dense
bed(s). In addition, the different pressure drop across the gas distributor between
bubbling and circulating dense beds lead to different expansion behaviours which
must be analyzed separately.
- The solids division between cluster and disperse phase in the freeboard I allows the
model to describe the two coexisting solid flow patterns in the freeboard of a CFB
furnace: a cluster phase which originates from bubble bursts which follows a
ballistic movement in the splash zone immediately above the dense bed surface
and a disperse phase following a core-annulus flow structure in the upper
freeboard. Proper description of solid flows in the freeboard is required for the
modeling of the heat transfer, size segregation and net circulating flow.
- The lateral fuel mixing in the bottom region II,IV,V (i.e. dense bed and splash zone) is
a critical phenomenon in the design of large FB boilers since too low lateral mixing
of fuel particles results in unbalanced local air-to-fuel ratios over the cross section
of the furnace and inefficient operation of the unit. Consideration of the lateral fuel
mixing allows calculation and analysis of horizontal heterogeneities in other
variables, e.g. in temperature and gas species concentration.
- The continuously varying physical properties of the fuel particles II as they dry,
devolatilize and combust imply that fresh fuel particles behave in a much different
way than those close to burn-out. These gradual changes in the flow of a fuel
particle as conversion advances must be accounted for.
- The gas flow fluctuations III originate from the bottom bed dynamics and strongly
characterize the gas flow in the bottom region of the furnace. The fluctuating
character of the gas flow is a key factor to describe combustion in the bottom region
and gas mixing in the splash zone.
- The gas dispersion III governs the gas flow in the upper freeboard and thereby
combustion in this region by controlling the fate of oxygen. This phenomenon is
enhanced by secondary air injections.

7
 Submodels: Governing phenomena in CFB combustion

- The size segregation I effect is observed along the height of the furnace, with
coarser particles populating the bottom region and a decreasing averaged solids
size with height, leading to that the solids inventory in the return leg consists of finer
solids than in the furnace. This effect has a strong influence on the calculation of
the solids distribution in the furnace and the net solids circulating flow in particular.
The particle size distribution of the bed material is needed to account for size
segregation.
- The so-called backflow effect I,II describes which part of the solids reaching the level
of the furnace exit will be entrained by the gas flow towards the cyclone and which
part will join the solids downflow in the wall layers. This effect plays an important
role in the calculation of both the net solids circulating flow and the locally increased
solids downflow in the wall regions coupled to the furnace exit(s).
- The population balance in the return leg I is a crucial term in the total population
balance (some units work with most of their solids inventory located in the return
leg). In this population balance, the solids contained in the existing fluidized solids
column in the downcomer might be of importance and cannot be neglected.

Combustion
- The kinetics of the fuel particle conversion II must be combined with the fuel mixing
in order to model the distribution of moisture and volatile releases and of char
combustion. These kinetics depend on a variety of parameters such as fuel
properties and characteristics of the surrounding atmosphere.
- Based on experimental observations, in the bottom region of the furnace the gas
division between throughflow and emulsion-only phase III can be characterized by a
high velocity oxygen-rich gas phase and a low velocity reducing gas phase (within
the emulsion phase). When included in the modeling, this phase division gives a
more accurate description of the real gas flow and thereby forms a basis for a more
realistic ground for the formulation of the combustion model.

Heat transfer
- The convective and radiative components of the heat transfer App. A differ in nature
and need different modeling approaches. While the convective term is determined
by particles and gas flow close to the heat transfer surface, radiation is more
complex to handle in a fluidized bed since modeling must include treatment of the
absorption of the particle suspension.
- The heat transfer to the fuel particles II has a direct influence on the fuel conversion
kinetics and need thus to be accounted for.

Thus, for successful modeling of CFB combustion the set of submodels chosen to
form the comprehensive model need to account (individually or through their
combination) for all of the above-listed phenomena.

8
 Submodels: Theory

2.2 Theory

This chapter summarizes the assumptions and expressions in the main submodels
used in the present work. A more detailed description is available in the appended
papers. The submodels are presented individually. The required inputs and the
balances through which they are coupled into the comprehensive model are given in
Chapter 3. The present chapter is structured in four sections: inert solids, fuel
particles, gas phase and heat transfer.

Inert solids
Inert solids form typically more than 95% of the solids in the circulating loop and are
thereby considered to govern the fluid dynamics of the solid phase. Submodels
related to the fluid dynamics of the furnace can be divided into three regions with
respect to the gas and solids flow: the dense (or bottom) bed, the splash zone and
the transport zone, as illustrated in Fig. 3. The dense bed and the splash zone is
here called bottom region while the splash and transport zone is called freeboard.

Figure 3: Representation of the solids distribution in the circulating loop.

Arrows indicate the main solids flow pattern in the riser.

The presence of a dense bed is common to almost all CFB units and is mandatory
in the formulation of the present model. Two phases with strong dynamics are usually
considered in the dense bed: a bubble phase often regarded as solids-free and an
emulsion phase containing solids and interstitial gas with a voidage which is usually
assumed to be under minimum fluidization conditions, εmf . Thus, the dense bed
voidage can be expressed as a weighted sum of the voidage of the two phases, i.e,
ε = δ + (1 − δ ) ⋅ ε mf (1)

where δ is the volumetric fraction of the bubble phase.

9
 Submodels: Theory

The dense bed is taken as the region with a constant vertical pressure gradient on
a time-averaged basis which, in addition, provides the time-averaged voidage, ε, in
the dense bed according to the expression:

= ρ s ⋅ g ⋅ (1 − ε )
dP
(2)
dh

Thus, the time-averaged dense bed voidage can be measured relatively

straightforward. The dense bed voidage has been the focus of several studies which
have shown the voidage to depend on fluidization velocity (see Figs. 4a and 4b) and
mean particle size. In addition, a dependency on the dense bed height (represented
through the variable ΔPref ) has also been found, but only for beds with gas
distributors providing a low pressure drop (typical for CFB furnaces), as observed in
Fig. 4b (see Svensson et al. (1996) for details).

a) Bubbling fluidized bed. Experimental b) Circulating fluidized bed. Experimental

data from Johnsson et al. (1991). data from Svensson et al. (1996).

Figure 4: Dense bed expansion. Experimental data compared to model fit.

It is shown by Johnsson et al (1991) that making use of a modified two phase flow
theory accounting for gas throughflow and of the widely used expressions for the
bubble size and rising velocity given by Clift and Grace (1977) and
Darton et al. (1985), respectively, the volumetric bubble fraction can be expressed
as:
1
δ= (3)
⋅ (u 0 − u mf )
1.3 − 0.8
1+
f

where f is empirically correlated and which from a fit to experimental data from BFB
(Johnsson et al , 1991) becomes:

( )
f BFB = 0.26 + 0.7 ⋅ e −3.3⋅d s ⋅ (0.15 + u 0 − u mf )
−1 / 3
(4)

This expression also holds for dense beds in the particle seals of CFB units, which
use gas distributors with high pressure drop. Paper I presents a corresponding
expression for bottom beds in CFB furnaces:
−1
f CFB = 0.31 + 0.13 ⋅ u 0 − 16.6 ⋅ d s − 2.61 ⋅ 10 −5 ⋅ Δp ref (5)

10
 Submodels: Theory

Thus, the voidage of a dense bed can be predicted by combining Eqs. 1, 3 and 4
or 5. A saturation value of the dense bed voidage in CFB units occurs at high
fluidization velocities (see Fig. 4b) and a correlation for this value is given in Paper I.
As explained in Papers I and III, modeling of the freeboard solids flow can be made
through a division of the solids flow into two phases: a cluster and a disperse phase.
The cluster phase, present in both BFB and CFB units, is considered to be formed by
particles ejected from the dense bed into the freeboard by means of bubble bursts.
During these bursts, locally high gas velocities (this is known as the throughflow
effect) eject into the freeboard particle clusters which would not be entrainable
otherwise. The clusters follow a ballistic movement, i.e. after reaching a certain
height, they fall back into the bed, as illustrated in Fig. 3. The cluster flow results in a
strong back-mixing of solids yielding a strong gradient in vertical solids concentration
gradient, following the expression:

dC cl
= −a ⋅ C cl (6)
dh

This leads to an exponential vertical decay in cluster phase concentration. The

decay constant a has been subject for several studies in both BFB and CFB
units e.g. Kunii and Levenspiel (1991), Adánez et al. (1994) and Hiller (1995). In the
present work, the expression given by Johnsson and Leckner (1995) is used:

ut
a = 4⋅ (7)
ug

The expression for the cluster phase given in Eq. 6 has been widely used to
describe the vertical solids concentration profile in the freeboard of BFB units as well
as in CFB units. However, in the freeboard of CFB units there is also a so-called
disperse phase. As illustrated in Fig. 3, particles in the disperse phase establish a
core-annulus flow structure, flowing upwards in the core at the solids slip velocity with
backmixing mainly at the furnace walls forming wall-layers with downward flowing
solids. A representation of the solids in the freeboard as a sum of the cluster and
disperse phase was introduced by Johnsson and Leckner (1995) which, in analogy to
the cluster phase, represented the vertical concentration of the disperse phase in the
core region as an exponential decay, i.e:

dC d
= −K ⋅ Cd (8)
dh

The decay constant K was correlated by Johnsson and Leckner (1995) to:

0.23
K= (9)
u g − ut

Disperse phase particles join the wall layers downflow as they gradually disengage
from the core region upflow. In the wall layers, which are relatively thin (several
expressions for its thickness are available in Paper I) and thereby present a higher
solids concentration than the core region. Measurements by Zhang et al. (1995)

11
 Submodels: Theory

indicate that particles in the wall layers flow downwards at roughly their terminal
velocities, which suggests the assumption of stagnant gas in the wall layers.
In summary, both cluster and disperse phase populate the freeboard, with the
bottom part dominated by the ballistic flow pattern of the cluster phase known as the
splash zone while the upper freeboard mostly populated by the more dilute disperse
phase is called transport zone or disperse zone. While the major part of the solids
inventory (mass of solids) in the freeboard belongs to the cluster phase, the disperse
phase dominates most of the extension of the freeboard and governs the net
circulating solids flow. Consideration of both phases is a must for an accurate
macroscopic description of the solids flow and concentration in the freeboard of CFB
units. This can be seen from Fig. 5, where experimental data on solids concentration
in a CFB freeboard is compared with results from several models considering only
one solid phase and one model considering two solid phases. While one-phase
models can predict the solids concentration in the freeboard of BFB units (splash
zone dominates), Fig. 5 shows that such models are not able to provide a satisfactory
prediction of the solids concentration in the freeboard of CFB boilers. Instead, two-
phase modeling accounting for both cluster and disperse phase (solid line in Fig. 5) is
required to the freeboard solids concentration in CFB units.

Figure 5: Vertical solids concentration profile. Comparison between one-phase models (dashed
lines), a two-phase model (solid curve) and measurements (Werdermann, 1993).

The denomination backflow effect is given to the fact that only a fraction of the
disperse phase core upflow reaching the furnace exit height follows the gas flow into
the cyclone, while the rest is disengaged at the exit ducts (mainly at their entrances)
and joins the wall layer downflow. Figure 6 shows experimentally measured values of
the particle entrainment probability for different particle sizes under different
operational conditions and in different large-scale CFB boilers. From these data, a
correlation for the entrainment probability, pent, of a particle at the level of the exit of a
furnace with a single exit duct is given in Paper I.

12
 Submodels: Theory

Figure 6: Entrainment probability of a particle at the exit height. Calculated from

experimental data found in Werdermann (1993) and Åmand et al. (1995).

Since values in Fig. 6 were sampled in CFB combustors with a single exit duct, an
assumption has to be made in order to estimate the particle entrainment probability,
pent’, in the general case of a furnace with n exit ducts. This is represented by the
following expression:
'
= 1 − (1 − pent )
n
pent (10)

Accounting for the backflow effect gives a more accurate representation of the
solids net circulating flow both in terms of magnitude and composition (note the
dependency of the particle entrainment probability on particle size through the slip
velocity, which yields a size segregation effect coupled to the backflow effect).

Fuel particles
Fuel particles follow the same fluid dynamical patterns expressed in the submodels
for inert solids. However, two facts make modeling of fuel mixing significantly more
complex than of inert bed material: the horizontal gradients in concentration and
release of moisture and volatiles and the continuous change in physical properties of
the fuel particles as they burn. Handling of these issues from a modeling point of view
is outlined below and details are given in Paper II.
Horizontal heterogeneity of fuel concentration and moisture and volatile releases is
due to the fact that the horizontal mixing rates of the fuel are limited compared to the
kinetic rate for fuel conversion and release of moisture and volatiles. Thus, for a
correct evaluation of the mixing, the mixing rate has to be compared with the kinetic
rate of the process analyzed. This can be done defining a Damköhler number, which
expressed in terms of characteristic times for the mixing (in [s·m-1]) and the kinetics
(in [s]) becomes:
τ mix ⋅ L
Da = (11)
τ kin
where L is the characteristic length of the system. Low values in Damköhler number
indicate a fast mixing compared to the kinetics and thus a rather homogeneous

13
 Submodels: Theory

distribution of the phenomenon studied. Note that the same mixing and the kinetic
rates might provide an homogeneous fuel distribution (i.e. low value of Da) in a given
unit but an heterogeneous fuel distribution (i.e. high value of Da) in a larger one. An
example of horizontal gradients coupled to the lateral fuel mixing is shown in Fig. 7
which plots two horizontal cuts of the simulated fuel concentration fields in the
Chalmers CFB boiler. The same fuel (wood pellets) has been used for both
simulations but a ten times higher fuel mixing coefficient, Ds,h, has been applied in the
simulation shown in Fig. 7b, leading to a more homogeneous (and generally more
desirable) fuel concentration field.

a) With Ds,h=0.05 m2/s b) With Ds,h=0.5 m2/s

Figure 7: Simulated steady-state fuel concentrations over the cross section of the Chalmers CFB
boiler for different values of the fuel mixing coefficient, Ds,h. Arrows indicate the fuel
feeding location. Simulations obtained with the model described in Paper II in this thesis.

Measurements of the horizontal fuel dispersion rate in FB boilers are scarce. While
measurements by Xiang et al. (1987) in a 100 MWth BFB boiler under cold conditions
and fluidization velocities below 1.3 m/s gave values of the fuel horizontal dispersion
coefficient between 0.001 and 0.01 m2/s, experiments in the Chalmers 12 MWth CFB
given by Niklasson et al. (2002) with a fluidization velocity of 2.3 m/s under hot
conditions gave values around 0.1 m2/s. However, experiments in cold units do not
give exact quantitative information of the mixing pattern under operation but only of
its magnitude.
In order to gain understanding of the fuel mixing process, an experimental method
has been developed which by means of digital image analysis records the trajectory
of a phosphorescent tracer particle (simulating a fuel particle) immersed in a cold 2-
dimensional bed. With this, not only the qualitative flow pattern of the tracer can be
studied but also the influence of several parameters on its mixing rate. This type of
experimental work has been carried out in Papers IV (in a 0.4 m-wide cold unit) and
V (in a 1.2 m-wide unit) and some of the main findings can be observed in Figs. 8
and 9. In both these figures the averaged velocity (vector plot) and the normalized
concentration (color plot) fields of the tracer particle are superimposed.
The first observation to note, common to all runs carried out, is the existence of
regions with relatively low tracer concentration and high upward velocities. These
regions are more clearly defined in some runs than others and there are two such
regions in the runs given in Figs. 8 and 9, located at about x=20 cm and x=100 cm

14
 Submodels: Theory

(easily seen in Figs 8a and 9a). Comparison with visual observation of video
recordings shows that these regions correspond to the main bubble paths.
Studying the velocity plot, it can be seen that the tracer particle follows a flow
pattern structured into horizontally-aligned vortices located around each of the
above-mentioned main bubble paths (these vortices are indicated by arrows in
Fig. 8b). The vortices can be seen as slice cuts of corresponding 3-dimensional
toroidal structures around each main bubble path. Thus, the solids flow pattern is
strongly coupled to the bubble flow. It is observed in other experiments that the
amount of vortices is not a function of fluidization velocity but of dense bed height
(the higher the dense bed, the lower the number of main bubble paths and, thereby,
of flow vortices).

a) Fluidization velocity u0=0.7 m/s b) Fluidization velocity: u0=2.7 m/s

-2 2 -2 2
Resulting dispersion: Ds,h=0.31·10 m /s Resulting dispersion: Ds,h=1.59·10 m /s
Figure 8: Normalized concentration [m-2] and velocity [m/s] of tracer particle with an increase in
fluidization velocity. The horizontal line indicates the dense bed height (Hb=0.33 m). See
Papers IV and V for similar results.

The results in Fig. 8 show a clear influence of the fluidization velocity on the mixing
behavior of the tracer particle. While at low gas velocities (Fig. 8a) the tracer particle
remains almost exclusively in a limited region of the bed (stagnation occurred for
some time periods and had to be removed from the result data for a more clear
representation), an increase in gas velocity (Fig. 8b) leads to a much higher mixing
rate, revealed by the more homogeneous concentration field in Fig. 8b and higher
magnitude of the velocity values (note the different vector plot scales used in Figs 8a
and 8b). In quantitative terms, an increase from 0.7 to 2.7 m/s results in a fivefold
increase in horizontal solids dispersion coefficient, Ds,h. This increase in solids mixing
is found to be a general result of an increase in fluidization velocity, as long as a
dense bed is maintained.
Another general observation from the experiments is the enhanced solids mixing
provided by an increase in dense bed height. This can be observed from Fig. 9,
where results from two runs at the same fluidization velocity but with different dense
bed heights, 0.18 and 0.33 m, are shown. The higher bed gives a 40% higher
horizontal solids dispersion coefficient and significantly higher values of the tracer
velocity (note the difference in scale of the vector plots).

15
 Submodels: Theory

The influence of the tracer particle size and density on the horizontal mixing within
representative ranges for fuel particles was also investigated and shown negligible in
comparison to the influence on fluidization velocity and height of dense bed.

a) Dense bed height: Hb=0.18 m b) Dense bed height: Hb=0.33 m

Resulting dispersion: Ds,h=1.10·10-2 m2/s Resulting dispersion: Ds,h=1.53·10-2 m2/s
Figure 9: Normalized concentration [m-2] and velocity [m/s] of tracer particle with an increase in
dense bed height (indicated by horizontal line) at a fluidization velocity of u0=1.5 m/s. See
Papers IV and V for similar results.

Finally, the pressure drop across the gas distributor was shown to give minor
influence on the mixing provided a proper fluidization could be maintained. Despite
this, low pressure drop might reduce the fuel-air contact, as shown in Paper V,
negatively influencing the combustion process.
In summary, phenomena related to the inert bed material can be treated more or
less satisfactorily with 1.5-dimensional models due to the rather homogeneous
horizontal distribution of the bed material within the dense bed and the core region in
the freeboard. It should be pointed out that the well-known heterogeneities in the wall
layers (due to corner effects and the backflow effect) cannot be handled with 1.5-
dimensional modeling. However, on the other hand, the horizontal gradients present
in crucial parameters coupled to the horizontal mixing of the fuel phase (such as char
concentration and moisture and volatile release rates) require a 3-dimensional
analysis in order to provide a proper description of the process. In the present model,
the horizontal mixing of fuel particles in the bottom region is modeled on a
macroscopic scale by means of horizontal diffusion:

∂C
= Ds ,h ⋅ ∇ 2 C + S (12)
∂t

with the source term S including all of the following: fuel feed to the bottom region,
i.e. fuel feeding ports, refeed of unburnt from the return leg, unburnt in the
downflowing wall layers and, in form of negative values, the core fuel upflow leaving
the bottom region.
In the transport zone, the core-to-annulus net solids flow also represents a lateral
solids mixing mechanism. With respect to the axial direction, perfect fuel mixing is

16
 Submodels: Theory

assumed by the model in the dense bed while the freeboard is, as for inert particles,
represented by Eqs 6 to 10.
In the return leg, fuel particles are assumed to be perfectly mixed with bed material
and are thereby assigned the same residential time as that found for the inert
particles.
Besides the consideration of the lateral mixing, modeling of the fuel mixing requires
that the continuous change in physical properties (i.e. size and density) of the fuel
particles as they burn is accounted for. This so that the corresponding changes in
fluid dynamical behavior of the fuel particles can be described. Solid fuel in a boiler
undergoes three potentially overlapping mechanisms: drying, devolatilization and
char combustion. Under drying, the moisture release decreases the density of the
fuel particle; analogously there is a further decrease in particle density during
devolatilization due to the volatile matter released together with a certain shrinkage
and, finally, char combustion can occur in two ways: through a shrink in size at
approximately constant density or at constant particle size and with decreasing
density. Neglecting the shrinkage due to devolatilization, the dynamical mass release
rates of each of these three steps govern the size and density of the fuel particle
under its conversion. As detailed in Paper II, these mass release rates can be
described through the fuel conversion submodels by Thunman et al. (2004) for drying
and devolatilization and by Field et al. (1967) for char combustion. The composition
of the volatile matter released is calculated through an appropriate combination of
energy and elementary mass balances and empirical ratios, as described in
Thunman et al. (2001).
The submodel for drying and devolatilization is based on the description of the
evolution of the temperature profile within the particle by solving the dimensionless
energy equation given by Thunman et al. (2004):

1 ∂ ⎛ r ∂θ ⎞ 1 ∂θ
⎜⎜ ξ ⎟⎟ + c ⋅ r =0 (13)
ξ ∂ξ ⎝ ∂ξ ⎠
r
ξ ∂ξ

The boundary conditions used in the solving process, expressed in non-

dimensionless form, are:

T r = drying front
= 100 ºC (14)

− λ part
∂T
∂r
(
= heff ⋅ T r = surface
− T∞ ) (15)
r = surface

where the effective heat transfer coefficient, heff, accounts for both convective and
radiative heat. A main advantage of this submodel given by Thunman et al. (2004) is
that it gives the existence of analytical expressions for the quasi-steady state
solution of Eq. 13, providing a fast solving procedure for a wide variety of fuel
particle geometries and compositions.
Concerning char combustion, a surface reaction process is most often established,
for which the effective char combustion rate, Reff, can be modeled according to
Field et al. (1967) through a combination of the kinetic rate, Rkin (determined through

17
 Submodels: Theory

an Arrhenius-type correlation) and the mass transfer coefficient, km (determined

through a Sherwood-type correlation):

1
Reff = (16)
1 1
+
Rkin km

Thus, the expression for char combustion (first order reaction is assumed) reads:

d mC
= −Ω ⋅ Reff ⋅ A part ⋅ C O2 , ∞ (17)
dt

The satisfactory agreement of the fuel particle conversion submodel outlined can
be seen in the comparison between in-house thermogravimetrical experiments and
modeled data shown in Fig. 10 for a wet coal particle. In this example, an
overestimation of the drying and devolatilization rates is the reason for the slight
overestimation of the modeled mass release in the initial time period. From measured
data it is seen that the mass release coupled to drying and devolatilization dominates
in the first 100 s and thereafter char combustion takes over. Finally, after 200 s, only
the non-combustibles corresponding to the fuel ash content remain. From
comparison with other modeling approaches, the present conversion submodel (Eqs
13 to 17) gives an agreement similar to that of discrete models but at a much lower
computational cost.

Figure 10: Coal particle conversion. Comparison Figure 11: Coal particle conversion. Change in
between model results and in-house physical parameters as obtained from
experiments (Thunman, 2005). the model (Eqs 13 to 17).

Having modeled the dynamical mass release during drying, devolatilization and
char combustion in a fuel particle, calculation of the change in particle size and
apparent density is rather straightforward neglecting shrinking during drying and
devolatilization, assuming char combustion to take place at the particle surface and
ashes to leave the particle during char burn out. With this, the modeled particle size
and apparent density corresponding to the same coal particle analyzed in Fig. 10 can
be calculated and are shown (in normalized values) in Fig. 11. With these, the
evolution of the terminal velocity of the fuel particle during its conversion can be
easily calculated (for example through the expressions by Haider and
Levenspiel, 1989) and is also plotted in Fig. 11. The continuous change in terminal
velocity gives an idea on how different the fluiddynamical behavior of a fuel particle
can be, depending on its degree of conversion.

18
 Submodels: Theory

Finally, it should be noted that the drastic and sudden change in the fuel particle
size entailed by fuel fragmentation plays an important role in both fuel conversion
kinetics and fuel mixing. The comprehensive model presented here does not include
a submodel for fragmentation (although such are available in literature, see
e.g. Salatino and Massimilla, 1989, or Chirone et al., 1991). Instead, this is an input
which can be determined experimentally through fragmentation tests. A brief
sensitivity analysis of the influence of fuel fragmentation on several fuel-related
parameters is given in Paper II. The reason for not implementing a fuel fragmentation
model is the further work needed in order to develop reliable fragmentation models
which can be applied under boiler conditions.

Gas phase
A key feature of the present work is the modeling of the gas phase, described in
detail in Paper III. The gas phase in a fluidized bed has a fluctuating character in
terms of both flow magnitude and composition, as shown respectively by the
oscillating pressure in the plenum (Sasic et al., 2004) and fluctuations in zirconia cell
measurements (Niklasson et al., 2003). In addition, measurements show that both
velocity and composition of the gas flow vary strongly in the bottom region of the
furnace depending on whether the gas flows through the bubble phase (throughflow
gas) or not (emulsion-only gas), as explained by Lyngfelt (1996).
However, neither a differentiation between throughflow and emulsion-only gas nor
the flow fluctuations have been included in the macroscopic models of the gas phase
given in literature. Instead, previous models have been limited to directly model time-
averaged values which lead to a need for empirical fitting of the combustion kinetics.
The present submodel divides the gas phase into a throughflow and an emulsion-
only gas and models their respective flow fluctuations, which then govern the oxygen-
fuel contact. This approach is based on the dynamical pressure balance presented in
Paper III. To begin with, the rise of a bubble through the dense bed must be
described. For this, there are expressions for bubble size and rise velocity available
in literature (Clift and Grace, 1977, and Darton et al., 1985). The pressure balance
over a representative bed portion, L*, at an arbitrary time step during the bubble rise
is illustrated in Fig. 12, with the gas divided into two phases: throughflow and
emulsion-only (orange and brown respectively). The pressure drops across the gas
distributor and the dense bed solids are considered for each gas phase, yielding
corresponding instantaneous gas fluxes (i.e. for gas flux in emulsion phase and in
through flow phase).

Figure 12: Scheme of the pressure balance for the two considered gas phases.

19
 Submodels: Theory

The calculated gas fluxes are combined with values of the combustible volatile
flows and char concentration obtained from other submodels (details on the link
between the models are given in Chapter 3) in order to compute the progress of
combustion. As mentioned previously, the combustion rates do not need to be
experimentally fitted in the present approach, and transport-controled combustion
(i.e. infinitely fast kinetics) can be assumed for the combustible volatile matter, while
char combustion can be calculated by Eqs. 16 and 17. Thus, in each cell of the bed,
the fluctuating flow of gas species calculated through the dynamical pressure
balance can be combined with the outputs from the fuel mixing submodel in order to
calculate how the combustion progresses. Modeled data showing the evolution of the
oxygen consumption with height during a bubble cycle is shown in Fig. 13. This figure
shows that, along the furnace centerline, at a height of 0.2 m above the gas
distributor the oxygen is fully consumed in the emulsion-only gas phase at some
point during the bubble cycle. Above this height, the time interval of flow without
oxygen increases with height. During absence of oxygen, released volatile
hydrocarbons will remain unburned and add to the gas flow, as observed in the
dotted curve corresponding to h=0.24 m.

Figure 13: Modeled cell flow of gas species (oxygen and hydrocarbons) in the
emulsion-only phase at different heights on the furnace centerline.

It is important to note that on a time-averaged basis the cell corresponding to a

height of 0.24 m at the furnace centerline yields non-zero concentrations of both
oxygen and hydrocarbons. Yet, coexistence of these is not allowed in the time-
resolved modeling.
As a general result from the pressure balance illustrated by Fig. 12, gas fluxes (and
thereby gas velocities) predicted for the throughflow are higher than those of the
emulsion-only flow at any time step. In addition, the throughflow flows across a
shorter emulsion path length, coming into contact with less fuel particles and thus
experiencing lower oxygen consumption. Therefore, two types of gas phases are

20
 Submodels: Theory

present at the dense bed surface: a high velocity oxygen-rich (throughflow) phase
and a low velocity oxygen-poor (emulsion-only) phase.
The velocity difference between gas phases, responsible for the bias observed in
gas probe measurements in the bottom region of fluidized bed units, decreases with
height in the splash zone. Also, the flow fluctuations within each gas phase are
gradually damped along the splash zone. As a result, there is a significant gas
backmixing in the splash zone which eventually leads to a rather constant (i.e. non-
fluctuating) gas flow in the transport zone.
As stated above, measurements by Zhang et al. (1995) suggest a more or less
stagnant gas flow within the wall layers of the transport region, supporting the
assumption of gas flow in the core region only. Having this, horizontal dispersion of
the gas in the core region of the transport zone is considered according to:

∂Gi
= D g , h ⋅ ∇ 2 Gi (18)
∂t

Gas species concentration, Ci, is usually the transported scalar in Eq. 18 instead of
the gas species flux, Gi , used here. While the use of both variables leads to the
same result under plug flow conditions, the use of the gas species flux is able to
describe the flattening of heterogeneities in gas velocity field when such are present.
Such heterogeneities are often present, mainly due to secondary air injections and
have been measured by Kruse and Werther (1995). From several experimental
studies available in literature on values for the gas horizontal dispersion coefficient,
Dg,h, the Peclet number expression given in Kruse et al. (1995) is used:

u ⋅ Lriser
Pe = = 387 (19)
D g ,h

In the cyclone perfect gas mixing is assumed, implying full combustion of any
combustible volatiles remaining unburned, provided enough oxygen is available.
Finally, all gas leaving the particle seals is assumed to flow towards the riser
through the solids recirculation ducts.

Heat transfer
In a fluidized bed boiler, heat is extracted from the solids and gas in the circulating
loop and from the flue gas in the convective pass after the cyclone. With a given
stack temperature for the flue gas, computation of the heat extraction from the
convective pass is straightforward once the composition and temperature of the flue
gas after the cyclone have been obtained from the comprehensive model. Thus,
focus is set on the more complex modeling of the heat transfer in the circulating loop.
While the formulation of a heat balance is given in Chapter 3, the different terms in
this modeling originate from the submodels explained in Chapter 2. With terms
related to solids (including the fuel fraction) and gas flow available as output from
other submodels, the heat transfer terms in the heat balance which describe the heat
extraction are addressed in this section.

21
 Submodels: Theory

Heat is transferred to the heat exchanging surfaces through two mechanisms:

convection and radiation. Simplified approaches estimating the heat transferred by
the sum of these two phenomena exist, but a separate treatment of convection and
radiation is desirable for a more accurate modeling of the heat transfer. This, since in
the rather dilute upper freeboard, radiation from regions away from the heat transfer
surfaces plays a more important role than in the bottom region. In the bottom region,
the denser solids suspension density has a stronger absorption effect and gas and
solids flowing near the heat transfer surfaces limit the influence of radiation from
further away from the surfaces. Thus, heat transfer is expressed as:

q = q conv + q rad (20)

where convective and radiative heat transfer from the gas-solid suspension to any
surface are treated separately and each expressed in the classical form given by:

q = h ⋅ A ⋅ (Tsusp − Tsurf ) (21)

where the temperature of the gas-solid suspension must be taken as the local
temperature close to the heat transfer surface in the case for convection and as
~
some average over the surrounding in the case of radiation ( T , which is explained
below). In the present work, only heat transfer between the gas-solid suspension and
the heat transfer surfaces is accounted for, i.e. heat transfer within the gas-solid
suspension is neglected.
Breitholtz et al. (2001) managed to differentiate convection and radiation in their
measurements and correlated the convective heat transfer coefficient to increase
with the local solids concentration as:

hconv = 25 ⋅ C s0,.58
susp (22)

The radiation heat transfer coefficient between a suspension and a heat transfer
surface reaches its maximum value when the suspension is free of solids, reading:

hˆrad =
1
1
1
(~2
⋅ σ ⋅ Tsusp + Tsurf
2
)( ~
⋅ Tsusp + Tsurf ) (23)
+ −1
ε susp ε surf

However, this coefficient decreases with the presence of solids in the suspension
due to the shadowing effect of the suspended particles. In order to face this,
Breitholtz et al. (2001) defined and correlated a radiative heat transfer efficiency as:
~
hrad ⎛ C s , susp ⎞
η rad = = 0.86 − 0.14 * atg ⎜ − 1.6 ⎟ (24)
hˆrad ⎜ 2.6 ⎟
⎝ ⎠

Note that suspension parameters (solids concentration and temperature) are

averaged when used in expressions which concern radiative heat transfer such as
Eqs. 23 and 24. This averaging stands for the fact that properties of the suspension
which is not in the vicinity of the heat exchange surface can also play an important
role in the radiative heat transfer. This is illustrated in Fig. 14, where the differential

22
 Submodels: Theory

weight of local values is represented by a continuous red line and the mesh grid at
the level studied is drawn in green with the two figures illustrating typical conditions at
low and high level in the freeboard (i.e. dilute and dense solids suspension,
respectively).

Figure 14: Weight of the local parameters in the Bouguer-averaging used in radiative
heat modeling. The green lines represent a mesh at a certain level in the
furnace, with the furnace wall (heat transfer surface) to the left.

As seen in Fig. 14, under dense flow conditions (such as in the lower level in the
freeboard) the suspension close to the heat transfer surface (the wall to the left in this
case) takes most of the area below the red curve representing the differential weight
of radiation parameters. A different situation is encountered in dilute suspensions
(upper freeboard positions), where cells also located relatively far away from the wall
have a certain weight according to the red curve. The weighting curve is given by the
Bouguer’s law for radiative properties of solids suspensions (see e.g. Baskakov and
Leckner, 1997, for an example on its use in fluidized bed units), which accounts for
the absorption effect of solids in a suspension according to:

⎛ 1 .5 ⋅ C v , s ⋅ L ⎞
α = 1 − exp⎜⎜ − ⎟
⎟ (25)
⎝ ds ⎠

For simplicity reasons, calculation of the radiative heat from the suspension to the
heat transfer surfaces in the circulating loop is limited in the sense that a certain cell
is assumed to only transfer radiative heat to cells at the same height in the furnace.
Details on the implementation of the Bouguer’s law and other details on heat transfer
modeling are given in Appendix A.

23
Chapter 3 – Comprehensive model

3.1 Coupling of submodels

The submodels summarized in Chapter 2 and given in Papers I, II and III and
Appendix A can be linked in order to build a comprehensive model covering several
fields in the process, i.e. fluid dynamics, combustion and heat transfer in the present
case. The submodels presented need inputs which can be either external (if provided
by the user of the model) or internal (if provided as outputs from other submodels or
balances in the comprehensive model). A scheme of the coupling between
submodels indicating the flow of internal and external inputs is given in Fig. 15.
Pressure, population and heat balances are required for the solving procedure of the
comprehensive model.
The external inputs required by the comprehensive model can be divided into four
categories:
• Geometry of the circulating loop.
• Characterization of the fuel and gas feeding through fuel proximate and ultimate
analyses, heating value, size, shape, temperature and feeding rate at each of
the feeding ports, and gas composition and temperature at all injection
locations. In addition, a fuel fragmentation pattern must also be provided to the
model, although this is not a known variable (but defined as an input to provide
transparency with respect to that this is an area where additional experimental
work is required).
• Physical properties of the in-bed inert solids, namely the PSD, density and
sphericity of each inert solid fraction considered in the bed material. As dicussed
above, the in-bed PSD is not known or easily measurable and its prediction
(through implementation of submodels for attrition and size segregation in the
cyclones and ash classifiers) should be included in further development of the
present model.
• Operational conditions, which in the present model (and generally in practical
operation) are the vertical pressure drop over the furnace (or part of it) and the
temperature of the heat-extracting surfaces (which, for most cases, is the
evaporation temperature at the steam pressure on the waterside).

24
 Comprehensive model – Coupling of submodels

C Concentration field Superscripts

F Flow field fuel Average experienced by fuel
H Height
h Heat transfer coefficient Subscripts
r Release rate field char Char fraction
t Wall layer thickness conv Convective
T Temperature field dev Devolatilization
u Velocity field fuel Fuel
δ Bubble fraction g Gas
τ Residence time inerts Inert solid fraction
ΔP0 Furnace pressure drop rad Radiative
vap Vaporization of fuel moisture
Figure 15: Data flow scheme used in the coupling of submodels. External inputs are indicated
by the four white ellipses. Papers describing the different submodels are indicated.

25
 Comprehensive model – Coupling of submodels

Note that the submodel for fuel conversion has two inputs (the average oxygen
concentration and temperature as experienced by the fuel particles) which are not
direct outputs from any of the submodels but calculated by combination of outputs
from other submodels.
The output of the overall model is a comprehensive set of data from the different
submodels, but the most relevant results for evaluation and/or validation of the
simulated process are generally the solids, gas and heat flow fields, the solids and gas
concentration fields and the temperature field.
In the following, the formulation of the different balances in the solving process is
given together with an outline of the strategy used to couple the fuel mixing and fuel
conversion submodels.

Heat balance
As seen in Fig. 15, the heat balance in the present model provides the temperature
field in the circulating loop given data on gas and solids flows, fuel releases (both
drying and devolatilization are endothermic processes) and heat transfer coefficients.
Individual heat balances are performed over different elements (either mesh cells in
the freeboard or regions considered to have perfect thermal mixing), providing an
output temperature value to each of these elements and thereby forming a complete
temperature field. In the present model, the bottom bed is assumed to be thermally
perfectly mixed and therefore a single heat balance is formulated over the bottom
bed. In the freeboard, individual heat balances are performed for each mesh element
except for the region above the exit ducts towards the cyclone, which is treated as a
single, thermally perfect mixed element. The return leg is divided into different
regions assumed to be perfectly mixed, representing the cyclones and particle seals.
The heat balance for each element reads:

∑ h&
s
s , in + ∑ h&g , in = ∑ h&s , out + ∑ h&g , out + h&vap + h&dev + qconv + qrad
g s g
(26)

The terms can be easily identified in Eq. 26 representing the enthalpy flows coupled
to the different solid fractions, s, and gas species, g, the heats of vaporization and
devolatilization from the fuel-released moisture and volatiles and the transferred
convective and radiative heat.
When applied to the different furnace regions considered in the model, the general
form of the heat balance given in Eq. 26 takes different forms which are summarized
below. For simplicity, only waterwalls have been considered as heat extraction
surfaces in the model (although any internals can be treated in an analogous way).
The general formulation of the heat balance for any of the regions in the circulating
loop which can be assumed to exhibit perfect thermal mixing, region (i.e. dense bed,
exit zone, cyclones and particles seals), is expressed as:

∑F s , in ⋅ hs (Ts , in ) + ∑ Fg , in ⋅ hg (Tg , in ) =
s g

∑F s , out ⋅ hs (Tregion ) + ∑ Fg , out ⋅ hg (Tregion ) (27)

s g

( ) (
rvap ⋅ hvap + rdev ⋅ hdev + hconv ⋅ Asurf ⋅ Tregion − Tsurf + hrad ⋅ Asurf ⋅ Tregion − Tsurf )

26
 Comprehensive model – Coupling of submodels

Except for the temperature of the region studied to be solved, Tregion, all data
required by Eq. 27 are obtained from different submodels, as shown in Fig. 15.
In the dense bed, several solid streams need to be considered as inflows: the
bottom wall layer solids downflow, the externally recirculated solids from the seal and
the fuel fed to the bed, each at different temperatures. Solids leaving the dense bed
are represented by the solids upflow in the bottom core region. In addition, a solids
flow (an input in the current version of the model) of inerts and char leaves the dense
bed at the dense bed temperature towards the ash classifier and reenters the bed at
a higher temperature due to char combustion in the classifier.
The exit zone, the furnace region between the exit ducts and the furnace roof of the
furnace, has the upflow in the top of the core region as inflow. The gas flow will leave
towards the cyclone and the solids flow will split due to the backflow effect into a
fraction entrained with the gas flow and a fraction internally recirculated through the
wall layers.
In the cyclones and particle seals, identification of the solids in- and outflows is
straightforward and thus is not described here.
For a core region cell both the gas and solids phase are assumed to enter the cell
through its bottom face. If the cell studied is within the secondary air penetration area
defined by the overall model inputs, secondary air is assumed to enter the cell at the
injection temperature. The net transfer of solids from core to wall layer, Flat, is
assumed to be at the cell average temperature. No convective heat transfer term is
considered for a core region cell. Finally, the radiative heat loss to the furnace wall
applies a cell-dependent weighting coefficient, w’, as detailed in Appendix A.

∑F s , bot ⋅ hs (Tbot ) + ∑ Fg , bot ⋅ hg (Tbot ) + ∑ Fg , inj ⋅ hg (Tinj ) =

s g g

⎛ Tbot + Ttop ⎞
∑F s , top ⋅ hs (Ttop ) + ∑ Fs , lat ⋅ hs ⎜⎜ ⎟⎟ + ∑ Fg , top ⋅ hg (Ttop ) + (28)
s s ⎝ 2 ⎠ g
⎛ ⎛ Tbot + Ttop ⎞ ⎞
rvap ⋅ hvap + rdev ⋅ hdev + w ' ⋅ hrad ⋅ Asurf ⋅ ⎜⎜ ⎜⎜ ⎟⎟ − Tsurf ⎟
⎟
⎝⎝ 2 ⎠ ⎠
In the wall region, no gas flow is considered (this is also assumed for gaseous fuel
releases and secondary gas injections). Downflowing solids enter the cells from the
top face and are joined by the solids laterally transferred from the core region. In the
lateral net solids flow from the core to the wall region each core cell is assigned a
target wall cell to which the solids disengaged from the upflow flow into. Thus, each
wall layer cell might have solids inflow from several core cells, with each inflow at the
respective temperatures of the different core cells they originate from. Heat transfer
to the waterwalls takes place by means of both convection and radiation. Thus, the
heat balance in a wall layer cell reads:

∑F s , top ⋅ hs (Ttop ) + ∑ Fs , lat ⋅ hs (Tcore cell ) =

s s

∑F s , bot ⋅ hs (Tbot ) + (29)

s

⎛ ⎛ Tbot + Ttop ⎞ ⎞ ⎛ ⎛ Tbot + Ttop ⎞ ⎞

hconv ⋅ Asurf ⋅ ⎜ ⎜⎜ ⎟⎟ − Tsurf ⎟ + w ' ⋅ hrad ⋅ Asurf ⋅ ⎜⎜ ⎜⎜ ⎟⎟ − Tsurf ⎟
⎜ 2 ⎟ 2 ⎟
⎝⎝ ⎠ ⎠ ⎝⎝ ⎠ ⎠

27
 Comprehensive model – Coupling of submodels

Pressure and population balances for the inert solids

The vertical pressure drop over the furnace (or part of it) is a usual operational
parameter in FB units and is therefore treated in the present model as an external
input to be provided by the user of the model (as listed above and shown in Fig. 15).
It is an indicator of the amount of solids in the furnace and, thereby, of the dense bed
height. Once a solids concentration field in the furnace is obtained, the pressure drop
over the furnace can be easily calculated through axial integration:
H furnace

ΔP0 = ∑C
h =0
s, h ⋅ g ⋅ dz h (30)

where C s , h is the cross-sectional average solids concentration at a certain height, h,

and dzh the height of the cells at that same height. Obviously, varying the summation
height limits in Eq. 30 provides the pressure drop over the given height interval.
Setting these summation limits to the heights corresponding to the recirculation ducts
and the furnace exit ducts (see respectively Prefeed and Pexit in Fig. 16) gives the
freeboard term in the pressure balance over the circulating loop as:

ΔPfreeboard + ΔPexit duct + ΔPcycl + ΔPdc + ΔPrecirc duct = 0 (31)

Generally, the pressure drop over the recirculation duct is neglected compared to
the other terms, i.e. Pseal and Prefeed in Fig. 16 are assumed equal. Models to estimate
the pressure drops over the exit duct and the cyclone exist in literature (see
Muschelknautz and Muschelknautz, 1991, and Rhodes and Geldart, 1987,
respectively). With such a model and given a solids concentration field in the furnace,
all terms in Eq. 31 can be either estimated or neglected except for the pressure drop
over the downcomer, ΔPdc, expressed as the difference between Pdc and Pseal in
Fig. 16. This pressure drop is caused by the presence of the fluidized solids column
(assumed to have the same voidage as that calculated for the particle seal) and is
calculated from the pressure balance over the circulating loop by using Eq. 31.

Figure 16: Key locations in the pressure balance over the circulating loop

28
 Comprehensive model – Coupling of submodels

From the calculated pressure drop across a solids column, the height of the column
is easily estimated through:

ΔPdc
H dc = (32)
C s , dc ⋅ g

Hereby, solids in the column formed in the downcomer (which might be an

important term depending on the geometry of the unit and the operational conditions)
can be calculated and taken into account in the formulation of a population balance
over the circulating loop. Thus, formulation of a pressure balance is used to provide a
value of the initially unknown term in the population balance.

Combined modeling of fuel mixing and conversion

A major difficulty encountered in the modeling of the fuel mixing is the continuously
changing fluid dynamical behavior of the fuel particles as they burn, due to the
variation in size and/or density. Thus solving the fuel mixing requires that the
characteristics of the fuel conversion are accounted for and, as shown in Fig. 15, the
two submodels of mixing and conversion must be combined in order to solve the fuel
flow and concentration fields, as explained in Paper II.
An efficient way to work out this problem is based on the transient simulation of the
behavior of a batch of fuel particles during the whole burnout time. The burn out time
is discretized in several time steps for which the fuel particles present the
corresponding values for the particle size and density provided by the fuel conversion
model. With this, the fluid dynamics of the fuel fraction are solved at each time step
applying the corresponding value for the fuel terminal velocity, providing the spatial
distribution of the fuel concentration at any time step, Cfuel,t, originated by the fuel
batch. This is illustrated by the first row in Fig. 17.

Figure 17: Scheme applied to the calculation of the steady-state fuel concentration field

Having this, a continuous feeding of fuel can be simulated by a ‘continuous batch’

approach, in which further fuel batches are fed to the unit at each time step. This
procedure is illustrated in Fig. 17, where the total fuel concentration at a certain time
step is equal to the sum of the values in the corresponding column. Thus, note that a
stationary concentration field is reached for the time steps after the burnout time of
the fuel particles, with a value equal to the sum of all intermediate concentration
values.

29
 Comprehensive model – Computational flow scheme

3.2 Computational flow scheme

The computational flow scheme required to solve the model is far from unique and
can be done in many different ways. Yet, optimization of the computational flow
scheme in order to minimize the computational time while ensuring a converging
solution is a demanding task. Generally, several self-containing iterative loops are
needed in the computation and sensitivity of each submodel is decisive for deciding
which submodels occupy the inner iterations. Computational techniques have been
adopted in the code in order to ensure and speed up convergence, such as
dynamical tolerance values and relaxation factors, but are considered outside the
scope of this manuscript. Thus, only a qualitative outline of the solving procedure of
the comprehensive model is given in Fig. 18.

Figure 18: Solving procedure of the comprehensive model.

Papers describing the different submodels are indicated.

As seen in Fig. 18, there are three iterational structures at an inter-submodel level,
with the convergence criteria for each of them being, respectively, the convergence
of the average oxygen concentration and temperature experienced by the fuel and
the temperature field in the circulating loop. While the internal computation of the
submodels in Fig. 18 is straightforward from what is given in Chapter 2 and the
respective papers dealing with each submodel, the internal computation of the
submodel for the inert solids mixing and the heat balance have a higher complexity
and are therefore explained further below.
Concerning the model for the inert solids mixing, its internal solving procedure is
illustrated in Fig. 19. As seen, the dense bed and the cluster and disperse phase in
the freeboard are first computed. This is done for each size interval of each solid
fraction considered, with initial values for their dense bed PSD. From the resulting
solids concentration field, the variable ΔPref is calculated by applying Eq. 30 between
h=0.14 and 1.5 m (as specified in Paper I). The value obtained is used in Eq. 5 to
estimate the dense bed voidage and this procedure is followed until a converging
dense bed voidage is reached. In the midst iteration, the calculated operational
pressure drop over the furnace is compared to the value provided as external input to
the model, and the dense bed height is adjusted accordingly. After this, the pressure
balance gives the height of the solids column in the downcomer, which can then be
included in the population balance. Comparing the PSD of the calculated population

30
 Comprehensive model – Computational flow scheme

balance over the unit with that given as external input, the PSD in the dense bed to
be used in the next iteration is chosen.

Figure 19: Solving procedure of the submodel for inert solids mixing

The solving procedure followed to couple the individual heat balances expressed by
Eqs. 27 to 29 into the overall heat balance for the entire circulating loop seen in
Figs 15 and 18 is illustrated in Fig. 20. The boxes in Fig. 20 referring to the core and
wall cells represent all the individual mass balances formulated for each of these
cells. For the core cells, the calculation starts with the heat balances of the bottom
freeboard cells and continues upwards, while the opposite applies for the wall layer
cells. Thus, while the Core cells and Wall cells boxes in Fig. 20 give temperature
fields with as many values as cells evaluated, the other outputs are single scalars.
Two iterative processes constitute the solving scheme of the heat balance over the
circulating loop. The inner iteration which is solved first gives the temperature field in
the furnace and uses as convergence criterion the temperature of the wall layer
downflow joining the dense bed.

Figure 20: Solving procedure of the heat balance over the circulating loop.

After obtaining a converged furnace temperature field, heat balances in the

elements of the return leg are formulated and the temperature of the solids externally
recirculated to the dense bed (i.e. the temperature of the particle seals) is used as
convergence criterion.

31
 Comprehensive model – Furnace mesh

3.3 Furnace mesh

In order to enable a 3-dimensional analysis, the furnace is discretized by a mesh

consisting of regular hexahedra. Finer meshing gives higher resolution in the results
but also increased calculation time, so some compromise has to be adopted in the
choice of the number of cells (calculations shown in this work have been performed
with a number of cells in the order of 105).
For simplicity, all cells at a given height in the furnace have the same vertical
extension, whereas the vertical extension varies with height in furnace. The mesh is
generated so that the dense bed surface exactly coincides with the interface between
two horizontal cell layers (note from Fig. 19 that the dense bed height is not fixed but
varies during the computations, which implies that the mesh also needs to be
regenerated during computation).

Figure 21: Meshing of the freeboard wall layers in the horizontal direction.
Wall layer cells are gray-colored.

While the mesh is regular in the dense bed (cell sizes in the three dimensions can
be chosen as input), the discretization of the freeboard is limited by the fact that no
heterogeneities in the wall layers are accounted for in the direction normal to the
furnace walls. Thus, at each height, the wall layer thickness becomes the size of the
wall layer cells in the direction normal to the furnace walls, as shown in Fig. 21 for
one of the furnace corners in an arbitrary horizontal cell layer (wall layer cells are
gray-colored and core region cells are white).
In FB units, the splash zone presents a strong gradient in solids concentration and
solids size and in fluctuating amplitude of the gas velocity, and its modeling is known
to be critical for the correct performance of the overall modeling. To achieve a higher
accuracy in this region, a refined mesh in the axial direction along the splash zone is
used.
Finally, Fig. 22 exemplifies a representation of a fraction (10%) of cells from the
mesh generated for the furnace of the 550 MWth biomass-fired CFB boiler located in
Alholmen (Finland).

32
 Comprehensive model – Furnace mesh

Figure 22: Mesh for the 550 MWth biomass-fired CFB boiler in Alholmen (Finland). Only 10% of the
nodes are plotted for better visualization. The total amount of cells is roughly 5·105.

33
Chapter 4 – Results
In order to illustrate its use and features, the comprehensive model presented
above is applied to analyze the differences between combustion of high- and low-
volatile content fuels in a CFB unit. Although fuel flexibility is one of the main
advantages of fluidized bed combustion, the characteristics of the combustion
process depend strongly on the fuel type. This makes the thorough analysis of
combustion of different fuel types a must for the correct understanding of the process
and thereafter for the optimization and scale up of the process.
Modeled data in this chapter are not compared to experimental data, but
comparisons of modeled results with measurements in large-scale CFB boilers are
available in Papers I, II and III.
Solid fuels can be classified according to the distribution of their combustible matter
into volatile and char fields, given in the proximate analysis. The fuel types used in
fluidized bed combustion sorted in descending char content (i.e. in ascending volatile
content) read: coal, peat, biomass and waste. Here, for the sake of comparison, CFB
combustion of a high-rank coal (anthracite) and biomass (wet, high-volatile wood
chips) are selected for the model simulations. These fuels have been used in the
experimental works by Lee et al. (2003) and Niklasson et al. (2007), respectively.
Proximate and ultimate analyses and heating values for both fuel types are given in
Table 2.
Table 2: Proximate and ultimate analyses and heating values
of the two fuel types used in the simulations.

Anthracite Wood
Moisture 3.2 34.9
[%] Volatiles 3.9 50.4
as received Char 65.6 14.2
Ash 27.3 0.5
C 95.61 49.50
H 1.11 6.21
[%]
O 1.15 44.17
dry ash-free
N 0.60 0.10
S 1.53 0.02
as received 22.02 11.38
LHV
dry 22.82 18.57
[MJ/kg]
dry ash-free 31.78 18.72

34
 Results

The boiler chosen for the simulations is the Chalmers CFB boiler, with a furnace
height of 13.5 m which is schematized in Fig. 23. The single fuel feed point is located
in the middle of the front wall at a height of 1.5 m and the secondary air injection
ports used in the simulations are at both front and rear walls at a height of 2.1 m. The
two sidewalls are waterwalls from 2.2 m above the distributor, with the remaining of
the riser walls being refractory-lined. The bottom part cross section is 1.42 × 1.42 m,
becoming 1.62 × 1.42 m as the side walls becomes bare membrane walls. Gases
from the seal and the ash classifier enter the riser through the rear wall at 0.99 and
1.29 m above the primary gas distributor respectively. The furnace exit duct to the
cyclone is placed in the rear wall, with its centerpoint at a height of 10.2 m above the
air distributor. Further details on the geometry of the unit are given in
Leckner et al. (1991). In the return leg, there is a heat exchanger in the seal system
for increased flexibility by allowing control of the furnace temperature.

Figure 23: Illustration of the furnace of the Chalmers CFB boiler. The model covers
the entire CFB loop, but only furnace is modeled 3-dimensionally.

The two fuel cases are simulated on the basis of a common heat power fed to the
unit of 8 MWth, at an air-to-fuel of 1.2. With the lower heating values given in Table 2,
the respective fuel flows can be calculated (i.e. higher volumetric flow for the wood
case, see Table 3).

35
 Results

Low-volatile fuels such as anthracite burn rather uniformly distributed over the
furnace cross section (due to the relatively low kinetics of char combustion) and the
fuel (char) is mostly present in the bottom region of the furnace (see Paper II). This
makes it possible to operate the boiler with more or less all combustion air injected as
primary air. Thus, no secondary air injection is applied in the simulations shown here
for the anthracite case. On the other hand, high volatile fuels are known to be
heterogeneously distributed over the furnace cross section. Introduction of secondary
air helps enhancing the burnout of the volatiles by improving the gas mixing,
i.e. breaking up the flow of so-called ghost bubbles (see Niklasson et al. (2003) for
details on the ghost bubble phenomenon). The optimization of the flow magnitude
and spatial distribution of the secondary air injections is often an issue of discussion,
and for the present simulation of the wood case a 2ary-to-1ary air-to-fuel of 0.57 has
been chosen, with 13 active secondary air nozzles located at a height of 2.1 m (7 in
the front wall and 6 in the rear wall).

Table 3: Main operational conditions used in the simulations.

Anthracite Wood

Heat power fed MWth 8

Fuel feeding rate kg/s 0.36 0.70

Air-to-fuel - 1.2

Fluidization velocity in furnace* m/s 4.48 2.74

2ary-to-1ary air ratio - 0 0.57

Fluidization velocity in seal* m/s 0.32

Furnace pressure drop Pa 9100

Mean particle size μm 348

Solids density kg/m3 2600

Dense bed aimed temperature K 1173 1123

Waterside temperature K 493

* calculated at 1123 K

For a more straight-forward comparison between the two cases, the same air flow
(0.17 kg/s) is supplied to the seal and the ash classifier in both cases. Also, the same
inert bed material, furnace pressure drop and fluidization velocity in the seal are used
in both cases.
Presence of internals and wing walls for controlling the furnace temperature is
common in large CFB boilers. In the Chalmers CFB boiler, the dense bed
temperature can be controlled through a heat exchanger located in the seal system.
Thus, under operation, an aimed dense bed temperature can be approximated
through the removal of more or less heat from the heat exchanger in the seal. The
maximum temperature in the dense bed is obviously reached when no heat is
removed from the seal system.

36
 Results

Finally, reactivity of anthracite char is much lower than that of wood. Furthermore,
the char fed to the unit for the anthracite case is more than double than for the wood
case. These (especially the difference in reactivity), make the char inventory in the
unit become very high for anthracite (see Paper II for data on this). Having too high
char concentrations in the unit implies a loss in combustion efficiency due to
entrainment of char fines out of the circulating loop. In order to avoid this, the furnace
is operated at a temperature above the usual 1123 K in order to speed up char
combustion kinetics and thereby reduce the char inventory. Thus, the dense bed
temperature is set to 1173 K. For the wood case, the standard dense bed
temperature around 1123 K is set as target.
Besides the above-described operational parameters, the model needs the
fragmentation pattern of each fuel type, the penetration of the secondary air gas jets
and the solids flow through the ash classifier, which are not modeled. For simplicity,
both anthracite and wood are assumed to fall apart in 8 pieces after 40 and 15
seconds respectively (this is arbitrarily chosen, but experiments have shown similar
values). The difference in fragmentation time is due to the fact that coal
fragmentation takes part mainly because of char combustion while wood
fragmentation occurs primarily during devolatilization and thus earlier. The secondary
air injection temperature is set to 493 K and the jet penetration in the furnace is
assumed to be of 0.4 and 0.3 m in the normal and tangential directions to the wall
respectively. A solids flow through the ash classifier of 1 kg/s is assumed.
With this, the model can provide the outputs for both cases considered. A selection
of the modeled outputs is shown in Table 4 and commented below.

Table 4: Modeled values for a selection of key parameters.

Anthracite Wood

Dense bed height m 0.39 0.54

Circulating net solids flux kg/m2·s 33.2 2.6

Mean particle size in seal μm 306 271

Temperature in furnace dense bed K 1173 1111

Temperature in seal K 1166 1143

Removed heat from seal MWth 0.76 0

With the same furnace pressure drop for both cases, a lower dense bed results for
the case with higher fluidization velocity, i.e. the anthracite case. As a consequence
of the higher fluidization velocity, the solids concentration in the freeboard for the
anthracite case is higher than for the wood case, which results in a higher flux of
externally recirculated solids observed in Table 4. Accordingly, since more of the
coarse particles are entrained in the anthracite case, the mean particle size of the
circulating solids, i.e. the mean particle size in the return leg, is larger.
According to the model, for the anthracite case the temperature in the dense bed of
the furnace (where most combustion takes place) can in the Chalmers CFB boiler be
adjusted to the targeted 1173 K by removing 0.76 MWth from the seal, i.e. by cooling

37
 Results

down the externally recirculated solids to 1166 K before these solids are fed back to
the furnace. For the anthracite case, the large char inventory in the unit (despite the
high temperature at which the unit is operated) together with the high circulating
solids flux (including char) makes char combustion in the seal high enough to
consume all oxygen supplied with fluidization gas to the seal. Also in the ash
classifier, all oxygen supplied is consumed.
For the wood case, less combustion takes place in the dense bed and the targeted
operational temperature of 1123 K for this region cannot be reached. A dense bed
temperature of 1111 K is obtained from the model, with no heat removal from the
seal system This, despite the contribution of the externally recirculated solids at
1143 K coming from the seal (from which no heat is removed in this case). For this
case, due to the low char inventory typical for high-volatile fuels, combustion in the
seal and in the classifier is practically negligible.
Figure 24 shows the modeled distribution of oxygen concentration in the furnace for
the two cases. For the anthracite case, the oxygen distribution in the horizontal
direction is almost homogeneous at all heights in the furnace, with a sharp vertical
decrease in the bottom region where most combustion occurs and a concentration
around 3% at the exit height of the furnace. In contrary, the oxygen distribution for
the wood case gives lateral heterogeneities already in the bottom region below the
secondary air injections. This is due to the relatively fast kinetics of the
devolatilization which, for this high-volatile fuel, implies a significant release of volatile
matter concentrated close to the fuel feeding point and thereby resulting in higher
oxygen consumption in this region. For the wood case, the limited amount of oxygen
in the relatively low primary air flow leads to the prediction of a height interval
immediately below the secondary air injections with strong reducing conditions. As
can be seen from Fig. 24b in the horizontal slice cut placed directly above the
secondary air injection, a larger part of the oxygen injected through the front wall is
consumed compared to the rear wall-injected air. With the assumed jet penetration
length of 0.4 m, the central region of the furnace gets oxygen depleted at the air
injection height, although this effect disappears with height due to the horizontal gas
dispersion in the freeboard (mixing of the secondary air further into the core region).
Also, it should be observed that there is an increase in oxygen concentration due to
the gas from the ash classifier injected close to the rear-right corner in the lower part
of the furnace (see Fig. 23 for the exact location). A similar effect occurs with the gas
from the seal, but this cannot be seen in Fig. 24b since this is located in a dead angle
of the figure. With almost no char combustion in the seal or classifier for the wood
case, the flue gas leaving these devices and injected to the riser is rich in oxygen.
For the coal case the model predicts char combustion to occur also in both the seal
and the ash classifier. Thus, injections of the (oxygen-free) flue gas from these
devices into the riser do not represent any contribution to the oxygen flow, as seen in
the horizontal slice in Fig. 24a.
Figure 25 gives the concentration fields for the combustible volatile matter for the
two cases (note the different orders of magnitude of the plotted values). For both
cases, the low Damköhler number for devolatilization in the horizontal direction leads
to high release of volatile matter close to the fuel feeding point, which eventually build
up regions of unburned volatiles (through the mechanism explained in relation to
Fig. 13). For the anthracite case shown in Fig. 25a the negligible amount of volatiles
is directly combusted. For the wood case, due to both the large mass flow of volatile
matter fed into the unit with the fuel and the relatively low oxygen flow provided as

38
 Results

primary air, an important build up of combustible volatile matter occurs in the bottom
region of the furnace, as seen in Fig. 25b. In the same figure, the horizontal
heterogeneities of this volatile build up can also be observed. The higher
concentration of combustible volatiles at the front wall than at the rear wall is the
reason for the different levels of oxygen obtained above the secondary air injections
in Fig. 24b. The model predicts a complete burnout of the volatiles in locations to
which the secondary air jets reach, while some volatile matter remains unburned in
the central part of the furnace. Obviously, results related to the injection of secondary
air are sensible to the assumed penetration length and width of the secondary air, for
which there is a notable lack of knowledge and experimental data. The remaining
volatiles in the center part of the furnace cross section are consumed as gas mixes
laterally on its way up through the freeboard. Note that with low enough values of the
modeled horizontal gas dispersion coefficient, the lateral gas mixing could become
slow enough to allow some of the unburned volatile matter to reach the cyclone.
Finally, the furnace temperature fields for both cases are given in Fig. 26. Solids in
a fluidized bed boiler act as thermal flywheel, with high values of the solids fluxes
improving the homogeneity of the temperature field over the circulating loop. Thus,
with the higher solids fluxes predicted for the anthracite case (see Table 4) the rather
homogeneous temperature field observed for this case in Fig. 26a is consistent. As
expected, the wood case presents heterogeneities in both the axial and lateral
directions, as seen in Fig. 26b. For this case, a resulting dense bed temperature of
1111 K is obtained instead of the targeted temperature of 1123 K (as commented
above), which increases with height in the splash zone mainly due to volatile
combustion induced by the gas mixing in this region. At the air injection height,
combustion of the volatile build up raises the temperature locally, with a predicted
maximum value of 1255 K by the front wall. Concerning the air injection through the
rear wall, excepting for a small region close to the furnace center, there is a balance
between the heat produced by volatile combustion and that needed to heat up the air
from injection temperature to surrounding temperature, and thus the temperature in
the core region close to the rear wall is predicted to remain roughly unchanged. With
the relatively dilute solids suspension in the freeboard for the wood case (mainly due
to the lower primary velocity than in the coal case), heat radiation from the hot
regions in the core to the waterwalls is significant and the temperature of the
ascending solid-gas flow in the core region decreases gradually with height. Note the
large temperature difference between solids downflowing by the front and the rear
walls. This is due to that the downflow in the front wall is fed with solids coming from
the locations with the highest temperatures in the core. From what is seen in
Fig. 26b, an optimization of the process could be achieved by decreasing the air
injection at the rear wall.

39
 Results

C v, O2 [-] C v, O2 [-]

0.212 0.212
10 0.2 10 0.2

9 9
0.18 0.18

8 8
0.16 0.16

7 7
0.14 0.14

6 6
0.12 0.12
z [m]

z [m]

5 5
0.1 0.1
4 4
0.08 0.08
3 3

0.06 0.06
2 2

0.04 0.04
1 1

0.02 0.02

-0.81 -0.81
0 0.71 0 0 0.71 0
0 0
0.81 -0.71 0.81 -0.71
x [m] y [m] x [m] y [m]

a) Anthracite case b) Wood case

Figure 24: Modeled distribution of volumetric concentration in of O2 in the furnace.

40
 Results

C v, vols [-] C v, vols [-]

-3
x 10
2.5 0.08
10 10

9 9 0.07

2
8 8
0.06

7 7

0.05
6 1.5 6
z [m]

z [m]

5 5
0.04

4 4
1
0.03
3 3

2 2 0.02
0.5
1 1
0.01

-0.81 -0.81
0 0.71 0 0.71
0 0 0 0
0.81 -0.71 0.81 -0.71
x [m] y [m] x [m] y [m]

a) Anthracite case b) Wood case

Figure 25: Modeled distribution of volumetric concentration of combustible volatiles in the furnace.

41
 Results

T [K] T [K]
1260 1260
10 10

9 1240 9 1240

8 8
1220 1220
7 7

6 1200 6 1200
z [m]

z [m]

5 5
1180 1180
4 4

3 1160 3 1160

2 2

1140 1140
1 1

0 1120 1120
-0.81 -0.81
0 0.71 1110 0 0.71 1110
0 0
0.81 -0.71 0.81 -0.71
x [m] x [m]
y [m] y [m]

a) Anthracite case b) Wood case

Figure 26: Modeled temperature field in the furnace.

42
Chapter 5 – Further work
Despite that the comprehensive model presented in this work is able to provide a
rather accurate description of the major phenomena governing the operation of CFB
boilers, potential improvements are planned to be implemented in the future
development of the model. This includes also improved understanding of some of the
underlying processes as expressed by the various submodels given in this work. A
selection of possible improvements is listed and briefly commented below.

- In-bed particle size distribution. Currently, the particle size distribution (PSD) of the
bed material is an input which the user of the model has to provide as input (although
the internal solids size segregation is included). In practical operation this is not a
known variable but results from a combination of several phenomena affecting the
material fed to the unit, namely: attrition, bottom ash removal and cyclone separation
efficiency. Proper description and combination of these processes through
submodels will give a resulting in-bed PSD.

- NOx and SOx chemistry. The present model puts its focus on the main combustion
reactions while those concerning sulphur capture and NOx chemistry have not been
included in the present version. Modeling of SOx emissions obviously requires
description of the mixing and kinetics of the sorbent, which should be carried out
analogously to the modeling of the fuel fraction.

- Heat transfer within the gas-solid suspension. While at present only heat transfer
from the gas-solid suspension to the heat transfer surfaces is considered in the
circulating loop, heat transfer within the suspension itself is not accounted for. While
this might not be of any major significance within the different fluid dynamical regions,
heat transfer between these regions (e.g. between the core and annulus regions in
the freeboard) might be significant under certain conditions.
- Lateral temperature gradients in the dense bed. The high axial solids mixing in the
dense bed makes the assumption of perfect thermal mixing in the vertical direction in
the dense bed a reasonable assumption. In relatively large units, however, the lateral
mixing of solids (both inert and fuel fractions), and thereby of heat, is limited which
might create regions of higher temperature in the dense bed close to the fuel feeding
locations. Thus, modeling the solid flows within the dense bed would enable
consideration of the temperature gradients in this region.

Finally, fuel fragmentation is not included in the modeling although models are
available in literature. The complex nature of the fuel fragmentation phenomenon and
its high sensibility on the fuel type and composition make experimental determination
of the main fragmentation patterns a required activity before it is meaningful to
implement a fragmentation submodel in the comprehensive model.

43
44
Chapter 6 – Conclusions
A model for the simulation of the combustion process in large-scale FB units has
been presented, to a large extent based on the submodels and subprocesses given
in the papers of this thesis. These submodels are combined with existing submodels
in literature to a so-called comprehensive model for which the key features are
presented above.
Most of the submodels developed in this work include innovative contents which
enable the consideration of phenomena such as: the changing fluid dynamics of the
fuel fraction as a function of the degree of conversion (including an experimental
study on its mixing patterns), the fluctuating character of the gas flow, the existence
of gas throughflow in the bottom region of the furnace, the influence of the pressure
drop across the gas distributor, the corner effects in rectangular furnaces, the
backflow effect and the separate modeling of radiative and convective heat transfer.
As a result, the modeled data show a good agreement with measurements carried
out in several large-scale CFB boilers (including units which were not used to verify
the submodels developed and used in this work).
Further work to improve the agreement and broaden the model targets is planned.
The model addresses FB combustion, but should also be able to handle other FB
applications such as FB oxy-fuel combustion, chemical looping combustion and FB
gasification.
In addition, a novel method for 2-dimensional particle tracking has been introduced
through an experimental study on the fuel mixing patterns. The mixing has been
investigated with respect to its dependence on main operational parameters and
physical properties of the fuel (tracer particle). A general structure in the flow pattern
of horizontally-aligned vortices induced by the main bubble paths is observed for all
cases. Increased dense bed height or fluidization velocity are found to enhance
mixing both vertically and laterally, while size and density of the tracer particle only
have a minor influence within the range of tracer particles studied (which correspond
to the properties of typical fuel particles). As far as good fluidization is maintained,
lowering the pressure drop across the gas distributor does not have any significant
influence the solids mixing rate, but reduces the gas-solids contact.

45
Notation

Abbreviations
BFB Bubbling fluidized bed
CFB Circulating fluidized bed
FB Fluidized bed

Greek letters
α Absorption coefficient [ - ] η Efficiency [ - ]
δ Bubble volumetric fraction [ - ] ρ Density [kg/m3]
ε Voidage [ - ] σ Stefan-Boltzmann ct. [J/K4m2s]
θ Dimensionless temperature [ - ] τ Residence time [s]
ξ Dimensionless length [ - ] Ω Conversion factor [kg C/mole O2]
λ Heat conductivity [W/m·K]

Subscripts
b Dense bed mf Minimum fluidization
bub Bubble mix Mixing
cl Cluster phase rad radiation
conv Convection riser Riser
cycl Cyclone s Solids
dc Downcomer settled Settled bed
dev Devolatilization slip Slip
disp Disperse phase surf Surface
eff Effective susp Suspension
em Emulsion t Terminal
eq Equivalent th Throughflow
freeb Freeboard vap Vaporization
g Gas 0 Primary
h Horizontal ∞ Surrounding
kin Kinetic

46
Variables
A Area [m2] c Constant coefficient [ - ]
a Decay constant [m-1] D Dispersion coefficient [m2/s]
C Concentration [kg/m3] d Particle size [m]
F Flow [kg/s] p Pressure [Pa]
f Empirical function pent Entrainment probability [ - ]
G Flux [kg/m2s] q Heat flow [W]
2
g Gravity acceleration [m/s ] R Rate [m/s]
2
h Convective coefficient [W/m K] r Release rate [kg/s]
h Height [m] S Source term [kg/m2s]
h Enthalpy [J] T Temperature [K]
-1
K Decay constant [m ] t Thickness [m]
k Mass transfer coefficient [m/s] u Velocity [m/s]
L Length [m] w Radiation cell coefficient [ - ]
m Mass [kg]

47
References
Adánez, J., Gayán, P., García-Labiano, F., de Diego, L.F. 1994. ”Axial voidage
profiles in fast fluidized beds”. Powder Technology, 81(3), pp.259-268.
Baskakov, A.P., Leckner, B. 1997. ”Radiative heat transfer in circulating fluidized bed
furnaces”. Powder Technology, 90(3), pp.213-218.
Breitholtz, C., Leckner, B., Baskakov, A.P. 2001. ”Wall average heat transfer in CFB
boilers”. Powder Technology, 120(1-2), pp.41-48.
Chirone, R., Massimilla, L., Salatino, P. 1991. “Comminution of carbons in fluidized
bed combustion”. Progress in Energy and Combustion Science, 17 (4), pp.297-
326.
Clift, R., Grace, J.-R. 1985. “Fluidization”. Davidson, Harrison and Clift Eds.,
Academic Press Inc., London (UK).
Darton, R., LaNauze, R., Davidson, J., Harrison, D. 1977. “Bubble-growth due to
coalescence in fluidized beds”. Transactions of the Institution of Chemical
Engineers, 55, pp.274-280.
Field, M.A., Gill, D.W., Morgan, B.B., Hawksley, P.G.W. 1967. “Combustion of
pulverized coal”. The British Coal Utilization Research Association, Surrey
(England).
Haider, A,, Levenspiel, O. 1989. ”Drag coefficient and terminal velocity of spherical
and nonspherical particles”. Powder Technology, 58(1), pp.63-70.
Hannes, J.-P. 1996. ”Mathematical modelling of circulating fluidized bed combustion”.
Ph.D. thesis, Delft University of Technology. City-Print Verlag GmbH, Aachen
(Germany). ISBN 3-88817-002-8.
Hiller, R. 1995. ”Mathematische Modellierung der kohleverbrennung in einer
circofluid-wirbelschichtfeuerungen”. Ph.D. thesis, University of Dortmund. Shaker
Verlag, Dortmund, (Germany).
Hupa, M. 2005. Lecture ”Fluidized bed combustion” in lecture series ”Options for
energy recovery from municipal solid waste” at Politecnico di Milano.
Johnsson, F., Andersson, S., Leckner, B. 1991. ”Expansion of a freely bubbling
fluidized bed”. Powder Technology, 68(2), pp.117-23.
Johnsson, F., Leckner, B, 1995. ”Vertical Distribution of Solids in a CFB-Furnace”.
Proc. of the 13th Int. Conf. on Fluidized Bed Combustion, pp.671-679.
Kruse, M., Werther, J. 1995. ”2D gas and solids flow prediction in circulating fluidized
beds based on suction probe and pressure profile measurements”. Chemical
engineering and processing, 34 (3), pp.185-203.
Kruse, M., Schoenfelder, H., Werther, J. 1995. ”Two-dimensional model for gas
mixing in the upper dilute zone of a circulating fluidized bed”. Canadian Journal of
Chemical Engineering, 73(5), pp.620-634.
Kunii, D., Levenspiel, O. 1991. ”Fluidization Engineering”. Butterworth-Heinemann
Eds., Boston (USA). ISBN 978-0409902334.

48
Leckner, B., Golriz, M.R., Zhang, W., Andersson, B.Å., Johnsson, F. 1991. “Boundary
layers - first measurements in the 12 MW CFB research plant at Chalmers
University”. Proc. of the 11th Int. Conf. on FBC (Montreal, Canada), pp. 771–776.
Lee, J.M., Kim, J.S., Kim, J.J. 2003. “Comminution characteristics of Korean
anthracite in a CFB reactor”. Fuel, 82, pp. 1349–1357.
Lyngfelt, A., Åmand, L.-E., Leckner, B. 1996. ”Progress of combustion in the furnace
of a circulating fluidized bed boiler”. Proc. of the 26th Int. Symp. on Combustion
(Naples, Italy), pp.3253-3259.
Lücke, K. 2003. ”On the influence of mixing on the performance of large-scale
atmospheric circulating fluidized bed combustors”. Ph.D. thesis, Technical
University of Hamburg-Harburg. ISBN 3-8322-1288-4.
Muschelknautz, U., Muschelknautz, E. 1991. ”Special design of inserts and short
entrance ducts to recirculating cyclones”. Proc. of the 4th Int. Conf. on Circulating
Fluidized Beds (Pittsburgh, USA), pp.597-602.
Niklasson, F., Thunman, H., Johnsson, F., Leckner, B. 2002. “Estimation of solids
mixing in a fluidized bed combustor”. Industrial and Engineering Chemistry
Research, 41 (18), pp.4663-4673.
Niklasson, F., Johnsson, F., Leckner, B. 2003. “Local air ratio measured by zirconia
cell in a circulating fluidised bed furnace”. Chemical Engineering Journal, 96(1-3),
pp.145-155.
Niklasson, F., Pallarès, D., Johnsson, F. 2007. “Biomass co-firing in a CFB boiler –
The influences of fuel and bed properties on in-furnace gas-concentration
profiles”. Proc. of the 19th Int. Conf. on Fluidized Bed Combustion (Vienna,
Austria).
Rhodes, M.J., Geldart, D. 1987. ”A model for the circulating fluidized bed”. Powder
Technology, 53(3), pp.155-162.
Salatino, P., Massimilla, L. 1989. “A predictive model of carbon attrition in fluidized
bed combustion and gasification of a graphite”. Chemical Engineering Science, 44
(5), pp.1091-1099.
Sasic, S., Johnsson, F., Leckner, B. 2004. ”Interaction between a fluidized bed and
its air-supply system: some observations”. Industrial and Engineering Chemistry
Research, 43(18), pp.5730-5737.
Svensson, A., Johnsson, F., Leckner, B. 1996. ”Bottom bed regimes in a circulating
fluidized bed boiler”. International Journal of Multiphase Flow, 22(6), pp.1187-
1204.
Thunman, H., Niklasson, F., Johnsson, F., Leckner, B. 2001 ” Composition of volatile
gases and thermochemical properties of wood for modeling of fixed or fluidized
beds”. Energy and Fuels,15 (6), pp.1488-1497.
Thunman, H., Davidsson, K., Leckner, B. 2004. “Separation of drying and
devolatilization during conversion of solid fuels”. Combustion and flame, 137 (1),
pp.242-250.

49
Thunman, H. 2005. Personal communication.
Werdermann, C.C. 1993. ”Feststoffbewegung und wärmeübergang in zirkulierenden
wirbelschichten von kohlekraftwerken”. Ph.D. thesis, Technical University of
Hamburg-Harburg. Verlag Shaker, Aachen (Germany). ISBN 3-86111-327-9.
Werther, J. 1993. ”Fluid mechanics of large-scale CFB units”. Proc. of the 4th Int.
Conf. on Circulating Fluidized Bed Technology, pp.1-14.
Xiang, Q., Huang, G., Ni, M., Cen, K., Tao, T. 1987. “Lateral dispersion of large coal
particles in an industrial-scale fluidized bed combustor”. Proc. of the 9th Int. Conf.
on Fluidized Bed Combustion (Boston, USA). pp.546-553.
Zhang, W., Johnsson, F., Leckner, B. 1995. ”Fluid-dynamic boundary layers in CFB
boilers”. Chemical Engineering Science, 50(2), pp.201-10.
Zhang, W., Leckner, B. 2001. ”Process simulation of circulating fluidized beds with
combustion/gasification of biomass”. Final report of project JOR3CT980306, Non-
Nuclear Energy Programme JOULE III, 4th EU framework.
Åmand, L.-E., Lyngfelt, A., Karlsson, M., Leckner, B. 1995. ”The reference case for
coal at the 12 MW CFB-Boiler at Chalmers. Characterisation of the gas and
particle phases in the combustion chamber”. Internal report A97-221, Dept. of
Energy Conversion, Chalmers Univeristy of Technology, Göteborg (Sweden).

50
 Paper I

Pallarès, D., Johnsson, F. (2006)

”Macroscopic modelling of fluid dynamics in
large-scale circulating fluidized beds”
Progress in Energy and Combustion Science, 32 (5-6), pp. 539-569
 ARTICLE IN PRESS

Progress in Energy and Combustion Science 32 (2006) 539–569

www.elsevier.com/locate/pecs

Review

Macroscopic modelling of fluid dynamics in large-scale

circulating fluidized beds
David Pallarès, Filip Johnsson
Department of Energy and Environment, Energy Conversion, Chalmers University of Technology, SE 412 96 Göteborg, Sweden
Received 31 May 2005; accepted 13 February 2006

Abstract
Macroscopic (semi-empirical) models for fluid dynamics of circulating fluidized bed (CFB) units are presented, with
emphasize on applications for conditions relevant to industrial units such as fluidized-bed combustors. In order to make a
structured analysis of the models, the CFB unit is divided into 6 fluid dynamical zones, which have been shown to exhibit
different fluid-dynamical behaviour (bottom bed, freeboard, exit zone, exit duct, cyclone and downcomer and particle
seal). The paper summarizes the main basis and assumptions for each model together with major advantages and
drawbacks. In addition, a practical example on how a selected set of these local models can be linked to an overall model of
the fluid dynamics of the entire CFB loop is presented. It is shown that it is possible to reach good agreement between the
overall model and experimental data from industrial units.
r 2006 Elsevier Ltd. All rights reserved.

Keywords: Circulating fluidized bed; Fluid dynamics; Modelling

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 540
1.1. Scope: large-scale CFB units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 540
1.2. Modeling of fluid dynamics in large-scale CFB units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543
1.3. The use of experimental data to verify models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544
2. Modeling of local CFB zones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544
2.1. Modeling of the bottom bed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545
2.2. Modeling of the freeboard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
2.2.1. Terminal velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
2.2.2. Wall-layer thickness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 551
2.2.3. Gas flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552
2.2.4. Solids concentration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552
2.2.5. Solid flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554

Abbreviations: BFB, bubbling fluidized bed; CFB, circulating fluidized bed; CFBC, circulating fluidized bed combustor; CFBG,
circulating fluidized bed gasifier; CFD, computational fluid dynamics; FCC, fluid catalytic cracking; HDCFB, high-density circulating
fluidized bed; LDCFB, low-density circulating fluidized bed; SFB, stationary fluidized bed; PSD, particle size distribution
Corresponding author. Tel.: +46 31 772 1449; fax: +46 31 722 3592.
E-mail address: fi[email protected] (F. Johnsson).

0360-1285/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.pecs.2006.02.002
 ARTICLE IN PRESS
540 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

2.3. Modeling of the exit zone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555

2.3.1. Backflow effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 556
2.3.2. Solids concentration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557
2.4. Modeling of the exit duct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557
2.5. Modeling of the cyclone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 558
2.6. Modeling of the downcomer and particle seal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 559
2.6.1. Pressure balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 560
3. Overall modeling of a CFB unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 561
3.1. Practical example of an overall CFB model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 562
4. Closure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 565
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566

1. Introduction in tall and narrow units, which together with the

operational conditions applied in these units, make
As concluded by for instance Werther [1], the results not applicable to CFB combustors.
circulating fluidized bed (CFB) units used in During the last decades there has been an increase
experimental investigations on CFB fluid dynamics in the use of large CFB units for energy production,
given in literature can be divided into two types, since the CFB technique has proved to be efficient,
depending on the aspect ratio (H0/Deq) of the riser clean and with a high flexibility concerning type of
fuel (e.g. wood, coal, municipal waste and mixtures
(1) Tall and narrow units, with an aspect ratio of of these), see e.g. [8]. Therefore, the development of
the order of 20 or higher. models to simulate and predict the behaviour of
(2) Large units, with a lower aspect ratio typically such units is needed for design and scale-up
less than 10. purposes. Prediction of the flow is crucial in
modeling heat transfer and combustion (mixing of
The first type is used in investigations directed fuel and combustion/gasification gas). The flow
towards chemical engineering applications (such as pattern in CFB gasifiers and combustors is similar
fluid catalytic cracking, FCC) and typically operates since the overall riser geometry, fluidization condi-
at high solids net fluxes. The latter (here called large tions and properties of the solids used are similar.
CFB units) is the scope of this review and Thus, the same models should be able to describe
corresponds to industrial-scale units with the the flow in both applications. Concerning combus-
purpose of producing heat, electricity, fuel-gas or tion, circulating fluidized bed combustors (CFBC)
combinations of these. For an introduction and have been built up to sizes of around 300 MWel (e.g.
overview of such fluidized bed systems see reviews the Gardanne and Turow plants, as described in
by Anthony [2], and Leckner [3] for combustion [9,10], respectively), while gasifiers (CFBG) are still
systems and Longwell et al. [4] for a general review at a more developing stage and are about to reach
including both combustion and gasification systems values of 100 MWth [11]. Combustors generally have
(not only fluidized-bed systems). a square or rectangular cross section and may partly
Issangya et al. [5] give an overview of experimental have bare membrane tube walls, whereas the riser of
investigations on CFB fluid dynamics concluding a gasifier normally has a circular cross section with
that ‘‘almost all reported CFB data are for low- plane (isothermal) walls. Except for the presence of
density circulating fluidized bed (LDCFB) systems corner effects in the former type of units, the overall
representative of CFB combustors’’ and therefore influence from the shape of cross section on the
not applicable to the high-density fluidized beds solids flow should be small, provided the riser has
(HDCFB) which are the scope of their work. The the characteristics listed in the next section.
authors of this paper agree that the data to which
Issangya et al. are referring are from investigations 1.1. Scope: large-scale CFB units
which may very well have had CFB combustors
as application. However, as discussed previously The first large-scale fluidized bed units were
(cf. [1,6,7]) much of this data are from experiments operated as stationary fluidized beds (SFB), i.e. at
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 541

Nomenclature Superscripts

a decay factor of cluster phase (m
1) d PSD size interval index

A area (m2) f solid fraction index
A0 gas-distributor area per nozzle (m2)
b entrance width of exit duct (m) Subscripts
C solids concentration (kg/m3)
b bottom bed
C1, C2 empirical constants
bub bubble
Ccycl empirical coefficient in cyclone model
bubN single bubble
CD drag-force coefficient
clust cluster phase
d particle diameter (m)
core core region
D diameter (m)
cycl cyclone
f empirical function
dc downcomer
F solids mass flow (kg/s)
disp disperse phase
G solids mass flux (kg/m2 s)
duct exit duct
h height (m)
emul emulsion phase
H height (m)
eq equivalent
H0 riser height (m)
g gas
k backflow ratio
i cell index
K decay factor of disperse phase (m
1)
lat net lateral flow from core-to annulus
L length (m)
mb minimum bubbling conditions
m mass (kg)
mf minimum fluidization conditions
p pressure (Pa)
n surrounding solids suspension
p0 (Dp0) riser pressure drop (Pa)
net net
pentr particle entrainment probability
p particle
t thickness (m)
ref riser height interval from 0.135 to
u velocity (m/s)
1.635 m above gas distributor
u0 superficial fluidization gas
refeed riser interval between return leg and exit
velocity (m/s)
rel relative
V_ gas volumetric flow (m3/s)
retleg return leg
w cloud settling velocity (m/s)
s solids
x horizontal coordinate (m)
seal particle seal
settled settled (non-fluidized) bed
Greek symbols
slip slip
sat saturation value
g solids loading ratio at exit-duct inlet
t terminal
d volumetric fraction
tf throughflow
e voidage
vis visible bubble flow
f sphericity
wall wall-layer (annulus) region
Z separation efficiency
m dynamic viscosity (kg/ms)
Dimensionless groups
n rise velocity (m/s)
r density (kg/m3) Ar Archimedes number
c dimensionless visible bubble flow Re Reynolds number

fluidization velocities well below the terminal maintained (stationary) as long as solid which are
velocity of the bed material resulting in that the discharged from the unit (e.g. ash discharge) is
amount of solids entrained out from the riser can be replaced with fresh bed solids (fuel and/or inert
neglected. Thus, there is no need for external solids). This type of bed is also known as bubbling
recirculation of bed solids and the bed can be fluidized bed (BFB). Under bubbling conditions,
 ARTICLE IN PRESS
542 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

bubbles rise through the bed and when erupting at regime is established (i.e. when the transport
the bed surface solids are thrown into the freeboard velocity is reached will mainly depend on the PSD
creating a so-called splash zone above the (time of the bed material, the fluidization velocity and the
averaged) bottom bed surface. There is a pro- amount of bed material), such a mapping is not
nounced decay in solids concentration up through known to the authors. In addition, the transport
the splash zone. velocity should also depend on the design of the
The CFB employs a higher gas velocity and/or CFB loop such as the height of the riser and the way
finer bed solids than used in an SFB. More the recirculation side of the unit is operated (e.g. to
specifically, the gas velocity exceeds the terminal what degree the loop seal is fluidized), i.e. this
velocity of all or a significant part of the bed makes it difficult to generalize the transport velocity
material, resulting in that solids are entrained out of only based on the above listed fluid dynamic
the riser. To maintain the bed solids, the entrained parameters.
solids must be recirculated back to the bed. This is The present modeling review is restricted to large
done in the return leg, which consists of solids CFB units working at atmospheric pressure with a
separation device (normally a cyclone), a dipleg and dense bottom bed present. Characteristics of such
a particle seal connected to the riser with a duct units are:
pipe. The particle seal is used to maintain the
pressure balance that prevents gas and solids from  A height to diameter (aspect) ratio of the riser
entering the return device. The particle seal is (H0/Deq) of the order of or less than 10.
normally fluidized so that it forms a low-velocity  A ratio of settled bed height (the bed formed if
SFB.1 The fluidization air distributor of a CFB has the solids are not fluidized) to riser diameter of
a characteristic pressure–velocity curve designed for less than 1 (Hb,settled/Deqo1).
operation at higher gas velocities, thus providing a  Fluidized solids belonging to group B in the
lower pressure drop than an SFB air distributor Geldart classification (a brief summary of this
operating at the same gas flow rate. The bottom bed classification is given below).
of a CFB is far more turbulent than in an SFB, with  A solids net flux typically ranging from 0.5 to
jet-formed gas voids (here called ‘‘exploding bub- 20 kg/m2 s.
bles’’) which can extend the whole way from the air
distributor up to the bottom bed surface. As a Primary operational parameters of the riser (with
result, there are occasions of by-pass (through flow) respect to fluid dynamics) are the riser pressure drop
of gas (or combustion air) during eruption of these and the gas flows (i.e. fluidization velocity, second-
voids, facilitating a low-pressure path for the gas. ary gas injection), whereas the solids net flux and
The term exploding bubble was originally proposed the total inventory of solids are normally not
by Fitzgerald [12] for bubbling beds but it was later known. The above-listed characteristics were shown
found that the characteristics are similar in the [1,6,7] to give a flow pattern different from that of
bottom bed of a CFB (e.g. [13]). Consequently, a the well-investigated tall and narrow laboratory
splash zone is present above the bed followed by a units directed towards chemical engineering appli-
so-called transport zone with a dilute solids suspen- cations. The latter type of units have a higher aspect
sion that extends all the way up to the riser outlet. ratio (H0/Deq typically 20) and are normally run at
Solids concentration and solids size decrease with a much higher solids net flux (Gs,net 50 kg/m2 s),
height in this zone, and at the top of the riser the i.e. are operated with finer (and often lighter) solids
mean particle size might differ strongly from that in (ds typicallyo100 mm) than CFBC and CFBG units.
the bottom bed due to solids size segregation. With The low aspect ratio of the riser in large CFB
a further increase in fluidization velocity (or low- units results in a solids-flow profile developing up
ering the total amount of solids in the CFB loop) through the freeboard (above the bottom bed),
the bottom bed disappears and the bed enters the i.e. the riser can be seen as an entrance zone with
pneumatic transport regime. Despite the importance respect to the flow (both solids and gas). This gives a
of determining the conditions under which this solids flux profile, which is fairly flat across the core
region, but with pronounced wall layers formed by
1
Instead of a particle seal a so-called L-valve could be used. the solids backmixing at the riser walls [1,7]. Thus, a
Since the L-valve does not contain a fluidized bed it is difficult to core/wall-layer structure is present. On the other
combine with an external particle cooler. hand, the above-mentioned tall and narrow risers
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 543

exhibit a more developed solids flux profile, bed. In addition, the gas flow becomes highly
typically with a parabolic shape [14,15] depending intermittent.
on operational conditions. Although varying with In laboratory risers operated with Group A
the solids net flux and the fluidization velocity, these solids, an increase in velocity from the slugging
risers also show a more or less pronounced back- conditions results in a transition to a turbulent
mixing at the riser walls forming a core-annular regime [22,23]. With Group B solids in this type of
structure of the flow, but at high enough gas unit, the picture is less clear, but it was argued that
velocities there may even be up-flow of solids an increase in velocity results in that transport
throughout the cross section (e.g. [16]). conditions are reached [23], i.e. it is difficult or not
The low Hb,fixed/Deq ratio (o1) in large CFB units possible to maintain a dense zone in the bottom
yields a non-slugging bed [13,17], whereas tall and region of narrow risers operated with Group B
narrow risers (with Hb,fixed/Deq41) give a slugging solids under circulating conditions [18].
bed [18]. As indicated above, the large CFB units In summary, the flow pattern of a large CFB unit
considered in this review are operated with Geldart differs significantly from that of tall and narrow
B solids according to the classification given by laboratory CFB risers, so the abundant literature on
Geldart which is a classification of particles in four the latter type of CFB units is seldom applicable for
groups according to their fluidizing behaviour [19] CFBC or CFBG.
(groups C, A, B and D in increasing size and density
order). The classification is based on empirical 1.2. Modeling of fluid dynamics in large-scale CFB
observations and Geldart gave the classification in units
a (rs
rg) vs. dp diagram which was later developed
by Grace [20] who draw the four groups in a Modeling of large CFB units can be divided into
dimensionless ug vs. dp diagram identifying different three main fields: fluid dynamics, chemistry (includ-
fluidization regimes. The Group B solids normally ing formation of pollutant emissions), and heat
used in large CFB units for combustion and balance (heat transfer). The relations between these
gasification consists of silica sand and/or primary three modeling fields are outlined in Fig. 1, where
ashes, and sorbent in the case of coal-fired units. the thickness of the arrows corresponds to the
However, depending on size distribution the bed sensitivity of the modeling field to the input
solids also consists of certain fractions of Group A represented by the arrow. As seen, all of the three
(fines generated by attrition) and Group D (rela- modeling fields need some input parameters that are
tively large fuel particles) solids. As will be shown outputs from the other modeling fields. Modeling of
below, the particle size distribution (PSD) is an the chemistry and the heat balance are sensitive to
essential feature of CFB units and must be included the values of the fluid dynamical parameters used as
in any model in order to correctly describe the inputs. On the other hand, concerning fluid
recirculation of bed solids as well as back-mixing of dynamics, most of the inputs needed are external
solids in the riser. Yet, when averaged quantities are operational parameters (known and adjustable
to be calculated, the mean volume-to-surface under operation), whereas only a few are related
diameter (also known as Sauter diameter) of the to heat balance and chemistry (temperature profiles
bed solids and the volume-averaged density are and internal gas generation, respectively). Under
usually used. normal operating conditions, the fluid dynamics are
Under CFB conditions with Geldart group B not especially sensitive to the latter type, i.e. fluid
solids, a low Hb/Deq ratio in combination with a low dynamics can be reasonably well modelled by
primary air-distributor pressure drop results in that assuming typical values of those inputs. Despite
a dense bubbling bed can be maintained also at high the fluctuations observed in most fluidized-bed
velocities with bubbles of a so-called exploding parameters even during steady-state operation, a
character [21]. Such flow results in large fluctuations time-averaged modeling approach is usually chosen
in the overall gas flow with a high throughflow of to simplify the analysis. With respect to modeling
gas in the bubbles, leading to high local gas CFB fluid dynamics, two types of models can be
velocities. Thus, the exchange of gas between the identified: macroscopic models (based on empirical
bubbles and the emulsion phase is low in relation to and semi-empirical expressions) and models from
the gas flow through the bed. In combustors, this first principles using computational fluid dynamics
results in strongly reducing conditions in the bottom (CFD). At present, only the former type of models
 ARTICLE IN PRESS
544 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

Fig. 1. Relationships between CFB modeling fields.

is practically applicable if aiming at both reasonable (not so much of gasifiers). In such large units—
calculation times and agreement. Although the typically 100 MW or more—there is for obvious
formulation of some critical terms in CFD modeling (practical) reasons little detailed experimental in-
is still at a developing stage and computational formation on cross-sectional distribution of entities
times of the simulations are long, simulations from such as solids concentrations and gas velocities. The
first-principle models will probably be the future aim of this work is to gather all available informa-
modelling tool. Yet, inclusion of chemistry and heat tion from large units and then use these data to
transfer modelling with fluidized-bed two-phase compare modelling results in the way that, when
flow in 3D CFD simulation will most likely require possible, models are applied to units that were not
significant development before reliable results can included in the development of the models.
be produced within reasonable calculation times. It should be mentioned that also cold models not
Thus, it can be assumed that there is a need for operated according to scaling laws can be of great
comprehensive semiempirical modelling of fluidized use when studying principal relationships between
bed combustors and gasifiers—the topic of this important parameters. Such experiments can be
paper—over at least the next decade. efficient when scanning over a broad range of
operational parameters (e.g. from bubbling condi-
1.3. The use of experimental data to verify models tions at low velocity to transport conditions). Then,
of course, only main trends are obtained and
The focus of this review is macroscopic modelling quantitative analyses will require full scale measure-
of CFBs, i.e. the modelling is of a semiempirical ments or measurements in a scaled down unit
type. An obvious drawback with semiempirical operated according to the above-mentioned scaling
models is that, since they are partly based on relationships.
empirical data, it may be difficult to apply these In summary, it is important to base macroscopi-
under conditions different from those for which cal semi-empirical modeling on robust and high-
they were developed. The experimental data shown quality experimental data obtained under condi-
in this review is measured in full-scale units or cold tions relevant for industrial units.
models operated according to scaling relationships
(e.g. such given by Glicksman [24]), i.e. transform- 2. Modeling of local CFB zones
ing industrial conditions to ambient conditions. As
indicated from the characteristics listed above, large In order to organize the present review, the CFB
cross sections are a key feature of CFB combustors loop is divided into six different zones, which, from
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 545

previous experiments, have been shown to exhibit constantly fluidized as well as some solids will be
different flow patterns and it is consequently present also in the bubble phase, but the above
assumed that these can be modelled separately assumption should be reasonable for the solids
(but linked by input/outputs between the zones). typically used in CFBCs and CFBGs, strongly
Therefore, each zone is analysed separately and this simplifying the analysis.
paper first presents the fluid dynamical models of Fluidization of the bed material occurs when the
each zone. The division in the different zones is fluidization gas velocity, u0, exceeds the minimum
shown in Fig. 2, i.e. the CFB unit is divided into a fluidization velocity, umf. Several expressions exist to
bottom bed, freeboard, exit zone, exit duct, cyclone calculate the Reynolds number at which fluidization
and downcomer and particle seal. is initiated. A widely used correlation is given by
This paper takes a selected set of models for each Wen and Yu [25]:
of the six zones and couples these in order to obtain qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
an overall model for the entire CFB unit therefore Rep;mf ¼ C 21 þ C 2 Ar
C 1 , (1)
making it possible to close the solids mass balance
where
using solids with a size distribution. This coupling
procedure must account for interactions between d 3p rg ðrs
rg Þg
the zones (i.e. the zones are not independent in the Ar ¼ . (2)
m2g
overall model) and is described in Section 3.
Grace [26] tabulated several pairs of values for C1
2.1. Modeling of the bottom bed and C2 proposed in literature, and suggested the
use of C 1 ¼ 27:2 and C 2 ¼ 0:0408. Thus, umf can
In the lower part of a CFB unit, a bottom bed is easily be obtained through the Reynolds number
formed provided enough bed material is present in expression:
the CFB loop. A CFB bottom bed consist of two Rep;mf mg
phases: a dense (or emulsion) phase, formed by the umf ¼ . (3)
d p rg
bed particles and the interstitial gas flow between
the particles, and a bubble phase, consisting of And the voidage under minimum fluidization
uprising gas bubbles assumed to be free of solids. In conditions is calculated by solving the equation
a strict sense, not all particles in the dense phase are derived from the Ergun equation [27]:
1:75 150ð1

mf Þ
Re2 þ Rep;mf ¼ Ar. (4)

3mf fs p;mf
2mf f2s
Finally, once fluidized, the pressure drop over a
vertical column of solids is obtained through:
Dp ¼ ð1

Þðrs
rg Þg H. (5)
A more detailed description of the pressure drop
under homogenous (bubble-free) fluidization and
for velocities below umf is given in [28,29]. However,
for the solids of interest in the present modeling
(Group B), there is no or little homogenous
fluidization, i.e. it can be assumed that umf ¼ umb .
Gas bubbles, which are the main responsible for
solids mixing in the bottom bed, appear in the bed
when the fluidization velocity exceeds the minimum
bubbling velocity, umb. Abrahamsen and Geldart
[30] gave a correlation to determine the ratio
umb/umf, which gives values close to 1 for particle
belonging to Group B (typical for large CFB units)
and Group D in the Geldart classification (see [19]).
Characterization of the properties of the gas
Fig. 2. The zone division used in the present model. bubbles is usually carried out by the set of
 ARTICLE IN PRESS
546 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

expressions briefly described below. Darton et al. expansion curve) differs between the two types of
[31] presented a model for determining the bubble units. Thus, in an intermediate range of primary-air
size based on the assumption that bubbles only velocities, corresponding to operation of an SFB at
grow due to coalescence, which is assumed to occur full load or a CFB at part-load, the rate of bed
after a bubble has travelled l times its own diameter expansion differs between the two types of units.
(l ¼ 1:17 in [31]), yielding: The bed expansion curve (bed voidage vs. superficial
pffiffiffiffiffiffi gas velocity) of a CFB unit [39] scatters more than
Dbub ¼ 0:54ðu0
umf Þ0:4 ðh þ 4 A0 Þ0:8 g
0:2 . (6) the corresponding curve for an SFB [17]. This is
Clift and Grace [32] summarized experimental believed to be due to that the lower air-distributor
data from several authors and correlated the rise pressure drop of the CFB results in stronger
velocity of an isolated bubble in a large bubbling interaction of the bed and the air-feed system (air-
fluidized bed: plenum, piping and fans) than in an SFB. As the
pffiffiffiffiffiffiffiffiffiffiffiffiffi velocity is increased, the CFB system reconfigures
nbub1 ¼ 0:71 g Dbub . (7) itself into a ‘‘new’’ system having a different
The above-given velocity is often corrected by a interaction between the bed and the air-feed system
term taking into account the presence of other (and the increase in bed expansion therefore does
bubbles and the increase in relative velocity between not follow a continuous curve). The influence of air-
the rising bubble and the downflow of surrounding distributor pressure drop on fluidization behaviour
solids. Davidson and Harrison [33] give the follow- is discussed in [21] and the interaction between the
ing expression for the bubble velocity: bottom bed dynamics (pressure fluctuations) and
the air-feed system of a CFB is given in [41–44].
nbub ¼ uvis þ nbub1 . (8)
The bottom bed is defined as the extension of the
However, the use of this expression is somewhat linear part of the time-averaged vertical pressure-
controversial and has been questioned in [34,35] and drop profile above the primary-air distributor.
experimental bubble rise velocities have been Thus,
successfully fitted to nbub,N instead of nbub [36]. 
dp
Since the above-mentioned expressions for esti- ¼ const. (9)
mation of bubble properties were obtained from dh b
experimental data measured at low fluidization A linear pressure drop was confirmed under
velocities, care should be taken when applying these circulating [13] as well as under bubbling conditions
expressions to CFB modeling, where fluidization [17] in units having the characteristics listed in
velocities are much higher. Section 1.1. The expression for the time-averaged
For the non-slugging beds treated in this work, pressure drop is given by
two types of behaviour were observed with respect 
to the time-averaged bottom bed voidage. The dp
¼
csb g ¼
rsb ð1

b Þg. (10)
difference is related to the type of fluidized bed dh b
studied: stationary (SFB) or circulating (CFB). Assumption of homogeneous mixing in the
Most studies in literature on the bottom bed region bottom bed does not allow a jetsam/flotsam
concern SFB e.g. [37,38], while there is little detailed segregation effect of the inert solids, provided the
data on CFB bottom bed fluid dynamics e.g. [39,40]. PSD is not too wide (see [45] for details on this
SFB units are characterized by operating at lower effect). Thus, the averaged solids density within the
fluidization velocities than CFB units, resulting in bottom bed, rsb , is generally assumed not to depend
zero or very low net solids flux. CFB units have a on height above the air distributor. Considering
lower pressure drop across the air-distributor than Eqs. (9) and (10), and assuming homogeneous
SFB at corresponding velocities. This is simply due mixing in the bottom bed, it can be concluded that
to the fact that CFBs are operated at higher
velocities than SFBs, but the affordable pressure d
b
¼ 0. (11)
drop is limited. As a consequence, CFBs cannot be dh
operated under stable conditions (even fluidization) With dbub being the bubble fraction, the bottom
below a certain velocity. Although it follows the bed voidage can be written as
same principal behaviour, the relation between bed
voidage and fluidization velocity (i.e. the bed
b ¼ dbub þ ð1
dbub Þ
emul . (12)
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 547

In terms of superficial gas velocities, the total Furthermore, the arrangement of the air feeding
fluidization gas flow supplied to the bottom bed, u0, system was shown to strongly influence the bottom
was originally divided into two terms by Toomey bed dynamics at low air distributor pressure drops
and Johnstone [46]: the flow in the emulsion phase, [44]. The semi-empirical expressions (Eqs. (18)–(20))
uemul, and the visible bubble flow, uvis. Toomey and given below have been obtained from experimental
Johnstone defined the flow in the emulsion phase as data sampled under single bubble and exploding
that required to fluidize the bed material, i.e. uemul ¼ bubble regimes, which are typical for the bottom
umf , while they assumed all excess gas supplied to bed of a CFB unit at low and high velocities,
pass the bed as bubbles, i.e. uvis ¼ u0
umf . It was respectively. However, it should be noted that the
later proven that the two-phase flow theory over- data on which these expressions are based come
estimate the bed expansion and, as a consequence, from certain test units with particular air-distributor
the presence of a third term in the flow balance, the characteristic curves and air-feeding arrangements
so-called throughflow, utf, was confirmed by several and, thus, data from other units, yet running under
authors [47,48]. The throughflow is gas which flows the same bottom bed regime, might differ from the
mainly through the low-resistance gas passages in expressions reported below.
and between the bubbles. Thus, the gas flow in the In summary, the excess gas, u0
umf, is divided
bottom bed can be expressed by a modified two- between uvis and utf. The division between these
phase flow model, which in terms of superficial gas flows can be given as a dimensionless visible bubble
velocities is written flow, c:
u0 ¼ umf þ uvis þ utf , (13) uvis
c¼ (15)
u0
uemul
where the visible bubble flow is
which expresses the deviation from the original two-
uvis ¼ dbub nbub . (14) phase flow model. Grace and Clift [47] presented a
Measurements of the bubble fraction were carried large collection of experimental values of c (ranging
out in a 2-dimensional bed by means of digital from 0.06 to 1) under different fluidization condi-
image analysis by Shen et al. [36]. These measure- tions, indicating that the through flow strongly
ments show that with an increase in the supplied gas depends on operating conditions as well as the
flow, u0, a decreasing proportion of the gas passes configuration of the fluidized bed and air feeding
the bed as visible bubble flow, uvis. Thus, the system (e.g. most likely air distributor pressure
proportion of throughflow of the total gas flow drop).
supplied increases with an increase in fluidization Werther and Wein [51] estimated c using a semi-
velocity. However, despite the importance of this empirical model for the bottom bed (later extended
flow term, it remains unclear in which proportions by Zijerveld et al. [52])
the throughflow distributes itself between the c ¼ 1:45 Ar
0:18 . (16)
emulsion and bubble phase, although this issue
However, outputs from this model result in a
has been experimentally investigated in [32,49,50].
height-dependence in bottom bed voidage, which, as
These three works conclude that the throughflow
discussed above, is in disagreement with measure-
must be taken into account already at fluidization
ments (of time-averaged pressure drop).
velocities just above the minimum fluidization
A semi-empirical modeling of c was proposed by
velocity, leading to a higher interstitial velocity in
Johnsson et al. [17], for which
the vicinity of bubbles than in bubble-free zones of
pffiffiffiffiffiffi
the emulsion. In addition, the influence of the c ¼ f ðh þ 4 A0 Þ0:4 (17)
bottom bed regime on the gas flow distribution
within the bed remains to a large extent unknown with
especially under conditions valid for industrial units f SFB ¼ ½0:26 þ 0:70  expð
3:3 d s;seal Þ
(which yield bubbles far from ‘‘text book’’ bubbles),
½0:15 þ ðu0
umf Þ
1=3 . ð18Þ
although Svensson et al. [39] showed that there is a
strong dependence of the bottom bed expansion Based on CFB bed-expansion data (bed-pressure
curve (bed voidage vs. velocity) on the pressure drop drop measurements) and the same assumptions
over the air distributor (and thereby also the bottom as discussed below Pallarès and Johnsson [53]
regime depends on air-distributor pressure drop). modified the correlation (Eq. (18)) for application
 ARTICLE IN PRESS
548 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

to CFB units likely that such an increase would be local, i.e. in-
between bubbles.
f CFB ¼ 0:3121 þ 0:129 u
1
0
16:6 d s;b In addition, the bottom bed voidage of a CFB

2:61  10
5 Dpref . ð19Þ furnace was found to level out at a certain
fluidization velocity [21] which could exceed the
Once a model for c is chosen and applied, uvis can terminal velocity of an average sized bed solids
be calculated from Eq. (15). Then, the bubble provided that enough bed solids are recycled to
density is determined by combining Eqs. (14), (6), maintain a bottom bed as can be seen from the
(7) and (10), and finally the bottom bed voidage is experimental data in Fig. 3. In this regime, bubbles
obtained from Eq. (12). The model by Johnsson are of an exploding type and during eruption they
et al. [17] gives a time-averaged bottom bed voidage extend all the way from the air distributor to the top
which is constant with height (in agreement with of the dense bed. This behaviour provides a large
the above-mentioned bed-pressure drop measure- and almost free passage of gas through the dense
ments). This is a consequence of assuming both the bed during the time of bubble eruption (i.e.
bubble fraction and the emulsion voidage constant throughflow). Thus, a further increase in primary
with height (see Eq. (9)). However, measurements gas velocity is restricted to an increase mainly in the
by Shen et al. [36] show a height-dependant bubble throughflow gas, utf, together with a small increase
density. In addition, Johnsson et al. [17] assumed in the visible bubble flow, uvis, due to the increase in
the voidage in the emulsion phase to equal the the bubble (growth) velocity. This saturation value
minimum fluidization voidage, which is a classical of the bottom bed voidage, found to depend on the
assumption used in this type of modeling (cf. above riser pressure drop and the mean particle size, is
equations) although part of the throughflow passes correlated as [53]
the bottom bed through the emulsion phase,

6
1
yielding an interstitial gas velocity higher than the
b;sat ¼ 0:5352 þ 496:5Dp
1
0 þ 4:9  10 d s;b . (20)
minimum fluidization velocity. It is uncertain to
what extent this increased velocity in the emulsion The fluidization velocity at which the levelling out
leads to an increase in the voidage of the emulsion. of eb takes place is taken as that for which the dense
The possibility of an emulsion voidage higher than bed voidage calculated by means of c (Eq. (17))
emf is discussed and defended by Gogolek [54] and equals the value obtained by Eq. (20) [53].
expressions for the voidage of an expanded emul- A comparison of results by the model given by
sion phase are given in literature [51,55], but the gas Eqs. (6)–(17) and (19) and experimental CFB data
velocity in the dense phase due to the throughflow is from Svensson et al. [39] is shown in Fig. 3.
required as input to these models. Also, if there is an Finally, it should be noticed that the mean
increase in voidage due to the throughflow, it is particle size in the bottom bed, which should

Fig. 3. CFB bottom bed expansion. Experimental data from Svensson et al. [39] compared with results from model calculations.
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 549

influence calculations of most of the variables given in a CFB riser show that the back mixing in the
above, is larger than the particle size as averaged transport zone occurs mainly in the downflowing
over the entire CFB loop, since fines tend to end up solids wall-layers, with an average concentration
in the return leg and the freeboard, while coarse several times higher than the upflowing solids
particles remain in the bottom bed. Thus, when an concentration in the core [7]. Furthermore, it has
overall modeling of a CFB is carried out, the size been experimentally determined that most of the
segregation effect can be taken into account and parameters in the wall layers have an heterogeneous
estimated through the formulation of a population horizontal distribution in the wall layer, as reported
balance over the whole CFB loop. in [7,60,61].

2.2. Modeling of the freeboard 2.2.1. Terminal velocity

The terminal velocity, i.e. the free fall velocity, of
In presence of a dense bottom bed with the a single particle falling in a fluid (terminal velocity),
characteristics mentioned in the previous section, ut, is derived as
the freeboard comprises the height interval !1=2
HbohpHduct with a solids suspension with a 4 d p ðrs
rg Þg
ut ¼ , (21)
decreasing solids concentration with height. Bed 3 rg C D
material is elutriated from the bottom bed to the
freeboard by the bursting of bubbles at the bottom where the drag coefficient, CD, is obtained from
bed surface ejecting particles into the freeboard. experiments. A study on drag coefficients was
Once in the freeboard, these particles might either carried out by Haider and Levenspiel [62], and they
fall down again to the bottom bed after reaching propose the following correlation:
different heights or be entrained up through the 24
CD ¼ ½1 þ ð8:1716 e
4:0655f Re0:0964þ0:5565f Þ
freeboard and be separated to the solids wall-layers Re
or externally recirculated through the cyclone and 73:69 e
5:0784f Re
return loop after having reached the exit duct height þ . ð22Þ
Re þ 5:378 e6:2122f
and followed the gas stream through the exit duct.
However, simpler correlations to estimate the
A division of the flow into two characteristic
terminal velocity exist in literature (see [62–65]).
zones with respect to the backmixing in the
One such correlation, also provided by Haider and
freeboard of CFB risers was suggested by Harris
Levenspiel [62], is
et al. [56]: a homogeneous particle clustering flow
!
1=3
and a core/annulus flow, with the former dominat- r2g

ing the flow pattern in the lower part of the riser and ut ¼ ut , (23)
mg ðrs
rg Þg
the latter dominating in the upper part of the riser.
Following this concept, Johnsson and Leckner [57] where
identified a splash zone and a transport zone in a
1
fluidized-bed boiler and in cold models operated ut ¼ , (24)
under conditions similar to those of boilers [57,58]. 18=d p 2 þ ð2:335
1:744 fÞ=d p 0:5
The splash zone, situated above the bottom bed !1=3
surface, is characterized by a strong backmixing due rs ðrs
rg Þg
1=3
to a ballistic-type of movement of clustered solids d p ¼ Ar ¼ d p. (25)
m2g
projected into the freeboard. This behaviour is
directly related to the eruption of bubbles, which Yet these correlations are intended for isolated
throw lumps of solids into the splash zone. The particles in a fluid. For calculation of the effective
clusters are immersed in a more dispersed phase. terminal velocity of particles in a solids suspension,
Few of the clusters reach high up in the freeboard. interactions between particles have to be taken into
Instead, the dispersed phase dominates the trans- account as done in the model by Win et al. [66],
port zone, adopting a core-annulus flow structure which was simplified by Palchonok et al. [67]
with upflow in the core and downflow in the assuming quasi-steady-state conditions applying
annulus in a solids wall layer, with a solids net flux averaged physical properties of the suspension.
from the core to the solids wall layer all along the These models are based on the momentum con-
transport zone [59]. Local solids flux measurements servation during collisions between particles flowing
 ARTICLE IN PRESS
550 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

at their slip velocity, making the effective terminal particle size. The figure illustrates the interaction
velocity of a particle different from the single effect for three different average sizes (100, 300 and
particle terminal velocity, due to the interaction 500 mm) of the particles forming the solids suspen-
with the neighbouring particles in the solids sion surrounding a particle with diameter dp. A
suspension. Applying the model by Palchonok value of
¼ 0:002 has been taken for the voidage of
et al. the effective terminal velocity for a particle the suspension surrounding the particle, (corre-
p, u0 t,p, immersed in a disperse solid suspension n is sponding to a particle concentration of 5 kg/m3). In
calculated solving the following polynomial: the example of Fig. 4, the particle density for the
2 suspension is assumed to have the same density as
a u0t;p þ b u0t;p þ c ¼ 0, (26) the surrounded particle (i.e. rs;n ¼ rp ).
where It can be seen that when a particle is surrounded
by a suspension of finer particles of equal density, its
pðd p þ d s;n Þ2 rs;n ð1

Þ effective terminal velocity is lower than the corre-
a¼ þ 0:055 p d 2s;n rg

4:75 ,
4ð1 þ ðmp =ms;n ÞÞ sponding single particle terminal velocity. In the
(27) core region of the freeboard, particles are assumed
to flow upwards at a velocity equal to the slip
pðd p þ d s;n Þ2 rs;n ð1

Þut;n velocity. Thus, a particle within a suspension of
ut;p
ut;n
b¼  finer particles in the core region moves upwards at a
jut;p
ut;n j 2ð1 þ ðmp =ms;n ÞÞ
higher velocity than in the case when it is
þ 3 p d p mg rg

4:75 , ð28Þ surrounded by gas only (i.e. it is pushed up by the
momentum of the surrounding particles, which flow
ut;p
ut;n pðd p þ d s;n Þ2 rs;n ð1

Þu2t;n upwards at a higher velocity). The opposite effects
c¼ 
ms;n g. take place when a particle is surrounded by coarser
jut;p
ut;n j 4ð1 þ mp =ms;n Þ
particles, and the above-mentioned effects are
(29) enhanced the higher the suspension density (for
The effect of considering particle interactions example in the wall layers).
through the use of a modified effective terminal In summary, particle interactions can be taken
velocity is shown in Fig. 4, where the ratio of into account through the use of an effective terminal
effective terminal velocity to single particle terminal velocity, allowing the solids suspension in the
velocity, u0 t,p/ut,p, has been plotted against the freeboard obtained from modeling to contain

Fig. 4. Effect of particles interactions on terminal velocity values according to Palchonok et al. [67].
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 551

particles coarser than the size corresponding to the CFB units, heights for which (H0
h)/Deq exceeds 6
slip velocity based on the terminal velocity of a are normally within the splash zone. In the splash
single particle which is in agreement with experi- zone, most of the solids belong to the cluster phase,
ments [66]. which, as mentioned above, has different fluid
dynamical behaviour than the disperse phase form-
2.2.2. Wall-layer thickness ing the core and wall layers structure. To experi-
The wall-layers thickness, twall, can be defined as mentally investigate the behaviour of the disperse
the distance from the wall to the position of zero net phase in the splash zone is difficult since measure-
local solids flux in the vertical direction. The ments are disturbed by the cluster phase. However,
thickness seems to be fairly independent of solids from a modeling point of view exact knowledge on
flux and the fluid dynamics of the unit once the dispersed phase in the splash zone should be of
circulating conditions are established, but depen- less importance since, in this zone, its concentration
dant on height in the riser [7,68]. This can be is low compared to that of the cluster phase.
observed in Fig. 5, where data from different units On the other hand, modeling of the disperse
under different fluidization conditions follow the phase (and the core-annulus structure formed by the
same trend. These experimental data from several separation of solids to the walls) in the splash zone
large-scale CFB units were summarized by Johns- can be carried out based on experimental data from
son et al. [6]. For large CFB units, there is an laboratory CFB units, which (having in mind that
increase in wall-layer thickness downwards along they may present different flow patterns than
the riser wall. However, data for large CFB units those of large CFB units) cover this missing range
are only available down to a certain distance from of data [6].
the top of the unit, Hsat, which can be defined as a From Fig. 5, it can be observed that the reduced
function of the equivalent diameter of the riser, Deq wall-layer thickness levels out for H0
h)/Deq46
(although the data scatter considerably which can
H sat ¼ H 0
6Deq (30)
be expected since they were obtained in different
with Hsat expressed as distance from the bottom of units under different operating conditions with
the riser. The reason for the lack of experimental respect to type of solids and gas velocity). Thus, a
data points for (H0
h)/Deq46 for large CFB units height-independent value of twall is assumed for
is that the riser height-to-width ratio of these units is (H0
h)/Deq46 (this assumption has little influence
typically less than 10 (and available data for this on the results, since the cluster phase is highly
ratio exceeding 6 is only available from small-scale predominant in this height interval and for this
units and these exhibit a different behaviour [6,69]. work the focus is on large-scale CFBs for which the
Thus, under typical operational conditions in large ratio is often less than 6). The expression for the

Fig. 5. Dependence of wall layer thickness on height in the riser. Measured data compared with results from model calculations with Eqs.
(31) and (32).
 ARTICLE IN PRESS
552 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

wall-layer thickness is written as Additional air is usually injected in the freeboard

(secondary or even tertiary air injections) with the
twall ðhÞ ¼ 0:0108ðH 0
hÞ for H sat ohpH duct ,
purpose to burn volatile matter released by the fuel
(31a) in the freeboard and to improve the gas mixing
above the bottom bed [73] and to lower NOx
twall;sat ¼ 0:0648 Deq for H b ohpH sat . (31b) formation. Despite the importance of the additional
Werther [1] also presented a correlation to air injections, no model in literature can reliably
estimate the thickness of the wall layer, reading predict their penetration and mixing with the main
 0:21   flow in the riser [74]. Two difficulties can be
twall H0 H 0
h 0:73 observed: the influence of the particular design of
¼ 0:55  Re
0:22
t . (32)
Deq Deq Deq the injector on the local gas flow, and the influence
of the downflowing solids wall layer through which
Another expression was derived by Johansson
the air jet has to flow (which varies with fluidization
et al. [70] under the assumptions that twall5De and
parameters as well as riser geometry). Obviously, in
that the solids downflow in the wall layers at the exit
an SFB there are hardly any solids wall layers and
duct height small enough to be neglected:
  these can be neglected in terms of influencing the
Deq H0
h secondary air penetration as done in [75]. In CFB
twall ¼ k  exp . (33)
4 H0 modeling one approach is to model air injections as
discretized in the horizontal plane assuming hor-
Note that a value for the backflow ratio, k, is
izontal diffusion of the injected air in a vertical plug
required to apply this expression (see Section 2.3 for
flow [76]. This does not account for the momentum
a detailed analysis and estimation of this ratio).
of the gas injected and the solution has to be
With the wall-layer thickness determined along
experimentally adjusted. In the present work, since
the entire freeboard, the cross-sectional area of the
no gas flow is assumed in the wall layers and
core and the wall layer region at any height in the
the core region is not discretized in the horizontal
riser, Ai core and Ai wall, can be determined.
direction, all gas injected is assumed to join the plug
flow at the injection height, as also assumed in other
2.2.3. Gas flow similar works [77,78]. This assumption becomes
Zhang et al. [7] measured velocities of down- critical in terms of mixing and combustion modeling
flowing solids in the wall layer for different particle for units with large cross sections, but, from a
sizes and fluidization conditions. The results ob- macroscopical point of view, it should be a reason-
tained are close to the values of the average terminal able assumption with respect to fluid-dynamics.
velocities of the particle suspension, which supports
the assumption of no or low gas flow in the annular
2.2.4. Solids concentration
region of the freeboard (wall layers). Measurements
In the literature, simple models are given for the
of vertical gas dispersion in CFB wall layers by
backmixing and decay in solids concentration above
Sternéus et al. [71] show vertical dispersion both in
bubbling beds as well as along CFB risers [79–84].
upward and downward direction. This supports that
These models assume a constant backmixing,
the net (time averaged) gas velocity within the wall
yielding an exponential decay in solids concentra-
layers is zero or near zero, but that the gas velocity
tion with height (with decay constant a0 ). Based on
may fluctuate in the vertical direction due to the
these single solids phase models, a general form to
time dependant character of the downward flowing
express the vertical solids concentration profile (for
wall-layer solids. The effect will depend on opera-
a given PSD size interval, d, of any of the solids
tional conditions, but the data of Sternéus et al.
fractions in the bed, f, at any height in the
were obtained under typical CFB conditions in the
freeboard, h) can be written as
Chalmers 12 MWth CFB boiler.
Concerning the core region, measured (horizon- C fh;d ¼ C fH;d0 þ ðC H
f ;d
b

C fH;d0 Þ expð
a0 fh;d ðh
H b ÞÞ.
tal) solids flux profiles are almost flat all over the
(34)
core (i.e. the solids flux [kg/m2 s] is more or less
independent of spatial location [1,68,72]). Thus, From experiments, the decay constant a0 for these
the gas can be assumed to follow a plug-flow in the models was found to be inversely proportional to
core region. the gas velocity as shown in [79,85] for fine and
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 553

coarse particles respectively. Thus, With the decay constants defined as

f ;d 1 utf ;d
a0 h / . (35) afh;d ¼ 4 , (39)
ug;h ug;h
Johnsson and Leckner [57] developed a similar 0:23
type of model. However, it differs from the above K fh;d ¼ . (40)
ug;h
uft ;d
models in that it considers both of the two above-
mentioned back-mixing phenomena (back-mixing of As indicated above, the formation of the splash
clusters dominating in the splash zone and back- zone is directly related to the bubbles erupting at the
mixing at the riser walls, mainly of the disperse top of the bottom bed, i.e. related to the bottom bed
phase). Thus, these back-mixing processes are super- characteristics (fluidization regime and height of
imposed in the model by Johnsson and Leckner [57]. bottom bed). However, nor Eq. (39) or any other
The back mixing of the cluster phase can be expression in literature for describing the splash
described as a ballistic type of solids back-mixing of zone solids concentration take into account the
clustered solids which gives a high backmixing rate bottom bed characteristics. Inclusion of the bottom
with the clusters originating from the eruptions of bed characteristics would most likely improve the
bubbles at the dense bed surface. Due to this origin, modelling of the decay in solids concentration in the
the cluster phase can be assumed to be equally splash zone (as expressed by the decay constant a).
distributed over the cross section (i.e. not to exhibit The gas velocity can, as a first approximation be
a core wall-layer flow structure). In the splash zone taken as ug ¼ u0 . In reality, the gas velocity varies
formed, not only the solids flux and solids along the riser height even if only considering the
concentration show a strong gradient with height time-averaged velocity. The variation is due to
above bed surface, but also the average particle size. variations in riser cross section and in the cross-
The strong gradients make modeling of the entire sectional area of the core region if it is assumed that
CFB loop sensitive to the splash zone modeling with the gas only flows upward through the core region. In
respect to pressure and mass balances [53]. addition, secondary gas injections, gas generation by
The dispersed solids phase occupies most of the fuel conversion, and temperature variations along the
height from the upper part of the splash zone to riser will add changes to the gas velocity. Therefore,
the top of the riser. This zone has been referred to as the decay constants should vary their values with
the transport zone [86,87] and, as mentioned above, height in furnace/riser and a discretized approach of
exhibits a core-annulus flow structure, with upflow the problem is required for an accurate analysis.
in the core with a net lateral flow to the down- Evaluation of the two types of modeling (single
flowing annulus (solids wall-layers). Solids concen- and two back-mixing terms) by comparison with
tration values in the transport zone are much lower experimental data from large CFB units is shown in
than in the splash zone and the decrease with height Figs. 6 and 7. Four models with only one back-
is less pronounced. mixing term (i.e. a single decay factor) from the
Consequently, the resulting backmixing is not literature [28,82,88,89] have been selected and
constant with height in the model by Johnsson and compared with the model by Johnsson and Leckner
Leckner [57], but the result of two back-mixing [57]. As can be seen, the experimental data from a
processes which both give an exponential decay in CFBC included in Figs. 6 and 7 yield an exponential
solids concentration with height with decay factors decay in solids concentration, but with strongly
a for the cluster phase and K for the disperse phase. different decay constants in the two different zones
Thus, if as for Eq. (34) written for a given PSD size (splash and transport zones) and accounting for the
interval, d, of any of the solids fractions in the bed, two back-mixing phenomena is required in order to
f, the vertical solids concentration profile in the obtain a good agreement between experimental data
freeboard becomes and model predictions. Although the model by
Johnsson and Leckner [57] is also based on
C fh;d ¼ C fclus;h
;d
þ C fdisp;h
;d
, (36) assumptions, it can be concluded that any modeling
of this type based on a single back-mixing expres-
C fclus;h
;d
¼ C fclus;H
;d
b
expð
afh;d ðh
H b ÞÞ, (37) sion (i.e. a single decay constant) cannot fit the
experimental data, whatever value for the decay
C fdisp;h
;d
¼ C fdisp;H
;d
0
expðK fh;d ðH 0
hÞÞ. (38) factor a0 is applied (see Figs. 6 and 7).
 ARTICLE IN PRESS
554 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

Fig. 6. Solids concentration profile. Experimental data from Johnsson and Leckner [57] vs. modelled data [28,57,82,88,89].

Fig. 7. Solids concentration profile. Experimental data from Werdermann [68] vs. modelled data [28,57,82,88,89].

Unfortunately, for large CFB units there is little by Adánez et al. [82] predicts with reasonable
experimental data on solids concentration covering accuracy the trend in the transport zone (h45 m)
both splash and the transport zones. In most cases, but not in the splash zone (ho5 m). Also in this
there is no experimental data on the splash zone, figure it can be seen that a 2-decay modeling is
while measured data on the transport zone is more required to reach reasonable agreement with mea-
common [68,90–92]. Single solid phase models are sured solids concentration profile along the entire
able to fit experimental data either in the transport freeboard, i.e. including both the splash and the
or the splash zone, and if enough experimental data transport zones.
from any of the two zones is lacking, fitting a one-
decay model may lead to that wrong conclusions are 2.2.5. Solid flows
drawn on the solids back-mixing process and The net solids flow out of the furnace (rate of
thereby on the characteristics of the vertical particles leaving the riser towards the cyclone) is
concentration profile. Fig. 7 shows how the model assumed to entirely consist of solids belonging to
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 555

the disperse phase, since no cluster phase is assumed wall region becomes
to reach the upper part of the freeboard. Measure-
ments show that particles in the core region of the F fwall;h
;d
cfwall;h
;d
¼ . (45)
transport zone travel at velocities close to their slip uft ;d Awall;h
velocity [72]. The use of the concept slip velocity is
analysed in [93]. Assuming particles to flow upwards
at their slip velocity (with the terminal velocity 2.3. Modeling of the exit zone
modified in order to account for particle interac-
tions, as explained above) and having calculated the The height interval HductohpH0 in a fluidized
vertical solids concentration profile, the upflow in bed riser is here referred to as exit zone. When an
the core region of the transport region becomes FB unit is operated under circulating conditions, the
exit zone exhibits a different and more uncertain
F fcore;h
;d
¼ cfcore;h
;d
Acore;h ðug;core;h
uft;core;h
;d
Þ. (41) solids flow mechanism from that in the freeboard
(see Section 2.2). The highly dilute conditions in the
Davidson [93] gives a detailed discussion on exit zone make pressure measurements unreliable.
modeling of the transfer of solids from the core to Thus, experimental knowledge concerning solids
the wall region based on a diffusion process in the concentration and flow in this zone of a large CFB
core region. Obviously, it is physically not a unit is usually limited to techniques based on visual
diffusion process since the solids concentration in observations in scale models. Due to the unclear
the annular region is higher than in the core (but the solids flow picture, a proper modeling of the exit
net transfer of solids is from core towards wall). zone requires the use of CFD tools [94].
There seems to be an agreement in literature on An important fact to point out concerning the
the existence of a net transfer of solids from the core solids flow pattern in the exit zone of large CFB
region to the wall layer in the transport zone. This units is that its picture differs from that of narrow
net transfer is the result of a balance where, at a laboratory units, so literature based on observations
certain height, there are more solids leaving the core in the latter type of units might be misleading.
towards the wall region than vice versa (thus, there For narrow laboratory units, it has been shown
is a flow in both directions). The lateral net solids that different configurations of the exit geometry
flow in the transport zone can be calculated by (i.e. smooth, extended or abrupt) can strongly
differentiation of the solids flow in the core region influence some of the main fluid dynamical para-
of the freeboard meters of a CFB unit, as the vertical solids
concentration profile, the solids net flow and the
F flat;h
;d
¼ F fcore;h
;d

F fcore;hþdh
;d
. (42) residence time. On the other hand, studies carried
out in large CFB units [58,95] show that the
Thus, note that the net lateral flow obtained is the
influence of the exit zone geometry is far less
result of the net balanceF fcoreXwall;h
;d

F fwallXcore;h
;d
,
important in CFBC and CFBG units (the most
although the individual values of these terms remain
common types of exit zone designs in operating
unknown. Knowing that the net solids flow
units were tested in these works). Zheng and Zhang
contributes to increase the downward flow in the
[96] carried out an analysis of the influence of the
wall layer, the downflow in the annular region of the
exit zone geometry on the performance of CFB
transport zone can be calculated as
combustors, but based on experiments in a riser
with an aspect ratio over 50 (i.e. far higher than the
F fwall;h
;d
¼ F fwall;hþdh
;d
þ F flat;h
;d
(43)
typical values below 10). Thus, their observations of
with the initial value given at the cell located at a strong influence of the exit geometry on the solids
h ¼ H duct by means of the definition of the backflow net flow and vertical concentration profile, typical
ratio, kb (see Section 2.3): for narrow laboratory risers, are hardly applicable
to CFB combustors. The reason that the exit
kf ;d geometry seems to be of less importance in large
F fwall;H
;d
¼ F fcore;H
;d
. (44) units than in laboratory units is most likely simply
duct
1 þ kf ;d duct

due to that the solids loading is much lower in the

Finally, with the above-mentioned assumption of exit region in large units than in the small units, with
zero wall gas-flow, the solids concentration in the the latter operated with high solids flux.
 ARTICLE IN PRESS
556 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

2.3.1. Backflow effect It has been shown [58,95] that there is little or no
Upflowing particles in the core region of the difference between an extended, abrupt and en-
freeboard can follow two paths when reaching the hanced exit with respect to the ratio of internal to
exit zone: they either enter the downflow in the riser external solids flow (k was similar in these three
wall layers (i.e. be internally recirculated) or they cases) in large CFB units. It was also found that a
follow the gas flow out of the riser through the exit reduction in the cyclone inlet area (increase in
duct (i.e. be externally recirculated, forming the so- entrance velocity) decreases k, resulting in an
called solids net flow of the unit). The proportions in increase of the solids flow into the cyclone (i.e. the
which these two effects take place are of great solids net flow). Addition of internals near the exit
importance, since the larger the internal recircula- duct was—as expected—found to lead to the
tion in the wall-layer downflow (backflow) the opposite effect, i.e. resulting in a higher residence
greater the particle residence time in the riser. The time and backmixing rate of the particles.
solids residence time in the riser is an important Despite the importance of k, little information is
parameter to adjust when optimizing the perfor- available on its magnitude and no simple and
mance of CFB units, and this parameter was found reliable modeling expressions are available in
to be strongly influenced by the exit configuration literature for its calculation. What is more, the
chosen, as discussed in [97]. Yet, as stated above, the backflow effect is often neglected in modeling of
low solids loadings in the top of industrially sized fluid dynamics. Experimental correlations for esti-
CFBs makes these less sensitive to exit geometry mating the backflow in standard exit configurations
than small units operated at high recirculation flows. are given in [1,53,68,98]. Only in the latter work by
The relative magnitudes of internal and external Pallarès and Johnsson [53] k is assumed not to only
recirculation of solids can be expressed through the depend on the vertical location of the exit duct but
backflow ratio, k, which, for a given size interval of also on the flow conditions. The Pallarès and
any solids fraction in the CFB unit is defined as Johnsson expression is based on fits to experimental
data of solids fluxes from three large CFB units and
F fwall;H
;d the correlation is shown in Fig. 8. The correlation
kf ;d ¼ duct
. (46) relates the probability of entrainment of a particle,
F fnet
;d
pent, to its slip velocity and the core region flux at the

Fig. 8. Entrainment probability of a particle in the exit zone calculated from experimental data found in [68] and [99].
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 557

beginning of the exit zone and it is written Finally, the backflow ratio can be expressed as a
function of the entrainment probability as
1
pfent
;d
¼ F fH;dduct F fcore;H
;d f ;d

F net
ð4:07
ufslip;H
;d
duct
Þmaxf0:5;3:057
0:129Gs;H duct g kf ;d ¼ ¼ duct

F fnet
;d
F fnet
;d
for ufslip;H
;d
p3:07 m=s; ð47aÞ
duct
F fcore;H
;d
duct
1
¼
1¼
1. ð49Þ
F fnet
;d
pfent
;d

pfent
;d
¼1 for ufslip;H
;d
duct
43:07 m=s: (47b)
2.3.2. Solids concentration
Unlike previous expressions in literature for
In contrast to investigations in narrow CFB risers,
evaluation of the backflow effect, the probability
where the vertical solid concentration profile is
of entrainment of a particle in Eq. (47) is not
observed to increase in the exit zone, solids
directly correlated to height in the furnace, but to
concentration values in the exit region of CFB
fluid dynamical variables which, however, vary with
combustors are low (as mentioned above). This
height in the furnace (thus, indirectly accounting for
implies that the solids inventory in the exit zone of a
the dependence on height). The data points in Fig. 8
large CFB unit can be neglected, when a global mass
are taken from [68,99]. However, since the perfor-
balance over the whole CFB unit is formulated.
mance of the correlation given by Eq. (47) depends
By means of fibre optical probes Lackermeier and
on exit configuration, it should not be used in units
Werther [95] observed that the exit zone flow exhibits
with an exit configuration significantly differing in
traces of the core-annulus flow structure governing
geometry from those units which were used as
in the transport zone flow of the freeboard, but with
experimental basis for Eq. (47) (although data
a superimposed horizontal velocity component
taken from three large CFB units, most of the data
towards the exit duct due to the drag of the gas
are from the Chalmers 12 MW CFB unit, cf. Fig. 2
flow exiting the riser. Furthermore, based on visual
and [58]).
observations made in a scale model by Johnsson et
In spite of the complex nature of the backflow
al. [58], particles were observed to follow a ballistic
effect as well as the fact that available data originate
movement in the exit zone. Harris et al. [100]
from different units, the resulting data points in
justified by means of force balances all of the
Fig. 8 fall on rather continuous curves, which
above-cited behaviours: particles follow the gas flow
makes it possible to simplify the modeling by
towards the exit duct until they eventually experience
identification of trends. The finer the particles, the
an inertial separation from the gas streamlines, then
higher the probability for them to be entrained and
following a ballistic movement towards the walls,
to become externally recirculated through the
where they join the downflowing solids annulus.
cyclone. There is also a dependence on the super-
ficial solids upflow due to the effect of the particle
2.4. Modeling of the exit duct
interaction, i.e. when a particle is surrounded by
other particles (high value of G s;H duct ), its probability
Particles entrained from the riser flow towards the
to be entrained is lower the more dilute the solids
cyclone through the exit duct due to the drag of the
suspension is.
gas stream, with the particles experiencing an
As indicated above, the probability of entrain-
acceleration that can be further enhanced by a
ment is known to depend also on the geometry of
reduction in the cross-sectional area of the duct
the exit configuration [58]. Thus, care has to be
(increase in velocity).
taken when extending the present probability values
Several works in literature deal with horizontal
to units with an exit geometry significantly different
conveying of solid particles [101–103], but the
from those of the three units from which the data
conditions in a CFB exit duct (dilute solids
were taken.
suspension, high acceleration and low length-to-
The solids net flow, Fnet, can be determined using
height ratio of the duct) are different from the
the definition of entrainment probability
conditions employed in these investigations.
Muschelknautz and Muschelknautz [104] devel-
F fcore;H
;d
F fnet
;d
¼ duct
. (48) oped a model especially aimed at CFB units, which
pfentr
;d
is used to determine Dpduct and the solids velocity
 ARTICLE IN PRESS
558 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

along the exit duct. These authors calculate the cyclone. Furthermore, studies on cyclones are
velocity of the solids at any vertical cross section in available in literature and available models are able
the exit duct, x, by solving the following differential to provide accurate results.
equation With respect to the pressure drop over the
  cyclone, several expressions are available in litera-
dus g ug;duct
us 2
z ture, all of them giving similar results [105–107]. The
¼ (50)
dx us ws latter expression [107] has been used in several
taking the initial value works in the field of large CFB units [77,84,108] and
ug;duct;0 reads:
us;0  , (51)
3 2
V_ g duct rg
where Dpcycl ¼ C cycl . (56)
A2cycl inlet
ug;core;H duct Acore;H duct
ug;duct ðxÞ ¼ , (52)
Aduct ðxÞ Based on comparison with experimental data
" #1=2
z [108] the empirical coefficient, Ccycl, can be set to
4ðrs
rg Þg d s 10.0, which yields values for Dpcyclone of the order of
ws ¼ ð1 þ ð0:25 þ zÞg0:25 Þ, 100 Pa under typical boiler and gasifier conditions.
3 Z mkg r1
k
g
With respect to the cyclone separation efficiency,
(53) several comprehensive models are available. Some
F net of them restrict their set of inputs to geometrical
g¼ parameters [109], while more accurate models also
ug;i¼nduct core Ai¼nduct core rg
account for flow conditions (gas velocity, solid load)
z  0:7; Z  18 for Rep ¼ 3 to 100. ð54Þ [110–112]. With the cyclone separation efficiency
Once the differential equation (Eq. (50)) is solved, determined for each size interval of each solid
the solids velocity profile is known. Thus, the fraction considered, the rate of solids separated by
pressure drop along the exit duct can be calculated the cyclone and externally recirculated through the
as return leg can be expressed as
 
ug;duct F fret;d leg ¼ Zfcycl
;d
F fnet
;d
, (57)
Dpduct ¼
rg ug;duct þ g Dus;duct . (55)
2
An additional loss of pressure due to the entrance where F fnet
;d
is the solids net flow of each size interval.
effect might be included in the calculations by Thus, in the return loop (i.e. downcomer, particle
taking the value of Dus,duct between the duct seal and, optionally, particle cooler), the mass
entrance (x0 ¼ 0) and a virtually extended length fraction of each size interval can be calculated as
of the duct equal to xf ¼ Lduct þ b (being b the
entrance width of the duct). f ;d F fret;d leg
xret leg ¼ P P f; d . (58)
f d F ret leg
2.5. Modeling of the cyclone
With these values, the other parameters re-
From a fluid dynamical point of view, two quired to the model the downcomer and particle
phenomena should be of main importance as far seal (e.g. mean particle size and mean solids density)
as the cyclone is concerned: the pressure drop, can be calculated. Since the separation efficiency
Dpcycl, and the separation efficiency for each size is high (near 1) for the inert solids present in
interval of each solid fraction present in the CFB the primary cyclone, the result will have little
unit, Zf,d. The pressure drop across the cyclone is effect on the overall fluid dynamical modeling
relatively small compared to the pressure drops in presented here. On the other hand, separation
the other parts of a CFB unit (see Fig. 15) and the efficiency is lower for fuel particles which are at
overall collection efficiency of the cyclone is close to the end of their conversion process (i.e. fine and
1 under typical operational conditions. Thus, nor light). Thus, these fines are to some extent entrained
the mass nor the pressure balance over the whole out of the cyclone, reducing the combustion
CFB unit are sensitive to the modeling of the efficiency [76].
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 559

2.6. Modeling of the downcomer and particle seal close as possible to minimum fluidization condi-
tions. This, since it is desired to minimize the gas
The system formed by the particle seal and the flow entering the cyclone through the cyclone leg
downcomer has the purpose of allowing the and the formation of bubbles (which may project
transportation of externally recycled particles solid particles into the cyclone). However, in
through the return loop once they have passed the practise a higher gas flow is employed in order to
cyclone, and preventing the fluidization gas to flow ensure a sufficiently good fluidization both in the
backwards through the return loop and into the downcomer and in the particle seal.
cyclone. This is usually achieved by inserting the Thus, assuming a dense bubbling bed in the
end of the downcomer in a bubbling bed and particle seal, models for bed expansion in SFB units
driving the fluidized particles back to the riser can be applied to model the voidage and the gas
through a return leg. In some CFB units, the flow division of the bed in the particle seal and the
bubbling bed acts as particle cooler, extending the fluidized solids column in the downcomer. A
range of operating conditions with respect to the description of the fluid dynamics of a dense bottom
heat balance/output power. Some CFB designs, bed including a modeling approach is given in
such as the one in Fig. 9, have a particle-cooler bed Section 2.1.
separated from the particle-seal bed. In addition, Fig. 10 compares experimental data from the
the pressure drop in the riser, exit duct and cyclone, dense bottom bed of a 16 MWth SFB boiler with
is compensated by a column of fluidized solids that results from the model by Johnsson et al. [17]. A
is formed in the downcomer. good agreement between experiments and modeling
In order to both ensure solids transportation and is shown, with the data obtained at different bed
avoid gas by-pass flow to the cyclone, the particle- pressure drops (Dp0 from 4 to 8 kPa), showing that
seal bed is normally under bubbling conditions. The the bed pressure drop has no or little influence on
fluidization regime in the column of fluidized solids the bed expansion (all data points follow the same
formed in the downcomer should theoretically be as curve). However, a CFB particle seal will typically
operate at lower gas velocities (below 1 m/s) and
smaller mean solids size than those corresponding
to the conditions in an SFB combustor. At this
point, it is important to recall the role played by
particle size segregation effects in a CFB unit.
Under typical operation conditions, the mean
particle size in the dense bed of the particle seal
(which is needed in the modeling of the particle seal)
may be less than half of the mean particle size of the
bed material used. Thus, as in the modeling of the
riser bottom bed, modeling the bed in the particle
seal by assuming a mean particle size equal to that
of the material in the whole CFB unit (which is
known) is not realistic and may lead to large errors.
Thus, the value of the mean solids size in the particle
seal should be obtained through an overall CFB
model accounting for the mass balance over the
entire unit (see Section 3), considering solids
segregation.
Since solids in the downcomer flow downwards at
a certain superficial velocity, us,dc, fluidization
conditions in the downcomer are defined by the
relative gas–solids velocity, urel,dc, which is the gas
velocity to be used as input when applying an SFB
bed-expansion model. Thus
Fig. 9. Overview of a CFB return loop with independent particle
cooler. urel;dc ¼ ug;dc
us;dc , (59)
 ARTICLE IN PRESS
560 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

Fig. 10. SFB expansion. Measured data compared to results from model calculations (from [17]).

P2

2 P1
3
Downcomer
1
prefeed (<0) solids height
1

4 2
pduct (<0)
p2 = p1+
prefeed+
pduct+
pcycl ;
3
pcycl (<0) ⇒p2 = p1-⏐
prefeed⏐-⏐
pduct⏐-⏐
pcycl⏐;

4 -
pdc (>0) ⇒p2 < p1

⇒p1 - p2 =
pdc = s·(1-dc)·g·Hs,dc

Fig. 11. Pressure balance terms in CFB loop, with focus on downcomer and particle seal.

where 2.6.1. Pressure balance

Modeling of the height of the column of fluidized
F ret leg solids formed in the downcomer requires the
us;dc ¼
. (60)
rs Adc formulation of a pressure balance over the circulat-
ing loop. Rhodes and Geldart [107] and Yang [108]
The superficial velocity of the gas in the down-
carried out some investigations on the pressure loss
comer is assumed to equal that in the bubbling bed
in different parts of FB units, and Leckner [3], Lei
of the particle seal. This assumption is supported by
[78] and Hannes [77] formulated pressure balances
the fact that the pressure that the gas has to flow
over the whole circulating loop in their works.
against is the same both through the bubbling bed
The value of the pressure drop over the solids
of the seal and the solids column in the downcomer
column formed in the downcomer, Dpdc, is obtained
(since p14p2, see Fig. 11). Thus
by formulating the pressure balance all over the
ug;dc ¼ ug;seal . (61) circulating loop. Assuming no pressure drop across
the return leg (where the particles are assumed to
However, this value of ug,dc is only a valid flow back to the riser by gravity), the pressure
estimation when the nozzles fluidizing the particle balance can be written
seal are equally distributed over the entire cross
section of the particle seal. Dprefeed þ Dpduct þ Dpcycl þ Dpdc ¼ 0. (62)
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 561

The meaning of the pressure drop terms in this as an input to be provided to the model. Hannes [77]
expression is illustrated in Fig. 11a, where Dprefeed is also presented an overall CFB model, which is not
the pressure drop caused by the solids suspension in limited to fluid dynamics but included also the other
the riser between the centerlines of the return leg modeling fields given in Section 1.2 (Fig. 1).
(inlet of externally recycled solids) and the exit duct However, the work by Hannes is restricted to the
(outlet of externally recycled solids). This is riser of the CFB unit, and the return loop (exit duct,
calculated as cyclone and downcomer and particle seal) is not
Z H duct modelled but substituted by a set of boundary
Dprefeed ¼ rs g C s dh. (63) conditions in the riser.
H retleg Due to the importance of the above-mentioned
The values of Dpduct and Dpcycl are determined by interaction of the zones of the entire CFB loop,
Eqs. (55) and (56), respectively. Thus, the pressure some key features are particularly expected to be
drop in the downcomer, Dpdc, is obtained from the included in a fluid dynamical overall CFB model:
pressure balance given by Eq. (62). In terms of
absolute values, Dpduct and Dpcycl are rather small  The overall model should exhibit a realistic input
compared to the terms Dprefeed and Dpdc. The scheme. If the model is intended to have a
pressure balance over the downcomer and the practical meaning and applicability, its input
particle seal is illustrated in Fig. 11b. Hence, parameters should be such which are both
the height of solids required in the downcomer measurable and independently adjustable under
solids column, Hs,dc, to achieve this pressure drop, operation, i.e. some models require the solids net
Dpdc, can be obtained by the balance flow as input, although this parameter is not
known under operation and is dependant on the
Dpdc ¼ rs ð1

dc Þg H s;dc . (64)
model inputs: geometry of the unit (specially of
Generally, the height of the solids column in the the riser exit), operational conditions and bed
downcomer increases as the fluidization velocity in material used. Hence, a reasonable set of inputs
the riser or the particle seal is increased. to an overall CFB fluid dynamical model can be
classified in three categories: geometry of the CFB
3. Overall modeling of a CFB unit unit, operational conditions (injected gas flows
and pressure drop over the riser) and physical
Separate modelling of any of the zones presented properties of the bed solids and the fluidizing gas.
in the previous section, schematically illustrated in  Discretization of the riser into a mesh is required
Fig. 2, might fail in giving satisfactory results, since to provide an appropriate treatment to geome-
strong interactions exist between the zones. Inter- trical and fluid dynamical discontinuities which
actions exist between the bottom bed and the rest of may be present due to changes in the riser cross-
the riser due to the elutriation of solids from the sectional geometry and secondary air injections,
bed, backmixing of solids from freeboard to bed; and possible discontinuities as a result of the
between the freeboard and the exit zone due to the existence of the above-mentioned (three) fluid
backflow effect; between the downcomer and the dynamical zones in the riser.
other zones due to the pressure balance; and  The well-established core-annulus flow structure
between all zones due to the population and in the freeboard should be included in the model,
pressure balances. Thus, changes in flow pattern in which implies a discretization of the freeboard
any given zone will to a certain extent affect the cells not only in the vertical but also in the
features of the flow in the remaining zones. Thus, horizontal direction.
for proper modelling of each of the six zones given  The particle size segregation is of importance and
above as well as of the entire CFB unit, an overall should be taken into account. This is done by
model taking into account interactions among the enabling the overall model to handle PSD for
different zones of a CFB unit has to be implemen- each of the solids fractions considered instead of
ted. Lei [78] provided such an overall CFB model, only assuming a mean particle size to represent
although expressions found in literature directed all solids. This modelling approach forces the
towards narrow laboratory units are used in some inclusion of a population balance over the CFB
zones of this model. The model does not account for unit in the calculations. As a result, the PSD of
solids size segregation and has the net solids flux, Gs, each solid fraction (and thereby the mean particle
 ARTICLE IN PRESS
562 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

size) in all zones of the CFB unit is obtained, i.e. An overall CFB model comprising the above
resulting in that particle size segregation is features will make it possible to close the pressure
modelled. As mentioned above, the mean particle balance in order to determine the amount of
size in the dense bed of the riser and in the material contained in the fluidized solids column
particle seal both usually differ considerably formed in the downcomer. This amount of solids
from that of the particle size averaged over the must also be considered in the population balance.
entire CFB loop, and there is also a considerable
variation in solids size within the freeboard. The 3.1. Practical example of an overall CFB model
crucial importance of including the return loop in
the overall modeling and not just the riser is In order to exemplify the possibilities that an
revealed in the evaluation of the population overall CFB model can offer, a set of local models
balance, where two facts can be observed: the have been selected from those presented in the
return loop contains an amount of bed material previous section and linked into an overall CFB
that is not negligible (typically from 10% to model through an implementation which takes the
30%) and that solids in the return loop have a four above-listed requirements into account. Details
PSD which typically differs significantly from the on the implementation itself are considered outside
average solids size in the CFB unit (i.e. PSD of the scope of this paper.
solids in the return leg is biased towards finer The local models selected are: Pallarès and
particle size intervals). Johnsson [53] for the expansion of the bottom bed

Fig. 12. Vertical profile of solids concentration in cross-section average. Experimental data from Werdermann [68] compared to model
results.
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 563

in the CFB riser; Johnsson and Leckner [57] for the however, is that certain key characteristics of solids
freeboard; Palchonok et al. [67] for the effect of flow in large-scale CFBs have been identified (see
interactions between particles; the correlation given criteria listed in Section 1.1) from which it can be
by Eq. (47) for the back-flow effect in the exit zone; expected that the same characteristics of the flow
Muschelknautz and Muschelknautz [104] for the should be present in all boilers fulfilling the criteria
riser exit duct; in the cyclone, the models by Dietz listed in Section 1.1. Fig. 12 shows the vertical
[110] for the collection efficiency and by Rhodes and profile of solids concentration in the riser of two
Geldart [107] for the pressure drop; and finally the large CFB combustors (with a cross section of more
model by Johnsson et al. [17] for the expansion of than 25 and 49 m2, respectively) investigated in
the dense bed in the particle seal. In addition, a detail in [68]. As seen, the agreement of the overall
pressure balance according to what is proposed in CFB model with the experimental data is good. The
Section 2.6 and a population balance all over the sudden drops in the solids concentration profiles are
CFB unit need to be formulated and fulfilled. due to changes in the secondary air injections or in
Finally, interaction between solids of different size the cross-sectional geometry of the riser. The drop
is accounted for according to Eqs. (26)–(29) and the in solids concentration at the top of the riser is due
riser has been discretized in a way accounting for to the different flow pattern in the freeboard and the
what is mentioned in the previous section. exit zone.
Fig. 12 exemplifies results obtained from the The value of the net solids flow, Fnet, is a valuable
overall CFB model together with experimental data. output from the overall CFB model, since it governs
It should be noted that the experimental data shown important features in an operating CFB unit, such
in Fig. 12 has not been part of the modeling process as the residence time of the bed material in the riser
of any of the local models used. As mentioned in and the return leg and the thermal stability around
Section 1.3 the data available from large units is for the circulating loop. The net solids flow cannot be
obvious reasons restricted. Much of the data is from directly measured in large CFB units, but it can be
the Chalmers 12 MWth boiler which, although of estimated through rough methods, such as integra-
industrial scale, is considerably smaller than a large tion of available cross-sectional data on the solids
power boiler of some 100 MWel. What is important, flux in the riser (as was done to obtain the

Fig. 13. Comparison between experimental [7,60,113] and modelled values of the superficial solids net flow.
 ARTICLE IN PRESS
564 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

experimental data in Fig. 13) or by using the integration of experimental data on solids flux. This
expression: should be due to the fact that the measured values
do not include measurements in the corners of the
F net ¼ C s;H duct ðug;H duct
ut ÞAH duct . (65)
cross section, which was shown to exhibit strong
Experimental values in Fig. 13 have been taken downflowing flux values [7], i.e. the so-called
from [7,60,113] in the case of the Chalmers 12 MWth ‘‘corner effect’’ is not taken into account in the
CFB unit, and from [68] for the other two units. experimental values, leading to an overestimation of
Concerning the empty symbols in Fig. 13, the values the experimental solids flow. This is seen when the
predicted by the model tend to be somewhat lower corner effect is taken into account, as in the filled
than the corresponding flow values calculated by symbols plotted in the same figure. The arrows show

Fig. 14. (a and b) Size segregation. Measured data from a 30 MWth CFB unit (unpublished) compared to results from model calculations.
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 565

that such inclusion improves the agreement of the sponds to the average solids concentration) is much
overall CFB model and experimental data. lower in the upper part of the riser (transport zone
Concerning the solids PSD, a good agreement and exit zone) than in the lower part of the riser
between model results and measured PSD in the (bottom bed and splash zone). The pressure drop
bottom bed of the riser and in the bubbling bed of along the exit duct is low compared to all other
the particle seal is shown in Fig. 14 (a and b, pressure drops in the loop Two parts of the
respectively). The measured values are from a downcomer can be identified in the pressure
30 MWth CFB boiler. Having the PSD of the inert diagram: an upper part, with almost no solids
solids in the CFB unit as input (with an average size and characterized by a loss of height at a constant
of 238 mm in this case), the measured average pressure value, and a lower part, occupied by
particle size in the bottom bed is 308 mm and this the fluidized solids column which maintains the
is estimated by the model to 273 mm (11% error). In pressure balance along the circulating loop (see
the dense bed of the particle seal, the measured Section 2.6.1). Finally, the pressure loop is closed by
value is 226 mm while the model estimates an the return leg recirculating the particles back into
average particle size of 193 mm (16% error). Con- the riser.
sidering the likely scatter in the data this should be
regarded as a satisfactory result. 4. Closure
Finally, the predicted pressure distribution
around the CFB loop for a run representing the This paper highlights the major features to take
Chalmers CFB unit (with u0 ¼ 4 m=s, d s ¼ 250 mm into account in semi-empirical modeling of macro-
and Dp0 ¼ 7 kPa as main input parameters) is scopic fluid dynamics in a large CFB unit. The CFB
shown in Fig. 15. As can be seen, most of the riser unit is divided into six zones, and models appro-
pressure drop takes place over the first metres of the priate for the modeling of each zone are described
riser (i.e. bottom bed and the splash zone). The and discussed. Special emphasis has been made in
point corresponding to the bottom bed surface is pointing out to the reader the differences in the flow
marked with an arrow in the pressure diagram (the pattern between large CFB units directed towards
bottom bed height, Hb, is estimated by the model to combustion or gasification applications (which are
0.27 m). The pressure drop gradient (which corre- the scope of the present work) and narrow CFB

Fig. 15. Pressure diagram predicted by the overall CFB model for a standard run in the Chalmers CFB boiler (a—bottom bed, b—
freeboard and exit zone, c—exit duct, d—cyclone, e—downcomer, f—return leg). The arrow indicates the location of the bottom bed
surface.
 ARTICLE IN PRESS
566 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

units used in many of the investigations given in of the international conference on fluidized bed combustion
literature. Literature for the latter is abundant but 2003:643–9.
hardly useful for modeling purposes for large CFB [9] Jaud P, Piedfer O, Jacquet L, Jestin L. Provence 250 MWe
unit: the largest CFB boiler ready for operation. Proceed-
units, for which, on the other hand, literature is ings of the 13th international conference on fluidized bed
more limited. combustion 1995:721–8.
The paper shows that integration of the local [10] Goidich SJ, Hyppänen J. Foster wheeler compact cfb
models into an overall CFB model for the entire boilers for utility scale. In: Proceedings of the 16th
CFB loop is possible, but requires that the popula- International conference on fluidized bed combustion.
Reno, NV; 2001.
tion balance of the solids is taken into account in [11] Greil C, Vierrath H. Fuel gas from biomass—utilisation
order for each model to have a sound link to the concepts. DGMK-Fachbereichstagung; 2000.
underlying physics of the fluid dynamics. The [12] Fitzgerald TJ. Cold modeling of fluidized bed combustors.
overall model given here is shown to exhibit good Report EPRI CS-1476, RP315-1; 1980.
agreement with experimental data and give reason- [13] Svensson A, Johnsson F, Leckner B. Fluid-dynamics of the
bottom bed of circulating fluidized bed boilers. Proceedings
able calculation times (in the order of a few of the 12th international conference on fluidized bed
minutes). Aspects known by experience to be key combustion 1993:887–97.
features in the establishment of an overall CFB [14] Rhodes MJ. Modelling the flow structure of upward-
model are: a set of inputs consisting only of flowing gas–solids suspensions. Powder Technol 1990;
parameters that are known and independently 60(1):27–38.
[15] Zhang Y-F, Arastoopour H. Dilute fluidized cracking
adjustable under operation, a discretized approach catalyst particles–gas flow behavior in the riser of a
of the model in the riser (both in the vertical and circulating fluidized bed. Powder Technol 1995;84(3):
horizontal directions) and consideration of the size 221–9.
segregation effect by including the entire circulating [16] Kim SW, Kirbas G, Bi H, Lim CJ, Grace JR. Flow
loop in the modeling. behavior and regime transition in a high-density circulating
fluidized bed riser. Chem Eng Sci 2004;59(18):3955–63.
[17] Johnsson F, Andersson S, Leckner B. Expansion of a
Acknowledgements freely bubbling fluidized bed. Powder Technol 1991;68(2):
117–23.
The work presented is a continuation of a work [18] Zijerveld RC, Johnsson F, Marzocchella A, Schouten JC,
originally funded by the EU Non Nuclear Energy van den Bleek CM. Fluidization regimes and transitions
from fixed bed to dilute transport flow. Powder Technol
Programme JOULE III (JOR3CT980306). Current
1998;95(3):185–204.
funding from the Swedish Energy Agency is greatly [19] Geldart D. Types of gas fluidization. Powder Technol
appreciated. 1973;7(5):285–92.
[20] Grace JR. High-velocity fluidized bed reactors. Chem Eng
Sci 1990;45(8):1953–66.
References [21] Svensson A, Johnsson F, Leckner B. Fluidization regimes
in non-slugging fluidized beds: the influence of pressure
[1] Werther J. Fluid Mechanics of large-scale CFB units. drop across the air distributor. Powder Technol 1996;86(3):
Circul Fluid Bed Technol 1993;IV:1–14. 299–312.
[2] Anthony EJ. Fluidized bed combustion of alternative solid [22] Avidan AA, Yerushalmi J. Bed expansion in high velocity
fuels; status, successes and problems of the technology. fluidization. Powder Technol 1982;32(2):223–32.
Prog Energy Combust Sci 1995;21:239–68. [23] Bi H, Fan L-S. Existence of turbulent regime in gas–solid
[3] Leckner B. Fluidized bed combustion: mixing and pollu- fluidization. AIChE J 1992:297–301.
tant limitation. Prog Energy Combust Sci 1998;24(1): [24] Glicksman LR. Scaling relationships for fluidized beds.
31–61. Chem Eng Sci 1984;39(9):1373–9.
[4] Longwell JP, Rubint ES, Wilson J. Coal: energy for the [25] Wen CY, Yu YH. A generalized method for predicting
future. Prog Energy Combust Sci 1995;21:269–360. the minimum fluidization velocity. AIChE J 1966;12:
[5] Issangya AS, Grace JR, Bai D, Zhu J. Further measure- 610–2.
ments of flow dynamics in a high-density circulating [26] Grace JR. Contacting modes and behaviour classification
fluidized bed riser. Powder Technol 2000;111(1–2):104–13. of gas–solid and other two-phase suspensions. Can J Chem
[6] Johnsson F, Zhang W, Leckner B. Characteristics of the Eng 1986;64:353–63.
formation of particle wall layers in CFB boilers. Proceed- [27] Ergun S. Fluid flow through packed columns. Chem Eng
ings of the second international conference on multiphase Progr 1952;48:89–94.
flow, 1995. p. 25–32. [28] Kunii D, Levenspiel O. Fluidization engineering, 2nd ed.
[7] Zhang W, Johnsson F, Leckner B. Fluid-dynamic bound- Amsterdam: Elsevier; 1991.
ary layers in CFB boilers. Chem Eng Sci 1995;50(2):201–10. [29] Gogolek PEG, Grace JR. Fundamental hydrodynamics
[8] Le Guevel T, Thomas P. Fuel flexibility and petroleum related to pressurized fluidized bed combustion. Prog
coke combustion at Provence 250 MW CFB. Proceedings Energy Combust Sci 1995;21(5):419–51.
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 567

[30] Abrahamsen AR, Geldart D. Behaviour of gas-fluidized [51] Werther J, Wein J. Expansion of gas fluidized beds in the
beds of fine powders part II: voidage of the dense phase in turbulent regime. AIChE Symp Ser 1994;90:31–44.
bubbling beds. Powder Technol 1980;26(1):47–55. [52] Zijerveld RC, Koniuta A, Johnsson F, Marzocchella A.
[31] Darton RC, LaNauze RD, Davidson JF, Harrison D. Axial solids distribution and bottom bed dynamics for
Bubble-growth due to coalescence in fluidized beds. Trans CFBC application. AIChE Symp Ser 1997;93:97–102.
Inst Chem Eng 1977;55(4):274–80. [53] Pallarès DJ, Johnsson F. Fluiddynamic modeling of large
[32] Clift R, Grace JR. Continuous bubbling and slugging. CFB units. In: Proceedings of the seventh international
Fluidization 1985:73–132. conference on circulating fluidized beds (CFB7). Niagara
[33] Davidson JF, Harrison D. Fluidised particles. Cambridge; Falls, Ont., Canada; 2002. p. 387–94.
1973. [54] Gogolek PEG. Mathematical modelling of fluctuations in
[34] Rowe PN, Yacono CXR. Bubbling behaviour of fine fluidized bed combustion. Kingston, Ont., Canada:
powders when fluidized. Chem Eng Sci 1976;31(12): Queen’s University; 1998.
1179–92. [55] Couderc J-P. Incipient fluidization and particulate systems.
[35] Yacono C, Rowe PN, Angelino H. Analysis of the Fluidization 1985:1–46.
distribution of flow between phases in a gas fluidised bed. [56] Harris BJ, Davidson JF, Xue Y. Axial and radial variation
Chem Eng Sci 1979;34(6):789–800. of flow in circulating fluidized bed risers. Proceedings of the
[36] Shen L, Johnsson F, Leckner B. Digital image analysis of fourth international conference on circulating fluidized
hydrodynamics two-dimensional bubbling fluidized beds. beds 1993:103–7.
Chem Eng Sci 2004;59(13):2607–17. [57] Johnsson F, Leckner B. Vertical distribution of solids in a
[37] Hepbasli A. Estimation of bed expansions in a freely- CFB-furnace. Proceedings of the 13th international con-
bubbling three-dimensional gas-fluidized bed. Int J Energy ference on fluidized bed combustion 1995:671–9.
Res 1998;22(15):1365–80. [58] Johnsson F, Vrager A, Tiikma T, Leckner B. Solids flow
[38] Kobayashi N, Yamazaki R, Mori S. A study on the pattern in the exit region of a CFB-furnace. Influence of
behavior of bubbles and solids in bubbling fluidized beds. geometry. In: Proceedings of the 15th international con-
Powder Technol 2000;113(3):327–44. ference on fluidized bed combustion. Savannah, USA; 1999.
[39] Svensson A, Johnsson F, Leckner B. Bottom bed regimes in [59] Johnsson F, Zhang W, Johnsson W, Leckner B. Optical
a circulating fluidized bed boiler. Int J Multiphase Flow and momentum probe measurements in a CFB furnace.
1996;22(6):1187–204. Circul Fluidiz Bed Technol 1997:652–7.
[40] Johnsson F, Stérneus J, Leckner B, Wiesendorf V, Hartge [60] Leckner B, Golriz MR, Zhang W, Andersson BA,
EU, Werther J, et al. Fluid Dynamics of the bottom zone of Johnsson F. Boundary layers-first measurements in the
CFB combustors. Sixth international conference on circu- 12 MW CFB research plant at Chalmers University.
lating fluidized beds. Würzburg; 1999. p. 191–8. Proceedings of the 11th international conference on
[41] Johnsson F, Zijerveld RC, Leckner B. Air-plenum pressure fluidized bed combustion 1991:771–6.
fluctuations in a circulating fluidized bed boiler. Proceed- [61] van der Meer EH, Thorpe RB, Davidson JF. The influence
ings of the second European conference on CFB 1997: of exit geometry for a CFB with a square sectional riser.
223–30. Proceedings of the 5th international conference on CFB
[42] Johnsson F, Larsson GBL. Pressure and flow fluctuations 1996:575–80.
in a fluidized bed—interaction with the air feed system. [62] Haider A, Levenspiel O. Drag coefficient and terminal
Chem Eng Sci 2002;57:1379–486. velocity of spherical and nonspherical particles. Powder
[43] Sasic S, Leckner B, Johnsson F. Dynamics of gas-solid Technol 1989;58(1):63–70.
fluidized beds—inclusion of the air supply system. In: [63] Geldart D. Estimation of basic particle properties for use in
Proceedings of the 11th International conference on fluid–particle process calculations. Powder Technol
fluidization. Ischia, Naples, Italy; 2004. p. 467–74 1990;60(1):1–13.
[44] Sasic S, Johnsson F, Leckner B. Fluctuations and waves in [64] Arsenijevic ZL, Grbavcic ZB, Garic-Grulovic RV, Zdanski
fluidized bed systems: the influence of the air-supply FK. Determination of non-spherical particle terminal
system. Powder Technol 2005;153(3):176–95. velocity using particulate expansion data. Powder Technol
[45] Wu SY, Baeyens J. Segregation by size difference in gas 1999;103(3):265–73.
fluidized beds. Powder Technol 1998;98(2):139–50. [65] Gibilaro LG, Di Felice R, Waldram SP, Foscolo PU.
[46] Toomey RD, Johnstone HF. Gaseous fluidization of solid Generalized friction factor and drag coefficient correlations
particles. Chem Eng Prog 1952;48(5):220–6. for fluid–particle interactions. Chem Eng Sci 1985;40(10):
[47] Grace JR, Clift R. On the two-phase theory of fluidization. 1817–23.
Chem Eng Sci 1974;141(70):53–62. [66] Win K, Nowak W, Matsuda H, Hasatani M, Bis Z,
[48] Werther J. Influence of the bed diameter on the hydro- Krzywanski J, et al. Transport velocity of coarse particles
dynamics of gas fluidized beds. Fluidiz Fluid-particle Syst in multi-solid fluidized bed. J Chem Eng Japan 1995;28(5):
1974;70(141):53–62. 535.
[49] Olowson PA, Almstedt AE. Influence of pressure and [67] Palchonok GI, Breitholtz C, Thunman H, Leckner B.
fluidization velocity on the bubble behaviour and gas flow Impact of heat and mass transfer on combustion of a fuel
distribution in a fluidized bed. Chem Eng Sci 1990;45(7): particle in CFB boilers. Procedings of the 14th interna-
1733–41. tional conference on FBC 1997:871–88.
[50] Sit SP, Grace JR. Effect of bubble interaction on interphase [68] Werdermann CC. Feststoffbewegung und Wärmeübergang
mass transfer in gas fluidized beds. Chem Eng Sci 1981; in zirkulierenden Wirbelschichten von Kohlekraftwerken;
36(2):327–35. 1992.
 ARTICLE IN PRESS
568 D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569

[69] Johnasson A. PhD. thesis, Chalmers University of Tech- [90] Horio M, Morishita K, Tachibana O, Murata N. Solid
nology, Göteborg, Sweden, 2005. distribution and movement in circulating fluidized beds.
[70] Johansson A, Johnsson F, Leckner B, Gadowski J. Solids In: P.a.L. Basu JF, editor. Circulating fluidized beds
back-mixing in CFB furnaces. In: Proceedings of the eighth technology II. New York, USA: Pergamon Press; 1988.
international conference on circulating fluidized beds. p. 147–53.
China; 2005. [91] Horio M, Ishii H, Nishimura M. On the nature of turbulent
[71] Stérneus J, Johnsson F. Gas mixing in the wall layer of a and fast fluidized beds. J Chem Eng Japan 1992;30:691–7.
CFB boiler. In: Proceedings of the 14th international [92] Andersson BÅ. Heat transfer in the Örebro circulaing
conference on fluidized bed combustion. Vancouver; 1997. fluidized bed boiler Internal report I. 91–28, Chalmers
p. 1237–46. University of Technology, Department of Energy conver-
[72] Zhang W, Johnsson F, Leckner B. Momentum probe and sion; 1991.
sampling probe for measurement of particle flow properties [93] Davidson JF. Circulating fluidised bed hydrodynamics.
in CFB boilers. Chem Eng Sci 1997;52(4):497–509. Powder Technol 2000;113(3):249–60.
[73] Koksal M, Hamdullahpur F. Gas mixing in circulating [94] Peirano E, Begis J, Johnsson F, Leckner B. Gas-solid
fluidized beds with secondary air injection. Chem Eng Res flow computations applied to circulating fluidized beds.
Design 2004;82(8):979–92. Göteborg, Sweden: Chalmers University of Technology;
[74] Savolainen K, Karvinen R. Experimental and numerical 1998.
studies of particle–turbulence interaction and jet penetra- [95] Lackermeier U, Werther J. Flow phenomena in the exit
tion in gas–particle flow. Adv Fluid Mech 2001;29: zone of a circulating fluidized bed. Chem Eng Process
87–96. 2002;41(9):771–83.
[75] Karlsson T, Svensson F. Penetration of air injection in the [96] Zheng Q-Y, Zhang H. Effect of geometry of bed exit (end
riser of a fluidized bed. Göteborg: Chalmers University of effect) on hydrodynamic behaviour of gas–solid flow in
Technology; 1999. CFB combustor. Fluidization VIII. C. L. E. J.-F. Large.
[76] Lücke K. On the influence of mixing on the performance of New York, Foundation; 1995. p. 453.
large-scale amospheric circulating fluidized bed combus- [97] Brereton CM, Grace JR. End effects in CFB hydrody-
tors. Technical University of Hamburg-Harburg; 2003. namics. Circulat Fluidiz Bed Technol 1993;4:137–44.
[77] Hannes JP. Mathematical modelling of circulating fluidized [98] Sanderson WE. The separation effect on the CFBC’s riser
bed combustion. Aachen, Germany: City-Print Verlag exit. DUP science. TU-Delft. Delft, The Netherlands: Delft
GmbH; 1996. University Press; 1994.
[78] Lei H. Hydrodynamic modelling of circulating fluidized [99] Åmand LE, Lyngfelt A, Karlsson M, Leckner B. The
bed reactors. PhD. thesis, Tokyo University of Agriculture reference case for coal at the 12 MW CFB-boiler at
and Technology Koganei, Tokyo, Japan, 1998. Chalmers. Characterisation of the gas and particle phases
[79] Lewis WK, Gilliland ER, Lang PM. Entrainment from in the combustion chamber. Report A97-221. Internal
fluidized beds. Chem Eng Prog Symp Ser 1962;58:65–78. report A97-221; 1995.
[80] Kunii D, Levenspiel O. Entrainment and elutriation from [100] Harris AT, Davidson JF, Thorpe RB. Influence of exit
fluidized beds. J Chem Eng Japan 1969;2:84–8. geometry in circulating fluidized-bed risers. AIChE J 2003;
[81] Kunii D, Levenspiel O. Entrainment of solids from 49(1):52–64.
fluidized beds. Powder Technol 1990;61(2):193–206. [101] Stoess HA. Pneumatic conveying. New York: Wiley
[82] Adanez J, Gayan P, Garcia-Labiano F, de Diego LF. Axial Interscience; 1983.
voidage profiles in fast fluidized beds. Powder Technol [102] Thomas D. Transport characteristics of suspensions.
1994;81(3):259–68. AIChE J 1961;7(3):423–30.
[83] de Diego LF, Gayan P, Adanez J. Modelling of the flow [103] Jones PJ, Leung LS. A comparison of correlations for
structure in circulating fluidized beds. Powder Technol saltation velocity in horizontal pneumatic conveying. Ind
1995;85(1):19–27. Eng Chem Process 1978;17(4):571–5.
[84] Lei H, Horio M. A comprehensive pressure balance model [104] Muschelknautz U, Muschelknautz E. Special design of
of circulating fluidized beds. J Chem Eng Japan 1998;31(1): inserts and short entrance ducts to recirculating cyclones.
83–94. Proceedings of the fourth international confrence on
[85] Walsh PM, Mayo JE, Beer JM. Refluxing particles in the circulating fluidized beds 1991:597–602.
freeboard of a fluidized bed. AIChE Symp Ser 1984:119–28. [105] Muschelknautz E. Fundamentals and practical aspects of
[86] Teplitskiy YS, Ryabov GA. Scaling in a circulating cyclones. Proceedings of the fifth international confrence
fluidized bed: particle concentration and heat transfer on circulating fluidized beds 1994:20–7.
coefficient in a transport zone. Int J Heat Mass Transfer [106] Krambrock W. Modellierung der zeitabhängigen Entga-
1999;42(21):4065–75. sung von kohlepartikeln in der zirkulierenden wir-
[87] Weinell CE, Dam-Johansen K, Johnsson JE. Single-particle belschicht. Aachen, Germany: The Aachen University of
behaviour in circulating fluidized beds. Powder Technol Technology; 1971.
1997;92(3):241–52. [107] Rhodes MJ, Geldart D. A model for the circulating
[88] Wein JW. Das expansionsverhalten von gas/feststoff- fluidized bed. Powder Technol 1987;53(3):155–62.
wirbelschichten bei höheren gasgesch windigkeiten. Har- [108] Yang WC. A model for the dynamics of a CFB loop.
burg, Germany: Technical University of Hamburg; 1992. Proceedings of the second international conference on
[89] Hiller R. Matematische Modellierung der Kohleverbren- circulating fluidized beds technology 1988:181–91.
nung in einer Circofluid-wirbelschichtfeuerungen. Dort- [109] Sundberg RE. The prediction of overall collection effi-
mund, Germany: Dortmund University; 1995. ciency of air pollution control devices from fractional
 ARTICLE IN PRESS
D. Pallarès, F. Johnsson / Progress in Energy and Combustion Science 32 (2006) 539–569 569

efficiency curves. J Air Pollut Control Assoc 1974;24(8): [112] Zhang R, Basu P. A simple model for prediction of solid
758–64. collection efficiency of a gas-solid separator. Powder
[110] Dietz PW. Collection efficiency of cyclone separators. Technol 2004;147(1–3):86–93.
AIChE J 1981;27(6):888–92. [113] Zhang W, Johnsson F. Particle flow pattern in circulating
[111] Avci A, Karagoz I. A mathematical model for the fluidized bed boilers. Thesis for the Degree of Licenciate,
determination of a cyclone performance. Int Commun Department of Energy Conversion in Engineering, Chal-
Heat Mass Transfer 2000;27(2):263–72. mers University of Technology, Göteborg, Sweden, 1992.
Errata in ”Macroscopic modelling of fluid dynamics in large-scale circulating fluidized beds”

Page Equation Reads Shall read

f,d
Fcore
557 48 F f ,d
net =
, H duct
f ,d
Fnetf , d = Fcore
f,d
, H duct ⋅ pentr
f,d

p entr

FHf duct
,d f,d
Fwall , H duct
557 49 k f,d = k f ,d =
Fnetf , d Fnetf , d

H duct H duct
561 63 ∆p refeed = ρ s ⋅ g ⋅ C s ⋅ dh ∆prefeed = g ⋅ Cs ⋅ dh
H retleg H retleg
 Paper II

Pallarès, D., Johnsson, F. (2008)

”Modeling of fuel mixing in fluidized bed combustors”
Accepted for publication in Chemical Engineering Science
 Modeling of fuel mixing in fluidized bed combustors
David Pallarès, Filip Johnsson*
Department of Energy and Environment, Chalmers University of Technology, SE-412 96 Göteborg (Sweden)

Abstract
This paper presents a 3-dimensional model for fuel mixing in fluidized bed combustors. The model accounts for mixing
patterns which were experimentally shown to govern mixing in risers with geometry and operational conditions representative for
furnaces in fluidized-bed combustors. The mixing process is modeled for three different solid phases in the furnace and the
model, which includes *the return leg, can be applied both under bubbling and circulating regimes. The semi-empirical basis of
the model was previously validated in different large-scale fluidized bed combustors and is combined with a model for fuel
particle conversion to obtain the fuel concentration field. Model results are compared with experimental data from the Chalmers
12 MWth CFB combustor, yielding reasonable agreement.
Keywords: Modeling; Mixing; Fuel; Combustion; Fluidization; Fluidized bed

1. Introduction
Fuel mixing has a great influence on the overall performance of fluidized bed combustors. The better the
horizontal mixing of the fuel the more homogenous is the local stoichiometric ratio over the cross section of the
furnace, which in its turn lowers the risk of occurrence of locations with unreacted fuel or oxygen. In the vertical
direction a high mixing rate is important in order to secure long enough contact time between the oxygen and the
fuel particles. Moreover, good mixing is a prerequisite for an even distribution of heat and gas release from the fuel.
Despite the importance of fuel mixing, there is at present a lack of mechanistic models describing this process and
research works have so far been limited to calculate experimental values of the dispersion coefficient [1-4] which is
occasionally correlated to operational parameters. Neither has there been much work done to incorporate mixing of
fuel particles in CFD calculations, excepting the attempt made by Tanskanen [5] which, however, gave unrealistic
results in some locations of the calculation domain. Concerning semiempirical modeling, fuel mixing was included
in a comprehensive model for a CFB combustor [6-8]. Yet, the fuel mixing process is complex and there is a need to
make a focused work on fuel mixing modeling where the fuel model is explicitly validated. Thus, the aim of the
work presented in this paper is to provide a model for fuel mixing in fluidized bed combustors validated with
experimental data representative for fluidized bed combustion obtained during runs with different fuel types. The
aim is to develop a fuel mixing model which can be used in a comprehensive fluidized-bed model as well as to
provide experimental basis for verification of CFD modeling.
The continuous physical changes of fuel particles (such as size and density, and thereby terminal velocity) as they
undergo conversion (i.e. drying, devolatilization and char combustion) lead to constantly changing fuel mixing
behaviours. Shortly after the injection into a fluidized bed combustor a fuel particle is likely to occupy the bottom part
of the riser (furnace) due to its relatively large size and high density, while it has an increasing tendency to populate
the upper freeboard or even being entrained to the return leg as it gets closer to its burn-out time (due to the smaller
size and lower density). Thus, in the present work, a fuel particle conversion model is combined with a (3-
dimensional) fuel mixing model in order to describe the transient fluiddynamics resulting from the constantly
changing physical properties during burnout of a batch of fuel particles. As detailed below the burnout of the fuel
batch is then applied in a continuous approach to simulate the effect of continuous fuel feeding.
It is important to stress the role of fragmentation on both fuel mixing and fuel conversion: With respect to fuel
mixing, fuel fragmentation enhances fuel concentration at higher levels in the freeboard and the flow of externally
recirculated fuel. Concerning fuel conversion, the burnout time of the fuel particles is drastically shortened by
fragmentation which thereby lowers the fuel inventory in the unit. Thus, fuel fragmentation will influence modeled
and measured results on in-furnace parameters such as temperatures, gas concentration and char distribution. Fuel
fragmentation is a complex process which, although is not known in detail, is known to be strongly dependent on the
fuel type and it has indeed been investigated in several studies, e.g. [9,10]. Since there is not yet any general model

*
Corresponding author: [email protected] . Tel.: +46 31 772 1449; Fax: +46 31 722 3592
available for any fuel type which has been verified under industrial conditions and since the focus of this work is on
fuel mixing, fuel fragmentation is not modeled in this present work. Instead fuel fragmentation is taken as input to the
modeling as a given fragmentation pattern. This is done in order to maintain a high degree of transparency in the
modeling procedure. A sensitivity analysis ion the fuel fragmentation is included.

2. Theory
2.1. Fuel mixing model

In general terms, mixing of fuel particles can be modeled on a semi-empirical

basis using the same principles and equations as for the fluiddynamics of the
inert (bulk) particles. Literature on macroscopical fluiddynamics of fluidized bed
boilers [6-8, 11-13] have roughly the same description of the overall solids
pattern in the furnace: in the bottom part a dense bottom bed is established
which, by means of bubble explosions, forms a splash zone of erupted solids in
the form of clusters (cluster phase) which follow a ballistic movement, hence
governing the back-mixing of solids in this region. Under circulating
conditions, a fraction of the solids (disperse phase) is entrained higher up in
the furnace creating a core-annulus structure in the freeboard, with upflow in
the core, net horizontal flow from the core to the downflowing annulus (i.e.
most of the solids back-mixing occurs at the furnace walls). Yet, both the
cluster and the disperse phase can exist all the way up through the furnace,
with the cluster phase dominating in the bottom freeboard (splash zone) and
the disperse phase in the upper freeboard (transport zone). Figure 1 illustrates
the main solids movements in the furnace. Table 1 lists the assumptions made Figure 1: Zones of a CFB riser with
corresponding mixing patterns
in this work with respect to the governing mixing mechanisms in the
horizontal and vertical directions for the bottom bed and for the two phases of
the freeboard (cluster and disperse phase). These fuel mixing mechanisms are further described below.
The 3-dimensional mesh used to discretize the furnace applies a finer mesh in regions known to exhibit large
gradients (e.g. splash zone). At present, perfect mixing is assumed in the return leg and this is therefore not
discretized. Note that, at present stage, the fuel mixing model does not account for interactions with the inert solids
flow, i.e. fuel mixing is modeled independently from the bulk flow. This results in the same type of error as in
fluiddynamical models which do not take into account
Mixing mechanism interactions between different size intervals of the bed
Vertical Horizontal
material. Although there might be a general increase of
the fuel hold up as a consequence of the interaction with
Bottom bed Perfect mixing
Diffusion the (mostly finer) inert phase, the extent of this effect is
Cluster phase Ballistic
limited and, as will be shown, the proposed model is
Disperse phase Core/annulus
able to provide reasonable agreement with experimental
data. However, further development of the model will
Table 1: Solids mixing mechanisms in a fluidized
bed furnace applied in this work include interactions between fuel and bulk particles
through a particle interaction model given in [14].
Despite that the real mixing mechanism behind horizontal mixing in the bottom part of the riser is strongly
convective, the experimentally verified existence of toroidal flow structures around each main bubble path [15,16]
makes it possible to macroscopically approximate the horizontal mixing in the dense bottom bed as a diffusion
process reading:

∂C fuel (x, y, z 0 )
∂t
( )
= ∇ ⋅ Dhoriz ∇C fuel (x, y, z 0 ) + S feed for z0 Hb (1)

with the source term Sfeed corresponding to the fuel feeding. This is a common method in literature (see e.g. [17])
with several experimental values and correlations for the horizontal diffusion coefficient Dhoriz available (cf. [2]) but
differing up to two orders of magnitude between investigations carried out under similar conditions. There is little
experimental fuel mixing data from fluidized bed combustors with the exception of the work given in [2,18] where
Dhoriz is estimated to be about 0.1 m2/s. Thus, this value is used in the simulations presented in the present work.

2
 This work assumes perfect vertical mixing in the bottom bed. This is a reasonable assumption according to
experimental works ([2] and references therein) showing that mixing in this region is at least one order of
magnitude faster in the vertical than in the horizontal direction together with the fact that dense bottom beds are
usually no higher than a few decimeters compared to several meters in the horizontal direction. In addition, an
experimental study [15] providing simulated steady-state fuel concentrations in a 2-dimensional bed showed almost
perfect vertical mixing in the dense bed for most types of tracer (fuel) particles and operational conditions, with the
only exception being runs for which the fuel-to-emulsion ratio Arf /Arem is higher than 104 at low fluidization
velocities (under 1.5 m/s). Under these conditions tracer particles show a tendency to float on the dense bed surface,
as reported also in [19].
As indicated above, the presence of the fuel in the freeboard is divided into a cluster and a disperse phase. Thus,
the total fuel concentration in the freeboard can be expressed as a sum of these phases:

C fuel = C clust
fuel + C fuel
disp
(2)
z>Hb

The cluster phase consists mainly of non-entrainable particles which are ejected into the freeboard from the bed
which then fall back into the dense bed. Thus, this phase corresponds to a vertical mixing phenomenon governed by
a ballistic movement, which leads to an exponential decay in vertical concentration of the fuel particles, i.e,

fuel ( x o , y o , z ) fuel ( x o , y o , H b ) ⋅ e
− a ⋅( z − H b ) (3)
C clust = C clust
z >Hb

Several experimental correlations for the decay constant a have been proposed in literature for fluidized bed
combustors operated under bubbling as well as circulating conditions [20-22], most of them adopting the form:
u (4)
a = c⋅ t
ug
Based on data from pressure measurements in several large-scale units, the coefficient c was correlated in [23] to
a value around 4 m-1 (although this value was later found by the authors to slightly increase as the bottom bed height
decreases and hosts less vigorous bubbles). Note that (as for the disperse phase decay constant, K, see below)
coefficients are obtained from pressure measurements, i.e. they represent the sum of all in-bed solids (including inert
phase). Under the assumption indicated above (i.e. neglecting interaction between phases), these empirical
coefficients can be applied also to the fuel phase.
The continuous separation of disperse-phase solids to the furnace walls gives the above-mentioned core-annulus
flow structure (cf. [24,25]). The thickness of the wall layers defining the border between the core and annulus
regions has been studied in several in several works, with the most exhaustive compilation of experimental data
given in [26] and correlations from different authors compared in [12].
Modeling the back-flow in terms of a falling film approach (which is widely accepted) where the disperse phase
in the core region flows at the slip velocity yields an exponential decay in solids concentration with height in the
upflowing core region (see [27] for details), i.e.

fuel ( x o , y o , z ) fuel ( x o , y o , H b ) ⋅ e
− K ⋅( z − H b )
C disp = C disp (5)
z >Hb

with (x0,y0) in the core region and the decay constant, K, taking the form:
4⋅k (6)
K=
Deq ⋅ (u g − u t )

In a study [23] gathering data from four different CFB boilers the mass transfer coefficient, k, in Eq. (6) is found
to depend linearly on Deq, which leads to the following correlation for the decay constant, K, in Eq. (5):
0.23 (7)
K=
u g − ut

The decreasing fuel upflow with height is due to the net lateral flow from core to wall region, i.e. fuel back
mixing is assumed to mainly occur at the furnace walls. Thus, in a core cell i with a given height dz, the net
differential fuel flow leaving the core cell upflow and joining the wall-layer downflow can be expressed as:
Flat , i = −dFcore, i = − Ai ⋅ (u g − ut , fuel ) ⋅ dC disp
fuel , i = Ai ⋅ (u g − u t , fuel ) ⋅ K ⋅ e
− K ( z −Hb )
⋅ dz (8)

3
 In net terms, this lateral flow at each
cross section of the riser freeboard is
assumed to follow the patterns illustrated in
Fig. 2, for circular and rectangular cross-
sectional geometries. The core-to-annulus
fuel flow feeds the downflow in the wall a) Circular cross section b) Rectangular cross section
region with fuel particles all along the riser.
When reaching the bottom bed, this Figure 2: Horizontal core to wall net solids transfer in the freeboard
downflow which joins the horizontal
diffusive mixing process described above is embodied within the source term Sfeed in Eq. (1).
Finally, some of the upflowing particles in the core region which reach the height of
the exit duct experience a backflow effect, through which only a certain fraction of
these particles reaches the cyclone. A proper modeling of this effect requires the use of
CFD tools, but experimental correlations for estimating the backflow in standard exit
configurations is given in [12,24,25,28]. From any of these correlations, an entrainment
probability p can be estimated. Since the correlations were found in CFB combustors
with a single exit duct an assumption has to be made in order to handle a more general
case where n exit ducts exist, which is represented by Eq.(9.a). Moreover, the backflow
effect can be assumed to take place in two steps in series: firstly, some of the upflowing
particles in the core follow the gas flow and leave the core upflow towards the exit duct
Figure 3: Mechanisms in the whereas the other part are separated to the walls-layers and secondly, once in the exit
backflow effect duct, some particles follow the gas stream all the way into the cyclone while the rest fall
back down into the riser and join the downflow in the wall-layers. These two steps can
be expressed as probabilities p1 and p2, as illustrated in Fig. 3. Thus, with the single-exit entrainment probability p
given by experimental correlations and knowing that p*= p1 p2, the assumption p1=p2, leads to the values of p1 and p2
given by Eq. (9.b):

p * = 1 − (1 − p ) (9.a) with p1 = p 2 = (9.b)

n
p*

As far as the externally recirculated fuel particles are concerned, their residence time in the return leg is
calculated as the sum of the residence times in the cyclone, downcomer and particle seal. The residence time of fuel
particles in the cyclone can be calculated according to [29]. The fuel residence time in the downcomer and particle
seal is easily calculated with the assumption that it equals the residence time of the bulk solids in the return leg
according to Eq. (10).
V fluidized ⋅ ρ s ⋅ 1 − ε return
τ downcomer &seal =
inreturnleg leg (10)
Fnet , s

In the present work, the net solids circulating flow required in Eq. (10) has been modeled according to [12].
There are also experimental methods to estimate this value in industrial CFB combustors as listed in [30]. Finally,
after flowing through the return leg, fuel particles (if not burned out) are refed into the bottom part of the riser. This
is implemented through the source term Sfeed in Eq. (1).

2.2. Fuel conversion model

In the present work, the fuel conversion model presented in [31] is chosen for modeling the drying and
devolatilization processes since this model yields low calculation times and yet is shown to give satisfactory
agreement with experimental data. Assuming quasi-steady state, this 1-dimensional fuel conversion model provides
an analytical transient solution to the formulation of the energy equation in a fuel particle whose geometry is
approximated through a plate, cylinder or sphere (r=0, 1 or 2 respectively in Eq. (11)). Thus, defining a
dimensionless spatial coordinate, , and temperature, , the energy equation for the quasi-steady state reads:
1 ∂ ∂θ 1 ∂θ (11)
ξr + αβ r =0
ξ r ∂ξ ∂ξ ξ ∂ξ

4
where and are constants expressing the magnitude of the convective term (see [31] for details). While the first
term represents the heat diffusion from the surroundings into the fuel particle, the convective term describes the
heating up of the moisture and volatiles as they flow towards the surface of the fuel particle. When applying
Eq. (11) the following boundary conditions (for simplicity not expressed here in dimensionless terms) are imposed:

T r = drying front
= 100 ˚C (12)

− λ part
∂T
∂r r = rpart
(
= heff ⋅ T r = rpart
− T∞ ) (13)

where heff accounts for both convective and radiative heat transfer between the particle and its surrounding.
Expressions for the analytical solution of Eq. (11) with the above boundary conditions are listed in [31].
Differentiation of the solution obtained provides temperature gradients which are used to calculate heat fluxes at the
drying front and particle surface, providing (once a time step is defined) the displacement of the drying front and
the new surface temperature (through Eq. (13)). Thus, this gives an updated set of boundary conditions for solving
the next time step.
From the dynamical evolution of the temperature field inside the fuel particle obtained after solving Eq. (11), all
of the drying, devolatilization and char combustion rates can be calculated. According to the approach described in
[32], the effective char combustion rate, Reff, is modeled through a combination of two processes which,
respectively, are governed by the kinetic rate, Rkin (determined through an Arrhenius-type correlation) and the mass
transfer coefficient, hm (determined through a Sherwood-type correlation) yielding:
1 (14)
Reff =
1 1
+
Rkin hm

which is to be used in the general expression for coal combustion (first order reaction is assumed) reading:
d mC (15)
= −Ω ⋅ Reff ⋅ Apart ⋅ C O2 , ∞
dt
With this, the dynamical mass loss of the fuel particle studied can be obtained (the modeled mass loss for a
10 mm i.d. brown coal particle is exemplified in Fig. 4). Neglecting particle shrinking and thus assuming no
changes in particle size due to drying and devolatilization, the progress of the fuel particle size and density during
combustion can be calculated and, from these, the terminal velocity is obtained (as shown in normalized values in
Fig. 5). As explained below, the progress of the fuel particle terminal velocity shown in Fig. 5 enables the mixing
model given in the previous section to account for the continuous change in the fluiddynamical properties of the fuel
particle. Depending on the fuel type, fuel fragmentation might play a major role in the conversion kinetics and
thereby significantly influence the final results and should be taken into account by the fuel conversion model (the
assumed fragmentation patterns applied in this work are given below).

Figure 4: Evolution of the relative mass of a fuel particle Figure 5: Evolution of the relative size, density and
terminal velocity of a fuel particle

5
 2.3. Method for a steady-state solution

Obviously, modeling the behavior of a batch of fuel particles requires a transient simulation during the whole
burnout time. An efficient way to proceed for the case of fuel particles in fluidized bed combustion consists in
modeling the mixing of a batch of fuel particles during its burn-out time. Applying the above-described model for fuel
mixing, the fuel distribution (expressed in [particles/m3]) at each time step can be calculated using the corresponding
value of the terminal velocity calculated from the fuel particle conversion model. Applying then the pertinent value for
the fuel particle mass, the concentration field can be converted to [kg/m3]. Thus, the spatial distribution of the fuel
concentration C originated by the fuel batch at any time step ti is known (given in the first row in Fig. 6). Having this,
a continuous feeding of fuel into a combustor can be simulated by a ‘continuous batch’ approach, in which a new
batch is fed to the unit at each time step. This procedure is illustrated in Fig. 6 and leads to a resulting total fuel
concentration in each time step equal to the sum of the values in the corresponding column. Thus, a stationary
concentration field is obtained after a time interval equal to the burnout time of the fuel particles, and this value is
equal to the sum of all intermediate values in each time step.

Figure 6: The scheme applied to calculate the steady state distribution of fuel concentration

It must be noted that the use of this approach is only suitable to problems where the influence of a certain batch
on the previous batches can be neglected, which is considered reasonable for fuel in fluidized bed combustors
(where it typically represents only between 1 and 5% of the total amount of solids in the combustor).

3. Experiments
The experimental data used for model validation in this work is taken from the Chalmers 12 MWth CFB combustor.
The size of the combustion chamber is 1.7×1.7×13.5 m with two of the walls covered with refractory lining (0.11 m
thick) up to 2 m above the air distributor and the other two covered all along their height. The fuel is fed by gravity
from the fuel chute at a height of 1.1 m above the air distributor (see [33] for a detailed description of the unit). The
mean size of the inert bed material is 320 m and the average height of the dense bottom bed during the tests was
estimated (from pressure drop) to be about 0.5 m.
The present work contains experimental data from Coal Peat Wood
runs with four different fuel types: two Polish Moisture 14.80 25.00 43.10
Proximate

bituminous coals, wood chips and peat (fuel analysis Volatiles 31.36 49.97 46.10
(%)

for these fuels are given in Table 2). Operational Char 47.24 22.03 10.40
conditions were kept as similar as possible for all runs: Ash 6.60 3.00 0.40
a fluidization gas velocity of 3.7 m/s, secondary air
injection at a rate of 1.3 kg/s through 13 ports located C 79.8 59.5 50.6
Ultimate

O 12.6 31.5 43.2

daf (%)

in the rear and front walls at a height of 2.1 m above

the gas distributor and a heat flow supplied with fuel H 5.3 6.4 5.9
of 9.4 MWth. This corresponds to an air ratio of about N 1.56 2.1 0.22
1.2, with a measured oxygen concentration on wet flue S 0.72 0.53 0.04
gases of approximately 3.3%. The temperature in the LHV [MJ/kg ar] 29.0 20.8 17.6
bottom of the furnace was kept at 850˚C. Part of the
Table 2: Analyses of fuels used in runs shown in
data sampled during these runs (including part of the Figures 11 to 13
experimental data used in this work) has been
previously published in other studies (see [34-36]).

6
4. Results
Figure 7a shows measured values from 9 points distributed across the cross-section at a low level (h=0.52 m) in
the furnace. Fuel is Polish coal with a mean size of 8 mm, with moisture, volatiles, char and ash contents (as
received) of 17%, 30%, 44% and 9%, respectively. The fuel was fed at a rate of 0.43 kg/s from the fuel chute, placed
at the center of the front wall (Fig. 7a), while the inlet of the return leg (fuel re-feed) is located to the right of the rear
wall. The data yield fuel concentration values from 2.3% to 4.3%, with a cross-sectional average of 3.2%. The
distribution of the fuel concentration given by the model in the same cross section is shown in Fig. 7b. In the model,
fuel particles are assumed to undergo fragmentation after 75% of the devolatilization time, yielding 10 pieces, which
is the pattern found in [37] for coal (although it should be emphasized that fragmentation is strongly fuel dependent
and difficult to generalize).

Fuel fraction in bottom region

0.05

Simulated [kg fuel/kg]

0.04

0.03

0.02

0.01

0
0 0.01 0.02 0.03 0.04 0.05
Experimental [kg fuel/kg]

a) Experimental data b) Modeled data

Figure 7: Fuel concentration at bottom region of the Chalmers combustor Figure 8: Comparison between modeled and
experimental data on fuel concentration

In the modeled fuel distribution, fuel concentration values range from 1.7 to 3.7 %, with a cross-sectional
averaged value of 2.5 %. The average error between modeled and experimental values is 19%. A comparison
between experimental and simulated values at the 9 sampling points is shown in Fig. 8, where a general tendency of
the model to slightly underestimate fuel concentration values is observed for all points. There may be several
reasons for this (or a combination of them). One probable explanation is an overestimation of the fragmentation
leading to an overestimation of the fuel conversion rate and thereby resulting in a lower fuel inventory.
Figure 9 shows the modeled fuel concentration values for the same test at a height of 7 m above the air
distributor. As seen, the fuel concentration in the core is much lower than in the walls (in agreement with literature,
e.g. [38]), where corner effects are significant. It is also seen that the wall with the fuel chute (to the left) gives
higher fuel concentration values than at the other walls, also this is an expected result.

Figure 9: Modeled fuel concentration at a height of 7m

7
 Modeled data from a case burning wet wood
chips (with an assumed proximate analysis of
45% moisture, 45% volatiles, 9% char, 1% ash)
is chosen to exemplify the importance of
accounting for the changes in the size and
density of fuel particles during their conversion.
During the first seconds in the furnace the wet
wood particles are relatively dense and remain
mostly in the bottom region of the riser while
releasing their moisture. As conversion
progresses the wood particles become lighter and
dryer and volatiles and later on char combustion
products are released at the same time that fuel
particles tend to occupy higher locations in the
riser. This is seen in Fig. 10, which shows the
modeled cumulative releases of moisture,
volatiles and char combustion products. It can be
observed that, in this simulated example, 95% of
Figure 10: Modeled cumulative fuel field releases
the moisture and 79% of the volatiles are
expected to be released below a height of 4 m in
the riser whereas only and 60% of the char combustion products are produced below his level. As far as coal is
concerned, its much lower moisture and volatile contents make physical changes in the fuel particles due to drying and
devolatilization of lower significance compared to those in wood, leading to a less pronounced effect than that shown
in Fig. 10.
Experimental data in Figs 11 and 12 (represented by symbols) were sampled under runs in the Chalmers 12 MWth
CFB boiler with different fuel types (with proximate and ultimate analyses as listed in Table 2). In all runs, the load
was kept around 8 MWth and bed material samples were taken at different heights in the riser and analyzed with respect
to their char content. In addition, the vertical pressure profile was measured along the furnace height with denser
spacing between the pressure taps in the lower region of the furnace where the strongest gradients in the solids
concentration are expected. Combining the results of
the bed samples and the pressure drop profile (i.e. the
vertical solids concentration) gives the experimental
cumulative mass of char in the vertical direction of the
riser shown in Fig. 11. The modeled data show a
satisfactory agreement with these data and follow the
trend which, as expected, gives coal-firing the largest
char inventory in the riser (about 51 kg measured and
44 kg modeled), much larger than that found under
peat- and wood-firing (with the latter yielding the
lowest char inventory: approximately 2 kg in both
experiments and simulations). Two main reasons lie
behind this large difference: the higher char content in
coal compared to wood (see Table 2) and the lower
reactivity of coal char compared to wood char (due to
the higher activation energy and lower particle voidage
and intrinsic area of coal char) which together make
the burnout time of coal char particles significantly
longer than that of wood char. The peat char content
and reactivity are between those of wood and coal
(although closer to wood), thus leading to an
intermediate inventory between those of wood and coal
in Fig. 11. As mentioned above, modeled data vary
strongly depending on the fragmentation pattern used,
Figure 11: Comparison between modeled and experimental data on so a fragmentation pattern could be used as fitting
vertical cumulative char mass
parameter between simulated data and experiments for

8
each particular run/fuel type. Thus, a high level of agreement would be obtained for all fuel types if such a fitting
exercise is applied separately to each run. Instead, the procedure has been based on using a common (i.e. regardless of
the type of fuel) fragmentation pattern as fitting parameter for all simulations shown in Figs 11, 12 and 13. The
obtained fragmentation pattern consists of the fuel particle falling apart into 5 equal pieces after 20 seconds (i.e. when
the fuel particle still undergoes devolatilization), which is a realistic pattern according to those found experimentally
(e.g. [37, 39]). For comparison and illustration of the sensitivity of the results to the fragmentation pattern fed to the
model, model results from the case burning coal are compared in Figs. 11 to 13 with modeled data in which
fragmentation of the fuel particle is assumed not to occur. As seen in the non-fragmented data curve (marked by an
arrow in Fig. 11), the char inventory in the unit increases substantially when fragmentation is not included (almost
100% increase in the case shown). This is due to that without fragmentation the burnout time predicted by the fuel
particle conversion increases as a consequence of the decreased char surface. Thus, since char combustion rate is
proportional to the char surface in the model, fragmentation (increased char surface) results in an enhanced conversion
rate and thereby to a decreased char inventory for a given fuel and operational conditions.
Figure 12 shows the dependency of the mean
size of the char particles on height in furnace. As
seen, the size distribution of wood particles shows a
strong difference between those in the dense bottom
bed which are close to the size at the feeding point
(about 1 cm) and those much smaller populating the
freeboard. This difference is well predicted by the
model but can only be confirmed by one
experimental data point (4.2 mm at a height of
0.56 m) and is not observed in the same proportions
under peat- or coal-firing (in this sense, the model
also follows the experimental data correctly).
Before analyzing the three curves presented, it must
first be noted that particle size is mainly decreased
due to fragmentation and char combustion,
assuming char combustion takes place on surface of
the char particle (which is the usual case in fluidized
bed combustion) and not inside the whole particle. Figure 12: Comparison between modeled and experimental data
Thus, fuel particle size (which is largest for wood on vertical distribution of char particle size
chips) does not notably decrease under the drying
and devolatilization processes, but the fuel particle density does (see Fig. 5). Bearing in mind this and the high moisture
content in the wood chips (see Table 2) for which there is a significant increase in density, wood fuel occupies the
bottom part of the riser until its size and density has decreased enough to be entrained up into the freeboard. On the
other hand, coal and peat particles have a lower terminal velocity than the large wood chips used and occupy both the
bottom bed and the freeboard directly from the feeding stage, leading to a more homogeneous fuel particle size
distribution. As seen in Fig. 12, when fuel fragmentation is not included, the fuel particle size increases at all heights in
the furnace compared to the case with fragmentation included, with the highest increase taking place in the dense bed.
This is an expected result, since without fuel fragmentation fuel particles remain relatively large for a longer time in the
dense bed. Concerning the freeboard, without fuel fragmentation the largest fuel particles are those with terminal
velocities not much below the gas velocity, which are larger than those found when accounting for fuel fragmentation.
For the case shown, omitting the fuel fragmentation results in a 36% increase in the average fuel size in the furnace.
Finally, Fig. 13 shows the normalized cumulative char surface along the height of the furnace of the Chalmers 12
MW CFB boiler, with experimental data taken from [36]. Bituminous coal with 40% of volatiles and a mean particle
size around 1 cm was used as fuel. Combining experimental data on fuel concentration (see above) and fuel mean size
at several heights, the cumulative char particle surface can be obtained. This variable, together with the oxygen
concentration gives information on how the char combustion is distributed along the riser. As seen in Fig. 13, a good
agreement is obtained between normalized experimental and simulated data for the fragmented coal case. When no
fragmentation is accounted for the curve becomes steeper, meaning that more char surface is located in the bottom
region which implies that the model predicts a higher proportion of the combustion taking place in the bottom region
(provided enough oxygen is available). While only about half (52%) of the total char surface is located in the dense
bottom bed when fragmentation is accounted for, omitting fragmentation results in that this value increases to 82%.

9
 For further validation of the model presented
in this work, and bearing in mind the importance
of fuel fragmentation on the fuel mixing model
results, there is a strong need for either
experimental fragmentation data of the fuels
used or a proper fuel fragmentation modeling
(although fragmentation is known to be very
specific to the fuel type and influenced by
different factors such as moisture and volatiles
content in fuel, geometry of the fuel particle,
temperature and gas velocity surrounding the
fuel particle). However, the model gives
reasonable agreement with experimental data
when typical fragmentation patterns are used
and the above possibility to make a common
fitting of the three different fuels indicate that, at
least for these fuels, the fragmentation process is
of the same character. It should also be noted
that although fuel fragmentation strongly
Figure 13: Comparison between modeled and experimental data on influences the in-furnace results, the importance
vertical cumulative char surface
for the overall CFB process may be less
provided similar overall burnout in the furnace. In addition, it should be noted that the dynamics of the inert solids are
not taken into account in the model either than within the bottom bed. A further development of this model should
allow interaction between the fuel and inert solid fractions.

5. Conclusions
A model for solids mixing in a fluidized has been applied to fuel particles in combination with a fuel conversion model
in order to account for physical changes that fuel particles undergo during conversion. The modeled fuel concentration
values differ around 20% compared to experimental data and a common fragmentation pattern can be adjusted so that
simulated distributions of fuel particles size show also a good agreement to measured values for 3 different fuel types.
Fuel fragmentation (which is not modeled but is an input in the form of a given empirical pattern from experiments) is
found to have a large influence on the results and its proper modeling is thereby crucial for a more accurate modeling
of the spatial distribution of the fuel inventory within the unit. Yet, the importance of fuel fragmentation on the overall
CFB process may be less than the one on fuel inventory and distribution which this paper focuses on.

Acknowledgments
The authors gratefully acknowledge financial support from the Swedish Energy Agency, the EU Research Fund for
Coal and Heat under contract RFC-CR-03001 and Metso Power Oy.

Notation

a Splash zone’s decay constant [m-1] F Solids flow [kg/m3]

A Area [m2] hm Mass transfer coefficient [m/s]
Ar Archimedes number [-] Hb Dense bed height
c Proportionality constant [-] k Mass transfer coefficient [m/s]
C Concentration [kg/m3] K Transport zone’s decay constant [m-1]
Deq Equivalent diameter [m] m Mass [kg]
2
Dhoriz Diffusion coefficient [m /s] n No. of exit ducts [-]

10
 p Entrainment probability with 1 exit duct T Temperature [C]
*
p Entrainment probability with n exit ducts ug Gas velocity [m/s]
p1 Disengagement probability at the riser ut Terminal velocity [m/s]
p2 Entrainment probability at the duct V Volume [m3]
r Geometrical scalar in Eq. (11) [-] Constant in Eq. (11)
R Rate [m/s] Constant in Eq. (11)
3
Sfeed Source term [kg/m s]

Greek letters
Voidage Dimensionless spatial coordinate
Conductivity [W/m·s] s Solids density [kg/m3]
τ Residence time [s] Dimensionless temperature
Conversion factor [kg C/mole O]

Subscripts and superscripts

clust Cluster phase i Cell index
core Core region kin Kinetical
disp Disperse phase lat Lateral
eff Effective part Particle
fuel Fuel Surrounding

References
1. Schlichthaerle, P., Werther, J. 2001. “Solids mixing in the bottom zone of a circulating fluidized bed”. Powder
Technology, 120 (1-2), pp. 21-33.
2. Niklasson, F., Thunman, H., Johnsson, F., Leckner, B. 2002. “Estimation of solids mixing in a fluidized bed
combustor”. Industrial and Engineering Chemistry Research, 41 (18), pp. 4663-4673.
3. Xiang, Q., Huang, G., Ni, M., Cen, K., Tao, T. 1987. “Lateral dispersion of large coal particles in an industrial-
scale fluidized bed combustor”. Proc. of the 9th Int. Conf. on FBC, Montreal, pp. 771–776.
4. Xiao, P., Yan, G., Wang, D. 1998. “Investigation on horizontal mixing of particles in dense bed in circulating
fluidized bed”. Journal of Thermal Science, 7 (2), pp. 78-84
5. Tanskanen, V. 2005. “CFD study of penetration and mixing of fuel in a CFB furnace”. Master Thesis, Dept. of
Energy and Environment, Lappeenranta Univ. of Tech. (Finland).
6. Knöbig, T. 1998. “Three-dimensional modeling of circulating fluidized bed combustion”. PhD Thesis, Tech.
Univ. of Hamburg-Harburg (Germany). Shaker Verlag, Aachen (Germany). ISBN 3-8265-4480-3.
7. Hartge, E.-U., Luecke, K., Werther, J. 1999 . “The role of mixing in the performance of CFB reactors”.
Chemical Engineering Science, 54 (22), pp. 5393-5407.
8. Luecke, K., Hartge, E.-U., Werther, J. 2004 . “A 3D model of combustion in large-scale circulating fluidized
bed boilers”. International Journal of Chemical Rector Engineering, 2, A11.
9. Chirone, R., Massimilla, L., Salatino, P. 1991. “Comminution of carbons in fluidized bed combustion”.
Progress in Energy and Combustion Science, 17 (4), pp. 297-326.

11
10. Salatino, P., Massimilla, L. 1989. “A predictive model of carbon attrition in fluidized bed combustion and
gasification of a graphite”. Chemical Engineering Science, 44 (5), pp. 1091-1099.
11. Hannes, J. 1996. “Mathematical Modelling of Circulating Fluidized Bed Combustion”. City-Print Verlag
GmbH, Aachen (Germany). ISBN 3-88817-002-8.
12. Pallarès, D., Johnsson, F. 2006. “Macroscopic modeling of fluid dynamics in large-scale circulating fluidized
beds”. Progress in Energy and Combustion Science, 32 (5), pp. 539-569.
13. Lei, H. 1998. “Hydrodynamic modelling of circulating fluidized bed reactors”. PhD Thesis, Tokyo Univ. of
Agric. and Tech. (Japan).
14. Palchonok, G., Breitholtz, C., Thunman, H., Leckner, B. 1997.“Impact of heat and mass transfer on combustion
of a fuel particle in CFB boilers”. Proc. of the 14th Int. Conf. on FBC, Vancouver, pp. 871-888.
15. Pallarès,D., Johnsson,F. 2006. “A novel technique for particle tracking in cold 2-dimensional fluidized bed -
simulating fuel dispersion”. Chemical Engineering Science, 61, pp. 2710-2720.
16. Xiao, P., Yan, G., Wang, D. 1998. “Investigation on horizontal mixing of particles in dense bed in CFB units”.
Journal of Thermal Science, 7 (2), pp. 78-84.
17. Berruti, F., Scott, D.S., Rhodes, E. “Measuring and modelling lateral solid mixing in a 3-D batch gas-solid
fluidized bed reactor”. Canadian Journal of Chemical Engineering, 64 (1), pp. 48-56.
18. Xiang, Q., Huang, G., Ni, M., Cen, K., Tao, T. 1987. “Lateral dispersion of large coal particles in an industrial-
scale fluidized bed combustor”. Proc. of the 9th Int. Conf. on FBC, Boston, pp. 546-553.
19. Stubington, J.F., Chan, S.W., Clough, S.J. 1990. “A model for volatile release into a bubbling fluidized bed
combustor”. AIChE Journal, 36, pp. 75-83.
20. Lewis, W.K., Gilliland, E.R., Lang, P.M. 1962. “Entrainment from fluidized beds”. Chemical Engineering
Progress Symposium Series 58, pp. 65-78.
21. Kunii, D., Levenspiel, O. 1990. “Entrainment of solids from fluidized beds”. Powder Technology, 61 (2),
pp. 193-206.
22. Lei, H., Horio, M. 1998. “A comprehensive pressure balance model of circulating fluidized beds”. Journal of
Chemical Engineering of Japan, 31 (1), pp. 83-94.
23. Johnsson, F., Leckner, B., 1995. “Vertical distribution of solids in a CFB furnace”. Proc. of the 13th Int. Conf.
on FBC, Orlando, pp. 671-679.
24. Werther J. 1993. “Fluid Mechanics of Large-Scale CFB Units”. Circulating Fluidized Bed Technology IV,
pp. 1-14.
25. Sanderson, W.E. 1994. “The separation effect on the CFBC'
s riser exit”. Delft University Press Science, TU-
Delft (The Netherlands).
26. Johnsson, F., Zhang, W., Leckner, B. 1995. “Characteristics of the formation of particle wall-layers in CFB
boilers”. Proc. of the 2nd Int. Conf. on Multiphase Flow, pp. 25-32.
27. Davidson, J.F. 2000. “Circulating fluidised bed hydrodynamics”. Powder Technology, 113 (3), pp. 249-260.
28. Werdermann, C.C. 1993. “Feststoffbewegung und Wärmeübergang in zirkulierenden Wirbelschichten von
Kohlekraftwerken”. PhD Thesis, Tech. Univ. Hamburg-Harburg (Germany).
29. Kang, S.K., Kwon, T.W., Kim, S.D. 1998. “Hydrodynamic characteristics of cyclone reactors”. Powder
Technology, 58 (3), pp. 211-220
30. Edvardsson, E., Åmand, L.-E., Thunman, H., Leckner, B., Johnsson, F. 2006. “Measuring the external solids
flux in a CFB boiler”, Proc. of the 19th Int. Conf. on FBC, Vienna.
31. Thunman, H., Davidsson, K., Leckner, B. 2004. “Separation of drying and devolatilization during conversion of
solid fuels”. Combustion and flame, 137 (1-2), pp. 242-250.
32. Field, M.A., Gill, D.W., Morgan, B.B., Hawksley, P.G.W. 1967. “Combustion of pulverized coal”. The British
Coal Utilization Research Association, Surrey (England).

12
33. Leckner, B., Golriz, M.R., Zhang, W., Andersson, B.Å., Johnsson, F. 1991. “Boundary layers - first
measurements in the 12 MW CFB research plant at Chalmers University”. Proc. of the 11th Int. Conf. on
FBC, Montreal, pp. 771–776.
34. Åmand, L.-E. 1996. “Nitrous oxide emission from circulating fluidized bed combustion”. PhD Thesis,
Chalmers Univ. of Tech. (Sweden).
35. Lyngfelt, A., Åmand, L.-E., Leckner, B. 1995. ”Low N2O and SO2 emissions from circulating fluidized bed
boilers”. ”. Proc. of the 13th Int. Conf. on FBC, Orlando, pp. 1049-1057.
36. Lyngfelt, A., Åmand, L.-E., Leckner, B. 1996. “Progress of combustion in the furnace of a circulating fluidized
bed boiler”. Proc. of the 26th Int. Symp. on Combustion, Naples, pp. 3253-3259.
37. Zhang, H., Cen, K., Yan, J., Ni, M. 2002. “The fragmentation of coal particles during the coal combustion in a
fluidized bed”. Fuel, 81 (14), pp. 1835-1840 .
38. Zhang, W., Johnsson, F., Leckner, B. 1995. “Fluid-dynamic boundary layers in CFB boilers”. Chemical
Engineering Science, 50 (2), pp. 201-210.
39. Sreekanth, M., Kolar, A. and Leckner, B. 2006. ”Effect of shape and size of wood on primary fragmentation in
a laboratory scale fluidized bed combustor”. Proc. of the 19th Int. Conf. on FBC, Vienna.

13
2
 Paper III

Pallarès, D., Johnsson, F. (2008)

”Macroscopic modeling of the gas phase in fluidized bed combustion”
Submitted for publication
 Macroscopic modeling of the gas phase in
fluidized bed combustion
David Pallarès*, Filip Johnsson
Department of Energy and Environment, Chalmers University of Technology, SE-412 96 Göteborg (Sweden)

Abstract
A model for gas phase mixing in fluidized bed boiler furnaces (risers) is presented. The model takes its basis in a
description of the dynamics of the dense bottom bed which strongly govern the gas mixing up through the furnace. Thus it is
the opinion of the authors that a time-resolved approach is required to link the modeling to the physics of the underlying
processes determining the gas mixing. As output, the model gives the fluctuating flux of gas species, in contrast to the
classical modeling approach which is limited to time-averaged gas fluxes.
The dynamical approach allows assumption of the volatile combustion system as transport-controled but at the same time
avoids complete consumption of either oxygen or combustible gases in each modeled cell. Instead, time-resolved analysis
enables application of a realistic criterion for the mixing: reactants must coincide in both space and time in order to react.
While fitted kinetics is strongly system and operational dependent, the present model integrates key system variables, such as
the bottom bed height and the characteristic pressure drop constant over the primary-air distributor.
The model divides the gas flow into two phases, a throughflow and an emulsion gas, and calculates their respective
fluctuations in velocity and composition. Having these, in-furnace gas probe measurements can be simulated and compared
with in-situ gas suction probe measurements (i.e. the probe is modeled). Such an approach is crucial in fluidized bed-boilers
furnaces since in-situ gas probe measurements in regions with high fluctuations in gas velocity are known to give results
which are biased towards reducing conditions. These bias results from the significant velocity differences between oxidizing
(throughflow) and reducing (emulsion) gas fluxes, with the latter flowing at a lower velocity. This effect is especially
pronounced in the lower part of the furnace, as differences in gas velocities decrease with height in furnace.
Model results, including simulation of the gas suction probe, are analyzed and compared with experimental data from the
Chalmers 12 MWth CFB boiler and a good agreement is obtained.

Keywords: Modeling; Gas; Fluidization; Fluidized bed; Fluctuations; Dynamics

1. Introduction
A thorough understanding of combustion applications requires knowledge on the behaviour of the gas phase,
which brings oxygen in contact with the fuel to be oxidized. As far as fluidized bed combustion is concerned
solid fuels (such as coal, peat, biomass or waste) is the usual fuel type, for which the combustion process is
characterized by the release of volatile gas species (part of which are combustible) and by char burn out. Oxygen
is mainly supplied to the fluidized bed furnace through gas injection in the lower part of the furnace, often
divided into primary and secondary air with primary air entering through a primary air distributor. Under typical
operating conditions there is a relatively dense bed directly above the primary air distributor. Due to the limited
pressure drop across the primary air distributor, the gas flow through the air distributor is highly intermittent
with strong fluctuations in gas velocity both in time and space, as shown by Svensson et al. (1996). Thus, the gas
supplied through the primary air distributor is responsible for the highly fluctuating dense bed dynamics which
strongly influence how oxygen is brought into contact with the fuel, both with respect to the char (which to a
large extent remains in the bottom region of the furnace during burnout) and the combustible volatile species.
There is a rather limited amount of experimental work with extensive gas concentration data in industrial
boilers. The work by Åmand (1994) provides the experimental data in Fig. 1, which shows vertical profiles of
oxygen and methane measured with a gas suction probe in the Chalmers CFB boiler.
Figure 1 shows that oxygen and a combustible gas such as methane seem to coexist at different heights despite
the high furnace temperature which should make kinetics for methane oxidation fast enough for this not to occur.
The values shown are time-averaged values from gas suction probe measurements over a certain period of time
(typically 20 minutes). Also, such probe measurements make cannot give information on the time scale
corresponding to typical times for mixing and reaction in the furnace, i.e. the response time of the gas sampling
probe with gas conveying lines is of the order of 10 s or more for the probe system used. The explanation for the

* Corresponding author: [email protected] ; Telf: (+46)317721435 ; Fax: (+46)317723592

coexistence of oxygen and methane on a time-averaged basis is found in the rapid fluctuations between oxidizing
and reducing conditions at the points of measurement.

Figure 1: Vertical gas concentration profiles in the Chalmers Figure 2: Time series of voltage signal from a zirconia
12 MWth CFB boiler. From Åmand (1994). cell probe. From Niklasson et al. (2003)

This can be illustrated by time-resolved experimental data, such as the measurements with zirconia cell probe
shown in Fig. 2. The output signal from a zirconia cell is an indicator of the oxidizing or reducing condition (low
and high values respectively) of the sampled environment. Studies on time series of signals from zirconia cells
inserted in a fluidized bed furnace can be found in Niklasson et al. (2003) and Stubington and Chan (1990). The
rapid fluctuations between oxidizing and reducing conditions seen in Fig. 2 indicate that there is no coexistence
of oxygen and combustible gases on a time-resolved basis. Yet, also the zirconia cell has a limited response time
and the measurements only give a qualitative illustration of these fluctuations, see Niklasson et al. (2003).
The fluctuating character of the gas phase concentration makes it difficult to apply the classical approach used
in macroscopic modeling of the gas phase in fluidized bed boilers (see e.g. Hannes (1996), Knöbig (1998) and
Okasha (2007)), namely a time-averaged modeling. This since time-averaged modeling will obviously not allow
for coexistence of combustion reactants with fast reaction kinetics in the same point (cell). This is usually solved
by fitting the combustion kinetics in the modeling to experimental data, which is an indirect way to account for
the fluctuating nature of the gas concentration. This fitting is strongly dependent on operational conditions,
nozzle characteristics and other parameters influencing gas fluctuations. This classical model approach is
illustrated in Fig. 3a, in which r represents empirically-fitted combustion kinetics while k represents inter-cell
gas mixing mechanisms.

a) Classical model b) Present model

Figure 3: Schemes for macroscopic modeling of the gas phase. Inter-cell gas mixing mechanisms are represented by k and
experimentally-fitted combustion kinetics are represented by r.

The present model is time-resolved and assumes volatile combustion to be transport-controled and thus does
not allow coexistence of combustion reactants at the same cell and time step. The model describes fluctuations in
the flux and species concentrations of the gas phases considered. Thus, comparison with time-averaged
experimental data requires the modeled values to be first time-averaged.

2
 A second phenomenon shown in Fig. 1 is the apparent contradiction in that oxygen concentration increases
with height in the region between the dense bed surface and the secondary air injection height. This is a
consequence of that gas flowing through the bubbles in the bottom bed region finds a low pressure drop path
through the bubbles and thus has a higher velocity (often called throughflow) than gas flowing through the dense
bed emulsion (called emulsion-only gas in this work). This is illustrated in Fig. 4, which shows a captured
videoframe from a cold 2-dimensional unit where two high velocity through flow areas can be seen: one with a
moderately high velocity to the left and one with a high velocity to the right creating what has been described as
an exploding-type bubble by Svensson et al. (1996). An experimental study evaluating this increased gas
velocity of the throughflow gas was carried out by Olowson and Almstedt (1990).

Figure 4: A videoframe from a 2-dimensional cold unit (1.2 x 0.02 x 2 m operated with Group B solids). The
white curve indicates the bed surface, the solid arrow the ejection of high-velocity throughflow gas
from an exploding type bubble and the dashed arrow a throughflow region with a lower gas velocity.

The pattern illustrated in Fig. 4 results in large fluctuations in gas velocity in and above the dense bottom bed.
The throughflow is characterized by a higher oxygen content due to the limited contact with char particles,
which predominantly stay and burn in the dense (low velocity) phase of the bed. This means that a gas suction
probe measurement in regions with such fluctuating gas flow conditions will be exposed to an environment
which alters between high-velocity oxygen rich gas flow and low-velocity oxygen lean flow. The flow
fluctuations will decrease gradually up through the splash zone (located above the dense bed), eventually
yielding a fairly homogeneous gas velocity distribution in the upper freeboard. The degree of bias in gas suction
probe measurements depends on the difference in velocity and gas concentration between the high and low
velocity gas phases, which, thus, is highest within the dense bed and decreases up through the splash zone
(i.e. the bias is reduced along the splash zone). The reduced velocity difference between the two phases up
through the splash zone explains the increase in oxygen concentration with height between the dense bed surface
and the secondary air injection height observed in Fig. 1. Thus, these effects prevent a proper evaluation of gas
concentration measurements (by suction probe) in the bottom region of fluidized bed furnaces. Lyngfelt et al.
(1996) proposed an empirical method to correct measured gas concentration values. Since the bias in measured
gas concentration values as well as the gas mixing in general are strongly influenced by the dynamics of the flow
in the bottom of the furnace a dynamic approach seems inevitable when modeling the gas mixing.
In the bottom region of the furnace, the model presented in this work describes the gas phase behavior in both
the throughflow and the emulsion phase of a fluidized bed with emphasis on circulating fluidized bed boilers,
i.e. for high velocity conditions with entrainment of solids from the bed. The model allows for simulation of
local gas concentration values as seen by a gas sampling probe in order to compare them with in-situ gas probe
measurements in line with the above given discussion. Description of the gas fluctuations takes into account key
parameters such as the pressure drop across the air distributor and dense bed height.
Finally, the list of inputs for the model presented contains the mass rate, composition and location of the
injected flows to the furnace and the penetrating length and width of the secondary air injections. Besides these,
it must be noted that the model presented is part of a comprehensive model for large-scale fluidized bed units
and requires additional input data (e.g. dense bed height, char concentration field, volatile release field) from
other submodels, but modeling of these are outside the scope of the present work. In summary, the model of gas
mixing presented in this work is combined with input data from other models (i.e. the fuel moisture and volatile
release and the char concentration fields, modeled according to Pallarès and Johnsson (2007) and, as a result, the
fields of gas species concentration and flux in the furnace are obtained.
In order to allow for comparison of modeled results with in-situ measurements, the modeled data are
transformed by means of a model of the gas suction probe used for the sampling.

3
2. Theory

The bottom region of a fluidized bed boiler furnace is typically occupied by a so-called dense (or bottom) bed,
which exhibits a linear drop in the vertical profile of the time-averaged pressure (see Johnsson et al. (1991)). The
dense bed is characterized by the presence of gas bubbles which become larger as they rise up through the bed.
Several authors have contributed with semi-empirical correlations for the bubble size, with the expression
proposed by Darton et al. (1977) accounting for bubble coalescence being amongst the most widely used:

Dbub = 0.54 ⋅ (u 0 − u mf ) ⋅ (h + 4
0 .4
A0 )
0 .8
⋅ g − 0 .2 (1)

Based on a survey of experimental data from several authors, Clift and Grace (1985) proposed the following
correlation for the rise velocity of a single bubble through an infinitely wide bed:

vbub = 0.71 ⋅ g ⋅ Dbub (2)

Combining Eqs. (1) and (2) on a time-discretized basis provides a description of the size and location of the
bubble as it raises through the bed.
The formulation of the present model is made for simplicity in a portion of the bed containing one single
bubble but which can be considered representative for the entire bed. Thus, the width of the part of the bed
which is analyzed, L*, is adjusted so that its bubble fraction will equal the bubble fraction, δ, in the entire bed.
This local value of the bubble fraction in the entire bed can be easily calculated assuming the dense bed emulsion
to remain under minimum fluidization conditions, which implies:
ε b − ε mf
δ = (3)
1 − ε mf
and estimating the voidages εb and εmf from some of the several submodels available in literature (see Pallarès
and Johnsson (2006) for the ones used in the present model). With this, having the modeled bubble size and the
width of the bed portion analyzed, the relative cross section of the throughflow gas phase can be defined as
(cf. Fig. 5):
Dbub
σ tf = (4)
L*
From this follows the relative cross section for the emulsion-only gas phase, σem =1-σtf. Note that σtf increases
as the bubble grows, while σem decreases.
The gas bubble rising in the representative bed portion analyzed represents a preferred low pressure-drop path
for a fraction of the fluidization gas (as illustrated in Fig. 4). This fraction is here called throughflow phase. The
gas flowing through the emulsion is called emulsion-only phase. In the model, it is assumed that there is no mass
transfer between these two gas phases. This should be a reasonable assumption since the throughflow phase has
been shown to be of high velocity whereas the emulsion phase flows at much lower velocity (see Olowson and
Almstedt (1990)). The regions of throughflow phase and emulsion-only gas phase are indicated in Figure 5 with
gray and dotted fields, respectively.
A dynamic pressure balance accounting for the above division between the two gas phases (the throughflow
and emulsion-only phase) is the basis of the modeling in this work, illustrated in Fig. 5. For each time step of the
rise of the bubble, the cross section of each gas phase is updated according to the calculated bubble size and a
pressure balance is applied which considers the same total pressure loss across the gas distributor and the dense
bed (ΔPbot = Pplenum – Pfreeb) for both gas phases. Thus, at each time step the total pressure drop is:
dPemul (u em , t ) dPemul (u tf , t )
ΔPbot = ΔPdist (u em , t ) + ⋅ H b = ΔPdist (u tf , t ) + ⋅ H emul tf , t (5)
dh dh
where Hb represents the dense bed height (an input to the present model) and Hemul tf the emulsion path length
seen by the throughflow gas phase (which is obtained from combination of the dense bed height and the bubble
size). The pressure loss across the gas distributor is a function of gas velocity (the so-called characteristic curve)
and takes the form:
ΔPdist (u ) = a ⋅ u 2 (6)

4
The coefficient a in Eq. (6) is known from experiments and thus an input to the model. The pressure loss across
the emulsion can be determined through the well-known Ergun equation (see Ergun (1952)), i.e,

dPemul (u ) μ ⋅ (1 − ε )2 ρ g ⋅ (1 − ε ) 2
= 150 2 2 3 ⋅ u + 1.75 ⋅u (7)
dh d p ⋅φ ⋅ ε d p ⋅φ ⋅ ε 3

Thus, assuming an initial value for the total pressure loss, ΔPbot, the pressure balance can be solved at each
time step, providing the evolution of the velocities of each gas phase during the bubble rise. Integration over
time of the gas flows corresponding to these velocities gives the time-averaged gas velocity, u , resulting from
the chosen ΔPbot.

u = ∑ u em, t ⋅ σ em, t + u tf , t ⋅ σ tf , t (8)

t

Having this, an iterative procedure is used to find the value for ΔPbot giving a time-averaged modeled gas
velocity equal to the fluidization velocity set as input.
It should be mentioned that for simplicity the model applies a constant value for the bottom region pressure
drop, ΔPbot. Indeed, this value is known to oscillate depending on several parameters related to not only the
operational conditions but also the air feeding system (e.g. pipe lengths, valve system), as studied by Sasic et al.
(2005). Yet, for typical boiler conditions (i.e. rather high velocities) these fluctuations should be limited.
To illustrate the feature of the dynamic pressure balance modeling an example is here taken of operation with
a 0.42 m-high dense bed fluidized by primary air at 850 °C with a superficial gas velocity of 3.1 m/s and two
different gas distributors (with values of a in Eq. (6) of 120 and 800 Pa·s2/m2, respectively). Figure 6 shows
the calculated flow dynamics in the part of the bed which is analyzed for the two gas phases (emulsion and
throughflow) obtained from the pressure balance expressed by Eqs (5) to (7).

Figure 5: Scheme of the pressure balance for the emulsion- Figure 6: Modeled flows of the emulsion-only and throughflow gas
only gas phase and the through flow gas phase. phases in a single bubble bed. Lines with bullets (•) are for
the high pressure-drop gas distributor whereas lines without
bullets are for low pressure-drop distributor.

Figure 6 shows the gas flow division between the throughflow phase and emulsion-only phase during one
bubble cycle. The lines without bullets represent the case with the low-pressure gas distributor (a=120 Pa·s2/m2),
which is representative for large-scale CFB boilers and predicts that a large part of the fluidization gas crosses
the dense bed as throughflow (solid line) whereas the emulsion gas flow (dotted line) is much lower and more
constant over time. These model results agree with what has been experimentally observed by Olowson and
Almstedt (1990). The throughflow gas flow reaches a maximum as the gas bubble reaches the dense bed surface.
The dotted curves for the high-pressure drop gas distributor (a=800 Pa·s2/m2) reveal a more even distribution of
the gas between the throughflow and emulsion-only phases than with the low-pressure drop gas distributor.

5
Indeed, during a considerable part of the bubble-rise cycle more gas joins the emulsion-only phase than the
throughflow (since emulsion-only phase has a larger cross section as long as the gas bubble is relatively small,
although gas velocity is always higher in the throughflow phase). In addition, it can be seen from Fig. 6 that a
higher pressure drop across the air distributor dampens the magnitude of the total gas flow fluctuations (the
amplitude of the total flow fluctuations for the low high-pressure distributor is 21% compared to the low-
pressure drop distributor). Thus, the air-distributor pressure drop has a large influence on the gas flow division
into a throughflow and emulsion-only phases, with lower pressure drops accentuating fluctuations and the
differences in gas velocity between these phases. The dense bed height plays a similar role, in the sense that
higher dense beds having an analogous effect to that of lower pressure drops across the gas distributor.
In the model, gas has the same statistical distribution between emulsion-only and throughflow phases at all
locations in the dense bed, namely the one obtained from the dynamic pressure balance. Thus, the so-called
“preferred bubble paths” (zones where the probability of presence of gas throughflow is higher than in the rest of
the bed, typical for fluidization with very low distributor pressure drop at low gas velocities) are not accounted
for in the present model. Thus, the gas phase division obtained from the above-described dynamic pressure
balance and exemplified in Fig. 6 can be applied as boundary condition for the entering primary air at the bottom
cells in the riser mesh. The data line for oxygen at h=0 m in Fig. 7 illustrates this showing the modeled flow of
oxygen injected with the primary gas in the emulsion-only gas phase of a bottom cell in the riser mesh.
The present model requires as input the spatial distribution of fuel release rates (i.e. moisture and volatile
species) and size and concentration of char (providing char surface), which in the present simulations are
provided by a fuel mixing submodel (see Pallarès and Johnsson (2008)). Thus, in each cell the fuel release fields,
r(x,y,z), are added to each gas phase proportionally to the relative cross section of the gas phase concerned.
Released moisture and non-combustible volatiles will increase the volumetric flow and thus the velocity of the
gas phase, while released combustible volatiles react with oxygen available at each time step (infinitely fast
kinetics are assumed) generating product gases (CO2 and H2O) which add to the gas phase flow.
In an analogous way, char particles in a given cell are assumed to get in contact with each gas phase in a
fraction proportional to the relative cross section of the two gas phases. Thus, together with the given average
char size this will provides the char surface as seen by the two gas phases (emulsion-only and throughflow) at
each time step in the cell studied (i.e. Ac in Eq.(9)). Thus, the char combustion rate in a specific cell can be
calculated as (first order reaction is assumed):
dmc (9)
= −Ω ⋅ Reff ⋅ Ac ⋅ cO2
dt
where, according to Field et al. (1967), the so-called effective combustion rate, Reff, can be expressed as a
combination of two processes governed by respectively the kinetic rate, Rkin (determined through an Arrhenius-
type correlation, see Field et al. (1967)) and the mass transfer coefficient, hm (determined through a Sherwood-
type correlation, see Field et al. 1967)), yielding:
1 (10)
Reff =
1 1
+
Rkin hm
Thus, combustion of volatile species and char can
be combined with the fluiddynamical inlet boundary
conditions for both gas phases provided by the
above-described pressure balance, yielding a model
for the transport-controled oxygen consumption in
the dense bed. Modeled data showing the evolution
of the oxygen consumption with height during a
bubble cycle is shown in Fig. 7. Figure 7 shows that,
along the furnace centerline, the lowest height above
the gas distributor for which no oxygen is available
in the emulsion-only gas flow at some point during a
bubble cycle is 0.2 m. Above this height, the
duration of emulsion-only gas flow with a zero
oxygen concentration lasts for longer periods.
During absence of oxygen released volatile
hydrocarbons will remain unburned and add to the
gas flow, as observed in the data line corresponding
Figure 7: Modeled gas species flows in the emulsion-only
phase at different heights in the furnace centerline. to h=0.24 m.

6
 However, it should be noted that, on a time-averaged basis, the cell corresponding to a height of 0.24 m at the
furnace centerline has non-zero concentrations of both oxygen and hydrocarbons, although coexistence of these
is not allowed in the time-resolved modeling.
While combustion in the freeboard is modeled in the same way as in the dense bed, modeling of the gas phase
mixing above the dense bed applies a widely-used concept of division of a CFB freeboard into two regions: a
splash zone and a transport zone. The solids flow of the splash zone, which is located immediately above the
dense bed surface, is dominated by cluster formations which are ejected into the freeboard as bubbles burst and,
following a ballistic movement, fall back to the dense bed. The transport zone is dominated by dispersed solids
suspension flowing more steadily in a core-annulus structure (i.e. back-mixing of solids mainly at furnace walls).
The height of the splash zone can be defined as the vertical position where the concentrations of the cluster phase
and the dispersed phase are equal, as proposed by Johnsson and Leckner (1995).
In the splash zone, the difference in gas velocity between gas phases and the gas velocity fluctuations decrease
with height until gas velocity becomes eventually fairly constant over the cross section at the interface with the
transport zone. In terms of modeling, this is approached through a gradual damping of the gas flow fluctuations
(so that both gas phases considered reach constant velocity values) together with an adjustment of the relative
cross section of the gas phases. This is done so that both gas phases reach the same constant velocity. The
expressions used for these model calculations implemented in a hexahedrical mesh are summarized below.
Firstly, a damping factor is defined at each cell which indicates how much of the gas velocity fluctuations
within each gas phase will be damped in the present cell and how much the difference in gas velocity between
phases will decrease. In the present model, this damping factor is calculated in each cell as the height of the
present cell divided by the distance from the bottom of the present cell up to the end of the splash zone:
hcell , top − hcell , bot
f = (11)
hsplash , top − hcell , bot

With this, damping of the gas velocity fluctuations across a splash zone cell in the two gas phases can be
expressed as:
u top = u bot − f ⋅ u bot
'
(12)

where u’ is the fluctuating component of the gas velocity, u. Damping of the gas velocity fluctuations implies a
rearrangement of the gas flow with mass transfer between the different time steps. However, if only damping of
the gas velocity fluctuations for each gas phase was considered, the time-independent gas velocity reached at the
top of the splash zone for the gas phases would differ. Thus, there is also a need to model the fade out in velocity
difference between the fast-throughflow and the slow-flowing emulsion-only gas. This is done knowing that the
phase-averaged velocity for a certain time step is:

u em ⋅ σ em + u tf ⋅ σ tf
u phase − avg = (13)
σ em + σ tf

which is the aimed velocity for both gas phases at the upper limit of the splash zone, where the relative cross
section for the gas phases then becomes (emulsion-only phase shown) :

u em
σ em, aim = σ em (14)
u phase−avg

Thus, making use of the damping factor defined by Eq. (11), the variation of the relative cross section of the
emulsion-gas phase through one splash zone cell is expressed as:

σ em , top = σ em , bot + f ⋅ (σ em , goal − σ em , bot ) (15)

Finally, the relative cross section for the throughflow gas phase is calculated knowing that σtf =1-σem.
With these two damping mechanisms for the gas velocity fluctuations within each gas phase (Eq.(12)) and for
the difference in gas velocity between the two gas phases (Eq.(15)), a constant gas velocity value common to
both gas phases is reached for cells at the top of the splash zone. Note that gas mixing associated to the damping

7
of the gas velocity fluctuations in the splash zone occurs on an intra-cell level, i.e. there is a rearrangement of the
dynamics of the gas flow within each cell (expressed by Eq.(12)), but no lateral gas mass transfer between cells.
In the transport zone extending from the top of the splash zone to the riser exit, the gas flows more steadily
and solids develop a core-annulus structure with upflow in the core region and downflow in so-called wall
layers. While gas upflow in the core region is clear, measurements in the wall layer region of the Chalmers
12 MWth CFB boiler by Sternéus and Johnsson (1997) indicate low gas velocity and both occasional upflow and
downflow of gas within the wall layer. Thus, in the present model, zero gas flow is assumed in the wall layers of
the transport zone. In the core region, gas mixing between cells is modeled by means of lateral diffusion of the
mass flux G of each gas species i considered, i.e,

∂Gi
= α g ⋅ ∇ 2 Gi (16)
∂t

Note that the mass flux is taken as transport variable in Eq.(16) instead of the concentration (which is the
traditional approach). Both formulations are equivalent in a homogeneous gas velocity field. However, in
presence of gas velocity gradients, Eq.(16) is able to model the lateral gas flows which lead to a flattening of the
gas velocity. This flattening of the gas velocity field with height in the furnace is reported in the experimental
work by Kruse and Werther (1995). Generally, the main reason for the existence of gas velocity gradients in the
transport zone is secondary gas injection (which in the present model is handled by having as input the
penetration length and width of the injected gas jet at the injection height).
Values of the gas lateral dispersion coefficient, αg,diff , have been studied in several experimental studies in
literature (see e.g. Werther et al. (1992), Gayán et al. (1997) and Sternéus et al. (2000)). This study applies the
Peclet number-based expression proposed by Kruse et al. (1995):

u ⋅ Deq
Pe = = 387 (17)
α g ,diff

With this final expression, a dynamical model for the gas mixing in fluidized bed boilers is formulated which
allows for introduction of transport-controled combustion. In summary, the model presented above divides the
gas into two phases and is based on different assumptions and approaches for the dense bottom bed, the splash
zone and the transport zone.

2.1. Gas probe simulation

As explained above in relation to Figure 1, the velocity differences between the throughflow and emulsion-
only gas phases prevent proper evaluation of in-situ gas probe measurements in the bottom region (i.e. dense bed
and splash zone) of a fluidized bed boiler. However, since the model accounts for the gas phase division which is
the reason for this effect and provides the velocity variations and relative cross sections of both gas phases,
simulation of the values measured by a gas probe is possible.
The time-averaged mass concentration for a certain gas species, i, in any cell is defined as the ratio between
the mass flow of the gas species i and the total gas mass flow, namely:

F ∑ (u C t
em
t
em , i σ em
t
, i + u tf C tf , i σ tf , i )
t t t

Ci = i = t (18)
FTOT ∑∑ (u C
i t
t
em
t
em , i σ em
t
, i + u tf C tf , i σ tf , i )
t t t

However, the time-averaged mass concentration of the gas species seen by a gas probe does not take into
account the effect of gas-velocity difference between the gas phases. Thus, the mass concentration of a gas
species i seen by a gas probe can be simulated by the following expression:

∑ (C t
em , i σ em
t
, i + C tf , i σ tf , i )
t t

C i , probe = t (19)
∑∑ (C
i t
t
em , i σ em
t
, i + C tf , i σ tf , i )
t t

8
 Note that in locations where utem= uttf (which is the case in the transport zone) the expressions in Eqs.(18) and
(19) give the same value. In such case, an in-situ gas probe measurement gives a representative value of the gas
species concentration. On the other hand, it is found in the dense bed and the splash zone that utem< uttf , which,
in accordance to Eqs (18) and (19), implies an under-representation of the throughflow gas-phase in the gas
concentration measured by a gas probe.

3. Experiments

The experimental part in this work consists of detailed in-situ measurements taken at the Chalmers 12 MWth
CFB boiler outlined in Fig. 8a. The riser (furnace) of the Chalmers CFB boiler is 13.5 m.-high with a top cross
section of 1.62×1.42 m. The location of the key parts of the riser is illustrated in Fig. 8b. There is a single fuel
feeding point located in the front wall and secondary air injection ports in both front and rear walls. The boiler
walls are membrane tube walls (water-tubes) and the boiler is a heat only boiler. The front and rear walls are
refractory lined all the way from bottom to top of furnace whereas the side walls are bare membrane walls except
for the heights below 2.2 m which are refractory lined. The air is introduced as primary air through the bottom
gas distributor and secondary air at 2.1 meters above the primary air distributor. Gas coming from the seal and
the particle classifier enter the riser through the rear wall at 0.99 and 1.29 m above the distributor respectively.
Further details on the geometry of the unit are given in Leckner et al. (1991). For the calculations in the present
work, the mesh discretizing the furnace contains about 100.000 cells (10% of which are plotted in Fig. 8.c).

a) Circulating loop scheme: (1) air plenum, b) Key elements in the furnace c) Furnace mesh
(2) fuel feeding, (3) secondary air nozzles,
(4) furnace, (5) cyclone, (6) particle seal,
(7) heat exchanger.
Figure 8: The Chalmers 12 MWth CFB boiler

Two runs are analyzed in this work, both using coal as a fuel. Burning similar coal types (but at a higher fuel
feeding rate for Run B), the air-to-fuel ratio is similar for both cases (around 1.2) and so does the air staging ratio
(around 58% of the air directly injected to the riser was supplied as primary air). The higher fuel feeding rate in
Run B represents a higher fluidization velocity and thus a higher operational pressure drop over the furnace was

9
established for this run in order to ensure a stable dense bed. Table 1 lists the experimental conditions applied in
this work and Table 2 gives the fuel analyses.

Table 1: Operational conditions Table 2: Fuel analyses

Run A Run B
Fluidization velocity m/s 2.22 2.90
Primary air kg/s 1.39 1.81
Secondary air kg/s 1.06 1.24 Coal A Coal B
Air injection height m 2.1 Moisture %, ar 8.5 16.9
Classifier air kg/s 0.14 0.15 Volatiles %, daf 34.2 40.2
Seal air kg/s 0.10 0.15 Ash %, df 9.7 10.7
Fuel rate kg/s 0.30 0.43 LHV MJ/kg, df 29.1 28.4
Fuel type Coal A Coal B C %, daf 80.7 78.4
Solids size μm 300 320 O %, daf 12.4 12.7
Solids density kg/m³ 2600 H %, daf 4.8 5.5
Bed temperature ºC 850 N %, daf 1.5 1.6
Riser pressure drop Pa 5110 7950 S %, daf 0.6 0.7
ar - as received, daf - dry and ash-free, df - dry fuel

4. Results

C vol [% vol] As mentioned above, the model of gas mixing and char
combustion of this work (Eqs (1) to (17)) are combined with
0.035 input data from other models (i.e. the fuel moisture and
10
volatile release and the char concentration fields, modeled
according to Pallarès and Johnsson (2007)) and the
9 calculated oxygen concentration field in the furnace is
0.03 modeled and compared to the in-situ measurements. For the
calculations in this work, the penetration length of all gas
8 jets injected in the freeboard (i.e. secondary air and gas from
seal and classifier) is set to 0.35 m.
7 0.025
Figure 9 shows the modeled volumetric concentration of
combustible volatile matter for Run B. As seen, all
6 combustible volatiles released immediately above the air
distributor are fully consumed and it is not until a few
0.02
centimeters above the distributor that some of the released
5 volatile matter remains unburned due to the lack of oxygen
z [m]

in some periods of the gas fluctuations, as explained above

in relation to Fig. 7. This effect takes place mainly close to
4 0.015
the fuel feeding point, where the volatile release rates are
much higher than close to the rear wall, due to the kinetic
3 rate of volatile release, which is fast compared to the
horizontal fuel mixing rate. Thus, close to the fuel feeding
0.01 point, the model predicts a build up of concentration in
2 combustible volatiles with height. According to the model,
these unburned volatiles are combusted by the secondary air
1 flow injected from the front wall at a height of 2.1 m.
0.005
Figure 10 shows the modeled oxygen volumetric
0 concentration for Run B. As observed, in the bottom region
-0.81
oxygen concentration decreases slightly faster with height
0 0.71 0 closer to the front wall, which is due to the higher release of
0.81 0 both volatile matter and moisture close to the fuel feeding
-0.71
x [m] y [m] point. The high volatile release implies a higher volatile
combustion and thus oxygen consumption and the high
moisture release leads to a more dilute, less concentrated
Figure 9: Modeled volumetric concentration of combustible oxygen flow. The lower horizontal slice in Fig. 10 shows the
volatiles (sum of hydrocarbons, H2 and CO) for effect of the injection of gas coming from the ash classifier
Run B.

10
(which is located at the rear wall, 0.2 m from the right wall). An increase in oxygen concentration is predicted
also close to the corner between the back wall and the left wall due to injection of gas coming from the particle
seal, although this is located in a dead angle in Fig. 10 and thus cannot be seen. The upper horizontal slice in
Fig. 10 shows the modeled oxygen concentration immediately above the secondary air injections. Secondary air
at a height of 2.1 m above the distributor is injected through the front and rear walls. As seen in the figure,
oxygen injected from the front wall is mostly consumed (by combustion of unburned volatiles, as seen in Fig. 9)
while oxygen from injected from the back wall hardly meets any unburned volatile matter (see Fig. 9) and
therefore remains mostly unconsumed (although some oxygen consumption through char combustion takes
place). Along the freeboard, horizontal convective mixing of the gas takes place due to the disturbance of the
velocity field created by the secondary air injection at the same time as horizontal diffusion of gas species
occurs. Both these phenomena lead to a significantly more homogeneous oxygen concentration in the upper
freeboard than immediately above the secondary air injection.
As explained above, gas concentration measurements sampled with a gas probe can be simulated through
Eq.(19). This has been done for the result data from the simulation of Run B and the obtained field is shown in
Fig. 10b. It can be observed how values in the upper heights of the furnace (i.e. in the transport region) are the
same as those shown in Fig. 10a, since both gas phases considered are assumed in the model to flow at the same
velocity in this region, which leads to the same outcome from the use of Eqs.(18) and (19). However,
concentration values for the simulated gas probe differ strongly from those in Fig. 10a in the bottom region of
the furnace. As seen in Fig. 10b the probe-measured oxygen concentration presents a much more accentuated
vertical gradient which leads to a bias in the results towards more oxygen-poor values.

C O [% vol] C O [% vol]
2 2

0.212 0.212
10 10
0.2 0.2

9 9
0.18 0.18

8 8
0.16 0.16

7 7
0.14 0.14
6 6
0.12 0.12
5 5
z [m]

z [m]

0.1 0.1
4 4

0.08 0.08
3 3

0.06 0.06
2 2

0.04 0.04
1 1

0.02 0.02
0 0
-0.81 -0.81
0 0.71 0 0.71
0 0 0 0
0.81 -0.71 0.81 -0.71
x [m] y [m] x [m] y [m]

a) Time-averaged oxygen concentration. b) Oxygen concentration in Figure 10a as measured

by a gas suction probe.
Figure 10: Modeled oxygen volumetric concentration for Run B.

11
 a) Experimental data b) Modeled data

Figure 11: Comparison between measured and modeled cross-sectional data

on oxygen volumetric concentration for Run B at h=3.7 m.

Figures 11 and 13 give comparisons between modeled and measured cross-sectional data on oxygen
concentration available from Run B, with nine experimental values sampled (with a gas suction probe) over
furnace cross sections at heights of 3.7 and 9.9 m respectively. Both these cross sections belong to the transport
zone, where both gas phases are modeled to have the same gas velcotiy, so the modeled gas concentration takes
the same values whether it is calculated through a true mass balance (i.e. Eq.(18)) or as seen by a gas suction
probe (Eq.(19)).
It can be seen that there is a relatively good agreement between modeled and measured data, except for the
value in the center of the front wall. The likely reason for this discrepancy is most likely related to blocking of
secondary air nozzles. In the Chalmers CFB boiler, the secondary air ports (13 active nozzles for the present run:
7 in the front wall and 6 in the back wall) should in principle all provide the same air flow (which is also the
assumption made in the modeling). However, partial and total plugging of the secondary air nozzles is a
phenomenon often occurring in the boiler. The picture in Fig. 12 is taken at the furnace of the Chalmers boiler
and shows partial plugging of a secondary air nozzle.
A rather severe partial plugging is shown in the
picture, but the different secondary air nozzles are
affected by plugging to rather different extent,
from no plugging to total plugging. This
redistributes the injected secondary air between
the different nozzles in a fashion which cannot be
measured and, thus, remains unknown during
operation. Measured values in Fig. 11a could
indicate that there is such a redistribution between
the nozzles. This, since the oxygen concentration
in the front wall varies strongly with the highest
value at the center of the wall. This pattern was
observed in other runs, which supports the
suggestion on the existence of plugging/partial
plugging of some of the secondary air nozzles.
Figure 13 gives a corresponding comparison in
the upper cross section (h=9.9 m). As seen from
Figure 12: Partial blocking of a secondary air nozzle. Photo taken the measured values (Fig. 13a), oxygen is more
from inside the Chalmers CFB boiler.
uniformly distributed at upper locations in the
furnace due to lateral mixing occurring along the transport zone and modeled by means of gas species mass flux
diffusion. However, note that at this height (almost 8 meters above the secondary air injection) the consequence
of the presumed effect of secondary air plugging is still seen: the highest oxygen concentration value is found at
the center of the front wall (where, theoretically, this value should be low due to the locally high volatile release
close to the fuel feeding point, in agreement with the modeled values in Fig. 13b).

12
 a) Experimental data b) Modeled data
Figure 13: Comparison between measured and modeled cross-sectional data on
oxygen volumetric concentration for Run B at h=9.9 m

Comparisons between modeled data and axial gas probe measurements from Runs A and B are shown in
Fig. 14. The comparisons are carried out along the centerline of the back wall, at 0.37 m from the back wall). As
seen from the experimental data for both plots, measured oxygen concentration increases along the splash zone
(i.e. approximately in the height interval between 0.5 and 2 m). The modeled oxygen concentration (represented
by a solid line) decreases steadily along the dense bottom bed and the splash zone, mainly due to volatile and
char combustion. A sudden increase in oxygen concentration is observed at a height of 2.1 m due to secondary
air injection followed by a decrease due to both char combustion and (mostly) horizontal dispersion of the
oxygen. As seen, the modeled data which includes simulation of the gas probe (represented by a dashed curve)
differs strongly from the modeled concentration curve in the bottom region of the furnace, showing lower values
of the oxygen concentration. It is also seen that the gas probe simulation agrees with the measurement data in the
bottom region, where there is a disagreement between experiments and raw (i.e. gas probe modeling not
included) modeled data, in analogy with the above given discussion.

a) Run A b) Run B

Figure 14: Comparison of measured and modeled vertical profiles of oxygen volumetric
concentration at 0.37 m from the center of the back wall.

As observed, the curve corresponding to the probe simulation through Eq.(19) gives the expected lower
oxygen concentration values (although it does not reach the almost zero values observed in Fig. 14 for Run A).
Note that the curves in Figs 14a and 14b approach a common value as the differences in gas velocity decrease
with height in the splash zone. In summary, it is shown that inclusion of a model for the gas sampling probe
accounting for the gas phase division expressed in this paper is crucial in order to compare with in-situ
measurements in the lower regions of a FB combustor, i.e. in regions where there are considerable fluctuations in
gas velocities.

13
5. Conclusions

A modeling approach for the gas phase mixing in fluidized beds is presented. The model has been
implemented and validated with experimental data from the Chalmers 12 MW CFB boiler. The model accounts
for both the fluctuating nature of the gas flow and the existence of two different gas phases with different
dynamics and composition, features which are necessary if the modeling should have a realistic coupling to the
underlying physics of the gas mixing. The present approach allows a formulation of the gas combustion reactions
as transport-controlled and thus, there is no need for an empirical fit of the kinetics. Good agreement between
modeled data and experiments is observed.
In addition, the gas phase division employed by the model allows the simulation of gas probe measurements in
the bottom region of the furnace, characterized by strongly varying gas velocities.

References

Clift, R., Grace, J.-R. 1985. “Fluidization”. Davidson, Harrison and Clift Eds., Academic Press Inc., London
(UK).
Darton, R., LaNauze, R., Davidson, J., Harrison, D. 1977. “Bubble-growth due to coalescence in fluidized beds”.
Transactions of the Institution of Chemical Engineers, 55, pp. 274-80.
Ergun, S. 1952. ”Fluid flow through packed columns”. Chemical Engineering Progress, 48, pp. 89-94.
Field, M.A., Gill, D.W., Morgan, B.B., Hawksley, P.G.W. 1967. “Combustion of pulverized coal”. The British
Coal Utilization Research Association, Surrey (England).
Gayán, P., de Diego, L.F., Adánez, J. 1997.” Radial gas mixing in a fast fluidized bed”. Powder Technology, 94
(2), pp. 163-171.
Hannes, J.-P. 1996. ”Mathematical modelling of circulating fluidized bed combustion”. City-Print Verlag
GmbH, Aachen (Germany). ISBN 3-88817-002-8.
Johnsson, F., Leckner, B., 1995. “Vertical distribution of solids in a CFB furnace”. Proc. of the 13th Int. Conf.
on FBC (Orlando, USA), pp. 671-679.
Johnsson, F., Andersson, S., Leckner, B. 1991. “Expansion of a freely bubbling fluidized bed”. Powder
Technology, 68 (2), pp. 117-123.
Knöbig, T. 1998. “Three-dimensional modeling of circulating fluidized bed combustion”. PhD Thesis, Tech.
Univ. of Hamburg-Harburg (Germany). Shaker Verlag, Aachen (Germany). ISBN 3-8265-4480-3.
Kruse, M., Werther, J. 1995. ”2D gas and solids flow prediction in circulating fluidized beds based on suction
probe and pressure profile measurements”. Chemical engineering and processing, 34 (3), pp. 185-203.
Kruse, M., Schoenfelder, H., Werther, J. 1995. ” Two-dimensional model for gas mixing in the upper dilute zone
of a circulating fluidized bed”. Canadian Journal of Chemical Engineering, 73 (5), pp. 620-634.
Leckner, B., Golriz, M.R., Zhang, W., Andersson, B.Å., Johnsson, F. 1991. “Boundary layers - first
measurements in the 12 MW CFB research plant at Chalmers University”. Proc. of the 11th Int. Conf. on
FBC (Montreal, Canada), pp. 771–776.
Lyngfelt, A., Åmand, L.-E., Leckner, B. 1996. ”Progress of combustion in the furnace of a circulating fluidized
bed boiler”. Proc. of the 26th Int. Symp. on Combustion (Naples, Italy), pp. 3253-3259.
Niklasson, F., Pallarès, D., Johnsson, F. 2007. “Biomass co-firing in a CFB boiler – The influences of fuel and
bed properties on in-furnace gas-concentration profiles”. Proc. of the 19thInt. Conf. on Fluidized Bed
Combustion (Vienna, Austria).
Niklasson, F., Johnsson, F., Leckner, B. 2003. “Local air ratio measured by zirconia cell in a circulating fluidised
bed furnace”. Chemical Engineering Journal, 96 (1-3), pp. 145-155.
Okasha, F. 2007. “Modeling combustion of straw–bitumen pellets in a fluidized bed”. Fuel Processing
Technology, 88 (3), pp. 281-293.

14
Olowson, P.A., Almstedt, A.-E. 1990. ”Influence of pressure and fluidization velocity on the bubble behaviour
and gas flow distribution in a fluidized bed”. Chemical Engineering Science 45 (7), pp. 1733-1743.
Pallarès, D., Johnsson, F. 2006. “Macroscopic modeling of fluid dynamics in large-scale circulating fluidized
beds”. Progress in Energy and Combustion Science, 32 (5), pp. 539-569.
Pallarès, D., Johnsson, F. 2008. ”Modeling fuel mixing in a fluidized bed combustor”. Accpeted for publication
in Chemical Engineering Science.
Sasic, S., Leckner, B., Johnsson, F. 2005. “Fluctuations and waves in fluidized bed systems: the influence of the
air-supply system”. Powder Technology, 153 (3), pp. 176-195.
Sternéus, J., Johnsson, F., Leckner, B. 2000. “Gas mixing in circulating fluidised-bed risers”. Chemical
Engineering Science, 55 (1), pp. 129-148.
Sternéus, J., Johnsson, F. 1997. “Gas mixing in the wall layer of a CFB boiler”. Proc. of the 14th Int. Conf. on
Fluidized Bed Combustion (Vancouver, Canada), pp. 1237-46.
Stubington, J.F., Chan, S.W. 1990. “Interpretation of oxygen-probe measurements in fluidised-bed combustors”.
Journal of the Institute of Energy, 63 (456), pp. 136-142.
Svensson, A., Johnsson, F., Leckner, B. 1996. “Bottom bed regimes in a circulating fluidized bed boiler”.
International Journal of Multiphase Flow, 22 (6), pp. 1187-1204.
Werther, J., Hartge, E.U., Kruse, M. 1992. “Radial gas mixing in the upper dilute core of a circulating fluidized
bed”. Powder Technology, 70, pp. 293–301.
Åmand, L-E. 1994.”Nitrous oxide emission from circulating fluidized bed combustion”. PhD Thesis, Chalmers
Univ. of Tech. (Sweden).

Acknowledgements
Financially support by the Swedish Energy Agency and Metso Power Oy is gratefully acknowledged.

Notation

A0 Nozzle density, m-2 H Height interval, m

Ac Char surface, m2 h Height, m
a Decay constant, m-1 hm Mass transfer coefficient, m/s
c Gas molar concentration, mole/m3 mc Char mass, kg
C Gas mass concentration, kg/m3 P Pressure, Pa
Dbub Bubble diameter, m Reff Effective combustion rate, m/s
Deq Equivalent diameter, m Rkin Kinetic combustion rate, m/s
dp Particle size, m u Gas velocity, m/s
f Damp factor, - u0 Fluidization velocity, m/s
G Mass flux, kg/m2s umf Minimum fluidization velocity, m/s
g Gravitational acceleration, m/s2 vbub Bubble velocity, m/s

Greek letters
α Gas dispersion coefficient, m2/s μ Gas viscosity, kg/m·s
δ Bubble fraction, - σ Relative cross section, -
εb Dense bed voidage, - ρ Apparent density, kg/m3
εmf Minimum fluidization voidage, - Ω Conversion factor, kg C/mole O
Ф Sphericity, -

Subscripts
em Emulsion-only gas phase
tf Throughflow gas phase

15
16
 Paper IV

Pallarès, D., Johnsson, F. (2006)

”A novel technique for particle tracking in cold 2-dimensional
fluidized beds - simulating fuel dispersion”
Chemical Engineering Science, 61 (8), pp. 2710-2720
 Chemical Engineering Science 61 (2006) 2710 – 2720
www.elsevier.com/locate/ces

A novel technique for particle tracking in cold 2-dimensional fluidized

beds—simulating fuel dispersion
David Pallarès, Filip Johnsson ∗
Department of Energy and Environment, Energy Conversion, Chalmers University of Technology, SE-412 96 Göteborg, Sweden

Received 1 April 2005; received in revised form 29 September 2005; accepted 16 November 2005
Available online 18 January 2006

Abstract
This paper presents a novel technique for particle tracking in 2-dimensional fluidized beds operated under ambient conditions. The method is
applied to study the mixing mechanisms of fuel particles in fluidized beds and is based on tracking a phosphorescent tracer particle by means
of video recording with subsequent digital image analysis. From this, concentration, velocity and dispersion fields of the tracer particle can be
obtained with high accuracy. Although the method is restricted to 2-dimensional, it can be applied under flow conditions qualitatively resembling
a fluidized-bed combustor. Thus, the experiments cover ranges of bed heights, gas velocities and fuel-to-bed material density and size ratios
typical for fluidized-bed combustors. Also, several fluidization regimes (bubbling, turbulent, circulating and pneumatic) are included in the runs.
A pattern found in all runs is that the mixing pattern of the tracer (fuel) solids is structured in horizontally aligned vortexes induced by the
bubble flow. The main bubble paths always give a low concentration of tracer solids and with the tracer moving upwards, while the downflow of
tracer particles in the dense bottom bed is found to take place in zones with low bubble density and at the sidewalls. The amount of bed material
(bed height) has a strong influence on the bottom bed dynamics (development and coalescence of bubbles) and, consequently, on the solids
mixing process. Local dispersion coefficients reach maximum values around the locations of bubble eruptions, while, in the presence of a dense
bottom bed, an increase in fluidization velocity or amount of bed material enhances dispersion. Dispersion is found to be larger in the vertical
than in the horizontal direction, confirming the critical character of lateral fuel dispersion in fluidized-bed combustors of large cross section.
䉷 2005 Elsevier Ltd. All rights reserved.

Keywords: Fluidization; Mixing; Dispersion; Fuel; Combustion; Digital image analysis

1. Introduction Peters and Bruch, 2001; Thunman et al., 2002). On the other
hand, no models of practical application based on the underly-
In fluidized-bed combustion, a high enough fuel dispersion ing physics are available for particle mixing, since numerical
is crucial in order to ensure good mixing of fuel and combus- modeling (CFD simulations) from first principles (e.g. Enwald
tion air (i.e., to obtain a satisfactory burnout while keeping the et al., 1996; van Wachem et al., 2001; Peirano et al., 2002) yield
excess air ratio as low as possible), at the same time as the long calculation times and still give uncertain results for condi-
number of fuel feed points must be kept small. Consequently, tions applicable to fluidized-bed combustors. In addition, such
fuel mixing is known to be critical in large fluidized-bed units simulations are still more or less restricted to beds with solids
which may have cross sectional areas up to several hundreds of one size and density. Some semiempirical models express
of square meters. the solids mixing in form of dispersion coefficients (Lin and
Evaluating fuel mixing requires modelling of both fuel Chyang, 2003; Du and Wei, 2002). These models focus mainly
dispersion and conversion (drying, devolatilization and char on modeling the vertical mixing in laboratory fluidized-bed
burnout). With respect to fuel conversion, literature gives a units where, due to their narrow geometry, this is the critical di-
number of particle combustion models shown to give satis- rection for solids mixing, although horizontal solids dispersion
factory results (Agarwal et al., 1986; Palchonok et al., 1997; coefficients have also been measured in narrow units (Mostoufi
and Chaouki, 2001; Fan et al., 1986). However, as mentioned
∗ Corresponding author. Tel.: +46 31 772 1449; fax: +46 31 722 3592. above, in large fluidized-bed units such as boilers solids mixing
E-mail address: fi[email protected] (F. Johnsson). is critical in the horizontal direction, due to the low bed-height
0009-2509/$ - see front matter 䉷 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2005.11.030
 D. Pallarès, F. Johnsson / Chemical Engineering Science 61 (2006) 2710 – 2720 2711

to bed-width ratio. There is some work on horizontal solids of solids mixing. However, the results are limited to qualitative
mixing in large fluidized-bed units which is based on estimat- interpretation, since quantitative analysis would require mea-
ing an effective horizontal dispersion coefficient (Highley and surements in a real combustor or a 3-dimensional unit operated
Merrick, 1971; Xiang et al., 1987; Schlichthaerle and Werther, according to scaling laws. Still, the present work is motivated
2001; Niklasson et al., 2002) showing that such large units by the low level of knowledge on the fundamentals behind the
give dispersion coefficients which are an order of magnitude fuel mixing process.
larger than those found in the above-mentioned narrow labo-
ratory CFB units. In summary, modeling of the mixing of fuel 2. Experiments and image analysis
particles in a large fluidized bed based on the phenomenol-
ogy behind the process is not available at present. In fact, little The particle-tracking technique is based on capturing the
is known about the basic physics behind the mixing process, phosphorescence of a single particle in a 2-dimensional
since it is experimentally difficult to study the mixing with the fluidized bed with a transparent front wall. To maximize
process obviously being highly dynamic, the latter especially phosphorescence, the riser is placed in a dark chamber. In
valid under industrial conditions. Thus, there is need to apply addition, a special digital camera equipped with an enhanced
experimental techniques which can resolve the particle mixing digital gain circuitry is used (up to −24 dB at a shutter speed
in both space and time. of 4 × 10−2 s). This technique presents two main advantages
Particle tracking techniques can provide valuable informa- compared to tracking of pigmented particles as was applied
tion on the governing mechanisms of solids mixing. In the case in a first step of this work (Shen et al., 2004): the contrast
of fluidized beds, several particle tracking techniques have been between a phosphorescent particle and its dark surrounding
used to investigate solids mixing. Although non-tracking tech- is much sharper than between a colored particle and the bed
niques providing local data, such as Laser Doppler Anemom- material, and the use of glass beads as bed material allows
etry (Ibsen et al., 2002) or the use of radio transmitters (van phosphorescence to be seen also in cases where the tracer par-
Barneveld et al., 1987) exist, these are not able to provide infor- ticle is behind the bed material. Both of these facts increase
mation on the trajectory of a single particle. For 3-dimensional the percentage of videoframes in which the tracer particle
units, various tomography-tracking techniques have been ap- is detected in the digital image analysis (95% compared to
plied both in narrow laboratory units and in scale models of around 65% when using pigmented particles in the same unit,
large fluidized beds. In the latter case, works are available based c.f. Shen et al., 2004). A recording time of 20 min per run
on
-ray emission (Parker et al., 1993; Larachi et al., 1994; (yielding around 30,000 videoframes to be used in the subse-
Dudukovic, 2002), X-ray (Grassler and Wirth, 2000) and elec- quent digital image analysis) was chosen based on an analysis
trical capacitance (Du et al., 2002). In scale models of large of three 40-min runs at three different fluidization regimes
fluidized beds, scaling relationships (see Glicksman, 1984, and which showed that recordings longer than 15 min provided no
references therein) have to be applied if the conditions should significant additional information.
resemble those of a fluidized-bed combustor. Stein et al. (2002) The experimental facility used (shown in Fig. 1) is a 2-
applied
-ray-based particle-tracking tomography to validate dimensional cold riser with a cross-section of 0.02 m × 0.4 m
a simplified set of scaling laws in the viscous limit. In 2- and a height of 2.15 m, having a perforated plate as air distrib-
dimensional units, solids mixing have been studied using op- utor. The front side of the riser is made of transparent Perspex,
tical techniques (Guo et al., 2001; Bokkers et al., 2004; Shen which enables visual observation and videorecording of the
et al., 2004) which enable direct observation of the bed dynam- flow. Since the solids mixing in the bottom region is the focus
ics simultaneously with measurements, but these 2-dimensional of this work, only the first 0.85 m of the riser is video recorded.
studies are obviously limited to qualitative analyses. Glass beads with size and density values similar to those of
The above experimental investigations are mostly directed sand particles typically used in fluidized-bed boilers are used
towards studies on flow and mixing of the bed material itself, as bed material; ds,bed = 330
m and s,bed = 2600 kg/m3 (be-
using a tracer particle that mimes the bed material. When mix- longing to the Group B in the Geldart classification and having
ing of a fuel particle is to be studied, as in the works by Lim a minimum fluidization velocity umf = 0.12 m/s and a termi-
and Agarwal (1994) and Rios et al. (1986), the tracer particle nal velocity ut = 1.76 m/s under ambient conditions). Hence,
should typically have larger size and lower density than the bed the bed solids are not scaled but the operational conditions (flu-
material. idization velocity, bed height and bed-height to bed-width ra-
There is a need to investigate the phenomenology of solids tio and air-distributor pressure drop) cover conditions with an
mixing under conditions relevant for fluidized-bed combustors, overall flow pattern similar to that in bubbling fluidized-bed
covering a large range of operational conditions and focusing on boilers as well as in the bottom bed of circulating fluidized-bed
the fuel mixing. This work introduces a novel particle tracking boilers.
method for 2-dimensional fluidized beds, with the tracer parti- Tracer particles used are cylindrical capsules made of trans-
cle simulating a fuel particle and the operational conditions em- parent plastic filled with a self-phosphorescent solution. The
ployed qualitatively resembling the dynamics of fluidized-bed so-called reference tracer particle (used in most of the exper-
combustors. The method is robust in that the dynamics of the imental runs) has a larger size and a lower density compared
mixing can be studied over a wide range of operational param- with the bed solids, making it suitable for simulating a fuel par-
eters allowing for a fundamental study on the phenomenology ticle in a fluidized-bed boiler. In order to study the influence
2712 D. Pallarès, F. Johnsson / Chemical Engineering Science 61 (2006) 2710 – 2720

Table 2
Conditions in experimental runs

Run Fluidization material Tracked Height of dense

velocity (m/s) (kg) particle bottom bed (m)

1 0.40 1.5 Reference 0.15

2 1.11 1.5 Reference 0
3 1.74 1.5 Reference 0
4 1.11 4.0 Reference 0.25
5 0.40 5.0 Reference 0.48
6 0.54 5.0 Reference 0.49
7 0.95 5.0 Reference 0.50
8 1.11 5.0 Reference 0
9 0.40 7.0 Reference 0.69
10 0.95 7.0 Reference 0
11 0.40 1.5 SS 0.15
12 1.11 1.5 SS 0
13 0.40 4.0 SS 0.32
14 1.11 4.0 SS 0.25
15 0.40 1.5 HD 0.15
16 0.40 4.0 HD 0.32

Or, expressed in the discrete form employed in digital image

analysis by using a total of N pixels:


N
Cfuel,n An = 1 with n = 1, . . . , N. (2)
1
Fig. 1. Experimental facility. The riser has a cross section of 0.02 m × 0.4 m
and a height of 2.15 m. A total of 4 parameters have been varied in the experiments
in order to study their influence on the mixing process: the
fluidization velocity, u0 , the amount of bed material, mbed , and
Table 1
the size and density of the tracer particle, ds,fuel and s,fuel ,
Properties of tracer particles used respectively. Table 2 lists the combinations of these parameters
applied in the experimental runs.
Tracked Height Diameter Density Terminal
The time-averaged height of the dense bottom bed is ob-
particle (mm) (mm) (kg/m3 ) velocity (m/s)
tained from an analysis of the video frames and the values are
Reference 12 10 985 10.9 included in Table 2 (a value of “0” indicates absence of a dense
SS 6 5 985 7.5 bottom bed, i.e., pneumatic transport regime). The fluidization
HD 12 10 2760 19.2
velocity, u0 , was always lower than the terminal velocity of
the average size of the bed particles forming the bed material
(ut,bed = 1.76 m/s). Still, circulating conditions or even pneu-
matic transport regime could be reached already at a fluidiza-
of size and density of the tracer particle, two additional types tion velocity of around 1 m/s. This can be explained by the
of tracer particles were used: one type smaller in size (type parabolic gas velocity profile established along the narrow di-
“SS”) and one with higher density, similar to that of the bed mension of the cross section, with local velocities in the center
solids (type “HD”). Table 1 summarizes sizes and densities of of the cross section considerably higher than the average ve-
the tracer particles used in the experiments. locity. This is one reason why results from 2-dimensional units
The method of evaluation is based on tracking a single parti- can only be qualitatively interpreted.
cle in each experimental run, so interactions between fuel par- With respect to the representation of the results, the velocity
ticles are assumed not to influence their mixing process, which field is given in form of a vector plot with the normalized
is reasonable knowing that the fuel fraction represents only be- concentration field plotted in the background. The height of the
tween 1% and 3% of the total bed material in fluidized-bed dense bottom bed, if existing, is indicated by a horizontal black
boilers. With this assumption, the probability of presence of line.
a single tracer particle can be seen as concentration value at Local values of the solids dispersion coefficients (at each
steady state, Cfuel , normalized so that: pixel) have been calculated as


Cfuel dA = 1. (1) lk,n
2
Dk,n = with k = x, y and for n = 1, . . . , N. (3)
A
2 · tn
 D. Pallarès, F. Johnsson / Chemical Engineering Science 61 (2006) 2710 – 2720 2713

Methods for calculation of the local dispersion coefficients from

experiments vary widely in literature. By setting tn equal to
the shutter time of the videocamera (i.e., very low, generally
in the order of 10−2 s or lower) and thus taking lk,n as the
displacement occurred from one videoframe to the following
implies values of the dispersion coefficient that do not account
for the history of the trajectory. On the other hand, setting
large values of tn in such a small unit as the one applied in
this work increases the risk of a value of tn larger than the
characteristic time for solids mixing in the unit, resulting in an
underestimation of the dispersion coefficient. A more robust and
reliable calculation strategy is to set the distance lk,n instead
of tn . For eminently convective flows (as the present case),
lk,n has to be given relatively high values in order to capture
the history of the trajectory and minimize the dependence of
the result on the election of lk,n . Thus, in the present work
lk,n is therefore set to 0.1 m for all cases analyzed. Thus, tn
is defined as the time it takes for the tracer particle at each point
on its trajectory to reach the distance of 0.1 m.
The global dispersion coefficient of a given run is calculated
by weighting the local dispersion coefficients with respect to
concentration of tracer particle, i.e.,

N
Dk = Cfuel,n · An · Dk,n with k = x, y and
1
for n = 1, . . . , N. (4)

3. Results

3.1. General pattern

Fig. 2. Schematic illustration of the flow structure of the tracer particle as
In all runs having a dense bottom bed, preferred bubble paths
obtained from the experiments in this work.
were observed as schematically illustrated in Fig. 2. This phe-
nomenon is typical for units with low pressure drop across the
air distributor, such as the present unit and also large industrial structure around each preferred bubble path, whose cut section
fluidized-bed units (Werther, 1977). These bubble paths, which in form of two vortexes along each preferred bubble path can
can easily be identified by visual observations, exhibit in all be identified in the present 2-dimensional observations. From
runs the lowest tracer particle concentration values in the dense this, the number of vortexes in the presence of a dense bottom
bottom bed as well as ascending velocities of the tracer parti- bed is even and depends on the amount of main bubble paths,
cle. This is exemplified in Figs. 3 and 4, which, respectively as can be seen in Fig. 4a (two main bubble paths and four vor-
show frames from the video recordings and corresponding re- texes) and Fig. 4b (one main bubble path and two vortexes).
sults from the digital image analysis. In Fig. 4a (corresponding Highest concentration values are found at the rotational cen-
to Run 1 in Table 2), two preferred bubble paths can be iden- terpoints (or centerlines) of the vortexes. The centerpoint (or
tified as vertical channels through the bottom bed, with low upper end of the vertical centerline) is located close to the dense
concentration and upward velocities. A video frame for the cor- bottom bed surface (see Fig. 8 below). Another general obser-
responding run is given in Fig. 3a in which these two bubble vation is that neighboring vortexes sharing downflowing parts
paths can be seen. do not have their points of highest tracer particle concentration
All runs exhibit flow pattern of the tracer particle which can in their respective rotational centers but in their common in-
be characterized by horizontally aligned vortexes with alter- terface. This is for example seen in Fig. 4a, with a total of 4
nated rotational direction, as indicated in Fig. 2. This vortex vortexes, where the second and third ones share a single high
structure is induced by the fact that preferred bubble paths act concentration zone (centerline of bed).
as ascending channels for the tracer particle through the dense
bottom bed. After reaching the bottom bed surface, the tracer 3.2. Influence of amount of bed material
particle is thrown into the splash zone, where it exhibits a
fast horizontal displacement before sinking again in the bottom One of the main operational parameters in fluidized-bed units
bed through the emulsion phase. In a 3-dimensional case, this is the pressure drop over the riser, which can be increased or
pattern presumably leads to the formation of a toroidal flow decreased by, respectively, adding or subtracting bed material.
2714 D. Pallarès, F. Johnsson / Chemical Engineering Science 61 (2006) 2710 – 2720

Fig. 3. Videoframes showing the development in the flow pattern with an increase in amount of bed material.

Fig. 4. Development of normalized concentration (m−2 ) and velocity of tracer particle (m/s) with an increase in amount of bed material (note the different
scales in the velocity vector plots). The horizontal line indicates the level of the time-averaged bottom bed height.

In practical terms, this has a direct effect on the dense bottom et al., 2001). Thus, adding bed material lowers, by means of
bed height, which becomes higher as bed material is added (see bubble coalescence, the number of preferred vertical bubbles
Fig. 3) provided the fluidization velocity is below the terminal paths, and, consequently, the number of vortexes in the flow
velocity of the bed material. The value of the dense bottom bed structure. All cases shown in Figs. 3 and 4 are characterized
height in each run is given in Table 2. by a bubbling regime. Data from Runs 1, 5 and 9 (shown in
Bubble coalescence is known to be enhanced by increasing Figs. 4a–c), for which the amount of bed material is varied
the dense bottom bed height (Darton et al., 1977; de Korte at a constant fluidization velocity, exemplify a change from a
 D. Pallarès, F. Johnsson / Chemical Engineering Science 61 (2006) 2710 – 2720 2715

Fig. 5. Development of local horizontal dispersion coefficients (m2 /s) with an increase in amount of bed material. The horizontal line indicates the level of
the time-averaged bottom bed height.

4-vortexes structure with 2 main bubble paths (Run 1) to a 2-

vortexes structure with one main bubble path (Runs 5 and 9).
Through the operational range covered by all runs, the num-
ber of vortexes forming the flow structure was found to depend
only on the amount of bed material and not on the regime or the
fluidization velocity. However, this behavior might not prevail
beyond the operational ranges tested and, therefore, cannot be
taken as a general statement. In the presence of a dense bottom
bed, this behavior is expected, since it is known (Darton et al.,
1977) that the amount of main bubble paths (and thereby, of
vortexes in the flow pattern of the tracer particle) has a higher
dependence on the dense bottom bed height (strongly related
to the amount of bed material) than on fluidization velocity. On
the other hand, observation of this behavior in the absence of
a dense bottom bed (as seen in Runs 2, 3 and 12, for which Fig. 6. Global horizontal dispersion coefficients (as obtained from Eq. (4))
tracer particle flow patterns are all governed by a 4-vortexes for different amount of bed material.
structure) is not available in literature and, as mentioned above,
this behavior might not prevail at higher fluidization velocities.
In addition to the number of vortexes in the flow structure, Fig. 5 shows the distribution of the local horizontal dispersion
the amount of bed material is also found to have a strong in- coefficient as obtained from Eq. (3). It can be seen for all three
fluence on the concentration and velocity fields of the tracer runs (Figs. 5a–c) that the highest dispersion values coincide
particle. The results shown in Figs. 4a–c show how the high- with the locations of bubble bursting (two main bubble paths
est concentration values of the tracer particle decrease drasti- are identified in Run 1 and bubble coalescence leads to a single
cally as the bed mass is increased, resulting in a considerable main bubble path in Runs 5 and 9). Also, the lowest values
increase in the velocity of the tracer particle although the flu- of the horizontal dispersion are obtained near the bottom bed
idization velocity is kept constant (note the change in scales surface close to the sidewalls, i.e., where the fuel particles are
between the figures). An increase in bed height results in bub- most likely to end up after being fed into the unit (with over
bles growing larger (from coalescence as well as from inflow bed feed) if no special measures against this are taken. Thus,
from the dense phase) and reaching a higher velocity and, as a these results may motivate installations of equipment avoiding
result, the tracer particle dragged upwards by the bubble flow fuel particles to end up in these zones after injection into the
reaches higher velocities. Bubbles from Runs 1, 5 and 9 shown riser from the fuel feeder.
in Fig. 3 show a large difference in size. As expected, bursts of Fig. 6 shows the dependence of the global horizontal disper-
the larger and faster bubbles in Fig. 3c throw the tracer particle sion coefficient (Eq. (4)) on the amount of bed material, with
higher up in the freeboard than what is obtained from smaller values corresponding to the above mentioned runs (Nos. 1, 5
bubbles in Run 1, indicating a more vigorous splash zone than and 9) marked with crosses. A complete list of values of the
for the low bed mass case (Fig. 3a). global dispersion coefficients obtained for all runs is given in
2716 D. Pallarès, F. Johnsson / Chemical Engineering Science 61 (2006) 2710 – 2720

Table 3 periodical pattern in the bubble flow. The solids suspension

Global dispersion coefficients above the bottom bed surface increases and pronounced down-
Run Dx × 102 (m2 /s) Dy × 102 (m2 /s) flowing wall layers are formed. This is the typical regime for
the dense bottom bed and splash zone of a CFB unit. Finally,
1 0.14 0.04
with a further increase in fluidization velocity, the transport ve-
2 2.06 2.93
3 0.93 1.77 locity is reached and a dense bed can no longer be maintained,
4 1.70 4.19 i.e., these conditions are similar to pneumatic transport (Run 8,
5 0.41 0.83 Fig. 7d). The flow regime is characterized by a flow structure
6 0.89 2.18 with a dilute core and denser regions near the riser walls. In
7 1.18 3.89
the central region the gas and solids flow are primarily directed
8 1.75 4.62
9 0.86 2.51 upwards and, provided the solids suspension is dense enough
10 1.46 4.22 to drag the tracer particle upwards, the tracer particle is mov-
11 0.18 0.21 ing upwards in the central part and descending in the denser
12 0.96 2.58 downflowing solids wall-layers, i.e., there is a net lateral solids
13 0.46 0.89
flow from the center of the riser to the side walls in the upper
14 1.57 6.74
15 0.57 0.61 part of the riser and in the opposite direction in the bottom part.
16 0.85 1.27 This structure extends up through the entire riser and results in
an increased flow of externally recirculated solids.
Results from the digital image analysis of Runs 5–8 are
shown in Fig. 8. Firstly, it is observed that a 2-vortexes flow
Table 3. Data in Fig. 6 correspond to runs under different flu- structure is kept under all runs (with upflow in the center of
idization regimes, and interconnected symbols correspond to the riser and downflow by the sidewalls), which is in agree-
runs with the same fluidization velocity and type of tracer par- ment with the above conclusion that the number of vortexes
ticle. An increased amount of bed mass results in an increased depends to a large extent on the amount of bed material. It is
global horizontal dispersion coefficient, except in one particu- also seen that, within the domain studied, the velocity of the
lar case. Thus, the higher the dense bottom bed, the higher the tracer particle increases with gas velocity as long as a dense
overall lateral mixing rate, i.e., a higher bed (with less frequent bed exists, but decreases in the freeboard when the dense bed
but more vigorous bubbles) spreads fuel more effectively than disappears at a sufficiently high fluidization velocity. Thus, ex-
a low bed (with more frequent but smaller bubbles). The ex- istence of a dense bottom bed with fast exploding bubbles
ception to this trend is represented by Run 2 (corresponding (Fig. 8c) gives high initial velocities of the particles projected
the highest value for Dx in Fig. 6), in which no dense bottom into the freeboard, i.e., higher than what the dispersed flow
bed is present (no bubbles are observed) but a solids suspension without a dense bed (Fig. 8d) can produce even at a higher
dense enough to displace the “Reference” tracer particle. gas velocity (1.11 m/s compared with 0.92 m/s). This should
be expected since the momentum of bed solids is considerably
3.3. Influence of fluidization velocity higher than that of the gas, i.e., the higher the concentration of
bed solids the higher is the momentum which can act on the
Runs number 5–8 (with typical video frames shown in Fig. 7) tracer particle.
correspond to different fluidization regimes which is obtained Fig. 9 gives corresponding distribution of local horizontal
by gradually increasing the fluidization velocity while all other dispersion coefficients for the above-mentioned series of runs
parameters are kept constant (mbed = 5 kg, with “Reference” (Runs 5–8). As for the vertical dispersion in the presence of a
particle). dense bottom bed, the horizontal dispersion increases with flu-
Run 5 (Fig. 7a) corresponds to a low velocity bubbling flu- idization velocity, with the highest values obtained around the
idized bed, with clear-cut bubbles rising through the bed and locations of bubble eruptions. Highest dispersion is seen in the
erupting at a rather constant frequency, resulting in a small dense bottom bed and low values are found in the vicinity of
splash zone above the bed surface. An increase in fluidization the sidewalls. There is a significant change in the picture as the
velocity (Run 6, Fig. 7b) results in larger, faster and more jet- fluidization velocity is increased further, so that the unit is op-
shaped bubbles that occasionally extend all the way from the erated without a dense bottom bed: local horizontal dispersion
air-distributor to the bed surface, creating a low pressure drop coefficients are higher and more evenly distributed. However,
channel for the fluidization air during the eruption of the bub- this high horizontal dispersion observed in Run 8 (note the dif-
bles yielding a high throughflow of gas (see Svensson et al., ferent scales used in Fig. 9) cannot be considered a general
1996 for a thorough description of this regime). Under these characteristic for pneumatic transport conditions, but decreases
conditions, bed dynamics are still rather periodic with a low as fluidization velocity is further increased. This, since the solid
density of solids suspension above the bottom bed surface. With phase becomes more dilute as the fluidization velocity is in-
a further increase in fluidization velocity (Run 7, Fig. 7c), the creased, resulting in a reduction in the momentum exchange
bed enters an exploding bubble regime where throughflow turns between bed material and tracer particle and, as a result, yield-
from occasional to a main characteristic of the flow pattern, ing a corresponding reduction in the global dispersion coeffi-
coexisting with erupting jet-shaped bubbles, with no clearcut cient as fluidization velocity is increased. This can be observed
 D. Pallarès, F. Johnsson / Chemical Engineering Science 61 (2006) 2710 – 2720 2717

Fig. 7. Videoframes showing the development in the flow pattern with an increase in fluidization velocity. The horizontal line indicates the level of the
time-averaged bottom bed height.

Fig. 8. Development of normalized concentration (m−2 ) and velocity of tracer particle (m/s) with an increase in fluidization velocity. The horizontal line
indicates the level of the time-averaged bottom bed height.

Fig. 9. Development of local horizontal dispersion coefficients (m2 /s) with an increase in fluidization velocity. The horizontal line indicates the level of the
time-averaged bottom bed height.
2718 D. Pallarès, F. Johnsson / Chemical Engineering Science 61 (2006) 2710 – 2720

3.4. Influence of size of tracer particle

Comparison between the “Reference” and the “SS” tracer

particles gives a picture of the influence of the size of the tracer
particle, illustrated in Fig. 11 (showing results from Runs 2 and
11, both corresponding to a bubbling regime). There is a signif-
icant increase in the velocity of the tracer particle as its size is
decreased. As expected, the 4-vortexes flow structure induced
by the bed material is maintained. From Fig. 11 it can also be
seen that the smaller tracer particle (“SS”) tends to occupy also
the lower part of the dense bottom bed (where there are nearly
no occurrences of the “Reference” tracer particle), and, conse-
quently, a more homogeneous concentration distribution in the
vertical direction is obtained with the smaller tracer particle.
Fig. 10. Global horizontal dispersion coefficients (as obtained from Eq. (4))
Comparison of runs under the exploding bubble regime (Runs
for different fluidization velocities.
4 and 14) gives a similar observation: a more homogeneous
distribution of the concentration field in the vertical direction
from a comparison of Runs 2 and 3, both under pneumatic when the smaller “SS” particle is used. Thus, in presence of a
transport regime and with a fluidization velocity of 1.11 m/s dense bottom bed, global horizontal dispersion coefficients re-
and 1.74 m/s, respectively: the highest velocity case gives a re- main nearly unaffected by changes in size of the tracer particle,
duction of 55% in the global horizontal dispersion coefficient. while vertical dispersion is strongly increased when a tracer
In all cases where there is no bottom bed, the distribution of particle with a smaller size is used.
the local horizontal dispersion coefficient is rather uniform. For transport conditions (no dense bottom bed) a change in
Fig. 10 shows the dependence of the global horizontal disper- size of the tracer particle gives no significant change in the
sion coefficient on fluidization velocity for different amounts results (Runs 2 and 12).
of bed material and types of tracer particle. The runs shown
in Fig. 9 (Runs 5–8) are marked with crosses. There is a clear 3.5. Influence of density of tracer particle
trend indicating that the dispersion increases with fluidization
velocity as long as a dense bottom bed is maintained. On the Results from Runs 1 and 15, with identical operational con-
other hand, under pneumatic transport conditions and beyond a ditions except for different density of the tracer particle, are
certain value of the fluidization velocity, dispersion of the tracer shown in Fig. 12. An increase in the density of the tracer particle
particle decreases with an increase in fluidization velocity, as (“HD” particle) leads to a drastic change in the concentration
discussed above (see transition from Runs 2 to 3). field with high concentration zones replaced from the bottom

Fig. 11. Development of normalized concentration (m−2 ) and velocity of tracer particle (m/s) with a decrease in tracer particle size (note the different scales
in the velocity vector plots).
 D. Pallarès, F. Johnsson / Chemical Engineering Science 61 (2006) 2710 – 2720 2719

Fig. 12. Development of normalized concentration (m−2 ) and velocity of tracer particle (m/s) with an increase in tracer particle density (note the different
scales in the velocity vector plots).

bed surface to just above the air distributor, coinciding with the and shown to be an efficient tool for characterization of the
preferred bubble paths. Thus, the tracer particle remains near particle mixing process. The technique has been used to sim-
the air-distributor until a bubble is able to drag it up to the dense ulate the mixing of a fuel particle in a fluidized bed, which,
bed surface and eject it into the splash zone, which is seldom as a general result, is found to follow a flow pattern in hori-
the case according to the low concentration values obtained in zontally aligned vortexes induced by main bubble paths which
the upper zone of the bed. In these occasions, however, the were characteristic for the conditions investigated.
tracer particle is accelerated all along the dense bed height, The influence of four parameters on the solids mixing has
differing from the low-density “Reference” tracer particle case been studied and the following conclusions can be drawn:
where the acceleration is typically limited to the upper half of
the dense bed (since the particle rarely resides in the lower half 1. Increasing the amount of bed material is found to reduce,
of the bed, see Fig. 12a). Thus, the high-density tracer particle by means of bubble coalescence, the number of vortexes
has a longer accelerating path and thereby reaches the dense forming the flow picture, increase the homogeneity of the
bottom bed surface with a higher velocity, making its occasional concentration distribution of the tracer particle, increase
trajectories into the freeboard reach higher positions than in the velocity of the tracer particle, and increase the global
the “Reference” tracer particle case. As expected, downflow horizontal dispersion coefficient.
velocities are considerably higher for the high-density tracer 2. Increasing the fluidization velocity leads to an increase in
particle than for the low density particle due to gravity. tracer-particle velocities and in global dispersion coeffi-
It should be emphasized that the 4-vortexes flow pattern with cients provided there is a dense bottom bed, while the op-
three downflowing channels (by the sidewalls and in the central posite result is obtained if there is no bottom bed present.
region of the bed) is maintained in both cases. 3. Reduction in the size of the tracer particle gives a more
Runs with a “HD” tracer particle in the absence of a dense homogeneous concentration distribution and increased ver-
bottom bed were carried out, but this only resulted in that the tical dispersion, whereas the horizontal dispersion remains
tracer particle ended up resting on the air distributor (i.e., these nearly unaffected.
cases are not reported here). This is expected since the terminal 4. An increase in density of the tracer particle gives high con-
velocity of the “HD” particle significantly exceeds the superfi- centration zones in the lowest part of the bottom bed and
cial gas velocity and the low solids suspension density cannot high velocities of the tracer particle above these zones.
exchange a high enough momentum on the tracer particle for
this to be entrained. The dispersion of the tracer particle is generally higher in
the vertical than in the horizontal direction. In the presence of
4. Conclusions a dense bottom bed, the horizontal dispersion coefficient ex-
hibits the highest values around the locations at the bottom bed
A novel technique for particle-tracking in 2-dimensional flu- surface where eruptions of bubbles or jets occur, and low val-
idized beds with subsequent digital image analysis is presented ues in the bottom bed surface near the sidewalls. Absence of a
2720 D. Pallarès, F. Johnsson / Chemical Engineering Science 61 (2006) 2710 – 2720

dense bottom bed enhances horizontal dispersion provided the Highley, J., Merrick, D., 1971. The effects of the spacing between solid
solids suspension is dense enough, but leads to a decrease in feed points on the performance of a large fluidized bed reactor. A.I.Ch.E.
dispersion when the solids suspension is dilute. Also, local hor- Symposium Series 67, 219.
Ibsen, C.H., Solberg, T., Hjertager, B.H., Johnsson, F., 2002. Laser Doppler
izontal dispersion coefficients are more uniformly distributed anemometry measurements in a circulating fluidized bed of metal particles.
than in the presence of a dense bottom bed. Experimental Thermal and Fluid Science 26 (6–7), 851–859.
Larachi, F., Kennedy, G., Chaouki, J., 1994. A gamma-ray detection system
Notation for 3-D particle tracking in multiphase reactors. Nuclear Instruments and
Methods in Physics Research Section A: Accelerators, Spectrometers,
Detectors and Associated Equipment 338 (2–3), 568–576.
A area, m2 Lim, K.S., Agarwal, P.K., 1994. Circulatory motion of a large and lighter
Cfuel normalized concentration of tracer particle, m−2 sphere in a bubbling fluidized bed of smaller and heavier particles.
Chemical Engineering Science 49 (3), 421–424.
D dispersion coefficient, m2 /s
Lin, Y.-C., Chyang, C.-S., 2003. Radial gas mixing in a fluidized bed using
N number of pixels, dimensionless response surface methodology. Powder Technology 131 (1), 48–55.
l reference displacement, m Mostoufi, N., Chaouki, J., 2001. Local solid mixing in gas–solid fluidized
t time interval, s beds. Powder Technology 114 (1–3), 23–31.
Niklasson, F., Thunman, H., Johnsson, F., Leckner, B., 2002. Estimation of
Subindexes solids mixing in a fluidized-bed combustor. Industrial and Engineering
Chemistry Research 41 (18), 4663–4673.
n pixel index Palchonok, G.I., Dikalenko, V.A., Stanchits, L.K., Borodulya, V.A., Werther, J.,
k dimension index Leckner, B., 1997. Kinetics of the main stages of fluidized bed combustion
of a wet biomass particle. Proceedings of the 14th International Conference
on Fluidized Bed Combustion, pp. 221–225.
Acknowledgements Parker, D.J., Broadbent, C.J., Fowles, P., Hawkesworth, M.R., McNeil,
P., 1993. Positron emission particle tracking a technique for studying
The authors gratefully acknowledge financial support from flow within engineering equipment. Nuclear Instruments and Methods in
the Swedish Energy Agency and the EU Research Fund for Physics Research, Section A: Accelerators, Spectrometers, Detectors and
Coal and Steel under contract RFC-CR-03001. Associated Equipment A 326 (3), 592–607.
Peirano, E., Delloume, V., Johnsson, F., Leckner, B., Simonin, O., 2002.
Numerical simulation of the fluid dynamics of a freely bubbling fluidized
References bed: influence of the air supply system. Powder Technology 122 (1),
69–82.
Agarwal, P.K., Genetti, W.E., Lee, Y.Y., 1986. Coupled drying and Peters, B., Bruch, C., 2001. A flexible and stable numerical method for
devolatilization of wet coal in fluidized beds. Chemical Engineering simulating the thermal decomposition of wood particles. Chemosphere 42
Science 41 (9), 2373–2383. (5–7), 481–490.
Bokkers, G.A., van Sint Annaland, M., Kuipers, J.A.M., 2004. Mixing and Rios, G.M., Dang, K., Masson, H., 1986. Free object motion in a gas fluidized
segregation in a bidisperse gas–solid fluidised bed: a numerical and bed. Chemical Engineering Community 47, 247–272.
experimental study. Powder Technology 140 (3), 176–186. Schlichthaerle, P., Werther, J., 2001. Solids mixing in the bottom zone of a
Darton, R.C., LaNauze, R.D., Davidson, J.F., Harrison, D., 1977. Bubble- circulating fluidized bed. Powder Technology 120 (1–2), 21–33.
growth due to coalescence in fluidized beds. Transactions of the Institution Shen, L., Johnsson, F., Leckner, B., 2004. Digital image analysis
of Chemical Engineers 55 (4), 274–280. of hydrodynamics two-dimensional bubbling fluidized beds. Chemical
de Korte, R.J., Schouten, J.C., Van den Bleek, C.M., 2001. Controlling bubble Engineering Science 59 (13), 2607–2617.
coalescence in a fluidized-bed model using bubble injection. A.I.Ch.E. Stein, M., Ding, Y.L., Seville, J.P.K., 2002. Experimental verification of
Journal 47 (4), 851–860. the scaling relationships for bubbling gas-fluidised beds using the PEPT
Du, B., Wei, F., 2002. Lateral solids mixing behavior of different particles technique. Chemical Engineering Science 57 (17), 3649–3658.
in a riser with FCC particles as fluidized material. Chemical Engineering Svensson, A., Johnsson, F., Leckner, B., 1996. Fluidization regimes in non-
and Processing 41 (4), 329–335. slugging fluidized beds: the influence of pressure drop across the air
Du, B., Fan, L.-S., Wei, F., Warsito, W., 2002. Gas and solids mixing in a distributor. Powder Technology 86 (3), 299–312.
turbulent fluidized bed. A.I.Ch.E. Journal 48 (9), 1896–1909. Thunman, H., Leckner, B., Niklasson, F., Johnsson, F., 2002. Combustion of
Dudukovic, M.P., 2002. Opaque multiphase flows: experiments and modeling. wood particles—a particle model for Eulerian calculations. Combustion
Experimental Thermal and Fluid Science 26 (6–7), 747–761. and Flame 129 (1–2), 30–46.
Enwald, H., Almstedt, A.-E., Peirano, E., 1996. Eulerian two-phase flow van Barneveld, J., Oosterhuis, N.M.G., Pragt, H.J., Smit, W., 1987. Measuring
theory applied to fluidization. International Journal of Multiphase Flow the liquid circulation time in a large gas–liquid contactor by means of
22 (Suppl. 1), 21–66. a radio pill—1: flow pattern and mean circulation time. Industrial and
Fan, L.T., Song, J.C., Yutani, N., 1986. Radial particle mixing in gas-solids Engineering Chemistry Research 26 (11), 2185–2192.
fluidized beds. Chemical Engineering Science 41 (1), 117–122. van Wachem, B.G.M., van der Schaaf, J., Schouten, J.C., Krishna, R., van
Glicksman, L.R., 1984. Scaling relationships for fluidized beds. Chemical den Bleek, C.M., 2001. Experimental validation of Lagrangian–Eulerian
Engineering Science 39 (9), 1373–1379. simulations of fluidized beds. Powder Technology 116 (2–3), 155–165.
Grassler, T., Wirth, K.-E., 2000. X-ray computer tomography—potential and Werther, J., 1977. Bubble chains in large diameter gas fluidized beds.
limitation for the measurement of local solids distribution in circulating International Journal of Multiphase Flow 3 (4), 367–381.
fluidized beds. Chemical Engineering Journal 77 (1–2), 65–72. Xiang, Q., Huang, G., Ni, M., Cen, K., Tao, T., 1987. Lateral dispersion
Guo, Q., Yue, G., Zhang, J., Liu, Z., 2001. Hydrodynamic characteristics of of large coal particles in an industrial-scale fluidized bed combustor.
a two-dimensional jetting fluidized bed with binary mixtures. Chemical Proceedings of the Ninth International Conference on Fluidized Bed
Engineering Science 56 (15), 4685–4694. Combustion, p. 546.
 Paper V

Pallarès, D., Díez, P., Johnsson, F. (2007)

”Experimental analysis of fuel mixing patterns in a fluidized bed”
Proc. of the 12th Int. Conf. on Fluidization (Harrison, Canada)
 th
12 International Conference on Fluidization. Harrison (Canada), 2007.

Experimental analysis of fuel mixing patterns in a fluidized bed

David Pallarès, Pedro A. Díez, Filip Johnsson

Department of Energy and Environment, Chalmers University of Technology, SE-41296 Göteborg (Sweden)
Telf: +46-317721449; E-mail: [email protected]

ABSTRACT

The mixing pattern of a tracer particle which simulates a fuel particle is studied in a cold 2-dimensional fluidized
bed with respect to the influence of fluidization velocity, bed height, tracer particle size and air-distributor pressure
drop under conditions typical for bubbling fluidized bed boilers as well as the bottom region of circulating fluidized
bed boilers. The results show that for all conditions studied, the tracer particle follows a flow pattern structured into
horizontally-aligned vortexes.

INTRODUCTION

Fuel mixing is a key phenomenon for the performance of fluidized bed (FB) boilers (bubbling as well as
circulating). In the vertical direction, good fuel mixing is important to ensure sufficient contact time between fuel
and combustion air. In the horizontal direction, the fuel dispersion determines to what extent there will be
homogeneous cross-sectional fuel distribution which is important for fuel burnout while allowing a low excess air
ratio (thus minimizing operational costs). For typical fluidization conditions, horizontal solids mixing in the bottom
region of an FB unit was found to be lower than in the vertical direction (1). This, together with the fact that the bed
of an FB boiler (where most of the fuel inventory is present) usually has a height-to-width ratio lower than 1, makes
the horizontal direction critical in terms of fuel mixing. Moreover, for economical reasons, the number of fuel
feeding points should be kept as low as possible (2), which is obviously strongly related to the degree of horizontal
fuel mixing.
To what extent a certain fuel mixing behaviour is sufficient or not depends on the fuel conversion time and the
characteristic mixing length. A comparison of the characteristic times for fuel dispersion and conversion can be
expressed by the Damköhler number:

L*
τ dispersion rdispersion
Da = = (1)
τ conversion τ conversion
Thus, the Damköhler number (Da) is a suitable parameter for evaluation of fuel mixing in FB units. The Da
number indicates whether the dispersion rate is high enough to ensure a sufficiently homogeneous distribution of the
fuel over the entire cross section of the unit (which is the case for low values of the Damköhler number, Da<1). It is
seen from Eq. (1) that operational conditions which yield a sufficient fuel mixing rate in a certain FB burning a
certain fuel may not be sufficient when changing fuel (e.g. to a fuel with a higher volatile content or which is more
reactive). It is evident that horizontal fuel mixing becomes a critical issue in large CFB boilers, which may have a
cross-sectional area of up to several hundreds of square meters.
To the authors knowledge there are no models on the fuel mixing which give satisfactory agreement with
experimental data under conditions applicable to FB boilers and which are based on the underlying physics of the
mixing process. Although, there has been significant progress in numerical modelling from first principles,
computational fluid dynamics (CFD) is generally limited to a mono-sized solid phase (3-5), while simulating fuel
mixing obviously requires accounting for a polydispersed solid phase (bulk particles and fuel particles). There are
 th
12 International Conference on Fluidization. Harrison (Canada), 2007.

some attempts in literature to account for polydispersed solids in CFD simulations (6, 7) but not really concerning
fuel mixing, which requires one of the phases (fuel) to have a much smaller fraction (less than 5%) than that of the
bulk and to be lighter as well. An attempt to implement conditions corresponding to fuel mixing was made by
Tanskanen (8), but more work is required until realistic results can be obtained. Thus, there is a need for
semiempirical models expressing the solids mixing as an overall dispersion coefficient to be used as a tool to
simulate horizontal fuel dispersion (the word “dispersion” is here used for simplicity since measured and modelled
solids mixing is normally expressed as an average dispersion coefficient, although the mixing process is highly
convective). Effective horizontal solids dispersion coefficients in FB boilers have been estimated by means of
experimental data from boilers (9, 10) and from cold rigs (11). Niklasson et al. (9) carried out experiments in the
Chalmers 12 MW FB boiler operated under bubbling conditions, obtaining a value of the horizontal fuel dispersion
coefficient of around 0.1 m2/s, a result which seems to be consistent with the boiler values reported by
Xiang et al. (10), ranging from 0.01 to 0.1 m/s2 for fluidization velocities lower than those applied in (9). However,
experimental evaluation of the dispersion only in the form of global dispersion coefficients is a limitation in the
sense that this gives no information on the mechanisms governing the mixing process. Application of particle
tracking techniques can provide an experimental basis for resolving the particle mixing process in both space and
time.
Several particle tracking techniques have been used to investigate solids mixing in FB units (cf. 12) for a survey
of experimental investigations on fluidized-bed solids mixing). Techniques measuring in a fixed point (Eulerian)
exist, such as Laser Doppler Anemometry or the use of radio transmitters, but no information on the trajectory of a
single particle can be obtained from such techniques. For tracking in 3-dimensional units, various tomographical
techniques have been developed based on X-ray, electrical capacitance or γ-ray emission. The latter was applied by
Stein et al. (13) in scale models of FB boilers under conditions accounting for fluiddynamical scaling relationships.
Experiments in 2-dimensional rigs have visual access to the dynamics also in dense beds as the main advantage but
are obviously limited to qualitative studies. Most experimental works focus on studying the mixing of the bed
material itself, i.e. using a tracer particle to mimic the bulk bed material. Accordingly, a tracer particle with larger
size and lower density than those forming the bulk bed material must be applied when mixing of a fuel particle is
studied, as done in (13-16). Yet, none of these works gives the velocity and concentration fields of the tracer particle
while varying main operational parameters. The authors of the present work studied the mixing pattern of a tracer
particle simulating a fuel particle in a cold 2-dimensional fluidized with the wide dimension being 0.4 m (12),
finding the flow pattern of the tracer particle to be structured into several horizontally-aligned vortexes with
alternated rotational direction. The question is to what extent this is also valid in a wider unit since the limited width
of the bed (0.4m) applied in the previous work may have influenced the horizontal spreading of the tracer particle.
Thus, the present work extends the previous work with the aim to further generalize the patterns of the fuel mixing
process, with focus on operational conditions typical for fluidized-bed boilers and with the experiments carried out
in a 2D bed with the wide dimension about three times (1.2 m) that applied in the previous work.

EXPERIMENTAL SETUP

This work applies the particle tracking technique developed in the previous work (12), which is suitable for
tracking particles coarser than the bed material (i.e. simulating fuel particles) in cold 2-dimensional fluidized beds and
is robust in that the dynamics of the mixing can be studied over a wide range of operational parameters, allowing for a
fundamental study on the phenomenology of solids mixing. The particle tracking technique is based on tracking a
single tracer particle (a plastic capsule filled with a phosphorescent solution) in a 2-dimensional fluidized bed with a
transparent front wall. The mixing process is then analysed and quantified by means of digital image analysis of the
trajectory of the capsule. To maximize phosphorescence, the riser is placed in a dark chamber. A special, high-gain
CCTV video camera with a time resolution of 4×10-2 s is used for filming the capsule in the bed. In addition, glass
beads are used as bed material, allowing phosphorescence to be seen through the bed material as when the tracer
particle flows close to the rear wall (opposite camera position). The 2-dimensional unit is illustrated in Figure 1. The
riser is 1.2 m wide with a depth of 0.02 m and a height of 2.05 m with a Plexiglas front wall. The gas flow is
controlled by a valve located close to the air plenum and the externally recirculated solids flow is refed into the riser
through the back wall (see item no. 9 in Figure 1). Two different perforated air-distributor plates are used, both with
2 mm i.d. holes and hole areas of 2% and 9% (called “high-ΔP” and “low-ΔP” air distributor, respectively), which
yields the ΔP vs u curves given in Figure 2. The glass beads forming the bed have a narrow particle size distribution
with an average size of 330 μm and a density of 2500 kg/m3, i.e. although no exact scaling is performed, these values
2
 th
12 International Conference on Fluidization. Harrison (Canada), 2007.

RUN
A B C D E F G H I
Fluidization velocity [m/s] 0.7 1.5 2.7 3.5 2.7 2.7 2.7 2.7 3.5

Fixed bed height [m] 0.33 0.33 0.33 0.33 0.33 0.18 0.18 0.18 0.33

Tracer size Large Large Large Large Small Large Large Small Large

Air distributor High-ΔP High-ΔP High-ΔP High-ΔP High-ΔP High-ΔP Low-ΔP High-ΔP Low-ΔP

Table 1: Test matrix

are similar to those of sand or ash particles typically used as bed material in fluidized bed boilers. The glass beads
belong to Group B in the Geldart classification with umf=0.12 m/s and ut=1.76 m/s (ambient conditions). For the
conditions applied, around 60 minutes of video recording of the tracer particle (providing approximately 90,000 video
frames) were shown to give high enough statistical significance in the digital image analysis. Thus, the method
requires significant number of data to reach statistical significance.
Table 1 lists the operational conditions for the 9 experimental runs carried out as basis for this work. The runs
cover flow conditions qualitatively resembling fluidization regimes typically found in a fluidized-bed boiler. Two sizes
of tracer particle were applied both with a size representative for fuels in FB boilers: 15 mm x 6 mm i.d. (“Large”) and
10 mm x 4 mm i.d. (“Small”). Although the bed is only 20 mm in depth, analysis of the trajectories of the tracer
particles do not indicate any significant influence of wall effects other than the above-mentioned general wall effect
which, of course, makes the results qualitative. Both tracers have an apparent particle density of around 650 kg/m3,
which is close to that of biomass, somewhat higher than that of coal, and lower than that of waste pellets. The different
regimes were reached by changing either operational conditions (fluidization velocity, amount of bed material and
tracer particle) or the air distributor.
Since the evaluation method is based on tracking a single particle in each run, interactions between fuel particles
are assumed negligible. This should be a reasonable assumption since the fuel fraction usually represents only between
1 and 5% of the total bed material in FB boilers. With this assumption, the probability of presence of the tracer particle
at some location can be interpreted as the concentration value at steady state, Cfuel, normalized so that:

∫C
A
fuel ⋅ dA = 1 (2)

Figure 1: The cold 2-dimensional FB unit Figure 2: Characteristic curves of the air distributor
used in the experiments. plates used in the experiments.

3
 th
12 International Conference on Fluidization. Harrison (Canada), 2007.

Table 2: Average dispersion coefficients as obtained in this work.

RUN
A B C D E F G H I
2 2
Dx [m /s]·10 0.31 1.11 1.59 2.30 1.48 1.16 1.08 1.20 4.24

Dy [m2/s]·102 0.11 0.53 1.67 4.01 2.32 1.14 0.90 1.22 3.00

This can be expressed in discrete form, more suitable for the digital image analysis:
N

∑C
1
fuel , n ⋅ An = 1 with n=1…N (3)

where N corresponds to the total number of pixels of the analyzed videoframe. It should be noted that the probability of
not capturing any phosphorescence from the tracer particle at a certain videoframe (around 30% in the present work) is
much higher when the tracer particle is immersed in the dense bottom bed than when it is flowing in the freeboard.
Thus, this uneven spatial distribution of the phosphorescence capture probability would bias the concentration plots
giving the bottom bed region lower concentration values. However, the location of the tracer during the short time
intervals (fractions of a second) in which its phosphorescence is not captured can be estimated by linear interpolation.
Although the method is restricted to 2D FB units and thereby to qualitative evaluation of the results obtained, tracer
particle dispersion is quantified to enable direct comparison of the runs. Thus, the local dispersion coefficient in
position n (i.e. pixel) in the direction k (horizontal or vertical in our case) at any location can be calculated as:

Δl k2,n
Dk , n = ; with k = x, y and for n = 1…N (4)
2 ⋅ Δt n
As discussed in (12), setting Δtn equal to a low value and thus taking Δlk,n as the displacement occurred in that time
interval would imply values of the dispersion coefficient that do not account for the history of the trajectory (which is a
prerequisite for defining a dispersion coefficient) and a pseudo-velocity value would be obtained instead. Setting Δtn to
a large value to avoid this effect would increase the risk of having Δtn larger than the characteristic time for the solids
mixing in the unit, also resulting in unrealistic values of the dispersion coefficients. Instead, the method used is to set
Δlk,n to a relatively high value in order to capture the history of the mixing process, which is shown to be a more robust
method for application of Eq (4). Thus, using this approach, Δtn is defined as the time it takes for the tracer particle to
flow a distance equal to Δlk,n, which has been set to 0.3 m in this work (a fourth of the riser width, as in (12)).
Finally, a global dispersion coefficient can be calculated by weighting the local dispersion coefficients with respect
to concentration values, i.e.
N
Dk = ∑ C fuel ,n ⋅ An ⋅ Dk ,n with k = x, y and for n = 1…N (5)
1

RESULTS

Table 2 lists the average dispersion coefficients obtained for the runs given in Table 1. It should be kept in mind
that evaluation of dispersion coefficients in eminently convective flows (as in the present work) has to be done with
care, since the values are strongly dependent on the calculation procedure (Δlk,n and Equations (4) and (5)), the 2D bed
flow and, as indicated above, the convective nature of the flow. Thus, the values in Table 2 have a restricted use only
for comparison between runs in this work and shall not be used in, for example, simulations based on a Brownian
diffusion process. The calculation procedure (inclusion of a concentration-weighting term in Equation (5) instead of a
spatial average) is the reason why the results show vertical and horizontal mixing coefficients in the same order of
magnitude (in contrast to (1)). Yet, fuel mixing in FB boilers is critical in the horizontal direction due to the low aspect
ratio of the dense bed (where most fuel particles are found). Removing the concentration-weighting term in
Equation (5) would significantly increase the values of vertical dispersion coefficient.

4
 th
12 International Conference on Fluidization. Harrison (Canada), 2007.

When it comes to the appearance of the concentration and velocity fields, the vortex structure of the tracer flow
observed in the previous work (in the 0.40 m wide bed) is also found in the present work (1.2 m wide bed) in all
runs (A to I), independent of fluidization regime. Thus, it seems to be a general feature that the flow of a tracer
particle of the size and density similar to a fuel particle is structured into several horizontally-aligned vortexes with
alternated rotational direction (cf. 12). Figure 3 exemplifies this for Run C, with the ascending channels provided by
the bubble paths located at the approximate coordinates x=0.20 m and x=1.0 m. Thus, the vortex structure is
induced by the bubble flow rising through the dense bottom bed, which provides ascending channels for the tracer
particle through the dense bottom bed and up to the bottom bed surface, where the tracer particle is projected into
the splash zone (experiencing a horizontal displacement) before sinking in the bottom bed through the emulsion
phase. This 2D vortex flow structure can be seen as a vertical cross section of what would be a toroidal flow
structure around each bubble path in a 3-dimensional case.
An increase in fluidization velocity clearly has a beneficial effect to the solids mixing as can be seen in Figure 4,
where the average dispersion coefficient increases in both the vertical and horizontal direction with an increase in
velocity, mainly due to the increased bubble flow.

Figure 3: Tracer particle concentration and velocity fields for Figure 4: Influence of fluidization velocity on dispersion
Run C. White line indicates the bed surface. (Runs A, B, C and D).

The tracer concentration and main velocity fields for Run A are shown in Figure 5, which is characterized by a
highly heterogeneous distribution of the tracer concentration field, compared to all other runs. Yet, this regime
yields four horizontally aligned vortexes induced by two main bubble paths at approximately the same horizontal
locations as in Run C in Figure 3 in spite of the fact that the fluidization velocity is four times as high in Run C.
Run A gives high-concentration regions at the downflowing interface of the two central vortexes, specially at the
height equal to the bottom bed height, i.e. there is a tendency for the tracer particle to remain most of the time at the
dense bed surface while moving horizontally (thus leading to a poor vertical dispersion compared to the other runs,
as indicated by the dispersion coefficients). In addition, comparison of runs A and C (and B and D, although not
shown here) confirms increasing tendencies of the tracer particle to occupy higher locations in the splash zone as
fluidization velocity is increased resulting in more uniform distributions of the concentration and velocity in both
directions.
In agreement with the trends shown in the previous work (0.40 m wide bed in (12)) and confirmed in an FB
boiler (17), the horizontal dispersion measured is enhanced as the solids inventory in the riser is increased (as seen
by comparing the results for runs F and C in Table 2). This is explained by the fact that an increased solids
inventory in the riser leads to a higher dense bottom bed, which extends a bit further the bubble rising path, ending
in faster and more vigorous bubbles reaching the dense bed surface. Even though a higher solids inventory in the
riser of an FB boiler helps homogenizing the fuel concentration over the cross section, thus improving oxygen-fuel
contact, it also represents a higher operating (fan power) cost.
The two different tracer particle sizes gave no significant difference in horizontal dispersion (see results for runs
C-E and F-H), while the vertical dispersion is enhanced when decreasing tracer particle size in the case with the
high-ΔP distributor (runs C-E).

5
 th
12 International Conference on Fluidization. Harrison (Canada), 2007.

Finally, the low-ΔP distributor gave a significantly different mixing pattern compared to the high-ΔP distributor.
This can be seen from Figure 6 corresponding to the low-ΔP distributor in Run G and which shows an asymmetry
in the tracer concentration field, with high values in the left side and low ones to the right, specially at x=1.05 m
(whereas Run F – not shown here - yields an even distribution in tracer concentration over the cross section). The
x=1.05 m position corresponds to the location of a high throughflow region, although bed was apparently fluidized
over the entire cross-section. In repeated experiments, the location of this throughflow channel varies in a random
fashion. The location of this intermittent by-pass strongly affects the flow pattern of the tracer particle: when it is
located close to the dense bed surface (y=0.17 to 0.23 m) it will flow away horizontally from the throughflow
location and when the tracer particle is located in the bottom of the bed (x=0-5 cm) it will move towards the
throughflow location. The large vortex originated by this effect can be seen in Figure 6. In addition, the video
recordings show that the tracer moves vertically through the dense bed in locations in the range x=0.10 to 0.80 m,
but (as also seen in Figure 6) there are no clear “only-rising” or “only-sinking” regions detected in this range, as is
the case with the high-ΔP distributor.

Figure 5: Tracer particle concentration and velocity fields for Figure 6: Tracer particle concentration and velocity fields for
Run A. White line indicates the bed surface. Run G. White line indicates the bed surface.

Figure 6 shows a rather uniform tracer concentration field when using a low-ΔP distributor. This would support
the use of low-ΔP distributors. However, while this leads to that fuel particles get more evenly distributed, the
throughflow effects enhance an uneven distribution of the gas flow (and thereby the oxygen). Thus, despite the high
horizontal dispersion values found for tests with the low-ΔP distributor (runs G and I), such conditions may still
yield insufficient oxygen-fuel contact.

CONCLUSIONS

A simulated fuel particle inserted in a cold 2D FB unit follows a flow pattern structured into horizontally-aligned
vortexes, which is a 2D equivalent of a toroidal structure in 3 dimensions. From a comparison with previous work in a
less wide bed than applied in this work it can be concluded that the vortex pattern is induced by the bubble flow and
not affected by the bed width.
Increasing the fluidization velocity enhances solids mixing in both vertical and horizontal directions. When the size
of the tracer particle is reduced, it reaches both higher velocities within the dense bed and higher locations in the splash
zone, increasing vertical dispersion while horizontal dispersion remains more or less the same. An increase in the
amount of bed material (constant velocity) enhances the dispersion coefficients by means of a higher dense bottom bed
which allows for the formation of larger and faster bubbles. Finally, lowering the air distributor pressure drop
significantly changes the mixing pattern, as a result of the formation of high-throughflow (gas) regions which
significantly reduces the solids mixing and the gas-fuel contact.

6
 th
12 International Conference on Fluidization. Harrison (Canada), 2007.

ACKNOWLEDGMENTS

The authors gratefully acknowledge financial support from the Swedish Energy Agency and the EU Research
Fund for Coal and Steel under contract RFC-CR-03001.

NOTATION

A Area [m2] r Mixing rate [m/s]

C Normalized concentration [m-2] umf Minimum fluidization velocity [m/s]
D Local dispersion coefficient [m2/s] ut Particle terminal velocity [m/s]
D Averaged dispersion coefficient [m2/s] k Subindex for direction
Da Damköhler number [-]
L* Characteristic length [m] Δl Tracer displacement [m]
n Subindex for pixel Δt Time interval [s]
N No. of pixels τ Characteristic time [s]

REFERENCES

1. Ito, O., Kawabe, R., Miyamoto, T., Orita, H., Mizumoto,M., Miyadera, H., Tomuro, J., Hokari, N., Iwase, T.
1999. “Direct measurement of particle motion in a large-scale FBC boiler model”. Proc.15th FBC Conf,
Savannah, 217.
2. Highley, J., Merrick, D. 1971. “The effects of the spacing between solid feed points on the performance of a
large fluidized bed reactor”. AIChE Symposium Series 67, 219.
3. Enwald, H., Almstedt, A.-E., Peirano, E. 1996. “Eulerian two-phase flow theory applied to fluidization”.
International Journal of Multiphase Flow 22, 21-66.
4. van Wachem, B. G. M., van der Schaaf, J., Schouten, J. C., Krishna, R., van den Bleek, C. M. 2001.
“Experimental validation of Lagrangian-Eulerian simulations of fluidized beds”. Powder Technology 116(2-
3), 155-165.
5. Peirano, E., Delloume, V., Johnsson, F., Leckner, B., Simonin, O. 2002. “Numerical simulation of the fluid
dynamics of a freely bubbling fluidized bed: influence of the air supply system”. Powder Technology 122(1),
69-82.
6. Ibsen, C.H., Helland, E., Hjertager, B.H., Solberg, T., Tadrist, L., Occelli, R. 2004. “Comparison of multifluid
and discrete particle modelling in numerical predictions of gas particle flow in circulating fluidised beds”.
Powder Technology 149 (1), 29-41.
7. Berres, S., Bürger, R., Tory, E.M. 2005. “On mathematical models and numerical simulation of fluidization of
polydisperse suspensions”. Applied Mathematical Modelling 29, 159-193.
8. Tanskanen, V. 2005. “CFD study of penetration and mixing of fuel in a CFB furnace”. Master Thesis, Dept. of
Energy and Environment, Lappeenranta Univ. of Tech. (Finland).
9. Niklasson, F., Thunman, H., Johnsson, F., Leckner, B. 2002. “Estimation of solids mixing in a FB combustor”.
Industrial and Engineering Chemistry Research 41(18), 4663-4673.
10. Xiang, Q., Huang, G., Ni, M., Cen, K., Tao, T. 1987. “Lateral dispersion of large coal particles in an industrial-
scale fluidized bed combustor”.Proc. 9th FBC Conf, Boston, 546.
11. Schlichthaerle, P., Werther, J. 2001. “Solids mixing in the bottom zone of a circulating fluidized bed”. Powder
Technology 120(1-2), 21-33.
12. Pallarès, D., Johnsson, F. 2006. “A novel technique for particle tracking in cold 2-dimensional fluidized beds”.
Chemical Engineering Science 61, 2710-2720.

7
 th
12 International Conference on Fluidization. Harrison (Canada), 2007.

13. Stein, M., Ding, Y. L., Seville, J. P. K., 2002. Experimental verification of the scaling relationships for bubbling
gas-fluidised beds using the PEPT technique. Chemical Engineering Science 57(17), 3649-3658.
14. Lim, K.S., Agarwal, P.K. 1994. “Circulatory motion of a large and lighter sphere in a BFB of smaller and
heavier particles”. Chemical Engineering Science 49, 421-424.
15. Shen, L., Johnsson, F., Leckner, B. 2004. “Digital image analysis of hydrodynamics two-dimensional bubbling
fluidized beds”. Chemical Engineering Science 59(13), 2607-2617.
16. Rios, G.M., Dang, K., Masson, H. 1986. “Free object motion in a gas fluidized bed”. Chemical Engineering
Community 47, 247-272.
17. Niklasson, F., Pallarès, D., Johnsson, F. 2006. “Biomass co-firing in a CFB boiler–Influence of bed properties
on in-furnace gas concentration profiles”. Proc. 19th FBC Conf., Vienna.

8
 Appendix A

Macroscopic modeling of heat transfer in circulating fluidized bed units

 Appendix A (D. Pallarès, 2008)

Macroscopic modeling of heat transfer in

circulating fluidized bed units

The heat balance over any control volume in the circulating loop of a CFB unit (i.e. riser and return leg)
contains enthalpy flow terms related to the in- and outflows of the solid fractions and gas species present, and
also terms which account for the loss or gain of heat flow derived from heat transfer mechanisms with the
surrounding. In a fluidized bed unit, the main heat transfer mechanisms are assumed to be radiation and
convection (where by convection, in this case, is understood the sum of both of gas convection and solids
conduction for the gas-solid suspension).
Firstly, some facts with important implications must be beard in mind: the low thermal inertia of the solid
particles (due to their size) and the relatively smooth temperature gradients which the gas-solids flow
experiences. These, together with the typical gas-particle heat transfer rate of a fluidized bed, support the general
and significant assumption of a common local temperature for the solids and gas phases. This assumption
significantly simplifies both the investigation of the heat transfer phenomena and the formulation of local heat
balances in fluidized beds.
Most experimental works in fluidized bed literature dealing with heat transfer bring the convective and
radiative heat mechanisms together to find correlations for effective heat transfer coefficients, which are
generally correlated as a function of some fluiddynamical parameter (e.g. the suspension density). However,
focusing on heat transfer to the heat transfer surfaces (i.e. waterwalls or internals), the role of radiation is
enhanced in dilute regions (such as the upper freeboard) and lower in dense regions at lower heights (such as the
bottom furnace region), due to varying solids absorption effect. The opposite applies to convection (in the sense
it is understood here, i.e. sum of gas convection and solids conduction), which increases in dense gas-solid
suspensions due to higher solids conduction.
Considering these differences, the two heat
transfer mechanisms should be modeled
separately in order to contribute to an improved
understanding of the global heat transfer
phenomenon.
Breitholtz et al. (2001) carried out an
experimental work which involved
measurements from three large-scale CFB
boilers in which radiative and convective heat
transfer to the furnace walls were differentiated.
This is represented by Fig. 1, which shows the
heat transfer coefficients obtained for
convective and total heat transfer as a function
of the suspension density. As can be seem, both
the convective and the total heat transfer
increase with an increase in suspension density,
while the radiative heat transfer (difference
between total and convective radiation) Figure 1: Convective and total heat transfer coefficients as a function of
suspension density. From Breitholtz et al. (2001).
decreases slightly.

From the estimated heat convection coefficients shown by the data points in Fig. 1, Breitholtz et al. gave the
following correlation for the convective heat transfer as a function of the suspension density:
hconv = 25 ⋅ C s
0.58
(1)
 Concerning radiative heat, the transfer coefficient between a solids-free gas suspension and a surface is
directly derived from established radiation theory as:

hˆrad =
1
1
1
(~2
⋅ σ ⋅ Tsusp + Tsurf
2
)( ~
⋅ Tsusp + Tsurf ) (2)
+ −1
ε susp ε surf

The averaged suspension temperature

~
is represented by Tsusp . Due to the absence
of solids absorption, this expression
represents the higher limit value for a gas-
solids suspension.
In a fluidized bed boiler, typical
emissivity values for the suspension
(accounting for typical solid fractions and
gas species) and the heat exchanging
surface (assuming oxidized steel) are
around 0.88 and 0.8, respectively.
Heat radiation coefficients in fluidized
beds were calculated as the difference
between the coefficients for the total and
the convective heat transfer and compared
to the theoretical maximum limit given by
Figure 2: Radiation efficiency as a function of suspension
density. From Breitholtz et al. (2001). Eq.(2) through the definition of a radiation
efficiency, ηrad. The obtained values for the
radiation efficiency are shown in Fig. 2 together with the curve for its correlation, which was given by
Breitholtz et al. (2001) as:

~
hrad ⎛ C s , susp ⎞
η rad = = 0.86 − 0.14 * atg ⎜ − 1.6 ⎟ (3)
hˆrad ⎜ 2.6 ⎟
⎝ ⎠

In the calculation of the parameters influencing radiation (such as the suspension density and temperature),
an appropriate type of averaging must be applied to the suspension. Due to the absorption by solids, properties in
the suspension closer to the surface studied will have a stronger weight in the process than those far away from
the surface. This absorption effect in fluidized bed radiative heat transfer was studied by Baskakov and
Leckner (1997), who described the absorption effect through the exponential curve proposed by the Bouguer’s
law for radiative properties of solids suspensions:

⎛ 1.5 ⋅ C v , s ⋅ L ⎞ (4)
α = 1 − exp⎜⎜ − ⎟
⎟
⎝ ds ⎠

Applying the above expression to a mesh allows consideration of the varying values of the solids volumetric
concentration and mean size. This has a significant effect in the case of heat transfer to the furnace walls, where
wall layers and the core region might present rather different values of the solids concentration and size. A
scheme for the application of the Bouguer’s law to the radiative heat transfer to the waterwalls in the furnace is
seen in Fig. 3. As seen in Fig. 3, the absorption phenomenon leads to that for denser suspensions (such as in the
bottom freeboard) the properties of the suspension close to the heat transfer surface dominate the radiation
mechanism. On the other hand, the lower absorption in more dilute suspensions makes properties of the
suspension relatively far from the surface play a certain role in the radiative heat transfer. Thus, weight
coefficients, wi , for each cell can be calculated according to the relative area below the red curve (i.e. the integral

2
of Eq. 4) in each cell. In a discretized mesh with varying suspension properties, the weight coefficient for a
certain cell i is calculated as:

n = i −1
wi = α i ⋅ ∏ (1 − α n ) (5)
n =1

with cell index 1 corresponding to the closest cell to the surface studied.

Figure 3: Weight of the local parameters in the Bouguer-averaging used in radiative heat modeling. The green lines
represent a mesh at a certain level in the furnace, with the furnace wall (heat transfer surface) to the left.

Thus, the cell weight coefficients are used to calculate the Bouguer-averaged suspension properties in
Eqs (2) and (3) and in the general equation for the radiative heat, i.e,

~
q rad = hrad ⋅ A ⋅ Tsusp − Tsurf ( ) (6)

With this, a certain radiative heat flow to the heat transfer surface can be calculated accounting for
absorption by the solids. However, what is required in the formulation of the heat balance for each individual
cell is the radiative heat flow exchanged between the cell and the heat transfer surface. This heat balance not
only depends on optical factor considerations (such as the Bouguer’s law) but also on the cell temperature. Cells
at higher temperatures obviously will contribute more to the radiative heat transfer than cells with a lower
temperature. To account for this, a further weighting coefficient accounting for the temperature effect has to be
defined, namely:

wi' =
(
wi ⋅ Ti 4 − Tsurf
4
) (7)

∑ w ⋅ (T )
N

i i
4
− Tsurf
4

i =1

This coefficient is then used in the individual cell heat balances to calculate the cell loss of radiative heat to
the heat exchanging surface as:

q rad i = wi' ⋅ q rad (8)

3
Notation

w Bouguer weight coefficient [-]

C Solids concentration [kg/m3]
d Particle diameter [m] w' Heat flow weight coefficient [-]

h Heat transfer coefficient [W/m2K] α Bouguer transmissivity [-]

L Length [m] ε Emissivity [-]

q Heat flow [W] η Efficiency [-]

T Temperature [K] σ Stefan–Boltzmann constant

[W/m2K4]
Subscripts

conv Convective surf Surface

rad Radiative susp Suspension

s Solids v Volumetric

References
Breitholtz, C., Leckner, B., Baskakov, A.P. 2001. ”Wall average heat transfer in CFB boilers”.
Powder Technology, 120(1), pp.41-48.
Baskakov, A.P., Leckner, B. 1997. ”Radiative heat transfer in circulating fluidized bed furnaces”.
Powder Technology, 90(3), pp.213-218.