Springer Theses

Recognizing Outstanding Ph.D. Research

Cheng Chen

Investigations on
Mesoscale Structure
in Gas–Solid Fluidization
and Heterogeneous
Drag Model
Springer Theses

Recognizing Outstanding Ph.D. Research

Aims and Scope

The series “Springer Theses” brings together a selection of the very best Ph.D.
theses from around the world and across the physical sciences. Nominated and
endorsed by two recognized specialists, each published volume has been selected
for its scientific excellence and the high impact of its contents for the pertinent field
of research. For greater accessibility to non-specialists, the published versions
include an extended introduction, as well as a foreword by the student’s supervisor
explaining the special relevance of the work for the field. As a whole, the series will
provide a valuable resource both for newcomers to the research fields described,
and for other scientists seeking detailed background information on special
questions. Finally, it provides an accredited documentation of the valuable
contributions made by today’s younger generation of scientists.

Theses are accepted into the series by invited nomination only

and must fulfill all of the following criteria

• They must be written in good English.

• The topic should fall within the confines of Chemistry, Physics, Earth Sciences,
Engineering and related interdisciplinary fields such as Materials, Nanoscience,
Chemical Engineering, Complex Systems and Biophysics.
• The work reported in the thesis must represent a significant scientific advance.
• If the thesis includes previously published material, permission to reproduce this
must be gained from the respective copyright holder.
• They must have been examined and passed during the 12 months prior to
nomination.
• Each thesis should include a foreword by the supervisor outlining the signifi-
cance of its content.
• The theses should have a clearly defined structure including an introduction
accessible to scientists not expert in that particular field.

More information about this series at http://www.springer.com/series/8790

Cheng Chen

Investigations on Mesoscale
Structure in Gas–Solid
Fluidization and
Heterogeneous Drag Model
Doctoral Thesis accepted by
Tsinghua University, Beijing, China

123
Author Supervisor
Dr. Cheng Chen Prof. Haiying Qi
Tsinghua University Tsinghua University
Beijing Beijing
China China

ISSN 2190-5053 ISSN 2190-5061 (electronic)

Springer Theses
ISBN 978-3-662-48371-8 ISBN 978-3-662-48373-2 (eBook)
DOI 10.1007/978-3-662-48373-2

Library of Congress Control Number: 2015948764

Springer Heidelberg New York Dordrecht London

© Springer-Verlag Berlin Heidelberg 2016
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made.

Printed on acid-free paper

Springer-Verlag GmbH Berlin Heidelberg is part of Springer Science+Business Media

(www.springer.com)
Parts of this thesis have been published in the following articles:

Cheng Chen, Fei Li, Haiying Qi. Modeling of the flue gas desulfurization in a CFB
riser using the Eulerian approach with heterogeneous drag coefficient. Chemical
Engineering Science, 2012, 69: 659–668. (SCI accession No.: 000298325800060,
Impact Factor: 2.386, EI accession No.: 20115114630129)
Cheng Chen, Joachim Werther, Stefan Heinrich, Haiying Qi, Ernst-Ulrich Hartge.
CPFD simulation of circulating fluidized bed risers. Powder Technology, 2013,
235: 238–247. (SCI accession No.: 000316039400028, Impact Factor: 2.024)
Cheng Chen, Fei Li, Haiying Qi. An improved EMMS-based drag model for gas–
solid flow in CFB risers with a novel sub-model for clusters. 8th International
Conference on Multiphase Flow, Jeju, Korea, 2013. (International Conference)
Cheng Chen, Haiying Qi. Modeling of cluster and its application in EMMS drag
model for heterogeneous gas–solid flows. 11th International Conference on
Fluidized Bed Technology, CFB-11, Beijing, China, 2014. (International
Conference)
Cheng Chen, Haiying Qi. Development and validation of cluster and EMMS drag
model. CIESC Journal, 2014, 65(6): 2003–2012. (In Chinese)
Cheng Chen, Feng Xu, Zhulin Yuan, Haiying Qi. Study on influence of
bubble-breaking structure upon the gas bubble properties in the scrubbing chamber.
Thermo Power Plant, 2010, 40(7): 42–47. (In Chinese)
Qunte Dai, Cheng Chen, Haiying Qi. Developing Heterogeneous Drag Model for
Fluidization Process and Its Generalization. Journal of Engineering Thermophysics,
2015, 4(36): 1–6. (In Chinese)
Qunte Dai, Cheng Chen, Haiying Qi. A Generalized Drag Law for Heterogeneous
Gas–solid Flows in Fluidized Beds. Powder Technology. (Accepted)
Supervisor’s Foreword

The modeling of drag between gas and solid phase has always been a challenging
issue in Eulerian simulation of heterogeneous fluidization processes for a long time.
It is generally known that existing drag models such as Wen-Yu and Ergun do
not apply to above processes because of their hypothesis of a homogeneous flow.
The key of the modeling is whether a drag model can account for effects of the
mesoscale structure, namely clusters that characterize the heterogeneous flow state.
It has been proved in recent years that the theory of Energy Minimization
Multi-scale (EMMS) is effective for developing heterogeneous drag models.
However, this theory still needs further improvement in hypothesis, description
of the mesoscale structure, definite conditions of conservative equations, and
generalization of drag models.
Dr. Cheng Chen devoted herself in last 5 years to drag modeling based on the
EMMS and made a series of significant progress. She focused on analysis of flow
heterogeneity and found a common feature—the “single peak” profile—in many
relationships, including cluster size dcl, cluster density ɛsc, and local gas–solid slip
velocity via local particle volume fraction ɛs, as well as overall slip velocity via
superficial velocity of circulating fluidized bed. She indicated that cluster size tends
to single particle diameter at the dense extreme and clarified the difference between
cluster size defined by the EMMS theory and measured in experiments. Compared
with the size, cluster density is the major factor that causes qualitative change of
drag. In addition, she established a relationship between the local heterogeneity and
operation parameters of fluidized bed reactors.
According to the above research, she developed a new drag model named
QC-EMMS and verified its prediction ability, accuracy, and universality by sim-
ulations and comparison with experiments.

vii
viii Supervisor’s Foreword

Overall, her work gives us valuable information and methods to deal with drag
issue and enlightens us that one should pay major attention to physical judgment
and understanding of mesoscale effects in fluidization process. I believe this work
will be helpful in simulating large-scale fluidization systems.

Beijing Prof. Haiying Qi

July 2015
Acknowledgments

I want to express my hearty gratitude to my dear teacher Prof. Qi Haiying for his
instructions and for his keen insight, persistence in science research, high respon-
sibility, and care toward students. His rigorous scholarship and impartial character
have greatly inspired me and will benefit my whole life.
I’m also thankful to Prof. Joachim Werther for his kind instructions and support
during the 6-month cooperative research at Technical University of Hamburg.
Thanks to Prof. You Changfu and Dr. Li Fei for their extensive instructions and
support for my project.
Thank you all my partners in the project including Wang Jinsheng, Luan Chao,
Zhu Junzong, Zhang Wei, Wang Haiming, and Xi Yingtao for their communication
on learning and their care and support in everyday life. I was so lucky to have a
chance of spending five unforgettable years with you and I wish you a brilliant
future.
Thank you my classmates and best friends during my doctoral career Mei
Zhequn, Wang Zhongnan, Li Xinzi, Huang Keman, and Gong Yuanyuan for adding
color to my doctorate studies and I wish our friendship will last forever.
Thank you my husband Zhong Xiaobo for loving me and growing up together
with me, for your constant understanding, tolerance, encouragement, and love. We
have had hardships accompanying each other and I hope we will continue to
treasure our time together.
Thank you my dear daughter Zhong Chenxi for coming to me during my doc-
toral dissertation. Your appearance has shown me how beautiful and valuable life is
and given me great courage and confidence. I wish you will grow up healthily and
happily.
Thank you my father Chen Chunshan and my mother Yin Zifeng for their
selfless contribution to the family and unconditional support and encouragement to
me. I wish you good health and happiness.
This project is funded by the National Natural Science Foundation of China
(NO. 91334111), hereby I express my thanks.

ix
Contents

1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Background and Significance. . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Heterogeneity of Gas–Solid Fluidization System . . . . . . . . . . . . 3
1.2.1 Overall Heterogeneity. . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 Local Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Heterogeneous Drag Models . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Empirical Models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.2 DNS-Based Statistical Model. . . . . . . . . . . . . . . . . . . . . 12
1.3.3 Statistical Model Based on Fine-Grid Simulation . . . . . . . 13
1.3.4 EMMS Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4 Numerical Simulation Methods for Dense Gas–Solid Flow . . . . . 19
1.4.1 Two-Fluid Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.4.2 Multiphase Particle-in-Cell (MP-PIC) Method . . . . . . . . . 22
1.5 Research Objectives, Contents, and Methodology. . . . . . . . . . . . 24
1.5.1 Objectives. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.5.2 Contents and Methodology . . . . . . . . . . . . . . . . . . . . . . 24
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2 Analysis on the EMMS Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.1 General. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 EMMS Theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.1 Basic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.2.2 Solving Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.3 Result Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3 Cluster Size and Its Effects on Drag . . . . . . . . . . . . . . . . . . . . . 41
2.3.1 Cluster Size Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.3.2 Effects on Drag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

xi
xii Contents

2.4 Cluster Density and Its Effects on Drag . . . . . . . . . . . . . . . . . . 50

2.4.1 Empirical Formula for Cluster Density . . . . . . . . . . . . . . 50
2.4.2 Effects on Drag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3 Cluster Model and Heterogeneous Drag Model . . . . . . . . . . . . . . . 55
3.1 General. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2 Local Heterogeneity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2.1 Slip Velocity Versus Cluster Parameters . . . . . . . . . . . . . 56
3.2.2 Slip Velocity Versus Flow Parameters . . . . . . . . . . . . . . 57
3.2.3 O-S Heterogeneous Drag Correction Factor . . . . . . . . . . . 59
3.3 Mesoscale Structure (Cluster) Model . . . . . . . . . . . . . . . . . . . . 61
3.3.1 Cluster Density Model . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.3.2 Cluster Size Correction . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.4 Heterogeneous Drag Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.4.1 Grid-Scale Validation of Drag Model . . . . . . . . . . . . . . . 66
3.4.2 Functional Form of Drag Model . . . . . . . . . . . . . . . . . . 67
3.5 System-Scale Validation of Drag Model . . . . . . . . . . . . . . . . . . 70
3.5.1 Geometric Model and Computational Conditions . . . . . . . 70
3.5.2 Grid-Size Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.5.3 Results Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.5.4 Analysis on Local Heterogeneity . . . . . . . . . . . . . . . . . . 77
3.5.5 Prediction of “Choking” State . . . . . . . . . . . . . . . . . . . . 81
3.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4 Condition Universality of Heterogeneous Drag Model . . . . . . . . . . 85
4.1 General. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.2 Evolution and Representation of Fluidization State . . . . . . . . . . . 86
4.2.1 “Ug-Gs” Fluidized Phase Diagram . . . . . . . . . . . . . . . . . 87
4.2.2 “Re*-Ar” Fluidization Phase Diagram . . . . . . . . . . . . . . . 90
4.3 Overall Heterogeneity Analysis . . . . . . . . . . . . . . . . . . . . . . . . 93
4.3.1 Overall Slip Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.3.2 Relationship of Overall Heterogeneity and Condition . . . . 94
4.4 Condition Universality of Model . . . . . . . . . . . . . . . . . . . . . . . 95
4.4.1 Relationship of Cluster Density Versus Fluidization
Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.4.2 Mathematic Description of Model Universality . . . . . . . . 96
4.5 Universal Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.5.1 Grid-Scale Validation . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.5.2 System-ScaleValidation. . . . . . . . . . . . . . . . . . . . . . . . . 101
4.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Contents xiii

5 Conclusions and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 115

5.1 Research Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 115
5.1.1 EMMS Theory and the Effects of Mesoscale Structure
on Drag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.1.2 Mesoscale Structure (Cluster) Model . . . . . . . . . . . . . . . 115
5.1.3 Heterogeneous Drag Model . . . . . . . . . . . . . . . . . . . . . . 116
5.1.4 Conditional Universality of Model . . . . . . . . . . . . . . . . . 116
5.2 Research Features and Innovations . . . . . . . . . . . . . . . . . . . . . . 117
5.3 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Nomenclature

ac , a f Internal particle acceleration in dense, dilute phase (m/s2)

a, b, c Coefficients in drag expressions (–)
Ar qg ðqp qg Þdp3 g
Archimedes number, Ar ¼ l2g (–)
C1, C2, C3 Empirical coefficients (–)
CD0 Single-particle drag coefficient (–)
CD Multi-particle drag coefficient (–)
dcl Cluster size (m)
dp Particle size (m)
 
qg ðqp qg Þdp3 g 1=3
d p* Dimensionless particle size, dp ¼ Ar 1=3 ¼ l2g (–)
D Diameter of fluidized bed riser (m)
dv Bubble diameter (m)
es Particle elastic collision coefficient (–)
f Volumetric fraction of particle dense phase (–)
fc Heterogeneity correction factor in O-S model (–)
Fc, Ff, Fi Single-particle drag in dense phase, dilute phase, and interaction
phase (N)
FD Grid-average drag force (N)
g Gravity acceleration (m/s2)
Gs Particle circulating mass flux (kg/m2s)
h Fluidized bed height (m)
H0 Initial particle accumulation height in fluidized bed (m)
Hd Heterogeneous drag correction factor (–)
Iinv Particle inventory in fluidized bed (kg)
n Empirical coefficient (–)
nc , n f , n i Numerical density of particles in dense phase, dilute phase and
interaction phase (1/m3)
np Number of single particles in computational parcel (–)
Np Number of computational parcels in grid (–)
Nst Suspended transport energy (J/kg)

xv
xvi Nomenclature

pg Gas pressure (Pa)

ps Particle pressure (Pa)
Δp/Δh Pressure drop per unit bed height (Pa/m)
q g dp

Re Local gas–solid slip Reynolds Number, Re ¼ ug  us (–)
lg
Re* Overall gas–solid
 slip Reynolds
 number,
 qg dp eg
Re ¼ l Ug  1eg  q (–)
Gs
g p

Ret qg d p
Terminal velocity Reynolds number, Ret ¼ lg Ut (–)
Δt Time step (s)  
U* Dimensionless gas velocity, U  ¼
qg dp e
Ug  1eg g  Gq s =Ar1=3 (–)
lg p

Uc Gas velocity in dense phase (m/s)

Uf Gas velocity in dilute phase (m/s)
UFD Critical gas velocity between fast and transport fluidization (m/s)
ug Grid-average gas velocity (m/s)
Ug Cross-sectional average gas velocity (empty-bed gas velocity) (m/s)
Umf Minimum fluidization gas velocity (m/s)
up Computational parcel velocity (m/s)
Up Cross-sectional average particle velocity (m/s)
Upc Particle velocity in dense phase (m/s)
Upf Particle velocity in dilute phase (m/s)
UPT Critical gas velocity between dilute and dense pneumatic transport
(m/s)
us Grid-average particle velocity (m/s)
uslip Local gas–solid slip velocity in grid, uslip = ug − us (m/s)
Uslip Overall gas–solid slip velocity, Uslip = Ug − Up (m/s)
Ut Single-particle terminal velocity (m/s)
Ut,cl Cluster terminal velocity (m/s)
UTF Critical gas velocity between turbulent and fast fluidization (m/s)
Vm Grid volume (m3)
Vp Particle volume (m3)
Vr The ratio of gas–solid slip velocity to terminal velocity
in O-S model (–)
xp Position coordinate of computational parcel (–)

Greek Character
β Drag function for heterogeneous flows (Ns/m4)
β0 Drag function for uniform flows (Ns/m4)
εg Gas volume fraction (Voidage) (–)
εmax Max voidage where clusters exist, εmax = 0.9997 (–)
εmf Minimum fluidization voidage (–)
εs Local time-averaged volume fraction of particle (local solids
concentration) (–)
Nomenclature xvii

εs,av Cross-sectional averaged volume fraction of particle (Cross-sectional

averaged solids concentration) (–)
εs,cr Critical solids concentration, εs,cr = εs + n · σs (–)
εs,in Solids concentration at fluidized bed return inlet (–)
εsc Solids concentration in clusters (Cluster density) (–)
εsf Solids concentration in dilute phase (–)
εsmf Solids concentration for minimum fluidization (–)
σs Standard deviation of solids concentration fluctuation (–)
φ Particle-wall specular reflection coefficient (–)
Φ Local heterogeneous index (–)
Ψ The overall system heterogeneous index (–)
γ Local heterogeneity defined by solids concentration fluctuation (–)
μg Gas kinematic viscosity coefficient (Pa s)
ρcl Cluster density, ρcl = εsc · ρp + (1 − εsc) · ρg (kg/m3)
ρg Gas density (kg/m3)
ρp Particle density (kg/m3)
τ g, τ s Gas/solid phase stress tensor (–)

Subscript
c, cl Particle dense phase (cluster)
f Particle dilute phase
g Gas
i Interaction phase
p, s Particle/Solid

Abbreviation
DNS Direct numerical simulation
EMMS Energy minimization multi-scale
MP-PIC Multiphase particle-in-cell
UDF User defined function
Chapter 1
Introduction

1.1 Background and Significance

Gas–solid fluidized bed reactor is widely used in industrial processes like oil cat-
alytic cracking, coal combustion, and flue gas desulfurization. On account of the
complexity of multiscale, multipattern, and multiphase coupling [1], in order to
guide the design, amplification, and operational optimization of reactor, mathe-
matical modeling, and numerical simulation have unshirkable responsibility.
The so-called “mesoscale structure” refers to extensive particle enrichment and
concentration variation formed in dense gas–solid two-phase flow under pneumatic
forces, that results in high flow heterogeneity. These particle enrichments are called
“clusters”. Unlike what we see on particle scale or fluidized bed scale, the
in-between mesoscale structure greatly intensifies the phase interactions and
directly influences the solids circulation and even the heat transfer, mass transfer,
and chemical reaction processes in fluidized bed [2]. Having a deeper insight on this
mesoscale structure, its characteristics and influence on flow and transfer is of great
scientific significance and practical application value.
Studies have indicated that accounting for the effects of mesoscale structure in
mathematical modeling is a key determinant for achieving accurate numerical
simulation [3]. So far, the Eulerian–Eulerian Approach (also called the two-fluid
model) still remains the primary method for simulating large-scale industrial flow
processes. Based on pseudo fluid hypothesis for solid particles, this method
describes solid mass and momentum transfer with transport equations having the
same form as gas phase, and incorporates particle kinetics to describe particles as a
discrete media and their constitutive relations [4]. However, this method has a fatal
deficiency in describing gas–solid momentum transfer by way of drag. The existing
drag models only apply to uniform flow, since they cannot consider the flow
heterogeneity resulted from the existence of mesoscale structure within the com-
putational grid-scale and its effects on drag.

© Springer-Verlag Berlin Heidelberg 2016 1

C. Chen, Investigations on Mesoscale Structure in Gas–Solid Fluidization
and Heterogeneous Drag Model, Springer Theses,
DOI 10.1007/978-3-662-48373-2_1
2 1 Introduction

People have tried to figure out the mesoscale effect through numerical means
such as grid fining (the minimum grid size is approximately ten times the particle
diameter), without modifying the mathematical model. However, practice has
proved that this is not a good idea and computational error still exists [5]. First, it is
still impossible to work out the mesoscale structure smaller than the fined grid.
Second, there is no theoretical explanation on the relationship between grid refining
and mathematical modeling. Last, grid refinement greatly increases the computa-
tional cost.
In reality, unless direct numerical simulation (DNS) is used without the support
of any stress submodel, the existing momentum transfer model (i.e., drag model)
has to be theoretically modified before accurate numerical simulation can be
achieved on a coarse grid [6]. To this end, it is necessary and important to inves-
tigate heterogeneous drag model in terms of mesoscale structure.
Existing drag models, such as the classical Wen-Yu model [7] and Ergun model
[8], are based on hypothesis of uniform flow or originated from experimental data
on uniform liquid–solid fluidization systems or fixed beds. Studies have indicated
that [9, 10], when these models and the Eulerian–Eulerian Approach are used to
simulate gas–solid two-phase flow through CFB riser, the particle circulating mass
flux is substantially higher than the experimental result and particles in bed are
uniformly distributed, making it impossible to reproduce the typical heterogeneous
distribution of particle concentration that is axially “dilute at top and dense at
bottom” and radially “dilute at center and dense at side”. The reason lies in that
uniform drag models overestimate the momentum transfer between gas and solid
phases and fail to account for the significant drag reduction under mesoscale effect.
The 1980s and 1990s saw the flourishing Eulerian approach, which started to be
popular in gas–solid two-phase flow simulation and gave rise to many research
teams in Europe and America. At that time, the European Economic Community
endeavored to boost testing, theoretical analysis, and numerical simulation on
pressurized CFB by initiating programs like JOULE II. Internationally, a
Benchmark topic was issued inviting teams to conduct back-to-back numerical
simulation on a specific fluidized bed system with their respective methods. Then,
the result from each team was evaluated by the topic issuer for error and prioritized
with experimental data. At that time, plagued by simulation errors, people tried a
number of countermeasures including curing bed pressure reduction, large-size
particle, optimized turbulence models, and grid fining mentioned above.
Unfortunately, neither of them secured a material solution, nor it was recognized
that the use of a uniform drag model is the essential cause of these errors [11]. For a
decade after that, people have devoted to study drag models and come to realize
what a crucial role the mesoscale structure, i.e., particle cluster, plays in drag
matters. One of the improvements made is the development of the theoretical
approach of using the Theory of Energy Minimization Multiscale (EMMS) to
establish a heterogeneous drag model [12]. So far, several models have come into
being, abbreviated as EMMS drag models. However, in the theoretical system,
many problems still exist. These include illogical hypothesis, fuzzy understanding
1.1 Background and Significance 3

on the characteristics of mesoscale structure, insufficient definite conditions for

solution, and the lack of flow pattern universality.
In view of this, this study focuses on how mesoscale structure interacts with
local flow parameters and overall operating conditions in dense gas–solid fluidi-
zation system, with mesoscale clusters as the object, and flow heterogeneity as the
principal line, and aim to develop a more accurate, more reliable,
condition-universal heterogeneous drag model. This has great significance on better
understanding the complex flow and transfer processes of multiscale gas–solid
fluidizations, and on establishing and refining numerical simulation methods for
large fluidization systems.

1.2 Heterogeneity of Gas–Solid Fluidization System

Conceptually, what is called “heterogeneity” is equivalent to “particle cluster”. In

other words, as long as the mesoscale structure of particle cluster exists, flow is
always heterogeneous and vice versa.
Gas–solid two-phase fluidization systems are typical nonlinear, heterogeneous
ones. Integrally, particle concentration is axially “dilute at top, dense at bottom” and
radially “dilute at center, dense at side”. Locally, it appears to be a nonsteady
enrichment of many particles, giving shape to particle clusters of different shapes
and sizes that keep breaking up and gathering together [13]. These processes, in
reality, are the result of the “gambling” and compromising between gas and solid
phases behind them. Therefore, studying the heterogeneous nature of flow is the key
and breakthrough point if ever we want to reveal the mechanism and rule of drag
variation.
To date, large amounts of fluidized bed test data have been published across the
world. Toward the drag problem, what is urgently needed to do will include:
(a) reexamine these data from the perspective of heterogeneity and identify how
flow heterogeneity (or the degree of heterogeneity) varies with operating parameters
and particle properties, (b) abstract key parameters that represent this pattern, and
(c) connect these parameters to mesoscale structure (i.e., cluster) characteristics and
drag force.

1.2.1 Overall Heterogeneity

A typical representation of overall heterogeneity is the diverse distribution forms of

particle cross-sectional average concentration along height (axial) and cross section
(radial) of fluidized bed.
For a fast fluidization system, its solids concentration is larger at the bed bottom
and gradually reduces with the axial height, presenting a profile similar to expo-
nential distribution. The distribution curve of axial solids concentration is also
4 1 Introduction

subject to equipment construction and operating conditions [14]. When gas–solid

flow is substantially exposed to constraints from the outlet member, the solids
concentration is considerable both at the bed bottom and the outlet, resulting in a
C-curve. When the solids circulating flux equals to its saturated entrainment, i.e.,
when flow is “choking”, the axial distribution of solids concentration is an S-curve
and the deflection point rises with increasing bed inventory, while the solids con-
centration remains unchanged at the bottom and upper parts.
For a dilute transport system where the gas flow velocity is quite high and the
average solids concentration in bed is low, breaking outweighs gathering, making it
hard to form particle cluster. Hence, the axial distribution of particle concentration
is quite uniform. Literature [15] gives a fairly complete description of various types
of axial solids distribution, and a systematic analysis on their influencing factors.
The radial heterogeneous distribution of solids concentration is primarily the
responsibility of the wall surface. In fast fluidization, this heterogeneity is the most
obvious which is attributable to the gas velocity gradient between the central region
and the near-wall region. Velocity difference leads to different entraining capacity
and then the particle backmixing under gravity. Particles tend to accumulate in the
near-wall region, i.e., the low-speed region, resulting in the so-called “ring-nuclear”
distribution structure of solids concentration that is low in the central region and
high in the near-wall region [16]. Of course, operating conditions, particle prop-
erties, and equipment construction can all contribute to the radial distribution profile
[17, 18].
The axial or radial distributions of solids concentration discussed above are both
time-averaged results. The radial distribution of time-averaged particle velocity is a
parabolic line similar to that of gas phase. In fast fluidization, particles in the central
region typically move upward and their time-averaged velocity is positive, while
those in the near-wall region primarily move downward and their time-averaged
velocity is negative. The point of zero velocity is what we call the “ring-nuclear”
borderline [19]. In dilute phase transport, the time-averaged particle velocity is
basically positive.
Instantaneous particle movements across the flow field appear to be highly
nonsteady. Both the gas–solid velocity and solids concentration fluctuate widely
with time and reach their maximums at the “ring-nuclear” borderline. Particle
turbulence, backmix, enrichment, and breakage also take place in the flow field,
during all of which the positive and negative particle velocities can be detected.
All these solids velocity and concentration distributions actually correspond to
different gas–solid two-phase flow patterns. As the instantaneous fluctuation level
of flow parameters differs, so do the accumulation level of particle clusters.
Therefore, different flow patterns differ largely by level of flow heterogeneity. For
example, the solids concentrations under bubbling fluidization and pneumatic
transport are both fairly uniform, though the former is dense flow while the latter is
dilute flow.
1.2 Heterogeneity of Gas–Solid Fluidization System 5

1.2.2 Local Heterogeneity

As discussed above, local heterogeneity is the intensive fluctuation of particle

velocity and concentration with time at local points of the flow field. This fluctu-
ation is resulted from the movement of particle clusters with varied sizes and shapes
and their continuous breakup and dissipation.
The coexistence of individual particle and cluster movement in flow field not
only leads to alternation of particles in dilute phase (i.e., mixture of a small amount
of particles with gas) and dense phase (i.e., clusters) [20, 21], but also changes the
phase interaction, as signaled by the great increase of time-averaged gas–solid
relative velocity (also called “slip velocity”) [22–25]. This indicates that, under
local heterogeneity, the average particle entraining capacity of gas flow is reduced
[26], what we call “drag reduction”. Studies have discovered that the gradual radial
increase of gas–solid slip velocity corresponds to velocity fluctuation and solids
concentration distribution.
The local heterogeneity of gas–solid two-phase flow has different degrees of
effects on the intraparticle circulation, heat and mass transfer, chemical reaction
rate, and even particle-wall abrasion [27]. Understanding the particle cluster
behavior forms the foundation for achieving better insight into the gas–solid
two-phase flow mechanism inside a fluidization system.

1.2.2.1 Cluster Definition

The earliest and widely accepted cluster definition method is the “Three Elements”
Method [28]. First, the instantaneous particle concentration is larger than the critical
concentration εs,cr, defined as the local time-averaged concentration εs plus n times
of the standard deviation of particle concentration fluctuation, where n is an
empirical coefficient ranging 1–3 [29]. Second, at a constant sampling interval, the
time when the instantaneous particle concentration is higher than the critical con-
centration must be three times the sampling interval. Third, the characteristic size of
the sampling range must be 1–2 orders of magnitude larger than the size of single
particles, but far smaller than that of the bed. Based on this definition, people often
directly assume that solid content in cluster (or cluster density) εsc equals to critical
concentration, i.e., εsc = εs,cr = εs + n · σs.
In recent years, people have started to doubt this cluster definition method. They
argue that fiber signals used to measure instantaneous particle concentration are not
normally distributed and are not steady. Therefore, the averages values and standard
errors are not statistically meaningful [30]. Some people suggest a new cluster
definition to be half of the average of the fiber measurement result [31]. To further
characterize the instantaneous, multiscale, and nonlinear dynamics of fluidization
processes, people also tried to investigate instantaneous particle concentration
fluctuations by Wavelet Multi-Resolution Analysis, and introduced a new
6 1 Introduction

identification criterion for clusters, the Wavelet-Threshold Criterion, which is a

value that varies with particle concentration [32].

1.2.2.2 Experimental Study on Cluster Characteristics

Research on solid clusters can date back to the 1970s and 1980s. Studies were
carried out mainly in the form of experimental observation and measurement, using
intrusive or nonintrusive approaches. The former include optical fiber [33, 34] and
capacitance probe [35] methods that yield local instantaneous particle concentration
and velocity, but have some interference with flow. The latter are based on visu-
alization principle or laser measurement technologies like high-speed photography
[36], phase Doppler particle analyzer (PDPA) [37], and radioactive particle tracking
technique (RPT) [38] that yield flow information across the field. These methods
mostly apply to dilute flow. Otherwise optical signals are easily blocked and
attenuated by particles, leading to higher signal-noise ratio and greater errors.
As clusters keep generating and breaking up, they present strong spatial heter-
ogeneity. As shown in Fig. 1.1, clusters come in different shapes like flocs, belts,
clusters, or flakes and keep on changing with their spatial position [39]. Generally,
time averages are used to characterize clusters, focusing on the clusters’ shape, size,
distribution, velocity, occurrence frequency, residence time, and other movement
behaviors [40, 41].
Recently, people have started to notice the characteristic parameters of clusters,
particularly the relations of density and size with local flow parameters. According
to the expression of cluster density ρcl, ρcl = εsc · ρp + (1 − εsc) · ρg, for given gas
and solid particles (with the gas and particle densities ρg and ρp staying unchanged),
the solid content of cluster εsc represents the cluster density. As it is hardly possible

Fig. 1.1 Particle clusters in gas–solid fluidization system

1.2 Heterogeneity of Gas–Solid Fluidization System 7

to express different cluster shapes in a uniform manner, the cluster size measured
during experiment is practically the axial length of the cluster.
Studies have revealed close correlation between the characteristic parameters of
cluster and local time-averaged solids concentration [42–44]. Several empirical
correlations have been fitted out based on experimental results [45–47]. Despite
divergences, all the correlations indicate that both cluster density and size increase
with the local solids concentration. Given this result, if solids concentration reaches
its maximum, i.e., the accumulation concentration, the cluster size will become
extremely large and may even oversize the characteristic size of the system, which
is both illogical and inconsistent with physical observations. This comes to a
question as to how cluster size is identified and defined. The measurement methods
used and the cluster divergences at different spatial measuring points have to be
investigated.
Besides that, the clusters’ characteristic parameters relate not only to local flow
parameters, but also to operating conditions of fluidized bed and particle properties.
1. Effects of operating conditions
Operating conditions refer to empty-bed gas velocity Ug, particle mass flux Gs, and
bed inventory Iinv. They decide different flow patterns, flow heterogeneities, and
consequently the mesoscale structure.
At present, no unified conclusion has been drawn as to how operating conditions
make a difference to clusters. Some people believe that the cluster characteristic
parameters reduce with the increase of Ug [48], or reduce first but tend toward
constant later [49], while Gs does not make much contribution [50]. Some studies
discover that, when the cross-sectional average solids concentration is constant, the
relation between the cluster characteristic parameters and the local solids concen-
tration is basically independent from operating conditions or particle properties
[51]. Some people even fitted out the relation between the cluster characteristic
parameters in the near-wall region and the cross-sectional average solids concen-
tration through mixing up experimental results under different operating conditions
and particle properties [52]. In addition, due to practical limitations, the local solids
concentrations measurable by existing experiments are smaller than 0.2, and no data
on dense flow is available.
Hence, it is necessary to find out the relation between cluster properties and
operating conditions before a condition-universal cluster model can be developed
beyond the restrictions from differences in experimental conditions.
2. Effects of particle properties
Particle properties typically include particle size, size dispersion, particle surface
viscosity, and elastic recovery coefficient. Generally speaking, few studies have
been made on the effects of particle properties on clusters.
Studies have revealed that particle diameter does not make much difference to
cluster density [53]. It is easier for smaller, lower density particles to get together,
while the sphericity does not play such a role [54]. Besides, the probability and
8 1 Introduction

duration of cluster in multiparticles, dilute fluidized bed have also been frequently
reported [55].
According to previous studies, the agglomeration of coherent particles is the
result of the joint action among a variety of interphase and interparticle forces,
including fluid drag, particle gravity, viscous forces (particularly Van der Waals’s
force), and collision forces. Whether particles are steadily aggregated or broken up
after collision is decided by the relative order of magnitude of these forces. By
examining changes of collision, elastic, and surface viscous forces during particle
collision and deformation, people have proposed criteria identifying whether par-
ticles are aggregated and models predicting cluster size [56, 57].
Particle size dispersion makes great difference to particle agglomeration. For
Geldart group C fine particles [58], proper particle size dispersion helps to weaken
coherent agglomeration and improve the fluidization performance [59]. Thus, it is
necessary to account the effects of particle size and distribution on mesoscale
structures.
In a word, understanding on this mesoscale structure or solid cluster is limited. It
is highly necessary to look deeper into and give a reasonable, quantitative
description on how cluster characteristic parameters interact with local flow
parameters and system operating conditions. Only on this basis, a more reasonable,
more accurate drag model on heterogeneous flow can be established.

1.2.2.3 Mechanism of Cluster Formation

The mechanism of cluster formation is very complex and subject to a range of

factors [60–62], including the effects of particle wake, particle–particle, and
particle-wall interactions and so on. No consensus has therefore been reached in this
respect [63, 64]. From the view of fluidization, two major factors can be identified
for the cluster formation. One is the vertical momentum transfer of gas–solid
two-phase flow; the other is the local average solids concentration.
The first factor comprises the entrainment and suspended transport of particles
by gas. There exist two opposite movement mechanisms. (a) Within a confined
space, gas flow always tends to select the least-resistance path to move upward.
This produces continuous, stable suspended transport, and instantaneous local
elutriation entrainment for particles. (b) Particles always tend to fall down while
maintaining the minimum gravitational potential. Leverage between these two
mechanisms inevitably leads to particle clusters, resulting in heterogeneous and
dynamic stability across the flow. Hence, the momentum transfer between gas and
solid is the prerequisite for cluster formation.
The second factor implies that mesoscale structures occur only within a certain
range of average solids concentration (it is called the kinetic region). This is
because the gas–solid interaction and the interparticle collision come to a balance,
and the gas–solid slip velocity is quite large in this range. On the contrary,
mesoscale structure disappears and flow tends toward uniform in dilute or dense
flow or when gas or particles prevail (e.g., under pneumatic transport or bubbling
1.2 Heterogeneity of Gas–Solid Fluidization System 9

fluidization). For example, in a dense region, interparticle friction enhances and

thus cluster development is restrained. Therefore, average solids concentration is
the sufficient condition for cluster formation.
Additionally, the stagnation of confined spaces to gas flow also forces particles
to accumulate in the near-wall region and fall down along the wall surface, i.e.,
clusters form at the sidewall, presenting a dense-side, dilute-center “ring-nucleus”
structure over the entire cross section.
In a word, the above-mentioned substantial prerequisites have to be satisfied for
the stability of mesoscale structure or cluster. At that time, the suspended transport
energy of the entire system is at its lowest level. That is why this study tries to
investigate cluster characteristics and model representation with the help of the
EMMS theory.

1.3 Heterogeneous Drag Models

As explained above, the existence of mesoscale structure is responsible for high

heterogeneity in gas–solid two-phase flow and great drag reduction [65, 66]. As
existing, drag models mostly apply to uniform flow without accounting for
mesoscale effect, it is hardly possible for them to give an accurate description of
drag reduction in heterogeneous flow [67, 68]. To solve this problem, people have
tried enlarging particle diameter, fining the grid size, and even lowering the pressure
of fixed bed, but none of them have been able to theoretically change the drag
description. As a result, simulation of fluidization is always subject to considerable
errors, overestimating the particle circulating mass flux [69]. That is why people
started to study the drag models suitable for heterogeneous systems.
The uncontrollable nature of particle movement in gas–solid two-phase flow
forms a big challenge to direct drag measurement. Research methods mainly
include four types. (a) Empirical methods based on indirect experiment and data
processing as a means to correct homogeneous drag models. (b) Statistical methods
based on DNS. (c) Correlation-based method that calculate the average drag within
a coarse grid using numerical approaches like fine-grid simulation. (d) Analysis
methods based on the EMMS theory.

1.3.1 Empirical Models

Commonly used empirical models include correction coefficient model,

cluster-based model, and empirical model based on experimental data.
1. Correction coefficient model
In correction coefficient model, the traditional uniform drag model is multiplied by
a coefficient smaller than 1 to account for the drag reduction caused by clusters.
10 1 Introduction

However, as the degree of clusters and drag reduction differs from condition to
condition, so does the correction coefficient used. It has to be determined through
extensive tentative calculation [70, 71]. Furthermore, even for the same condition,
the degree of drag reduction also varies depending on the local position and solids
concentration. It is therefore hardly possible to describe the correct drag simply
with one empirical correction coefficient. Thus, correction coefficient model is not
so condition universal and also lacks the support of perfect, reliable theoretical
basis.
2. Cluster-based model
Considering that uniform fluidization is the flow of discrete particles while heter-
ogeneous fluidization is characterized by clusters, some researchers tried to treat
clusters as a whole and calculate drag using cluster size instead of single particle
diameter [72]. Initially, the calculation was conducted with a single cluster size
assuming that all the clusters in bed were of the same size. Later, a cluster size
formula was introduced to address the variation of cluster size with solids
concentration.
Although cluster-based drag model is already used to calculate and predict gas–
solid fluidization processes [73–75], the prerequisite for its accuracy and reliability
is that the cluster size formula used is accurate. Clusters in fluidized bed keep
generating and breaking up and are subject to a range of factors including gas–solid
property parameters, bed geometric dimensions, and operating conditions. At
present, there is no uniform consensus definition concerning the cluster size, cluster
density, and even the cluster definition [76]. Thus, whether a cluster-based model
can provide accurate calculation and prediction of gas–solid fluidization processes
is still a disputed topic.
3. O-S drag model
This is a drag relation concluded by O’Brien and Syamlal after regressing exper-
imental data on fluidized flows [77]. It is so far the only model available that reflects
the essence of heterogeneous drag. It has great significance for this study. For
facilitate description, we call it hereafter the O-S model.
This model contains two main parts, a uniform drag model and a heterogeneity
correction coefficient fc. Equations (1.1)–(1.5) is the model expression, where Vr
[Eq. (1.3)] is physically the ratio of gas–solid two-phase slip velocity Uslip to
particle terminal velocity Ut.

1 3


b ¼ CD0  2
 qg eg ð1  eg Þ
ug  us
ð1:1Þ
Vr 4dp
rffiffiffiffiffiffi!2
Vr
CD0 ¼ 0:63 þ 4:8 ð1:2Þ
Re
1.3 Heterogeneous Drag Models 11

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
C1  0:06 Rep þ ð0:06ReÞ2 þ 0:12Reð2C2  C1 Þ þ C12
Vr ¼ ð1:3Þ
2
(
0:8e1:28 if eg \0:85
C1 ðeg Þ ¼ e4:14
g ; C2 ðeg Þ ¼ g
ð1:4Þ
e2:65
g if eg  0:85

dp qg
Re ¼ ðug  us Þ ð1:5Þ
lg

Heterogeneity correction coefficient fc:

Vr;cl ¼ Vr  fc ð1:6Þ
2 2
fc ¼ 1 þ C3  Re es e0:005ðRe5Þ 90ð0:08es Þ ð1:7Þ
 
250 98
C3 ¼ ; Gs ¼ kg/m2 s ð1:8Þ
1500 147

Equation (1.6) indicates that the mesoscale effect of clusters causes the gas–solid
slip velocity in heterogeneous flow much larger than that in uniform flow.
The second term of heterogeneity correction coefficient fc [Eq. (1.7)] is the
heterogeneity correction term, where the empirical coefficient C3 represents the
degree of mesoscale effect and varies depending on the operating conditions of
fluidization (here it is Gs). This suggests that the degree of flow heterogeneity (i.e.,
the mesoscale effect) is different across different conditions. Equation (1.8) only
gives the C3 values under two Gs values, suggesting that the universality of the O-S
drag model is limited [78].
Figure 1.2 shows the O-S model curve, exhibiting some of the important
characteristics of heterogeneous gas–solid two-phase flow.
1. Compared with traditional uniform drag function curves, heterogeneous drag
reduction only occurs within a certain range of local solids concentration,
approximately εs = 0.01–0.4.
2. The rate of drag reduction comes to its maximum nearby εs = 0.1, which is a
reduction of almost three orders of magnitude, suggesting that the mesoscale
effect is the strongest at this point.
3. The drag comes back to uniform at the dilute (εs < 0.01) and dense (εs > 0.4)
ends, suggesting that flow also tends toward uniform.
Besides that, the rate of drag reduction increases with particle mass flux Gs. This
is because increased particle mass flux results in more intense solids cluster and
consequently higher heterogeneity. Whether a monotonous increase always exists
between them is yet to be studied.
12 1 Introduction

Fig. 1.2 Uniform part and 10

6

heterogeneous part of O-S

drag model (ρp = 1714 kg/m3, Uniform state

Drag function β (Ns/m4)

5
10
dp = 76 μm, ρg = 1.205 kg/m3,
μg = 1.815 × 10−5 Pa s, 10
4

uslip = 1.0 m/s, εsmf = 0.6)

3 (3.7, 98)
10
O-S Model
2
10
2
1
Ug(m/s),Gs(kg/m s)
10
4.3, 147
0
10
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Local solids concentration εs (-)

Based on the characteristics above, the O-S drag model can be used as a
benchmark for evaluating other models [19, 79].

1.3.2 DNS-Based Statistical Model

Theoretically, DNS does not rely on any model. It obtains drag information using
an ultrafine grid smaller than particle size (the integral of the projection of viscous
shear stress and pressure on particle surface in the direction of movement). The drag
is then correlated to the local solids concentration and thus a drag correlation is
established [80, 81]. As long as the calculation is accurate enough, DNS can
disclose the real physical process and mechanism as a substitute for the hardly
achievable experimental measurement. It is therefore a very promising, highly
expected approach. It should be true for any microscopic research to eventually
provide a general description of macroscopic physical processes. However, no
usable result is available yet.
From Fig. 1.3, the drag curve from DNS [82] already bears the typical char-
acteristics of heterogeneous drag, i.e., drag reduction exists within a given range of
solids concentration and reaches minimum near εs = 0.1. However, compared with
the O-S model curve, considerable divergences still exist. There are three principal
reasons. (a) The number of particles used as statistical samples is too small to
“traverse” all possible flow states. (b) The two-dimensional (2D) DNS result is yet
to agree with realities. (c) The wall effect on the flow in confined spaces is not
eliminated. Obviously, there is still a long way before achieving a macroscopic
heterogeneous drag correlation by DNS-based statistical method.
1.3 Heterogeneous Drag Models 13

Fig. 1.3 Chart comparing 5

10
uniform model, DNS-based
statistical model, and O-S Uniform state

Drag function β (Ns/m4)

drag model (ρp = 1714 kg/m3, 4
10
dp = 76 μm, ρg = 1.205 kg/m3,
μg = 1.81 × 10−5 Pa s,
uslip = 1.0 m/s, εsmf = 0.5, 3
10 DNS
Gs = 98 kg/m2s)

2
10
O-S Model

1
10
0.0 0.1 0.2 0.3 0.4 0.5
Solids concentration εs (-)

1.3.3 Statistical Model Based on Fine-Grid Simulation

The fundamental statistical model based on fine-grid simulation is to reveal the

heterogeneous flow structure within the micro-element space of grid by means of
fine-grid simulation, and then use them to close larger scale conservation equations
[83].
Some researchers believe that the reason why existing uniform drag models
cannot capture mesoscale structure is that the grid used is too coarse; and a fine grid
smaller than the cluster size will be able to do it [84–87]. Hence, people tried to
develop drag models and particle stress models that contain mesoscale effect and
are suitable for course grid out of the fine grid two-fluid model [88–90]. As cluster
size spans widely and the minimum is approximately ten times the single particle
diameter, the fine-grid simulation of entire fluidized bed needs a large amount of
calculation. Thus, it is a general practice to perform fine-grid simulation on local
regions before deducting a drag and particle stress correlation applicable for coarse
grid [91–93].
Figure 1.4 shows the simulation region, filter region, and grid involved in sta-
tistical model based on fine-grid simulation. These steps are followed. (a) For a
region with spatial size far smaller than the size of the fluidized system (domain
region), perform fine-grid simulation with a traditional uniform drag model. When
instantaneous calculation is stabilized, record the flow attributes in each fine grids at
different times. (b) Within the above-mentioned domain region, simulate a filter
region or subgrid region which size is far smaller than the domain region to neglect
the effect of wall boundary conditions. (c) Average, the parameters of a number of
times from step (b), and establish how these parameters vary with local solids
concentration, i.e., the drag correlation applicable for course grid. Here, the coarse
grid size is larger than fine grid while smaller than the filter region.
The key of this method is whether fine-grid simulation is able to work out the
movement characteristics of gas–solid fluidization. If it is, with the rapid
14 1 Introduction

Fig. 1.4 Grid schematics Domain region

Fine grid

Filter region

Sub-grid region

development of computer and parallel machine technologies, it will be possible to

simulate large industrial-scale fluidized bed systems on a traditional Eulerian
two-fluid model.
However, there are still objections to this method [94]. First, grid fining is
somewhat stochastic and hardly applies to different flow state. Hence, whether it is
able to capture true heterogeneous structure is still a question. Second, fine grid is
far from the fineness required by DNS. It cannot abstract single particle’s drag
information or work out the mesoscale structure inside the grid. Hence, using a
uniform drag model for fine-grid simulation will obviously lead to errors. Third,
boundary condition and 2D simplification of the simulation region also result in
considerable errors. As we know, description of physical processes depends on
mathematic models, whereas numerical means (like grid fining here) is only a tool
that solves mathematical models and equations but cannot be used to describe
physical processes. Therefore, trying to work out mesoscale effect by fine-grid
method lacks necessary theoretical support.
Figure 1.5 discloses that, the drag function curve deflects slightly with the filter
region size. However, they always feature a uniform drag model, i.e., it increases
monotonously with local solids concentration (as shown in the uniform curve in
Fig. 1.2). This precisely confirms to the analysis above that it is impractical to
describe physical processes simply by numerical means without changing the
mathematical model itself.

1.3.4 EMMS Model

There are two implications behind the Theory of EMMS [95]:

1. Flow decomposition based on mesoscale analysis. A heterogeneous flow system
is decomposed into three hypothetic uniform subsystems, the particle dense
phase, the particle dilute phase, and the dense-dilute interaction phase. As shown
in Fig. 1.6, the dense phase are composed of many particles and are equivalent to
a particle cluster, while the dilute phase comprise a small amount of single
1.3 Heterogeneous Drag Models 15

6
10
Filter region size: 0.257

Drag function β (Ns/m4)

5
10

4
10
16.448
3
10

2
10

1
10
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)

Fig. 1.5 Statistical model result based on fine-grid simulation (ρp = 1714 kg/m3, dp = 76 μm,
ρg = 1.205 kg/m3, μg = 1.81 × 10−5 Pa s)

particles in fluid. The interaction phase is a virtual subsystem introduced to

analyze the interaction between dense and dilute phases. Then, each subsystem is
described with a number of state variables. The relations among all these vari-
ables are described by separate mass and momentum conservation equations. The
significance underlying this practice is that a uniform flow drag model can be
used to describe the gas flow phase interaction in the three subsystems.
2. System stability condition. It is assumed that the condition for dynamic stability
of gas–solid two-phase flow is that the suspended transport energy Nst of the
flow system per unit particle mass reaches its minimum. It is the result of the
interaction and coordination between “gas control” and “particle control”. The
former means that gas flow always tends to select the minimum-resistance path,
and the latter means that particles always tend to move downwards maintaining
the minimum gravity potential. There is absolutely a mesoscale structure out of

+ +

Dense phase Dilute phase Interaction phase

Fig. 1.6 Decomposition of local heterogeneous gas–solid flow field

16 1 Introduction

Region G Point A

Temporal
compromise Grid D Region G
Regional
Compromise
Global
Spatial
compromise Compromise

Grid D
Point B
Gas

Particle-Fluid EMMS model Radial EMMS model

Systems
Local (Micro-scale) structure Meso-scale structure Global (Macro-scale) structure
Point A &B Region D Region G

Fig. 1.7 Changes in suspended transport energy on different scales. Reprinted from Ref. [96],
Copyright 2004, with permission from Elsevier

this mechanism, and this structure will never disappear with the reduction of the
space studied (e.g., the numerical computational grid). This theory is
undoubtedly an important means to investigate how mesoscale structure affects
flow and transport processes.
The physical existence of minimum suspended transport energy, the extreme
condition for stability, has been validated by Pseudo Particle Model (PPM) [96, 97]
and Lattice Boltzmann Simulations (LBM) [98].
As shown in Fig. 1.7, at local microscale positions, the suspended transport
energy of gas to particles fluctuates widely and no extremum exists. For mesoscale
grid space, the suspended transport energy continues to fluctuate modestly with
time and gradually presents a stable minimum. For a macroscale cross section, the
average suspended transport energy does not fluctuate much and obviously tends
toward a stable minimum. This not only verifies the physical existence of a stability
condition, but also suggests that this stability condition exists in a certain grid space
and over the cross section of the entire fluidized bed system.
As the EMMS theory is a theoretical system independent from the existing
Eulerian–Eulerian multiphase flow numerical simulation, our research group pro-
posed the idea of coupling these two methods in the 1990s [19]. Using the
Eulerian–Eulerian strategy to provide gas–solid field information and then the
EMMS theory to derive the drag value at which the minimum suspended transport
energy stabilizes, and finally returning to the momentum equation of the Eulerian
strategy for the next step of iteration. This is actually using the EMMS theory to
solve the drag model as a supplementary energy equation of mechanical energy and
flow stability condition, and then closing the Eulerian–Eulerian strategy. On this
basis, many kinds of multiscale heterogeneous drag models, EMMS drag models
for short, have already been developed [99–104].
1.3 Heterogeneous Drag Models 17

The history of EMMS drag models can be summarized as a process involving

continuous improvement and refinement of the particle momentum equation.
However, no physical understanding of the heterogeneous flow drag has been
provided.
The earliest EMMS model assumed a balance between particle drag and gravity
with no inclusion of time evolution or involvement of spatial distribution. Only
with the help of axial and radial kinematic models was it possible to predict the
heterogeneous solids distribution characteristics. Studies have indicated that the
EMMS model is able to successfully predict, and give a logical physical expla-
nation of, the “choking” phenomenon in gas–solid fluidization processes [105].
Accurate description of this “choking” phenomenon is crucial to the optimal design
and stable operation of industrial fluidization systems. However, this EMMS model
only applies to steady calculation to predict the overall flow state at the reactor level
of the fluidized bed, but does not involve local flow information [106].
The model that followed (here named as “GY-EMMS model”) corrected the
particle momentum equation by incorporating particle acceleration [107, 108]. The
particle acceleration is assumed to be the same in dense, dilute, and interaction
phases. In a real gas–solid fluidized bed reactor, particles pass through an accel-
eration section, a transitional section, a stable section, and an outlet section along
the height of the bed. Among all the sections, particles have acceleration with the
exception only of the stable section. Hence, the initial hypothesis of drag-gravity
balance no longer exists.
Based on GY-EMMS model, researchers further assumed cluster density to be a
constant value (εsc = 0.31) and established the mathematic correlation between drag
function and the average solids concentration in local cells. They also coupled this
correlation with the Eulerian–Eulerian Approach and worked out the heterogeneous
flow characteristics in gas–solid fluidized bed. However, some quantitative errors
still existed to compare the experimental results [109, 110].
Considering that particles in dense, dilute, and interaction phases do not share
the same particle acceleration, researchers further corrected their respective particle
momentum equations by introducing accelerations of three subsystems and estab-
lished a new EMMS model termed the EMMS/matrix model [111, 112]. Instead of
assuming a constant εsc, this model used a two-step method for solution. The first
step analyzes macroscopic operating parameters and finds out the relation of cluster
density and size with solids concentration. The second step uses this relation inside
the grid and yields the corresponding drag function by accounting for the contri-
bution made by solids concentration and slip velocity in local cells. Simulation of
gas–solid fluidized bed under different operating conditions with EMMS/matrix
model obtained the bed pressure reduction curve at different empty bed gas
velocities and particle mass fluxes, and successfully predicted the “choking” phe-
nomenon [112, 113]. Besides that, coupling of EMMS/matrix with Eulerian–
Eulerian Approach also enabled three-dimensional (3D) full-loop numerical sim-
ulation of semi-industrial-scale CFB [114]. The resulting particle velocity distri-
bution agrees well with the experimental result while the solids concentration error
is relatively large.
18 1 Introduction

Based on the EMMS/matrix model, people further refined the momentum

equation for particles in dense phase. They considered that gases in dilute phase
also apply forces to particles in dense phase and developed an even more accurate
QL-EMMS model [115, 116].
In view of the constant controversy upon cluster size in EMMS models, people
established an EMMS drag model that implicitly solves cluster size in the light of
the relation of additional mass forces with the internal particle accelerations in
dilute and dense phases [117]. Here, we call it the WG-EMMS model. Besides, the
literature [118] studied the effects of cluster size equations by introducing different
cluster size equations into the EMMS model and discovered that the cluster size
equation in the model resulted in quite considerable computational errors and has to
be improved.
Although these EMMS models have already been used in the numerical simu-
lation of gas–solid fluidization processes [119–124], there still remain problems of
model accuracy and condition universality [125]. Figure 1.8 compares a few typical
EMMS drag models and the experiment-based, relatively reliable O-S model.
As shown in Fig. 1.8, a commonplace among EMMS models is that drag
reduction is given within a certain range of solids concentration, which partially
validates flow heterogeneity caused by mesoscale structure. However, model
deviation from experiment is still significant and the result does not agree with the
heterogeneous drag variation demonstrated by the O-S model. Looking into the
GY-EMMS and WG-EMMS model curves, there are sudden turns of no physical
significance. The QL-EMMS model is comparatively continuous and smooth, but it
still does not present the essential characteristic of the O-S curve.
From the problems within the EMMS models, several deficiencies in the EMMS
theory can be concluded.
(a) This theory mathematically constitute a problem of constrained nonlinear
programming, since it is builted from a number of state variables, a limited
number of basic equations, and one extreme condition. The constraint function

Fig. 1.8 Chart comparing a 10

6

number of heterogeneous drag

GY-EMMS
models (ρp = 1714 kg/m3,
Drag function β (Ns/m4)

5
dp = 76 μm, ρg = 1.205 kg/m3, 10 Uniform state
μg = 1.81 × 10−5 Pa s,
uslip = 1.0 m/s, εsmf = 0.5, 10
4 QL-EMMS
Gs = 98 kg/m2s)
3
10
WG-EMMS
2
10
O-S
1
10
0.0 0.1 0.2 0.3 0.4 0.5
Solids concentration εs (-)
1.3 Heterogeneous Drag Models 19

is the basic equation that constrains the range of the extremum to be solved.
The objective function is the stability condition that decides the uniqueness of
the solution. However, the feasible domain for solving extremums is too broad
to determine the true value, since the basic equations is much fewer than state
parameters.
(b) This theory lacks an understanding of the relation of the characteristic
parameters for mesoscale structure, i.e., cluster size and density, with local
solids concentration and there are illogical hypotheses. Cluster size is believed
to be inversely related to the system suspended transport energy. Yet this is not
completely expressed by the equation. Cluster density was once assumed to be
constant, i.e., εsc = minimum fluidization concentration εsmf, and was solved
by “Traverse” later. However, the accuracy of cluster density is unknown.
Previous researches mostly focused on cluster size, but neglected the impor-
tance of cluster density [126].
(c) The condition universality of the models has never been theoretically clarified.
The expressions of cluster size and density for different flow states with dif-
ferent levels of heterogeneity need further investigation.
(d) The applicability of the stability condition for solving energy extremums is not
identified. It is used both on the reactor and the computational grid-scale with
no reasonable basis or explanation. The selection of grid size also needs
further studies.
(e) The effects of particle size distribution are not covered yet.

1.4 Numerical Simulation Methods

for Dense Gas–Solid Flow

Numerical simulation of gas–solid two-phase flow can be divided into two basic
types by way of the treatment of solid particles.
One is the continuity method or the Eulerian–Eulerian Approach, which regards
particles as a continuous medium or another kind of fluid (“pseudo fluid” for short).
It is therefore also called the Quasi Fluid Method. For two-phase flow, it is a
two-fluid method which describes particle movement in the same form of a con-
servation equation as for gas phase, except that the particle property parameters and
constitutive relation are described by the kinetic theory of granular medium [127].
The quantity of particles in space is represented by particle volumetric fraction or
solids concentration.
As the Eulerian–Eulerian Approach aims at the macroscopic transport of mass
particles in movement, i.e., the spatial distribution of macroscopic quantities like
the pressure, velocity, concentration, and temperature of gas–solid phases, it is
widely used in the numerical studies on large fluidization systems, particularly
dense gas–solid two-phase flow [128–130].
20 1 Introduction

The defect of the Eulerian–Eulerian Approach mainly lies in its limitation to

uniform particle sizes and densities. For mixed particles with size distribution or
different densities, a separate group of conservation equations has to be established
for each of the particle groups. This not only significantly increases the solution
complexity and computational amount. More importantly, a model describing the
interaction among these particle groups has to be established [131].
The other one is the discretization method or the Eulerian–Lagrangian
Approach. It describes the movement of each particle (group) with Newton’s
Second Law, and obtains the trace coordinates of particles by integrating the
resulting velocity distribution to time [132, 133]. As under the Eulerian–Lagrangian
description framework, different discrete particles can have their respective attribute
characteristics (e.g., density and velocity), this method can apply to polydisperse
particle groups.
The greatest defect of the Eulerian–Lagrangian Approach lies in that the com-
putational time increases exponentially with the increase of particles. It is highly
sensitive to the computational memory. For systems with a large number of par-
ticles (with average solids concentration larger than 5 %), economical and accurate
solution is generally impossible under the present level of collision treatment
because of the frequent and complex interparticle collisions [134]. This is partic-
ularly true for Discrete Element Method which calculates interparticle collision on a
soft sphere model [135, 136]. The particle time step is limited to less than 1 × 10−5 s
due to the stability requirement for the particle collision force integer. The com-
putational amount is quite large for tracing the stress and trajectory of each particle
in the system. A general-purpose computer can simulate a maximum of tens or
hundreds of thousands of particles, which is far smaller than the quantity of par-
ticles in a real fluidization system. Therefore, the Eulerian–Lagrangian Approach
does not apply to the numerical simulation of dense gas–solid flow processes at
present.
In recent years, a new multiphase flow numerical simulation method, the
Multiphase Particle-In-Cell (MP-PIC) strategy has been developed. This method
blends the continuous Eulerian strategy and the discrete Lagrangian strategy
together to describe solid particles. It combines the respective merits of both the
Eulerian–Eulerian Approach and the Eulerian–Lagrangian Approach [137–140].
The treatment of particle–particle interaction by the MP-PIC strategy is closer to
the Eulerian–Eulerian Approach. It obtains continuous particle stress by solving the
Eulerian control equation for solid particles on grid and uses it to describe the
particle–particle coupling [141]. As the particle–particle interaction is solved on
grid, this method largely improves the computational efficiency with nearly no cost
of the computational accuracy. Therefore, the MP-PIC strategy can apply to the
numerical simulation of dense gas–solid fluidization processes.
In next section, the Eulerian–Eulerian two-fluid approach with the MP-PIC
strategy, the two numerical methods applicable for dense gas–solid fluidization
systems, will be compared. The focus is how these two methods deal with the drag
term under gas–solid interphase interaction and how important drag models are to
numerical simulation.
1.4 Numerical Simulation Methods for Dense Gas–Solid Flow 21

1.4.1 Two-Fluid Method

In two-fluid method, a separate group of mass, momentum (i.e., N-S equation), and
energy conservation equations is provided for gas and solid phase, respectively. The
respective internal stress of the two phases, the gas–solid interaction and the tur-
bulence effect can all be addressed by introducing different physical models.
Mass conservation equation:

@ðek qk Þ
þ r  ðek qk~
uk Þ ¼ 0; k ¼ g; s ð1:9Þ
@t

where ε represents the volume fraction, ρ represents the density, u represents the
velocity. Subscripts g and s represent gas phase and solid phase. The sum of the gas
and solid volume fraction is 1.
Momentum conservation equation:

@ðeg qg~
ug Þ
þ r  ðeg qg~ ug Þ ¼ eg rpg þ r ~
ug~ ~ g~
sg þ eg qg~ FD ð1:10Þ
@t
@ðes qs~
us Þ
þ r  ðes qs~ us Þ ¼ es rpg  rps þ r ~
us~ ~ gþ~
ss þ es qs~ FD ð1:11Þ
@t

where pg is the gas-phase pressure, ps is the particle pressure, g is the gravity

acceleration, τg and τs are the stress tensors of gas phase and solid phase, FD is the
average drag of grid that describes the momentum transfer between gas and solid
phases.
Accurate description of drag force, the interaction force between gas and solid
phases, is the key determinant for the accuracy of the Eulerian–Eulerian Approach
[142–144]. Drag force signifies the ability of fluids to drive and entrain particles to
movement. The drag force in the momentum equation is expressed by Eq. (1.12).
Drag is positively related to gas–solid two-phase slip velocity. Its proportional
coefficient β is called “drag function”, expressed by Eq. (1.13).

~ ug  ~
FD ¼ bð~ us Þ ð1:12Þ

3 es eg qg



b¼ ~ us
CD0 f ðeg Þ
ug  ~ ð1:13Þ
4 dp

Here, CD0 is the standard drag coefficient of a single isolated particle, f(εg) is the
drag correction function in a multiparticle system that is expressed differently by
different drag models like Wen-Yu [7], Ergun [8], Gidaspow [67], and Syamlal and
O’Brien [68] models.
A commonplace among these models is that they are all uniform flow models
assuming that drag increases with local solids concentration (see uniform state
curve in Fig. 1.2). When simulating dense heterogeneous gas–solid flows, uniform
22 1 Introduction

drag models usually overestimate the drag, assuming that more particles are
entrained by gas flow. This results in far larger solids circulating mass flux than
reality, and makes it impossible to reproduce the real heterogeneous flow charac-
teristics [145–147]. Hence, establishing an accurate, reliable heterogeneous drag
model is the key to accurate Eulerian–Eulerian numerical simulation.

1.4.2 Multiphase Particle-in-Cell (MP-PIC) Method

In the MP-PIC strategy, the gas-phase control equation is the continuous equation
and momentum equation under the Eulerian description while solid-phase particles
are treated with the Lagrangian strategy. Unlike the traditional Eulerian–Lagrangian
Approach that traces all the particles, the MP-PIC strategy regards particles having
the same attributes (like density, size, position, and velocity) as a computational
parcel that corresponds to fluids of the same parameters, thereby saving a significant
amount of computation [148].
Please note that in the paragraphs below, subscript p (particle/parcel) stands for
the attribute of discrete particles and subscript s (solid) represents the average
attribute of local discrete particles under continuous description. That is to say the
particle parcel is regarded as a quasi-continuous medium.
The gas-phase control equation of the MP-PIC strategy is the same as that of the
two-fluid method [Eq. (1.10)]. The only difference, however, is found in the drag
term,

XNp
np Vp  
~
FD ¼ b ~ xp Þ  ~
ug ð~ up ð1:14Þ
p¼1
Vm

where Np is the quantity of computational parcel in grid, np is the quantity of single

particles contained in each computational parcel, Vp is the volume of single parti-
cles, Vm is the volume of grid, ug (xp) is the gas flow velocity at the position of the
computational parcel, xp is the locational coordinates of the computational parcel,
up is the velocity of the computational parcel, β is the drag function decided by the
drag model.
From Eq. (1.14), the MP-PIC strategy adds up the contribution of each discrete
particle per unit volume to interphase momentum transfer and obtains the level of
the total drag. Compared with the two-fluid method in which drag is simply
assumed to be proportional to the gas–solid interphase average velocity difference,
the MP-PIC strategy provides more accurate treatment of drag and is more likely to
obtain detailed information.
1.4 Numerical Simulation Methods for Dense Gas–Solid Flow 23

The Lagrangian description of solid particles,

d~
xp
¼~
up ð1:15Þ
dt
d~
up b  1 1
¼ ~ xp Þ  ~
ug ð~ up  rpg þ ~
g rsp ð1:16Þ
dt qp qp es q p

where ρp is the particle density, t is the time coordinate. τp is the particle stress
describing particle–particle interaction which is decided by the stress model [149].
Equation (1.16) suggests that particle acceleration is originated from interphase
drag, fluid pressure gradient, gravity, and particle stress gradient. The drag terms
include drag function β which, again, is decided by the drag model.
The solving procedure of the MP-PIC strategy can be summarized as follows
[150],
1. Interpolate particle attributes and other parameters from the particle position to
the Eulerian grid.
2. Iterate and solve the control equation of fluid phase under the Eulerian frame-
work to obtain the continuous flow field and other parameters of fluid phase, and
calculate the solid-phase stress distribution under the Eulerian grid.
3. Interpolate these continuous parameters from the Eulerian grid back to the
positions of the discrete particles to close the interaction between the drag to
which particles are exposed to and particles themselves.
4. Explicitly update the velocities and positions of all particles in the computa-
tional domain according to the Lagrangian kinematic model. Repeat these steps
on the next time step.
From this solving procedure, the MP-PIC strategy can not only deal with a wide
range of particle loading from dilute to dense phases, but also trace each particle
separately, making it convenient to deal with polydisperse particle parcels.
Given all these merits, the MP-PIC strategy is already being used in the
numerical simulation of some gas–solid fluidization systems such as the flow and
reaction characteristics in bubbling beds, turbulence beds, and gasifiers [151–153].
As particle concentration in these systems is relatively limited and the gas–solid
two-phase flow processes are relatively uniform, it is easier for a uniform drag
model to achieve accurate simulation.
Nevertheless, numerical simulations of dense gas–solid two-phase flows in
fluidized bed risers indicate that, the MP-PIC strategy still overestimates the particle
mass flux using uniform drag model. Therefore, the MP-PIC strategy itself is
impossible to reproduce heterogeneous flow characteristics [154]. Only coupling
the heterogeneous EMMS drag model with the MP-PIC strategy, is it possible to
improve the computational accuracy significantly [155]. Besides that, through
simulating spaces with biperiodical boundaries, researchers have discovered that the
MP-PIC strategy has to introduce a mesoscale model before it can simulate dense
heterogeneous gas–solid flow processes accurately [6].
24 1 Introduction

That is to say, for the MP-PIC strategy, the key for accurate simulation of dense
gas–solid flow processes also lies in an accurate, reliable heterogeneous drag
model.
In summary, the existence of mesoscale structure (cluster) in a gas–solid
two-phase fluidization system leads to high flow heterogeneity and significant drag
reduction. For both the traditional two-fluid method and the latest MP-PIC strategy,
an accurate, reliable heterogeneous drag model is the key for accurate numerical
simulation. The EMMS theory is an important theoretical means that helps to build
a heterogeneous drag model. Previous EMMS drag models have turned out to be
defective due to insufficient definite conditions, illogical hypothesis, and the failure
to describe mesoscale structure. In view of this, it is necessary to examine meso-
scale clusters in gas–solid fluidization, deepen our physical understanding on the
formation and evolution of clusters, clarify the relation between cluster properties
and flow heterogeneity, further refine the EMMS theory and develop a more
accurate, more reliable, flow-universal heterogeneous drag model. This will con-
tribute to a better understanding of complex, multiscale gas–solid fluidized flow and
transfer processes, and establishing and refining a numerical simulation method for
large dense gas–solid fluidization systems.

1.5 Research Objectives, Contents, and Methodology

1.5.1 Objectives

By refining the EMMS theory, disclose the formation and evolution pattern of
mesoscale structure in heterogeneous gas–solid two-phase flows, understand the
interphase force, build an accurate drag model, and enable accurate simulation of
fluidization processes.

1.5.2 Contents and Methodology

1. EMMS theory analysis. Analyze problems within the EMMS theory and the
reasons behind them. Study the effects of mesoscale structure—cluster density
and size, on drag through sensitivity analysis. Find out the most crucial cluster
property parameters so as to guide the development and refinement of the
EMMS theory.
2. Mesoscale structure (cluster) modeling. Investigate the local slip velocity and
flow heterogeneity in gas–solid flow processes. Study the relation of cluster
density and size with local solids concentration. Establish a mathematical model
on mesoscale structure and validate its accuracy with experimental results.
3. Heterogeneous drag mode. On the basis of the mesoscale structure model, refine
the EMMS theory relating to mathematical optimization. Develop a
1.5 Research Objectives, Contents, and Methodology 25

heterogeneous EMMS-based drag model, examine its ability to give an accurate

description of the essence of heterogeneous drag reduction, and validate it with
the accurate O-S drag model.
4. Condition universality of model. Examine the relation of mesoscale structure
characteristics with system-scale flow, i.e., the variation of cluster parameters
under different levels of heterogeneous flow. This includes the parametric rep-
resentation of fluidization conditions as well as the relation between flow het-
erogeneity and conditions. Establish the mathematical correlation between
overall heterogeneity and condition parameters to make the model condition
universal.
5. Numerical Verification of drag model. Couple the heterogeneous drag model
with the Eulerian–Eulerian Approach to simulate the heterogeneous gas–solid
fluidization processes under a range of conditions including high and low fluxes,
type A and B particles, 2D and 3D, and various system sizes. Compare the
results with the calculations of other heterogeneous drag models and the
experimental results to validate the accuracy and condition universality of the
drag model.

References

1. Ge W, Liu X, Ren Y, Xu J, Li J (2010) From multi-scale to meso-scale: new challenges for

simulation of complex processes in chemical engineering. CIESC J 61(7):1613–1620 (in
Chinese)
2. Wang W, Hong K, Lu B, Zhang N, Li J (2013) Fluidized bed simulation: structure-dependent
multiscale CFD. CIESC J 64(1):95–106 (in Chinese)
3. Wang J (2009) A review of Eulerian simulation of Geldart A particles in gas-fluidized beds.
Ind Eng Chem Res 48(12):5567–5577
4. Gidaspow D (1994) Multiphase flow and fluidization: continuum and kinetic theory
descriptions. Academic Press, Boston
5. Lu BN, Wang W, Li JH (2009) Searching for a mesh-independent sub-grid model for CFD
simulation of gas-solid riser flows. Chem Eng Sci 64(15):3437–3447
6. Benyahia S, Sundaresan S (2012) Do we need sub-grid scale corrections for both continuum
and discrete gas-particle flow models. Powder Technol 220:2–6
7. Wen CY, Yu YH (1966) Mechanics of fluidization. AIChE Symp 62:100–111
8. Ergun S (1952) Fluid flow through packed columns. Chem Eng Prog 48(1):89–94
9. Sundaresan S (2000) Modeling the hydrodynamics of multiphase flow reactors: current status
and challenges. AIChE J 46(6):1102–1105
10. Wang W, Li Y (2004) Simulation of the clustering phenomenon in a fast fluidized bed: the
importance of drag correlation. Chin J Chem Eng 12(3):335–341
11. Ullah A, Wang W, Li J (2013) Evaluation of drag models for cocurrent and countercurrent
gas–solid flows. Chem Eng Sci 92:89–104
12. Li J (1987) Multiscale model for two-phase flow and energy minimum method. Doctoral,
The Chinese Academy of Sciences, Beijing (in Chinese)
13. Li J, Kwauk M (2001) Multiscale nature of complex fluid-particle systems. Ind Eng Chem
Res 40(20):4227–4237
14. Jin Y, Zhu JX, Wang ZW, Yu ZQ (2001) Fluidization engineering principles. Tsinghua
University Press, Beijing (in Chinese)
26 1 Introduction

15. Bai DR, Jin Y, Yu ZQ et al (1992) The axial distribution of the cross-sectional averaged
voidage in fast fluidized-beds. Powder Technol 71(1):51–58
16. Zhang W, Tung Y, Johnsson F (1991) Radial voidage profiles in fast fluidized-beds of
different diameters. Chem Eng Sci 46(12):3045–3052
17. Xu GW, Sun GG, Gao SQ (2004) Estimating radial voidage profiles for all fluidization
regimes in circulating fluidized bed risers. Powder Technol 139(2):186–192
18. Tortora PR, Ceccio SL, Mychkovsky AG et al (2008) Radial profiles of solids loading and
flux in a gas-solid circulating fluidized bed. Powder Technol 180(3):312–320
19. Qi HY (1997) Euler/Euler Simulation der Fluiddynamik Zirkulierender Wirbelschichten.
Verlag Mainz, Wissenschaftsverlag, Aachen, Germany (ISBN3-89653-224-3)
20. Bi HT, Grace JR (1995) Effect of measurement method on the velocities used to demarcate
the onset of turbulent fluidization. Chem Eng J Biochem Eng J 57(3):261–271
21. Zhu HY, Zhu J (2008) Characterization of fluidization behavior in the bottom region of CFB
risers. Chem Eng J 141(1–3):169–179
22. Yang YL, Jin Y, Yu ZQ et al (1993) Local slip behavior in the circulating fluidized bed.
AIChE Symp Ser 89:81–90
23. Yang YL, Jin Y, Yu ZQ et al (1992) Investigation on slip velocity distributions in the riser of
dilute circulating fluidized-bed. Powder Technol 73(1):67–73
24. Tanner H, Li JH, Reh L (1994) Radial profiles of slip velocity between gas and solids in
circulating fluidized beds. AIChE Symp Ser 90(301):105–113
25. Yerushalmi J, Cankurt NT (1979) Further-studies of the regimes of fluidization. Powder
Technol 24(2):187–205
26. Huang W, Zhu J (2001) An experimental investigation on solid acceleration length in the
riser of a long circulating fluidized bed. J Chem Eng Chin Univ 9(1):70–76
27. Lim KS, Zhu JX, Grace JR (1995) Hydrodynamics of gas-solid fluidization. Int J Multiph
Flow 21:141–193
28. Soong CH, Tuzla K, Chen JC (1994) Identification of particle clusters in circulating fluidized
bed. In: Avidan AA (ed) Circulating fluidized bed technology IV. AIChE, New York,
pp 615–620
29. Tuzla K, Sharma AK, Chen JC et al (1998) Transient dynamics of solid concentration in
downer fluidized bed. Powder Technol 100(2):166–172
30. Guenther C, Breault R (2007) Wavelet analysis to characterize cluster dynamics in a
circulating fluidized bed. Powder Technol 173(3):163–173
31. Ren J, Mao Q, Li J, Lin W (2001) Wavelet analysis of dynamic behavior in fluidized beds.
Chem Eng Sci 56(3):981–988
32. Lu X, Li H (1999) Wavelet analysis of pressure fluctuation signals in a bubbling fluidized
bed. Chem Eng J 75(2):113–119
33. Qi X, Zeng T, Huang W et al (2005) Experimental study of solids holdups inside particle
clusters in CFB risers. J Sichuan Univ (Eng Sci Ed) 5:46–50 (in Chinese)
34. Lin Q, Wei F, Jin Y (2001) Transient density signal analysis and two-phase micro-structure
flow in gas-solids fluidization. Chem Eng Sci 56(6):2179–2189
35. Brereton CMH, Grace JR (1993) Microstructural aspects of the behavior of circulating
fluidized beds. Chem Eng Sci 48(14):2565–2572
36. Lackermeier U, Rudnick C, Werther J et al (2001) Visualization of flow structures inside a
circulating fluidized bed by means of laser sheet and image processing. Powder Technol 114
(1):71–83
37. Liu X, Gao S, Li J (2004) Characteristics of particle clusters in gas-solids circulating
fluidized beds by using PDPA. J Chem Ind Eng (Chin) 55(4):555–562 (in Chinese)
38. Mostoufi N, Chaouki J (2004) Flow structure of the solids in gas–solid fluidized beds. Chem
Eng Sci 59(20):4217–4227
39. Bi X, Zhu J, Jin Y, Yu Z (1993) Forms of particle aggregation in CFB. In: Proceedings of the
6th Chinese conference on fluidization, Huazhong University of Science and Technology,
Wuhan, pp 162–167 (in Chinese)
References 27

40. Cocco R, Shaffer F, Hays R et al (2010) Particle clusters in and above fluidized beds. Powder
Technol 203(1):3–11
41. Noymer PD, Glicksman LR (2000) Descent velocities of particle clusters at the wall of a
circulating fluidized bed. Chem Eng Sci 55(22):5283–5289
42. Qi X (2003) Gas-solids two-phase flow dynamics in circulating fluidized bed risers. Doctoral
dissertation, Sichuan University, Chengdu (in Chinese)
43. Liu M (2004) Study of cluster formation in dense gas-solid flow. Doctoral dissertation,
Tsinghua University, Beijing (in Chinese)
44. Subbarao D (2010) A model for cluster size in risers. Powder Technol 199(1):48–54
45. Zou B, Li H, Xia Y, Mooson K (1993) Statistic model for cluster size distribution in fast
fluidized bed. Eng Chem Metall 14(1):36–42
46. Gu WK, Chen JC (1998) A model for solid concentration in circulating fluidized beds.
Fluidization IX 501–508
47. Gu WK (1999) Diameter of catalyst clusters in FCC. AIChE Symp Ser 95(321):42–47
48. Manyele SV, Parssinen JH, Zhu JX (2002) Characterizing particle aggregates in a
high-density and high-flux CFB riser. Chem Eng J 88(1–3):151–161
49. Afsahi FA, Sotudeh-Gharebagh R, Mostoufi N (2009) Clusters identification and
characterization in a gas–solid fluidized bed by the wavelet analysis. Can J Chem Eng 87
(3):375–385
50. Liu X, Gao S, Li J (2005) Characterizing particle clustering behavior by PDPA measurement
for dilute gas–solid flow. Chem Eng J 108(3):193–202
51. Yang TY, Leu LP (2009) Multiresolution analysis on identification and dynamics of clusters
in a circulating fluidized bed. AIChE J 55(3):612–629
52. Harris AT, Davidson JF, Thorpe RB (2002) The prediction of particle cluster properties in
the near wall region of a vertical riser. Powder Technol 127(2):128–143
53. Sharma AK, Tuzla K, Matsen J et al (2000) Parametric effects of particle size and gas
velocity on cluster characteristics in fast fluidized beds. Powder Technol 111(1–2):114–122
54. Xu J, Zhu JX (2011) Visualization of particle aggregation and effects of particle properties on
cluster characteristics in a CFB riser. Chem Eng J 168(1):376–389
55. Chew JW, Parker DM, Cocco RA et al (2011) Cluster characteristics of continuous size
distributions and binary mixtures of Group B particles in dilute riser flow. Chem Eng J
178:348–358
56. Zhang W (2002) Study on mechanism of collision and agglomeration of fluidized particles.
Doctoral dissertation, Tsinghua University, Beijing (in Chinese)
57. Zhang W, Qi H, You C, Xu X (2002) Study on mechanism of agglomeration and
fragmentation of solid particles in fluidized bed. In: Proceedings of the annual meeting-2002
of Chinese society of particuology and particle technology seminar of both sides of the
Taiwan straitson (in Chinese)
58. Geldart D (1973) Types of gas fluidization. Powder Technol 7(5):285–292
59. Wang Y, Cheng Y, Jin Y et al (2007) On impacts of solid properties and operating conditions
on the performance of gas-solid fluidization systems. Powder Technol 172(3):167–176
60. Horio M, Kuroki H (1994) Three-dimensional flow visualization of dilute dispersed solids in
bubbling and circulating fluidized beds. Chem Eng Sci 49(15):2413–2421
61. Grace JR, Tuot J (1979) A theory for cluster formation in vertically conveyed suspensions of
intermediate density. Trans Inst Chem Eng 57(1):49–54
62. Sundaresan S (2003) Instabilities in fluidized beds. Annu Rev Fluid Mech 35(1):63–88
63. Matsen JM (1982) Mechanisms of choking and entrainment. Powder Technol 32(1):21–33
64. Li HZ (2004) Multi-scale aggregation of particles in gas-solids fluidized beds. Chin Particuol
2(3):101–106
65. Bai D, Jin Y (1991) The interaction between gas and particles in a vertical gas-solid flow
system. J Chem Ind Eng 42(6):697–703 (in Chinese)
66. Mueller P, Reh L (1993) Particle drag and pressure drop in accelerated gas-solid flow. In:
Avidan AA (ed) Preprint volume for circulating fluidized beds IV. AIChE, Somerset, New
York, pp 193–198
28 1 Introduction

67. Gidaspow D (1986) Hydrodynamics of fluidization and heat transfer: supercomputer

modeling. Appl Mech Rev 1(39):1–22
68. Syamlal M, Obrien TJ (1988) Simulation of granular layer inversion in liquid fluidized beds.
Int J Multiph Flow 14(5):473–481
69. Van Wachem BGM, Schouten JC, Van den Bleek CM et al (2001) Comparative analysis of
CFD models of dense gas-solid systems. AIChE J 47(5):1035–1051
70. Mckeen T, Pugsley T (2003) Simulation and experimental validation of a freely bubbling bed
of FCC catalyst. Powder Technol 129:39–152
71. Ye M, Wang J, Van der Hoef MA, Kuipers JAM (2008) Two-fluid modeling of Geldart A
particles in gas-fluidized beds. Particuology 6:540–548
72. Zou LM, Guo YC, Chan CK (2008) Cluster-based drag coefficient model for simulating gas–
solid flow in a fast-fluidized bed. Chem Eng Sci 63(4):1052–1061
73. Liu X, Xu X (2009) Modeling of dense gas-particle flow in a circulating fluidized bed by
distinct cluster method (DCM). Powder Technol 195(3):235–244
74. Wang S, Liu G, Lu H et al (2012) A cluster structure-dependent drag coefficient model
applied to risers. Powder Technol 225:176–189
75. Wang S, Yang Y, Lu H et al (2012) Computational fluid dynamic simulation based cluster
structures-dependent drag coefficient model in dual circulating fluidized beds of chemical
looping combustion. Ind Eng Chem Res 51(3):1396–1412
76. Ye M, van der Hoef MA, Kuipers JAM (2005) The effects of particle and gas properties on
the fluidization of Geldart A particles. Chem Eng Sci 60:4567–4580
77. O’Brien TJ, Syamlal M (1993) Particle cluster effects in the numerical simulation of a
circulating fluidized bed. In: Avidan AA (ed) Preprint volume for CFB-IV. AIChE, New
York, pp 430–435
78. Cruz E, Steward FR, Pugsley T (2006) New closure models for CFD modeling of
high-density circulating fluidized beds. Powder Technol 169:115–122
79. Li F, Qi H, You C (2009) Basis of moderate temperature flue gas desulphurization processes
in CFB-Drag force model and numerical research of two-phase flows. J Eng Thermophys 30
(8):1335–1338 (in Chinese)
80. Ma J, Ge W, Xiong Q et al (2009) Direct numerical simulation of particle clustering in gas–
solid flow with a macro-scale particle method. Chem Eng Sci 64(1):43–51
81. Wang X (2011) Study on fluidization and bond characteristics of coarse cohesive particles in
fluidized process. Doctoral dissertation, Tsinghua University, Beijing (in Chinese)
82. Wang X, Liu K, You CF (2011) Drag force model corrections based on non-uniform particle
distributions in multi-particle systems. Powder Technol 209:112–118
83. Van der Hoef MA, Ye M, Van Sint Annaland M et al (2006) Multiscale modeling of
gas-fluidized beds. Adv Chem Eng 31:65–149
84. Agrawal K, Loezos PN, Syamlal M et al (2001) The role of meso-scale structures in rapid
gas-solid flows. J Fluid Mech 445:151–185
85. Wang J, Van der Hoef MA, Kuipers JAM (2009) Why the two-fluid model fails to predict the
bed expansion characteristics of Geldart A particles in gas-fluidized beds: a tentative answer.
Chem Eng Sci 64(3):622–625
86. Shah S, Ritvanen J, Hyppänen T et al (2012) Space averaging on a gas–solid drag model for
numerical simulations of a CFB riser. Powder Technol 218:131–139
87. Benyahia S (2009) On the effect of subgrid drag closures. Ind Eng Chem Res 49
(11):5122–5131
88. Igci Y, Andrews AT, Sundaresan S et al (2008) Filtered two-fluid models for fluidized
gas-particle suspensions. AIChE J 54(6):1431–1448
89. Igci Y, Sundaresan S (2011) Constitutive models for filtered two-fluid models of fluidized
gas-particle flows. Ind Eng Chem Res 50(23):13190–13201
90. Parmentier JF, Simonin O, Delsart O (2012) A functional subgrid drift velocity model for
filtered drag prediction in dense fluidized bed. AIChE J 58(4):1084–1098
91. Arthur T, Andrews IV, Loezos PN, Sundaresan S (2005) Coarse-grid simulation of
gas-particle flows in vertical risers. Ind Eng Chem Res 44:6022–6037
References 29

92. Igci Y, Sundaresan S (2011) Verification of filtered two-fluid models for gas-particle flows in
risers. AIChE J 57(10):2691–2707
93. Igci Y, Pannala S, Benyahia S et al (2012) Validation studies on filtered model equations for
gas-particle flows in risers. Ind Eng Chem Res 51:2094–2103
94. Wang W, Lu BN, Zhang N et al (2010) A review of multiscale CFD for gas-solid CFB
modeling. Int J Multiph Flow 36:109–118
95. Li J, Ge W, Wang W et al (2013) From EMMS model to EMMS paradigm. Springer, Berlin,
Heidelberg
96. Li JH, Zhang J, Ge W et al (2004) Multi-scale methodology for complex systems. Chem Eng
Sci 59:1687–1700
97. Zhang Y, Ge W, Li JH (2005) Simulation of heterogeneous structures and analysis of energy
consumption in particle-fluid systems with pseudo-particle modeling. Chem Eng Sci
60:3091–3099
98. Zhang Y, Ge W, Wang X et al (2011) Validation of EMMS-based drag model using lattice
Boltzmann simulations on GPUs. Particuology 9:365–373
99. Xiao H (2001) Theoretical and numerical investigation on gas-solid drag model within
eulerian scope. Master’s thesis, Tsinghua University, Beijing (in Chinese)
100. Xiao H, Qi H, You C, Xu X (2003) Theoretical model of drag between gas and solid phase.
J Chem Ind Eng (Chin) 54(3):311–315 (in Chinese)
101. Qi HY, Li F, Xi B et al (2007) Modeling of drag with the Eulerian approach and EMMS
theory for heterogeneous dense gas-solid two-phase flow. Chem Eng Sci 62(6):1670–1681
102. Li JH, Cheng C, Zhang Z et al (1999) The EMMS model-its application, development and
updated concepts. Chem Eng Sci 54(22):5409–5425
103. Ge W, Chen F, Gao J et al (2007) Analytical multi-scale method for multi-phase complex
systems in process engineering—bridging reductionism and holism. Chem Eng Sci 62
(13):3346–3377
104. Ge W, Wang W, Yang N (2011) Meso-scale oriented simulation towards virtual process
engineering (VPE)—the EMMS Paradigm. Chem Eng Sci 66:4426–4458
105. Cheng C, Ge W (2008) Further analysis on the choking criteria in the EMMS model. Process
Chem 4:620–624 (in Chinese)
106. Xu GW, Li JH (1998) Analytical solution of the energy-minimization multi-scale model for
gas-solid two-phase flow. Chem Eng Sci 53(7):1349–1366
107. Ge W, Li JH (2002) Physical mapping of fluidization regimes-the EMMS approach. Chem
Eng Sci 57(18):3993–4004
108. Yang N, Wang W, Ge W et al (2003) CFD simulation of concurrent-up gas-solid flow in
circulating fluidized beds with structure-dependent drag coefficient. Chem Eng J 96
(1–3):71–80
109. Shah MT, Utikar RP, Tade MO et al (2011) Hydrodynamics of an FCC riser using energy
minimization multiscale drag model. Chem Eng J 168(2):812–821
110. Armstrong LM, Luo KH, Gu S (2010) Two-dimensional and three-dimensional
computational studies of hydrodynamics in the transition from bubbling to circulating
fluidised bed. Chem Eng J 160(1):239–248
111. Wang W, Li JH (2007) Simulation of gas-solid two-phase flow by a multi-scale CFD
approach—extension of the EMMS model to the sub-grid level. Chem Eng Sci 62
(1–2):208–231
112. Wang W, Lu B, Dong WG et al (2008) Multi-scale CFD simulation of operating diagram for
gas–solid risers. Can J Chem Eng 86(3):448–457
113. Wang W, Lu B, Li JH (2007) Choking and flow regime transitions: simulation by a
multi-scale CFD approach. Chem Eng Sci 62(3):814–819
114. Zhang N, Lu BN, Wang W et al (2008) Virtual experimentation through 3D full-loop
simulation of a circulating fluidized bed. Particuology 6(6):529–539
115. Li F (2009) Investigations on the turbulent gas-solid two-phase interactions in fluidized
desulfurization process. Doctoral dissertation, Tsinghua University, Beijing
30 1 Introduction

116. Li F, Chen C, Wang J, Qi H (2011) QL-EMMS drag model and its revision for fluidized
dense gas-solid two-phase flow. J Eng Thermophys 1:75–79
117. Wang JW, Ge W, Li JH (2008) Eulerian simulation of heterogeneous gas-solid flows in CFB
risers: EMMS-based sub-grid scale model with a revised cluster description. Chem Eng Sci
63:1553–1571
118. Shah MT, Utikar RP, Tade MO et al (2011) Simulation of gas–solid flows in riser using
energy minimization multiscale model: effect of cluster diameter correlation. Chem Eng Sci
66(14):3291–3300
119. Lu BN, Wang W, Li JH et al (2007) Multi-scale CFD simulation of gas-solid flow in MIP
reactors with a structure-dependent drag model. Chem Eng Sci 62:5487–5494
120. Hartge E, Ratschow L, Wischnewski R et al (2009) CFD-simulation of a circulating fluidized
bed riser. Particuology 7(4):283–296
121. Wang X, Jiang F, Lei J et al (2011) A revised drag force model and the application for the
gas–solid flow in the high-density circulating fluidized bed. Appl Therm Eng 31(14):
2254–2261
122. Nikolopoulos A, Papafotiou D, Nikolopoulos N et al (2010) An advanced EMMS scheme for
the prediction of drag coefficient under a 1.2 MWth CFBC isothermal flow—part I: numerical
formulation. Chem Eng Sci 65(13):4080–4088
123. Nikolopoulos A, Atsonios K, Nikolopoulos N et al (2010) An advanced EMMS scheme for
the prediction of drag coefficient under a 1.2 MWth CFBC isothermal flow—part II:
numerical implementation. Chem Eng Sci 65(13):4089–4099
124. Chen C, Li F, Qi HY (2012) Modeling of the flue gas desulfurization in a CFB riser using the
Eulerian approach with heterogeneous drag coefficient. Chem Eng Sci 69(1):659–668
125. Chen C, Li F, Qi HY (2013) An improved EMMS-based drag model for gas-solid flow in
CFB risers with a novel sub-model for clusters. In: 8th international conference on
multiphase flow, Jeju, Korea
126. Naren PR, Lali AM, Ranade VV (2007) Evaluating EMMS model for simulating high solid
flux risers. Chem Eng Res Des 85(A8):1188–1202
127. Balzer G (2000) Gas-solid flow modeling based on the kinetic theory of granular media:
validation, applications and limitations. Powder Technol 113(3):299–309
128. Enwald H, Peirano E, Almstedt AE (1996) Eulerian two-phase flow theory applied to
fluidization. Int J Multiph Flow 22:21–66
129. Wang JW, Ge W (2005) Collisional particle-phase pressure in particle-fluid flows at high
particle inertia. Phys Fluids 17(1):281–312
130. Wei Wang, Youchu Li (2000) Progress of the simulation of particle-fluid two-phase flow.
Process Chem 12(2):208–217 (in Chinese)
131. Rizk MA (1993) Mathematical modeling of densely loaded. particle-laden turbulent flows.
Atomization Sprays 3:1–27
132. Lyczkowski RW (2010) The history of multiphase computational fluid dynamics. Ind Eng
Chem Res 49(11):5029–5036
133. Deen NG, Van Sint Annaland M, Van der Hoef MA et al (2007) Review of discrete particle
modeling of fluidized beds. Chem Eng Sci 62(1):28–44
134. O’Rourke PJ (1981) Collective drop effects on vaporizing liquid sprays. Ph.D. thesis,
Princeton University, USA
135. Tsuji Y, Kawaguchi T, Tanaka T (1993) Discrete particle simulation of two-dimensional
fluidized bed. Powder Technol 77(1):79–87
136. Kawaguchi T, Tanaka T, Tsuji Y (1998) Numerical simulation of two-dimensional fluidized
beds using the discrete element method (comparison between the two-and three-dimensional
models). Powder Technol 96(2):129–138
137. Andrews MJ, O’rourke PJ (1996) The multiphase particle-in-cell (MP-PIC) method for dense
particulate flows. Int J Multiph Flow 22(2):379–402
138. Snider DM, O’Rourke PJ, Andrews MJ (1998) Sediment flow in inclined vessels calculated
using a multiphase particle-in-cell model for dense particle flows. Int J Multiph Flow 24
(8):1359–1382
References 31

139. Patankar NA, Joseph DD (2001) Lagrangian numerical simulation of particulate flows. Int J
Multiph Flow 27(10):1685–1706
140. Snider DM (2001) An incompressible three-dimensional multiphase particle-in-cell model
for dense particle flows. J Comput Phys 170(2):523–549
141. O’Rourke PJ, Zhao PP, Snider D (2009) A model for collisional exchange in gas/liquid/solid
fluidized beds. Chem Eng Sci 64(8):1784–1797
142. Breault RW (2012) An analysis of clustering flows in a CFB riser. Powder Technol
220:79–87
143. Li JH, Kuipers JAM (2003) Gas-particle interactions in dense gas-fluidized beds. Chem Eng
Sci 58(3):711–718
144. Du W, Bao X, Xu J et al (2006) Computational fluid dynamics (CFD) modeling of spouted
bed: assessment of drag coefficient correlations. Chem Eng Sci 61(5):1401–1420
145. Zhang Y, Reese JM (2003) The drag force in two-fluid models of gas–solid flows. Chem Eng
Sci 58(8):1641–1644
146. Helland E, Occelli R, Tadrist L (2000) Numerical study of cluster formation in a gas–particle
circulating fluidized bed. Powder Technol 110(3):210–221
147. Sundaresan S, Beds CF (2011) Reflections on mathematical models and simulation of
gas-particle flows. In: Tenth international conference on circulating fluidized beds and
fluidization technology-CFB-10, Sunriver, Oregon, USA, pp 21–40
148. Benyahia S, Galvin JE (2010) Estimation of numerical errors related to some basic
assumptions in discrete particle methods. Ind Eng Chem Res 49(21):10588–10605
149. O’Rourke PJ, Snider DM (2010) An improved collision damping time for MP-PIC
calculations of dense particle flows with applications to polydisperse sedimenting beds and
colliding particle jets. Chem Eng Sci 65(22):6014–6028
150. Zheng K (2012) A hybrid approach to particle breakage simulation in dense particulate flows.
Master thesis, Tsinghua University, Beijing (in Chinese)
151. Karimipour S, Pugsley T (2012) Application of the particle in cell approach for the
simulation of bubbling fluidized beds of Geldart A particles. Powder Technol 220:63–69
152. Snider DM, Clark SM, O’Rourke PJ (2011) Eulerian-Lagrangian method for
three-dimensional thermal reacting flow with application to coal gasifiers. Chem Eng Sci
66(6):1285–1295
153. Abbasi A, Ege PE, De Lasa HI (2011) CPFD simulation of a fast fluidized bed steam coal
gasifier feeding section. Chem Eng J 174(1):341–350
154. Chen C, Werther J, Heinrich S et al (2012) CPFD simulation of circulating fluidized bed
risers. Powder Technol 235:238–247
155. Li F, Song F, Benyahia S et al (2012) MP-PIC simulation of CFB riser with EMMS-based
drag model. Chem Eng Sci 82:104–113
Chapter 2
Analysis on the EMMS Theory

2.1 General

In order to develop an accurate Energy Minimum Multiscale (EMMS) drag model,

it is necessary to clarify the effects of mesoscale structure on drag force and identify
the key parameter that are decisive to drag. This chapter aims to give the research
results of this aspect in three sections. Section 2.1 examines the solving methods
and results of cluster characteristic parameters (size and density) and drag function
based on a relatively reliable heterogeneous drag model QL-EMMS. Problems
within the EMMS theory itself and the reasons behind them are discussed.
Section 2.2 focuses on cluster size based on theoretical analysis and physical
judgment. Cluster size correlation conforming to physical judgment is introduced
into the QL-EMMS model to investigate its effects on drag function. Section 2.3
analyzes the relationship between cluster density and solids concentration. Examine
how cluster density acts on drag function and clarify cluster characteristic param-
eters that are essentially determinative to drag.

2.2 EMMS Theory

As mentioned above, several heterogeneous drag models have been developed from
the EMMS theory. They are collectively called “EMMS models.” Among them, the
QL-EMMS model not only considers the contribution of local solids acceleration,
but also discriminates the difference among particle accelerations in dense, dilute,
and interaction phases. Besides that, its stress equation treatment for particle in
dense phase is more accurate. It is therefore believed to be a relatively reliable
EMMS model [1, 2]. The following studies are based on the QL-EMMS model to
discuss problems with the EMMS theory itself and the reasons behind them.

© Springer-Verlag Berlin Heidelberg 2016 33

C. Chen, Investigations on Mesoscale Structure in Gas–Solid Fluidization
and Heterogeneous Drag Model, Springer Theses,
DOI 10.1007/978-3-662-48373-2_2
34 2 Analysis on the EMMS Theory

2.2.1 Basic Equations

The key of the Theory of EMMS lies in system decomposition and extreme stability
condition. Figure 2.1 gives a vivid description of the mesoscale structure and phase
interaction in heterogeneous gas–solid fluidization processes after system decom-
position. The entire heterogeneous system is decomposed into three hypothetic
uniform subsystems, particle dense phase (cluster), particle dilute phase, and their
interaction phase. This gives rise to three scales of phase interaction. The micro-
scale effect is the interaction between gas flow and single particles inside each
subsystem. The mesoscale effect is the interaction between the dilute and dense
phases resulted from gas flowing round the dense phase. The macroscale effect is
the interaction between the entire gas–solid fluidization system and the reactor wall.
Clusters between microscale single particles and macroscale reactor are also called
“mesoscale structure.”
From Fig. 2.1, the state of the entire system can be represented by ten variables:
solid content in dilute/dense phase (εsf/εsc), gas velocity in dilute/dense phase
(Uf/Uc), particle velocity in dilute/dense phase (Upf/Upc), particle acceleration in
dilute/dense phase (af/ac), the volume fraction of dense phase in the entire system
(f). Besides, the dense phase is regarded as a cluster and the equivalent cluster size
is expressed as dcl. By integrating these variables, we get the state variable repre-
!
senting the entire heterogeneous system, X

Dense phase εsc, f, dcl, Uc, Upc, ac + Dilute phase εsf, Uf, Upf, af

Micro-scale Micro-scale
Slip velocity between particle Slip velocity between particle
and fluid in dense cluster and fluid in dilute phase

Meso-scale Macro-scale
Slip velocity between dilute Global average slip velocity
phase and dense cluster within equipment boundaries

Fig. 2.1 Decomposition of mesoscale structures and phase interaction in heterogeneous gas–solid
fluidization system [5]
2.2 EMMS Theory 35

!

X ¼ esc ; esf ; Uc ; Uf ; Upc ; Upf ; dcl ; f ; ac ; af ð2:1Þ

where solids concentration in dense phase (cluster) εsc has the following relation
with cluster density ρcl,
qcl ¼ qp  esc þ qg  ð1  esc Þ ð2:2Þ

For a given gas and solid particle, the gas and solid densities (ρg and ρp) are
constant and thus εsc is equivalent to ρp. To simplify the expression, we will use
“cluster density” for “solids concentration in cluster” in the next paragraphs.
Six basic equations including mass conservation equation, force equilibrium
equation, and pressure drop equilibrium equation for dilute and dense phases,
respectively, plus a cluster size equation, are established to describe the relation
among these ten variables.
(1) Particle force equilibrium equation in dense phase per unit volume


fnc Fc þ ni Fi ¼ f esc qp  qg ðg þ ac Þ ð2:3Þ

(2) Particle force equilibrium equation in dilute phase per unit volume


nf Ff ¼ esf qp  qg ðg þ af Þ ð2:4Þ

(3) Pressure drop equilibrium equation per unit bed height

ðDp=DhÞi
ðDp=DhÞf þ ¼ ðDp=DhÞc ð2:5Þ
1f

ni F i
nf F f þ ¼ nc Fc ð2:6Þ
1f

(4) Gas mass conservation equation

U g ¼ U f ð1  f Þ þ U c f ð2:7Þ

(5) Particle mass conservation equation

Up ¼ Upf ð1  f Þ þ Upc f ð2:8Þ

(6) Equivalent cluster size
h  i
Up emf Up
 Umf þ 1e
1emax g
dcl ¼   dp ð2:9Þ
mf

qp emf Up
Nst q q  Umf þ 1e mf
g
p g
36 2 Analysis on the EMMS Theory

where Nst is the suspended transport energy defined as the energy consumed by gas
to transport unit mass particle in suspension, and equals to the sum of the suspended
transport energy of all the subsystems.

Nst ¼ðNst Þc þ ðNst Þf þ ðNst Þi

 
q p  qg esc  esf 2 ð2:10Þ
¼ g Ug  f ð1  f ÞUf
qp 1  eg

Besides, the EMMS theory assumes that a stability condition exists in a heter-
ogeneous gas–solid fluidization system. That is, the suspended transport energy has
reached its minimum. This stability condition is the result of interaction and
coordination between “gas control” and “particle control.” The former signifies that
gas flow always tends to select to flow along the minimum-resistance path. The
latter means that particle always tends to maintain the minimum gravity potential
energy.

Nst ! min ð2:11Þ

2.2.2 Solving Procedure

Mathematically, the ten state variables, six basic equations, and one extreme con-
dition contained in the EMMS theory as discussed above is a problem of con-
strained nonlinear programming. Classical solutions to this kind of problem include
penalty function, feasible direction, and general reduced gradient, which are quite a
challenge to numerical solution and difficult to converge. To solve this problem, a
series of heterogeneous drag models together with some simplified solution
methods were developed.
The earliest EMMS model assumed that particle drag is balanced with effective
gravity. That is the particle acceleration af and ac equals 0, which makes a problem
of nonlinear programming composed of eight variables, six equations, and one
extreme condition. Using general reduced gradient [3], the result shows that, when
target function “Nst = min” is satisfied, εsc equals to εsmf and εsf equals to 1 − εmax.
Here, εsmf is the minimum fluidization particle concentration and εmax is the max-
imum voidage where cluster exists. εmax = 0.9997, which is the exact time when
cluster disappears and particles are uniformly distributed in gas [4]. Subsequent
researchers [5, 6] directly used “εsc = εsmf, εsf = 1 − εmax” as the simplified stability
condition. In this way, the problem of nonlinear programming is transformed into a
problem of linear equation group and the remaining six variables are solved with
the six basic equations in closed form. Nevertheless, this EMMS model only applies
to the overall analysis of fluidized bed since particle acceleration is neglected.
The latest QL-EMMS model does not have any hypothetic or simplified con-
dition and can solve the gas–solid flow parameters in local cells. It is therefore more
2.2 EMMS Theory 37

logical and reliable. When performing EMMS analysis on local cells, the average
solids concentration in grid is the weighted sum of the respective solids concen-
tration of dilute and dense phases,

es ¼ f esc þ ð1  f Þesf ð2:12Þ

Given the average solids concentration in grid, this equation can be added into
the basic equation group of the EMMS theory as the seventh one.
In other words, the QL-EMMS model is a nonlinear programming problem
composed of ten state variables, seven basic equations, and one extreme condition.
The QL-EMMS model assumes that cluster characteristic parameters (size and
density) are the main determinants for local drag and conducts two-step EMMS
analysis [7].
The first step conducts EMMS analysis on macro operating conditions (oper-
ating gas velocity Ug and particle circulating mass flux Gs), traverses cluster density
εsc, cluster size dcl, and particle acceleration in dilute phase af, calculate the
remaining seven variables with seven basic equations in closed form, and finally
obtains the only solution group by minimum energy constraint. The solution
includes the correlation of cluster size and density, dcl(εs) and εsc(εs).
The second step conducts EMMS analysis on the local flow parameters (gas
velocity ug, particle velocity us and solids concentration εs) in cells given by
computational fluid dynamics (CFD). The cluster size correlation dcl(εs) achieved in
Step 1 is used as a substitute for the original size formula (Eq. 2.9). The cluster
density correlation εsc(εs) achieved in Step 1 is used as a supplementary equation.
This is equivalent to correcting one basic equation and supplementing one basic
equation. Now still there are ten variables while the number of equations increases
to eight. Finally, all the local state variables are derived by traverse solution and
extreme constraint.
With all state variables solved, the average drag function in local grid is derived
by summing up the weighted drags in the three subsystems. Thus, the average drag
coefficient per unit volume, i.e., the drag function β, is obtained.
eg eg
b¼ FD ¼ ½fnc Fc þ ð1  f Þnf Ff þ ni Fi  ð2:13Þ
uslip uslip

2.2.3 Result Analysis

With a two-step solution, QL-EMMS model combines global operating parameters

with local flow parameters. In the numerical simulation of gas–solid fluidization
processes, for each computational grid, as long as CFD simulation gives the average
flow parameters ug, us, and εs, the drag function can be derived from EMMS
analysis and then returned to the CFD momentum conservation equation for next
step calculation. According to this coupling process, people usually apply the
38 2 Analysis on the EMMS Theory

QL-EMMS model to the numerical simulation of gas–solid flow, and then compare
the calculation and experimental results. Nevertheless, CFD numerical calculation
often covers the defects intrinsic to drag models.
From the model solving procedure, the cluster size and density resulted from the
EMMS analysis in the first step are the input parameters for the second step, thus
directly decide the accuracy of the local grid drag function, which in turn is the
input parameter for the CFD calculation in the next step. Therefore, the cluster size
and density finally decide the accuracy of flow calculation.
From the view of multiscale validation, it should be performed in turn on
“mesoscale” cluster characteristics, “grid-scale” drag function, and “system-scale”
gas–solid flow characteristics. The accuracy of each scale relies on the accuracy of
the previous scale as the prerequisite.
As a result, before applying a drag model to any numerical simulation of gas–
solid fluidization, it is necessary to first conduct mesoscale and grid-scale testing.
That is to say, the accuracy of cluster characteristic parameters and drag function
should be validated before CFD simulation. However, the analyses of QL-EMMS
model reveal some problems.

2.2.3.1 Illogical Description of Cluster Characteristic Parameters

As shown in Fig. 2.2a, cluster size presents a unimodal profile, reaching its max-
imum near the point where solids concentration is 0.45 but being dozens of times of
single particle size both at the dilute end and dense end. This does not agree with
physical judgment, especially at the dilute end. It is impossible for particles to
gather into clusters when solids concentration approaches zero. If any, only a very
modest number of particles are scattered in gas flow. So it is impossible for the
so-called cluster size to be 50–100 times the single particle size. What’s more, no
reasonable explanation has ever been given as to why cluster size reaches its
maximum at εs = 0.45.
The reason why QL-EMMS model gives such a result lies in the defects in its
cluster size formula. In the EMMS theory, cluster size is assumed to be inversely
related to system suspended transport energy. That is, the larger energy con-
sumption means the smaller cluster size. This formula has been widely doubted and
regarded as one of the aspects to be improved within the EMMS theory.
As shown in Fig. 2.2b, cluster density increases monotonously with increasing
solids concentration, but suddenly deflects at solids concentration of 0.45, and finally
stabilizes at the dense end. In reality, gas–solid fluidization is a varying process in
which cluster density should change smoothly and continuously with local solids
concentration without any sudden deflection. Further, why cluster density remains
constant in a high solids concentration region lacks physical implication.
The reason must be that the EMMS theory lacks an equation describing cluster
density which is a traverse solution and minimum energy constraint. Therefore, it is
necessary to find other solving method or introduce an accurate, reliable cluster
density model.
2.2 EMMS Theory 39

Fig. 2.2 Cluster (a)

characteristic parameters 300

Dimensionless cluster size dcl /dp (-)

predicted by QL-EMMS
model (ρg = 1.144 kg/m3, 250
μg = 1.848 × 10−5 Pa s,
ρp = 1714 kg/m3, dp = 76 μm, 200 (4.3, 147)
εsmf = 0.6). a Dimensionless
cluster size. b Cluster density 150

100
(3.7, 98)
50

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentraiton εs (-)

(b) 0.6

0.5
εsc(-)

0.4
Cluster density

0.3

0.2 Ug(m/s) Gs(kg/m2s)

3.7 98
0.1
4.3 147

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentraiton εs (-)

2.2.3.2 Inability to Reflect the Essential Feature of Drag Reduction

Figure 2.3 compares the drag functions from the QL-EMMS model, the uniform
model, and the O-S model. Compared with the uniform drag function curve, the
QL-EMMS can reflect the drag reduction in connection with heterogeneous flow
characteristics, but the pattern of reduction does not agree with the
experiment-based O-S curve. As shown by the O-S curve, drag function changes
continuously and smoothly with local solids concentration, first decreasing, then
increasing, and approaching to the uniform curve at the dilute end (εs → 0) and the
dense end (εs > 0.4). The QL-EMMS model, on the other side, rises monotonously
with local solids concentration and suddenly deflects at εs = 0.45, increasing the
ascending rate and even exceeding the uniform drag function. The deflection point
of drag function corresponds to that of cluster characteristic parameters (Fig. 2.2).
This also confirms the importance of cluster characteristic parameters on drag force.
40 2 Analysis on the EMMS Theory

Fig. 2.3 Chart comparing the 6

10
drag coefficient of QL-EMMS
model with uniform model 5
Uniform state
and O-S result 10

Drag function β (Ns/m )

4
(ρg = 1.144 kg/m3,
μg = 1.848 × 10−5 Pa s, 4
10 QL-EMMS
ρp = 1714 kg/m3, dp = 76 μm,
εsmf = 0.6) 3
10

2
10 2
Ug (m/s) G s (kg/m s)

1
10 3.7 98
O-S 4.3 147
0
10
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs(-)

In a word, although the QL-EMMS model is able to reflect drag reduction in

heterogeneous flow, it cannot capture the essential rule of drag reduction.

2.2.3.3 Poor Condition Universality

As shown in Fig. 2.3, the O-S curves under two conditions are nearly one order of
magnitude apart from each other, and the drag function is smaller under high
particle circulating mass flux Gs. This is because the increase of Gs increases the
solids concentration, flow heterogeneity, and thus reduce drag in fluidized bed.
Nevertheless, the QL-EMMS drag function curves under two conditions are almost
completely overlapped. It cannot capture the effects of condition changes.
This must be related to the cluster characteristic parameters. The cluster density
curve in Fig. 2.2b hardly changes with the condition. Though the cluster size in
Fig. 2.2a increases with Gs, it does not make much difference to the drag function.
This, on the one hand, suggests that cluster size does not increase much and, on the
other hand, drag function is not sensitive to cluster size. The possible key factor lies
in cluster density. Thus, it is necessary to examine the effects of cluster charac-
teristic parameters on drag function, and identify a parameter that is decisive to drag
function. Only this key parameter has the ability to automatically reflect the effect of
condition changes, the model can be condition universal.
In order to solve the problems mentioned above, and to improve the EMMS
theory, it is necessary to incorporate more equations in addition to improving the
existing equations. The EMMS theory contains a number of variables, a limited
number of basic equations, and one extreme condition, making it mathematically a
problem of constrained nonlinear programming. The constraint function is the basic
equation that restricts the range of extremum solution, and the target function is the
stability condition that decides the uniqueness of the solution. When the number of
basic equations is much smaller than the state parameters, the feasible domain for
2.2 EMMS Theory 41

extremum solution is too broad to find out the true value definitely although it is
mathematically solvable. This is because it is hard for a few constraint equations to
tell whether they can fully reflect the true characteristics of a gas–solid fluidization
system. The trueness of experimental test solution is the “ultimate” validation, but it
does not solve how the true value is obtained.
Therefore, it is important to add as many basic equations as practically possible
to reduce the feasible domain and help to obtain the true solution. Meanwhile, basic
equations should reflect the real physical processes to the largest extent. That is,
basic equations must be both logical and reliable.
The two-step solution of the QL-EMMS model is actually also a process of
“improving the existing equations and incorporating more ones.” That is, it uses the
cluster size from the first step to correct the original cluster size equation, and the
cluster density from the first step as a supplementary equation in the second step. As
a result, the cluster characteristic size and density results inevitably decide the
accuracy of the drag function result. Furthermore, as cluster is the essential source
of flow heterogeneity, accurate descriptions of cluster characteristics are absolutely
crucial to accurate prediction of heterogeneous drag.
So far, researchers across the globe have highly recognized the importance of
cluster size, believing that it makes critical contribution to drag. They believe that
developing an accurate cluster size model is an important direction for improving
drag models. Yet, few researches have been reported concerning cluster density.
In view of this, this study examines the effects of cluster size and density on drag
function respectively, and conduct comparative analysis to identify which is the key
decisive factor. This will guide the improvement of the EMMS theory.

2.3 Cluster Size and Its Effects on Drag

In real gas–solid flow processes, clusters appear in varying shapes like flocs, belts,
clusters, or flakes and change continuously with time as shown in Fig. 2.4. Hence,
cluster size is not always the spherical diameter, but may also indicate its axial or
radial length or equivalent diameter. This nonuniform definition makes it difficult to
truly compare the size of different shapes of clusters.
Generally, cluster size is measured by fiber photography or double-probe fiber.
The results are often limited to the axial size of clusters when they are in vertical
movement, but do not reflect their exact horizontal scale. This often gives rise to
deviation in understanding.
Existing cluster size models mainly include the empirical model based on
experimental measurement, the equivalent diameter model based on simple
hypotheses, and the model within the EMMS theory itself (Eq. 2.9), as will be
discussed in the next subsection.
42 2 Analysis on the EMMS Theory

Fig. 2.4 Cluster size by

shape

2.3.1 Cluster Size Model

2.3.1.1 Empirical Models and Equivalent Diameter Model

There are a number of empirical formulas describing cluster size [8–10],

h i1:3889
dcl ¼ 1 þ 1:8543  ð1  es Þ1:5 e0:25
s = ð 0:6  e s Þ 2:41
ð2:14Þ

dcl ¼ dp þ ð0:27  10dp Þes þ 32e6s ð2:15Þ

!!0:333
qg
dcl ¼ 1:125  ð1 þ 3:8  ð1  eeÞ  ggÞ  l0:667  9:8  1 ð2:16Þ
g
qp

where  
ee ¼ 0:6  1  ð1  es =0:6Þ3:4
gg ¼ 1ðe =0:6
1
Þ0:333
s

Figure 2.5 shows the cluster size curves from the empirical formulas above.
Their commonplace is that cluster size increases monotonously with local solids
concentration and even tends toward infinite. Cluster size is generally several orders
of magnitude larger than single particle size. For example, from the “ring-nucleus”
structure in fluidized bed, in the near-wall “ring” region where the solids concen-
tration is high and moves down the wall, possibly giving shape to belt-like cluster,
thus the measured size is larger. In the “nucleus” region at the center where the flow
is quite dilute, the cluster size is naturally smaller.
2.3 Cluster Size and Its Effects on Drag 43

Fig. 2.5 Cluster size versus 6

10

Dimensionless cluster size d cl /dp (-)

local solid concentration
(ρg = 1.144 kg/m3, 5
10
μg = 1.848 × 10−5 Pa s,
ρp = 1714 kg/m3, dp = 76 μm) 4
10
(2.15)
3
10

2
10 (2.16)

10
1 (2.14)

0
10
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentraion εs (-)

Some scholars summarized the experimental data of clusters in the near-wall

region and worked out the correlation between cluster size and average
cross-sectional particle concentration εs,av [11],
es;av
dcl ¼ ð2:17Þ
40:8  94:5es;av

As shown in Fig. 2.6, the variation of cluster size with average cross-sectional
particle concentration is similar to that shown in Fig. 2.5. That means that high
solids concentration, whether local or cross-sectional average, is statistically good
for cluster expansion, especially in the near-wall region. However, as size is defined
and used differently, data in these figures actually contain information of different
radial positions and do not truly reflect the internal relation between dcl * εs. This
results in disagreed understanding on dcl and also proves that the cluster size model
is not logical, accurate mathematical model either. Of course, the restriction of this
model to specific experimental conditions also makes it hardly condition universal.

Fig. 2.6 Cluster size versus 4

10
Dimensionless cluster size dcl /dp (-)

average cross-sectional
particle concentration
3
10

2
10

1
10

0
10
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Cross-section averaged solids concentration s,av (-)
44 2 Analysis on the EMMS Theory

One model worthy of notice is the equivalent diameter model based on theo-
retical analysis [12]. It assumes that a bubble exists at the tail of each cluster in a
gas–solid fluidization system, and the volumetric ratio of them equals to the ratio of
the dense phase volumetric fraction to dilute phase volumetric fraction,

dcl3 f
¼ ð2:18Þ
dv3 1  f

where, dv is the bubble diameter.

Considering that cluster size must be no smaller than single particle size, this
formula is written as,
 1=3
f
dcl ¼ dv þ dp ð2:19Þ
1f

Assuming particles are all in clusters and no particle is contained in the bubble,
there is,

es ¼ f esc ð2:20Þ
 1=3
es
dcl ¼ dv þ dp ð2:21Þ
esc  es

Then, introducing the bubble diameter formula, the equivalent diameter formula
for clusters is obtained,
 1=3  1
es 2u2t u2t
dcl ¼ 1þ þdp ð2:22Þ
esc  es g 0:1225ðgDÞ

where D is the fluidized bed riser diameter and ut is the single particle terminal
velocity.
This model connects dcl to local solids concentration εs, cluster density εsc,
particle properties, and fluidized bed size. Technically, it should be able to reflect
physical realities. However, as shown in Fig. 2.7, the model curve (assuming εsc is
constant and equals to εsmf) actually gives similar results as other models.
Obviously, there are two loopholes in this model. One is that Eq. (2.20) assumes a
linear relation, whereas in reality, it is impossible for all particles to come into
clusters. The other is that, if Eq. (2.20) applies, the bubble volume should vary with
flow. Yet, the dv formula introduced by Eq. (2.22) only relates to particle terminal
velocity and fluidized bed pipe diameter, and constitutes a basically constant
parameter. These loopholes decide that dcl can only be inversely related to
ðesc  es Þ1=3 and present monotonous variation.
Analyzing this result from the equivalence between cluster and flow heteroge-
neity, there will be a conflict. As well known, a larger cluster size signifies a higher
2.3 Cluster Size and Its Effects on Drag 45

Fig. 2.7 Resulting equivalent 4

10
cluster diameter
(ρg = 1.144 kg/m3,

Dimensionless cluster size

μg = 1.848 × 10−5 Pa s, 3
10
ρp = 1714 kg/m3, dp = 76 μm,
εsmf = 0.6, D = 305 mm)

dcl/dp (-)
2
10

1
10

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentraiton εs (-)

level of flow heterogeneity. When flow tends toward dilute, this is understandable.
But when flow tends toward dense end, the flow should tend to be uniform, and the
solids concentration at different spatial points should tend toward consistent.
Obviously, the results in Figs. 2.5, 2.6, 2.7 disagree with this physical analysis.
Furthermore, from the perspective of drag, as can be seen from Fig. 1.2, drag
function gradually returns to a uniform state value with increasing solids concen-
tration. This suggests that mesoscale effect at extremely large solids concentration
should gradually disappear and no cluster exists. Hence, the cluster size should be
equal to the particle size. Then why the fact that very large cluster size is read in the
high solids concentration in experiments? The only explanation is that the so-called
“size” measured in experiment is actually the vertical length between the mea-
surement discontinuity points.
If we directly apply this model to numerical simulation without analysis or
judgment, we will discover that the cluster size is not only far larger than the grid
size, but also even exceed the device size. In such case, if we try to correct drag
with such a size parameter, we will find ourselves troubled by these conflicts.
Based on the analysis above, it is believed that the essential source of all these
conflicts is the fact that, up to date, the variation of cluster size dcl in a dense region
has never been clarified. As such, it is necessary to further examine the stress
condition during particle clustering.
First, the accurate cluster definition in its traditional sense must be identified.
According to the definition in Horio et al., a cluster is the particle collection formed
under gas flow, represented by spatial heterogeneous distribution of solids con-
centration [13, 14]. Once a cluster takes form, it will immediately change the local
flow condition and result in resistance to gas flow. Gas naturally selects to flow
toward a more dilute area, i.e., where the resistance is smaller, around the cluster,
rather than through the cluster. This leads to a velocity slip that is much larger than
single particle velocity between the falling particles and the gas flow. In other
words, the drag on the cluster as a whole is far smaller than that on single particles,
46 2 Analysis on the EMMS Theory

what we call drag reduction. If we take clusters as a loose large particle, it will be
easier to understand this phenomenon.
Second, the formation of clusters relies on particle and gas flow conditions. Only
when there are a sufficient number of particles (εs) in space with room for free
movement (free path), can cluster forms under gas flow entrainment. The maximum
level of this cluster formation is the accumulation state, which is generally the
minimum fluidization solids concentration εsmf.
Furthermore, during particles collection, intensive collision and friction takes
place between particles, between particles and clusters, and between clusters, which
also cause clusters to break up and even dissipate the next moment. Hence, in
fluidization process, forces boosting clusters (gas flow) and degrading clusters
(collision and friction) both exist.
Obviously, in dense flows (εs > 0.2), as solids concentration increases, the free
path gradually reduces and the collision effect degrades while the friction effect
gradually intensifies and becomes a constraint on flow and an obstacle to particle
movement. The gradual increase of flow resistance continues to degrade the driving
effect of gas flow. In this way, the cluster is impeded and gradually degrades and
disappears. In other words, the cluster size should tend to reduce. When εs is close
to εsmf, there will be no room for particles to move freely, and it will be impossible
for them to come into clusters. This results in uniform flow typically in the form of
particle flow. As a result, the cluster size dcl should eventually tends toward single
particle size dp. The conflicts mentioned above are readily solved.
The reason why experiments gave those unreasonable results lies in the incon-
sistent definition of cluster size. As discussed above, in the near-wall region of
fluidized bed, almost continuous particle descending flow is formed. The mea-
surement taken at this time is simply not the cluster size in its true sense. In reality,
in the vertical flow of fluidization, similar to single particle deposition, the key size
deciding cluster movement is not its axial length, but rather, its cross-sectional size.
As solids concentration increases, the cluster is already broken into a very small
size by friction. This explains why results from the models are physically incon-
sistent. They can neither help to achieve a deeper understanding of mesoscale effect,
nor be used as a mathematical model to correct drag.

2.3.1.2 Correction Model Based on EMMS Theory

Look back on the previously mentioned O-S drag model [15]. In Eq. (1.7), the
second term of heterogeneous coefficient fc is precisely the heterogeneous correc-
tion term. The remaining part other than empirical coefficient C3 is actually a
unimodal function. It is right that this term decides the unimodal profile of the
heterogeneous drag curve in O-S model.
As analysis stated above, the cluster size given by the latest QL-EMMS drag
model (Eq. (2.9) and Fig. 2.2a) does not conform to physical understanding. Hence,
Eq. (2.9) is corrected with the unimodal function form in fc, i.e., Eq. (2.23), and
obtained Eq. (2.24) [1].
2.3 Cluster Size and Its Effects on Drag 47

2 2
fcl ¼ Re es e0:005ðRe5Þ 90ð0:08es Þ ð2:23Þ
8 h  i 9
emf Up
< f 1e
Up
 Umf þ 1e g =
dcl ¼
cl
   þ 1  dp ð2:24Þ
max mf

:fcl;max N qp  U þ emf Up g ;
st q q mf 1emf p g

The cluster sizes before and after model correction are shown in Fig. 2.8.
As shown in Fig. 2.8, the corrected cluster size also presents a unimodal profile.
The peak value is moved from the original εs = 0.45 to nearby 0.1. The value is
slightly descended. At the εs → 0 dilute end and εs → 0.6 dense ends, dcl tends
toward single particle diameter, dcl/dp → 1. This conforms to the physical reality
and our previous analysis.
Further studies are still needed to find out what difference this correction will
make to drag. The result is given below.
To ensure the comparability, the condition for subsequent calculations is unified
as, ρg = 1.144 kg/m3, μg = 1.848 × 10−5 Pa s, ρp = 1714 kg/m3, dp = 76 μm,
εsmf = 0.6, Ug = 3.7 m/s, Gs = 98 kg/m2 s.

2.3.2 Effects on Drag

Figure 2.9 shows that, both cluster density εsc and drag function β significantly
changed, after introducing the corrected cluster size into the QL-EMMS drag
model.
The overall shape of cluster density εsc stays the same, except that the deflection
point moves forward from the original εs = 0.45 to nearby 0.3. Some important
changes take place in drag functions.

Fig. 2.8 Cluster size before 200

and after model correction A
Dimensionless cluster size

(ρg = 1.144 kg/m3,

160 C
μg = 1.848 × 10−5 Pa s,
ρp = 1714 kg/m3, dp = 76 μm, Origion
εsmf = 0.6, Ug = 3.7 m/s, 120
dcl/dp (-)

Gs = 98 kg/m2 s)
80
Corrected
40
B
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentraiton εs (-)
48 2 Analysis on the EMMS Theory

7
0.6 10

Drag function β (Ns/m )

4
Cluster density εsc (-)

6
0.5 10
5
0.4 10 Uniform
4
0.3 10
0.2 Before dcl correction 3
10
After dcl correction Before dclcorrection
2
0.1 10
O-S After dclcorrection
0.0 10
1

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentraiton εs (-) Solids concentraiton εs (-)

Fig. 2.9 Effects of cluster size correction. a Cluster density. b Drag function

(1) The overall shape presents concave function with characteristics similar to O-S
model, and the minimum value also appears nearby the solids concentrate
where the maximum dcl is located, suggesting that the corrected cluster size
distribution does reflect the characteristics of heterogeneous flow.
(2) Drag function rises about two orders of magnitude within εs = 0.2–0.6, which
is obviously the reflection of the divergence before and after dcl within the
same range. But the drag function is larger than that of uniform flow, which
has no physical significance.
(3) The deflection point of the drag function curve before and after correction
corresponds in position to the dcl curve. That is respectively the maximum
positions of the dcl curve before correction (point A in Fig. 2.8), and the
transition from a unimodal curve toward a horizontal one after correction
(point B). After point B, dcl tends toward dp, implying that cluster gradually
disappears and flow tends toward uniform. Drag function also presents this
change. That is, like the O-S curve, its ascending tendency gradually disap-
pears and tends toward gentle after the deflection point. Besides, the convex
and concave character of the dcl curve at points A and B are precisely inverted
in the drag function curve into concave and convex character. Further
observation on the curve within εs = 0–0.3 also discovers this “specular”
correspondence. Before correction, dcl curve is close to a straight linear and
the drag curve is similar. After correction, it becomes a convex function at
εs = 0.13 (point C), and at the same position the drag curve becomes a concave
function. This is right the change we discussed in article (1).
(4) From another perspective, the drag curve after the dcl correction looks very much
like the result of moving the O-S model curve laterally upwards. We cannot help
to wonder if this lateral movement implies that the absolute cluster size after
correction is too small, and this explains why the drag reduction is not as sig-
nificant as the O-S curve. Further, to remove the nonphysical character of the
curve after εs > 0.3 beyond the uniform value, what we need to do seems to be
simply cause the gentle part to overlap the uniform curve. This reminds us the
necessity to further examine what difference the absolute dcl value makes to drag.
2.3 Cluster Size and Its Effects on Drag 49

(a)
Dimensionless cluster size
(b)
4
3x10 0.6

0.5

Cluster density εsc (-)

×100
4
0.4 x1
dcl/dp (-)

2x10
0.3 x10
4 x100
1x10 0.2

×10 0.1

0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentraiton εs (-) Solids concentraiton εs (-)

(c) 7
10
Drag function β (Ns/m )
4

6
10
x1
10
5 Uniform state
4
10 x10
3
10 x100
2
10
O-S
1
10
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentraiton εs (-)

Fig. 2.10 Effects of increasing cluster size. a Dimensionless cluster size. b Cluster density. c Drag
function

Thus, we directly multiply dcl Eq. (2.24) with 10 and 100 and keep the size
distribution unchanged. The results are shown in Fig. 2.10.
Figure 2.10 shows that, when cluster size is amplified, both cluster density and
drag function reduce, but no “lateral movement” appears as expected. Instead, the
drag function curve is gradually distorted, gradually becoming more divergent from
the O-S curve, and appears to be less sensitive. At εs = 0.1, though dcl is already 104
times of dp, the β value is still one order of magnitude higher than O-S model. After
εs > 0.3, the curve still exceeds the uniform flow values, except that the deflection
point has slightly moved toward the dense region. This is because dcl has remained
close to dp within this range.
Therefore, unlike what was previously assumed that dcl is the principal and only
factor contributor to mesoscale effect and drag reduction, relying on it alone is not
enough to provide accurate heterogeneous drag. It can be inferred that there is
definitely another factor that makes a substantial difference to drag. Hence, it is
necessary to examine the contribution made by the other cluster characteristic
parameter, the cluster density.
50 2 Analysis on the EMMS Theory

2.4 Cluster Density and Its Effects on Drag

The EMMS theory provides no formula that solves εsc directly, but uses the “tra-
verse” algorithm and minimum energy constraint. Therefore, it is in an “enslaved”
driven position. This also mirrors the deficiencies of current research and the lack of
understanding of this problem.
As shown in Figs. 2.2, 2.9, and 2.10, the variation of εsc with solids concen-
tration is virtually sublinear monotonous rise, followed by a deflection, before
stabilizing at εsmf. Some author directly assumed that it constantly equals to the
minimum fluidization solids concentration, εsc = εsmf, which cannot be supported by
substantial research basis [16].
In view of this, it is necessary to study the variation of cluster density with solids
concentration and its effects on drag.

2.4.1 Empirical Formula for Cluster Density

The following equations are empirical cluster density formulas from regressing
experimental data. The only difference is that Eq. (2.25) uses the cross-sectional
average solids concentration εs,av [11], while the other two use the local solids
concentration εs [17, 18].

0:58e1:48
s;av
esc ¼ 0\es;av \0:55 ð2:25Þ
0:013 þ e1:48
s;av

0:385
esc ¼ 0:389  0\es \0:35 ð2:26Þ
1 þ 66:451e1:536
s

0:595e1:143
esc ¼ s
0\es \0:4 ð2:27Þ
0:165 þ e1:143
s

Due to experimental and measurement limitations, no experimental data is

available in dense regions where solids concentration is large. All these empirical
formulas only apply to a limited range of solids concentration.
Figure 2.11 compares the cluster densities from the empirical formulas and from
the QL-EMMS model. All the three formulas provide similar nonlinear monotonous
rise with smooth curve. This again indicates that the straight line plus deflection
point given by the QL-EMMS model is not reasonable.
Besides, the divergence among the three formulas shows close correlation
between εsc and the experimental condition. That is to say, flows under different
conditions have different levels of heterogeneity and result in different mesoscale
structures. As a characteristic parameter of mesoscale structure, εsc will definitely
2.4 Cluster Density and Its Effects on Drag 51

Fig. 2.11 Comparision of 0.6

cluster density formula and
Eq.(2.25)
QL-EMMS model result 0.5
QL-EMMS

Cluster density εsc (-)

0.4
Eq.(2.27)
0.3

0.2 Eq.(2.26)

0.1

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentraion εs, εs,av(-)

change with it. Thus, cluster density should be condition-dependent which cannot
be represented by the EMMS theory (Fig. 2.2b).

2.4.2 Effects on Drag

Of these three empirical formulas discussed above, the latter two agree with the
definition in the EMMS theory. That is, they describe the relation between cluster
density and local solids concentration. Hence, we can introduce them into the
QL-EMMS drag model for calculation and analysis.
Figure 2.12 presents the resulting drag functions under different cluster densities,
including the cluster densities from empirical formulas (2.26), (2.27) and from the
original “traverse” method. In all these results, cluster size is calculated on the
correction model based on the EMMS theory Eq. (2.24).
In Fig. 2.12, after introducing the empirical formula, the resulting drag function
immediately presents the essential characteristics of the O-S curve. This is because
the curves of εsc empirical formulas present a convex function shape. However,
drag function has no solution in the εs > 0.3 dense region, since empirical formulas
only apply to a limited range of solids concentration.
Comparing the two εsc empirical curves, the εsc from Eq. (2.26) is marginally
larger while β is significantly smaller, suggesting that drag function is inversely
related to cluster density. Again at εs = 0.1, εsc has increased by approximately
66 %, and this causes β to reduce by one order of magnitude, suggesting that drag
function is much more sensitive to cluster density than it is to cluster size dcl
(compared with Fig. 2.10).
Comparing the εsc from empirical formulas and from “traverse” method, the
reduction of εsc at εs = 0.3 causes the drag function move downward substantially.
This is unachievable by changing dcl as discussed above. Now, looking back on the
52 2 Analysis on the EMMS Theory

(a) (b)
0.6
6 εsc-traverse
Traverse 10

Drag function β (Ns/m )

Cluster density εsc (-)

4
0.5
5
10 Uniform
0.4
Eq.(2.26) 10
4

0.3
εsc-Eq.(2.27)
Eq. (2.27) 3
0.2 10
2
εsc- Eq.(2.26)
0.1 10
1
O-S
0.0 10
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentraiton εs (-) Solids concentraiton εs (-)

Fig. 2.12 QL-EMMS model results using various cluster density model. a Cluster density. b Drag
function

downward movement of the drag curve in Fig. 2.10c, one will immediately realize
that this is not the direct result of changing dcl, but rather, the indirect result of
reducing cluster density (Fig. 2.10b).
Analysis above confirms again that the cluster density makes essential difference
to drag function. Thus, it is necessary to establish a logical, accurate cluster density
model, and introduce it into the EMMS theory as a new equation. This will refine
this theory and obtain a drag function that qualitatively and quantitatively agrees
with the O-S curve.

2.5 Summary

(1) Through analysis on the QL-EMMS drag model, revealed defects of the
EMMS theory in describing cluster characteristics as its equations are not
closed. It is identified that the direction for further refining this theory is to
establish a new mesoscale (cluster) theoretical model.
(2) Clarified the fuzzy understanding on cluster variation in dense gas–solid
two-phase flow and clearly pointed out that, cluster will break up and even-
tually return to single particle state at extremely dense region. It is identified
that the cluster size is unimodally distributed and tends toward single particle
size at the dilute and dense ends.
(3) Discovered that drag function is inversely related to cluster size, while it is
hardly possible to improve the drag function result simply by changing cluster
size.
(4) Cluster size makes a great difference to and is inversely related to drag
function. It is necessary to establish a logical, accurate cluster size density to
replace the “traverse” solution pattern in the EMMS theory.
References 53

References

1. Li F (2009) Investigations on the turbulent gas–solid two-phase interactions in fluidized

desulfurization process. Doctoral dissertation, Tsinghua University, Beijing
2. Li F, Chen C, Wang J, Qi H (2011) QL-EMMS drag model and its revision for fluidized dense
gas–solid two-phase flow. J Eng Thermophys 1:75–79
3. Li J (1987) Multiscale model for two-phase flow and energy minimum method. Doctoral, The
Chinese Academy of Sciences, Beijing (in Chinese)
4. Matsen JM (1982) Mechanisms of choking and entrainment. Powder Technol 32(1):21–33
5. Li JH, Cheng C, Zhang Z et al (1999) The EMMS model-its application, development and
updated concepts. Chem Eng Sci 54(22):5409–5425
6. Xu GW, Li JH (1998) Analytical solution of the energy-minimization multi-scale model for
gas–solid two-phase flow. Chem Eng Sci 53(7):1349–1366
7. Wang W, Li JH (2007) Simulation of gas–solid two-phase flow by a multi-scale CFD
approach—extension of the EMMS model to the sub-grid level. Chem Eng Sci 62(1–2):
208–231
8. Bin Zou, Hongzhong Li, Yashen Xia, Kwauk Mooson (1993) Statistic model for cluster size
distribution in fast fluidized bed. Eng Chem Metall 14(1):36–42
9. Gu WK, Chen JC (1998) A model for solid concentration in circulating fluidized beds. In:
Fluidization IX, Durango, Colorado, 1998, pp 501–508
10. Gu WK (1999) Diameter of catalyst clusters in FCC. AIChE Symp Ser 95(321):42–47
11. Harris AT, Davidson JF, Thorpe RB (2002) The prediction of particle cluster properties in the
near wall region of a vertical riser. Powder Technol 127(2):128–143
12. Subbarao D (2010) A model for cluster size in risers. Powder Technol 199(1):48–54
13. Qi HY (1997) Euler/Euler Simulation der Fluiddynamik Zirkulierender Wirbelschichten.
Aachen, Germany: Verlag Mainz, Wissenschaftsverlag. ISBN 3-89653-224-3
14. Horio M, Kuroki H (1994) Three-dimensional flow visualization of dilute dispersed solids in
bubbling and circulating fluidized beds. Chem Eng Sci 49(15):2413–2421
15. O’Brien TJ, Syamlal M (1993) Particle cluster effects in the numerical simulation of a
circulating fluidized bed. In: Avidan AA (ed) Preprint volume for CFB-IV. AIChE, New York,
pp 430–435
16. Naren PR, Lali AM, Ranade VV (2007) Evaluating EMMS model for simulating high solid
flux risers. Chem Eng Res Des 85(A8):1188–1202
17. Manyele SV, Parssinen JH, Zhu JX (2002) Characterizing particle aggregates in a high-density
and high-flux CFB riser. Chem Eng J 88(1–3):151–161
18. Yang TY, Leu LP (2009) Multiresolution analysis on identification and dynamics of clusters in
a circulating fluidized bed. AIChE J 55(3):612–629
Chapter 3
Cluster Model and Heterogeneous
Drag Model

3.1 General

Given that cluster density is the principal mesoscale determinant for drag function,
this chapter will establish a new cluster density model according to understandings
on flow heterogeneity. The cluster density curve features and their effects on drag
function will be found out. The original cluster size equations will be corrected to
agree with physical judgment both at the dilute and dense ends.
To ensure that the new model is scientific and accurate, the model is investigated
and validated by experiments on three scales. (a) Mesoscale. The existing experi-
mental results have already made the mesoscale cluster density validation possible.
(b) Grid-scale. The gird-average heterogeneous drag function developed out of
mesoscale models is validated with O-S drag results. (c) System-scale or
Macroscale. The numerical computational results are tested with experimental data
of gas–solid fluidized bed.
A heterogeneous drag model is established by introducing the new cluster
density model and corrected the cluster size equation into the QL-EMMS drag
model. The new model is named “QC-EMMS” with the initials of the main
researcher for distinguishment.

3.2 Local Heterogeneity Analysis

Previous analysis has shown that the heterogeneity of gas–solid two-phase flow is
reflected on mesoscale structures or clusters. These mesoscale structures cause
significant gas–solid slip velocity that is far larger than single particle terminal
velocity and thus lead to drag reduction. As a result, flow heterogeneity, Mesoscale
structure, gas–solid slip velocity, and drag reduction are actually incorporated.
Examining the pattern of slip velocity and drag reduction can help better to identify

© Springer-Verlag Berlin Heidelberg 2016 55

C. Chen, Investigations on Mesoscale Structure in Gas–Solid Fluidization
and Heterogeneous Drag Model, Springer Theses,
DOI 10.1007/978-3-662-48373-2_3
56 3 Cluster Model and Heterogeneous Drag Model

flow heterogeneity and the mesoscale characteristics. This will provide basis for
building a new cluster model.

3.2.1 Slip Velocity Versus Cluster Parameters

As well known, the slip velocity uslip of single particles in vertically ascending gas
flow equals to its terminal velocity Ut. This is the same in a uniform multiparticle
flow system. As shown in Fig. 3.1, single particle terminal velocity Ut increases
monotonously with particle size dp and density ρp. The larger the dp is, the more
sensitive Ut is to particle density. On the contrary, when dp < 0.1 mm, the effect of
ρp becomes very little.
In a heterogeneous multiparticle flow system, the mesoscale clusters causes the
particles inside the cluster to be much less exposed to gas flow entrainment than the
particles in uniform flow and fall at very high rate with the entire cluster, resulting
in very significant slip from the local average gas flow velocity. At this time, the
movement of the cluster looks like a large parcel that also has a terminal velocity
Ut,cl similar to single cluster and should also be subject to the variation like Fig. 3.1.
That is, Ut,cl is also positively related to cluster size and density. A large gas–solid
slip velocity means a denser and larger cluster, and vice versa.
As the cluster formation is closely related to local flow state, its terminal velocity
is not decided by its own size and weight alone like that of single particles. Here,
the term “cluster terminal velocity,” Ut,cl, is just used to help us understand the
movement behavior of clusters. Note that, Ut,cl is not completely the same as gas–
solid slip velocity uslip, since Ut,cl refers to a single cluster itself while uslip is the
gas–solid velocity difference when there are clusters in the flow field. The uslip
includes the time averages statistical results of single particles, clusters, and many
other conditions.

Fig. 3.1 Particle terminal

velocities Ut of various ρp =1490 - 2640 kg/m3 (Sankar et al.1986)
particle density and size [1–3] ρp =2470 - 2630 ρp=7620
16
Terminal velocity Ut (m/s)

14 Density ρp (kg/m3)
12 2200 (bed mater.)
mater )
1800 (lime)
10 1400 (coal)
8 700 (coke)

6
4
2 Haider et al.(1989)
0
0.1 1.0 5.0
Particle size dp (mm)
3.2 Local Heterogeneity Analysis 57

In view of this, studying how slip velocity relates to local flow parameters,
particularly solids concentration helps better understand local heterogeneity and
clusters.

3.2.2 Slip Velocity Versus Flow Parameters

3.2.2.1 Model Results

Some scholars regarded cluster as a large parcel and calculated cluster terminal
velocity by analogizing the formula of single particle terminal velocity. They
believed it to be the gas–solid slip velocity in heterogeneous flow and derived a
formula below [4],

1=2

uslip CD;cl dcl esc esc  es n1
¼ ð3:1Þ
ut CD;p dp esc

where slip velocity (cluster terminal velocity) and single particle terminal velocity
are,
!1=2
4 dcl gðqcl  qg Þ
uslip ¼ ð3:2Þ
3CD;cl qg

!1=2
4 dp gðqp  qg Þ
ut ¼ ð3:3Þ
3CD;p qg

432
CD;k ¼ 0\ Ark \ 36
Ark
56:9
¼ 0:4242 36 \ Ark \ 83750 ð3:4Þ
Ark
¼ 0:44 Ark [ 83750; k ¼ p; cl

n ¼ 4:6 0\Arcl \36

¼ 5:665  0:684 log Arcl 36 \ Arcl \ 83750 ð3:5Þ
¼ 2:3 Arcl [ 83750

where εsc is cluster density assumed to be constantly equal to εsmf, dcl is single
particle size calculated by Eq. (2.22), CD,k is single particle and cluster drag
coefficient, subscript k = p or cl, representing single particle or cluster, n is a
correction coefficient considering the contribution of other clusters.
58 3 Cluster Model and Heterogeneous Drag Model

Fig. 3.2 Dimensionless slip 3

ut (-)
velocity model result
(ρp = 2500 kg/m3,
dp = 500 μm,

Dimensionless slip velocity uslip/

ρg = 1.144 kg/m3,
2
μg = 1.848 × 10−5 Pa s)

0
0. 0 0. 1 0 .2 0 .3 0. 4 0 .5
Solids concentration εs (-)

Figure 3.2 shows the dimensionless slip velocity from the formula above, i.e.,
the slip velocity to terminal velocity ratio.
As shown in Fig. 3.2, the slip velocity curve is obviously a unimodal profile,
suggesting that flow heterogeneity is quite modest at the dilute and the dense ends,
and comes to its maximum at a certain solids concentration in the center.

3.2.2.2 Experimental Result

The axial and radial distribution of the respective velocity and volume fraction of
gas and solid phases can be obtained by point-to-point measurement using fiber or,
sampling tube or laser technology [5–7]. By subtracting the velocities of the two
phases, their slip velocity distribution can be derived as shown in Fig. 3.3.
In Fig. 3.3, local slip velocity first increases with radial position r/R, and then
reduces in the near-wall region, with its maximum located between r/R = 0.8–1.0,
which corresponds to the maximum particle fluctuation rate. Particle velocity
fluctuation comes to its maximum, suggesting the most intensive solid clustering. In
the near-wall region, the stagnation on the wall causes the reducing gas and solid
velocities. Consequently, the slip velocity becomes smaller near wall.
Nevertheless, it is impossible to identify the full essence of flow heterogeneity by
simply based on the radial distribution of flow parameters alone. Eliminating the
spatial coordinates and converting the original relation between flow parameters
and spatial coordinates into the relations between different parameters, a series of
important rules are immediately discovered [1]. Figure 3.4, for example, shows how
local slip velocity relates to solids concentration.
In Fig. 3.4, the uslip (εs) curve is a unimodal profile with its peak located between
εs = 0.1–0.15 regardless of the conditions. The slip velocity gradually reduces at
both the extremely dilute and dense ends. Thus, the peak value of the slip velocity
model curve in Fig. 3.2 is somewhat reasonable. However, the peak locates at a
3.2 Local Heterogeneity Analysis 59

Fig. 3.3 Radial distribution of gas–solid slip velocity (left) and particle fluctuation velocity (right)
under different conditions [8]

fairly small solids concentration. This is because of the model hypothesis that “all
the particles come into cluster phase while the bubble phase contains no particles,
εsc = εsmf”. It is clear that, describing cluster density logically and reliably will help
to increase model accuracy.
Combining Fig. 3.3 with Fig. 3.4, it is found that the uslip (εs) distribution curve
presents the pattern of flow heterogeneity. The peak of the curve corresponds to
the interface of the radial “ring-nucleus” structure inside fluidized bed. There, the
gas–solid phase interaction is quite intense and solids concentration is fairly high,
providing necessary conditions for the formation of mesoscale structure.
Meanwhile, between particles, there is also sufficient room for free movement and
can give rise to intense collision. Particle velocity fluctuation is also very intense.
From the peak toward the dilute ends, particles become rare and flow becomes
uniform. From the peak toward the dense ends, friction intensifies, gas flow effect
diminishes, clusters disappears, and flow also becomes uniform. This coincides
with what have been observed in the previous sections.

3.2.3 O-S Heterogeneous Drag Correction Factor

Flow heterogeneity not only causes high gas–solid slip velocity, but also great drag
reduction. Hence, studying drag reductions under different solids concentrations
also helps to investigate heterogeneity.
60 3 Cluster Model and Heterogeneous Drag Model

Fig. 3.4 Experimental result 7

of local slip velocity
(assuming Ar = 18, 6
I = 218 kg) Ug=4.62 m/s, h=2.325 m

Slip velocity uslip (m/s)

5

3 4.44/3.933

1
2.4/2.325
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)

The reliable O-S drag model (Eqs. 1.1–1.8) contains a uniform and a hetero-
geneous drag function [9]. Now, define heterogeneous drag correction factor Hd to
be the ratio between them,

b
Hd ¼ ð3:6Þ
b0

where β and β0 are the heterogeneous and uniform drag function, respectively.
Hence, Hd varies between 0 and 1, reflecting drag reduction caused by flow
heterogeneity.
In Fig. 3.5, the variation of local heterogeneity with local solids concentration
reflected by heterogeneous drag correction factor is the same as the local slip
velocity described above. Hd presents a unimodal distribution with solids concen-
tration and its minimum is also located between εs = 0.1–0.15. The minimum Hd
suggests the most intensive particle clustering, the highest flow heterogeneity, and
the maximum drag reduction. At the dilute and dense ends, Hd tends toward 1,
suggesting no drag reduction and thus the flow tends toward uniform there.
In summary, at the dilute (εs → 0) and dense (εs → 0.6) ends, the local slip
velocity tends toward single particle terminal velocity, and the heterogeneous drag
correction factor tends toward 1. That is, drag function equals to that under uniform
flow. Hence, flow tends toward uniform and clusters disappear. At a point in
“εs = 0.1–0.15” region, the flow heterogeneity is highest and particle clustering
becomes the most intensive.
3.3 Mesoscale Structure (Cluster) Model 61

Fig. 3.5 O-S drag correction 10

0

factor (ρg = 1.144 kg/m3,

μg = 1.848 × 10−5 Pa s,

factor Hd (-)
ρp = 1714 kg/m3, dp = 76 μm, -1
εsmf = 0.6) 10 (3.7, 98)

Drag correction
-2
10

t
-3
Ug=4.3 m/s
10
Gs=147 kg/m2s

-4
10
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)

3.3 Mesoscale Structure (Cluster) Model

Experimental measured cluster characteristic parameters are usually restricted to

specific experimental apparatuses and operating conditions. Also, limitations of the
measuring technique used also prevent us from obtaining results under high solids
concentrations.
Direct Numerical Simulation can predict cluster characteristic parameters but is
restricted by the computational amount. It only applies to very small number of
particles and cannot reflect results in a real fluidization system.
In the next section, a mesoscale structure (cluster) model is established on the
basis of theoretical analysis of local heterogeneity.

3.3.1 Cluster Density Model

This study first proposes a 45° straight line with the εsc as X-axis and εs as Y-axis,
in order to describe the relation between cluster density and local solids concen-
tration εsc (εs).
This line represents the borderline between uniform and heterogeneous flow
state. On the 45° straight line, εsc always equals to local solids concentration,
εsc ≡ εs. That is, no clusters exist since the solids concentration is the same ever-
where. The area under the line has no physical solution, since εsc (solids concen-
tration inside particle clusters) can never be smaller than local ambient
concentration. The area above the 45° straight line represents heterogeneous state,
since εsc is always larger than local ambient concentration, indicating particle
62 3 Cluster Model and Heterogeneous Drag Model

Fig. 3.6 Cluster density 0.6

versus solids concentration εsmf
0.5

Cluster density εsc (-)

0.4

0.3
Uniform state
0.2

0.1

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)

clustering. Therefore, the 45° straight line identifies the lower boundary of the εsc
curve.
Next, the maximum accumulation concentration line, the “εsc = εsmf level line” is
proposed. This level line identifies the upper limit reachable for εsc. That is, εsc can
never exceed the minimum fluidization concentration.
Most importantly, the unimodal distribution character of cluster density curve is
suggested. According to analysis on local heterogeneity above, at the dilute and
dense ends, flow tends toward uniform, so the cluster density curve tends toward
the 45° line representing uniform state. At an intermediate solids concentration,
there is definitely a maximum heterogeneous state, i.e., the peak position of cluster
density.
The resulting cluster density curve is shown in Fig. 3.6.
Assuming gas–solid flow reaches its maximum heterogeneous state at εs of
0.15, the cluster density reaches its maximum value of εsmf there. At the dilute end
(εs = 1−εmax = 0.0003) and the dense end (εs = εsmf = 0.6), flow reaches uniform,
and thus the cluster density curve returns to the 45° line. Therefore, a mathematical
expression for cluster density model can be builded,

30:35e2s
esc ¼ es þ 2 ðesmf  es Þ ð3:7Þ
e15:37ðes þ0:03Þ  0:96

Figure 3.7 validates the accuracy of the cluster density model with experimental
data [10]. The result of QL-EMMS model is also added for comparison.
In Fig. 3.7, within the region of solids concentration less than 0.3, the cluster
density model curve agrees well with the experimental data, showing an increase
followed by a reduction. However, the QL-EMMS model result presents a
monotonous increase and is closer to the uniform state, or the 45° line. This
signifies that the particle clustering degree predicted by the QL-EMMS model is
3.3 Mesoscale Structure (Cluster) Model 63

Fig. 3.7 Experimental 0.6

validation of cluster density
model 0.5

Cluster density εsc(-)

0.4
Eq.(3.7)
0.3
QL-EMMS
0.2
Uniform state
0.1

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)

weaker and the flow heterogeneity is smaller. This also explains why the drag
function of QL-EMMS model is larger than that of O-S model as shown in Fig. 2.3.
There lack experimental results when εs is larger than 0.3. The cluster density
model curve gradually tends toward the 45° line representing uniform state, which
agrees with our previous analysis on heterogeneity. The QL-EMMS model result,
on the other side, continues to increase until its sudden turning at εs = 0.45, which
lacks theoretical supports.
Additionally, as shown in Fig. 3.7, experimental data is obviously scattered,
suggesting that cluster density is not only related to solids concentration, but also is
affected by other factors such as operating conditions. Previous researches
neglected the effects of operating conditions and obtained empirical formulas by
fitting experimental data under a number of operating conditions, resulting in
considerable errors. Furthermore, since the cluster density of QL-EMMS model
cannot reflect the effects of condition changes either (Fig. 2.2b), the model’s
condition universality is poor. The variation of cluster properties with operating
conditions will be discussed in the next chapter.

3.3.2 Cluster Size Correction

As described in the previous chapter, people use the second term (fcl) of the het-
erogeneous correction coefficient (fc) in O-S drag model to correct the cluster size
equations in the EMMS theory. As fc1 is unimodal, the corrected cluster size can
automatically tends toward single particle size (uniform) at the dilute and dense
ends. However, this correction formula has no definite physical significance, but is
only a mathematical processing.
In view of this, this study defines a local heterogeneous index and corrected the
cluster size equation with this index. The index not only has definite physical
64 3 Cluster Model and Heterogeneous Drag Model

significance, but also can make the cluster size automatically to tend toward single
particle size at the dilute and dense ends.
Define local heterogeneous index Φ,
esc  es
U¼ ð3:8Þ
esmf  es

The basis for this definition is that, as cluster is composed of a large amount of
particles, the solids concentration in clusters εsc must always be larger than local
solids concentration εs, and its maximum can never exceed minimum fluidization
concentration εsmf.
According to this definition, Φ represents the deviation level of cluster density
εsc from local solids concentration. Φ varies between 0 and 1. The larger the Φ
value, the higher the level of deviation of εsc from local solids concentration, the
higher the level of particle clustering, and vice versa. Φ = 0 stands for the uniform
flow state where clusters disappear and εsc is precisely εs. Φ = 1 stands for the most
intensive particle clustering, εsc = εsmf. Therefore, Φ is actually the flow hetero-
geneity level at different local solids concentrations and is thus called the local
heterogeneous index here.
Based on cluster density model (Eqs. 3.7 and 3.8), the mathematical expression
of Φ is obtained,

Ae2s 30:35e2s
U¼ 2 ¼ 2 ð3:9Þ
eBðes þCÞ  D e15:37ðes þ0:03Þ  0:96

According to this formula, the correlation curve of local heterogeneous index with
local solids concentration is as shown in Fig. 3.8.
As shown in Fig. 3.8, the local heterogeneous index is a unimodal profile. At the
dilute and dense ends, flow tends toward uniform, and Φ tends toward 0. When the
solids concentration equals to 0.15, flow tends toward the most heterogeneous state

Fig. 3.8 Local heterogeneous 1.0

Local heterogeneous index Φ (-)

index versus solids

concentration
0.8

0.6

0.4

0.2

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)
3.3 Mesoscale Structure (Cluster) Model 65

Fig. 3.9 Corrected cluster 160

Dimensionless cluster size dcl/dp (-)

size (ρg = 1.144 kg/m3,
μg = 1.848 × 10−5 Pa s,
ρp = 1714 kg/m3, dp = 76 μm,
εsmf = 0.6, Ug = 3.7 m/s,
120
εsc-Traverse
Gs = 98 kg/m2s)

80

40
εsc model

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)

and Φ reaches its maximum. This conforms to our analysis result on flow hetero-
geneity in the previous section.
Take this unimodal local heterogeneous index Φ to correct the cluster size
equation in the EMMS theory,

dcl ¼ U  dcl;EMMS þ dp
8 h  i 9
emf Up ð3:10Þ
< Up
 Umf þ 1e
1emax g =
¼ U   þ1  dp
mf

: N qp  U þ emf Up g ;
st q q mf 1emf p g

As this formula contains suspended transport energy Nst and other flow parameters,
the cluster size can be obtained after conducting complete EMMS analysis. The
resulting cluster size is not a fixed curve, but closely related to cluster density.
As shown in Fig. 3.9, after the cluster size equation is corrected by the formula
above, whether cluster density is solved by the original “traverse” method or by the
cluster density model (Eq. 3.7), the cluster size out of the QL-EMMS model, and
presents a unimodal distribution. It tends toward single particle size (dcl/dp = 1) at
the dilute and dense ends, which agrees with the physical reality that “clusters
disappears and flow becomes uniform at the dilute and dense ends.”
Within the intermediate solids concentration, cluster size is affected to cluster
density and its peak position changes. However, as observed in Chap. 2, changes in
cluster size will not make great difference to drag function.
Equation (3.10) is merely a simple correction to the cluster size equation in the
EMMS theory. To obtain a more accurate cluster size equation, more efforts should
be employed in the understanding of the formation and evolution of mesoscale
structure. Further experiments should be conducted to observe and measure the
clusters’ characteristics and movement behavior.
66 3 Cluster Model and Heterogeneous Drag Model

3.4 Heterogeneous Drag Model

In the previous section, we proposed a cluster density model and a corrected cluster
size equation. In this section, a new drag model out of the QL-EMMS drag model,
named QC-EMMS model will be established, using the cluster density model as a
supplementary equation and the corrected cluster size equation as a substitute for
the original size equation.
QC-EMMS model is still a problem of constrained nonlinear programming and
contains ten variables, eight basic equations, and one definite condition. Hence, a
two-step solution the same to QL-EMMS model is also employed [11–13]. The
addition of a basic equation describing cluster density reduces the feasible domain
and increases the model accuracy.

3.4.1 Grid-Scale Validation of Drag Model

In the previous section, the accuracy of the cluster density model has been directly
validated with experimental data. That is, a “mesoscale” validation has been
conducted.
Next, whether the model gives an accurate description of the average drag
coefficient on grid, i.e., drag function, will be examined. This constitutes a
“grid-scale” validation of the model.
Figure 3.10 compares the QC-EMMS model with the reliable O-S drag model.
The uniform model and the previous QL-EMMS model are also involved for
comparison.
In Fig. 3.10, differently from the monotonous rise of the QL-EMMS model
curve, the QC-EMMS drag function shows a consistent curve with the O-S curve.

Fig. 3.10 Grid-scale test on 10

7

QC-EMMS drag model

(ρg = 1.144 kg/m3,
Drag function β (Ns/m )

6
10
4

μg = 1.848 × 10−5 Pa s,
ρp = 1714 kg/m3, dp = 76 μm, 5 Uniform state
εsmf = 0.6, Ug = 3.7 m/s, 10
Gs = 98 kg/m2s) 4
10 QL-EMMS
3
10
QC-EMMS
2
10
O-S
1
10
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)
3.4 Heterogeneous Drag Model 67

It is a concave curve featuring a reduction followed by an increase with increasing

solids concentration. The minimum drag function locates nearby the solids con-
centration of 0.1. This suggests that QC-EMMS model can successfully predict the
essence of heterogeneous drag, and also provides an indirect confirmation that the
cluster density proposed in the previous section is reasonable.
Despite the qualitative agreement of QC-EMMS with O-S model, the former is
quantitatively larger, attributable to the difference of slip velocity. The O-S drag
model assumes slip velocity to be constant, uslip = 1 m/s. QC-EMMS model deals
with slip velocity in a more accurate manner, uslip = Ug/(1 − εs) − Gs/(ρp · εs),
relating it to solids concentration and operating conditions.
In addition, the flow state is jointly decided by the average gas, solid velocity
(ug, us), and solids concentration (εs) for a local grid. Hence, drag function is not
only related to εs, but also closely related to (ug, us). Studies have indicated that drag
function is decided by the slip velocity (uslip = ug – us) and solids concentration, i.e.,
β = f (εs, uslip) [12]. Hence the relation of drag function with solids concentration is
actually not a curve but a curved surface comprising a number of curves each
corresponding to one different slip velocity.
Next, the functional form of drag model will be explained. That is the mathe-
matical expression relating drag function to slip velocity and solids concentration.

3.4.2 Functional Form of Drag Model

A heterogeneous drag function model should be represented by a complete math-

ematical formula with both theoretical significance and facilitate application.
Previous representation of an EMMS drag model was in the form of subpro-
grams (e.g. EMMS/matrix [11]), which uses and finds out all new drag values in
grid in each CFD time step until the end of computation. This makes sense only to
numerical simulation, yet its frequent use of subprograms greatly degrades the
computational efficiency.
This study decouples the EMMS subprogram analysis from the entire CFD
computation. A complete mathematical expression of the drag model is concluded
in advance. The heterogeneous drag function is given by multiplying the uniform
one with the heterogeneous correction factor Hd = f (εs, uslip). The mathematical
expression is directly put into the CFD program. This greatly simplifies the com-
putational process.
Here is the solving method of Hd. First, define the range of a real local grid flow
parameters, (uslip, εs). Then, work out the drag function value β and its ratio to
uniform drag function β0 under each group of (uslip, εs). At last, fit out the
expression of Hd (uslip, εs) according to the data matrix above.
Therefore, as long as CFD calculation gives grid parameters derived, the Hd is
directly derived and then multiplied by uniform drag function to get the hetero-
geneous drag function, which is then substituted into the momentum equation for
subsequent iteration.
68 3 Cluster Model and Heterogeneous Drag Model

Fig. 3.11 Drag correction

factor predicted by
QC-EMMS model
(ρg = 1.225 kg/m3,
μg = 1.789 × 10−5 Pa s,
ρp = 930 kg/m3, dp = 54 μm,
εsmf = 0.5, Ug = 1.52 m/s,
Gs = 14.3 kg/m2s)

For the operating condition of gas–solid fluidized bed to be calculated in the next
section, the relation of heterogeneous drag correction factor with local solids
concentration and slip velocity is worked out through QC-EMMS model analysis.
Figure 3.11 shows the result. Here, Reynolds number Re represents the slip velocity
uslip,

q g dp
Re ¼ uslip ð3:11Þ
lg

In Fig. 3.11, Hd is greatly affected by solids concentration. Hd tends toward 1 at

both the dilute and dense ends with extremely small or large solids concentration,
suggesting that flow is uniform and there is no drag reduction. Between solids
concentration of 0.1 and 0.3, Hd is quite small, suggesting high flow heterogeneity
level and large drag reduction rate.
Figure 3.12 gives the Hd(εs) and the Hd(Re) curve, so as to better clarify the
relation of drag correction factor with solids concentration and Reynolds number.
As shown in Fig. 3.12a, the Hd(εs) curve is similar under various slip velocities.
It is a unimodal profile that makes up a “concave” curve surface featuring a
reduction followed by an increase.
In Fig. 3.12, the Hd(Re) curve is almost the same under all solids concentrations
and present a relatively simple exponential distribution that can be expressed by an
exponential function,

Hd ¼ a  ðRe þ bÞc ð3:12Þ

where coefficients a, b, and c are all functions of solids concentration.

The correlation curves of a, b, and c with εs are obtained using Matlab pro-
gramming, as shown in Fig. 3.13.
3.4 Heterogeneous Drag Model 69

Fig. 3.12 Drag correction factor versus solids concentration and Reynolds number a drag
correction factor versus solids concentration b drag correction factor versus Reynolds number

Comparing Fig. 3.12a with Fig. 3.13a reveals that, the coefficient a presents the
relation of drag correction factor with solids concentration at a constant slip
velocity. As shown in Fig. 3.13b, the coefficient c is 0 both at the dilute and dense
ends, and coefficient b has no solution. At this time, the flow is almost uniform, and
the drag correction factor is just the coefficient a, and equals to 1.
Fitting the three curves in Fig. 3.13 and establishing the mathematical equations
of coefficients a, b, and c with the solids concentration, the mathematical expression
of drag correction function is obtained, as shown in Table 3.1. Note that, this
expression is for the definite operating condition in Fig. 3.11, the fluidized bed
conditions to be calculated in the next section. When operating condition changes,
another expression should be presented through solving QC-EMMS model.

(a) (b)
0
10 0.4

0.3
-1
10
b / c (-)
a (-)

0.2
c

-2 0.1
10
b
0.0

-3
10 -0.1
0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
εs (-) εs (-)

Fig. 3.13 Fitting coefficient versus solids concentration; a coefficient a b coefficients b and c
70 3 Cluster Model and Heterogeneous Drag Model

Table 3.1 Calculation formula of drag correction factor Hd

Hd ¼ a  ðRe þ bÞc


1 1
a ¼ 0:016 þ 0:984  þ
1þ 10ðes 0:015Þ189:5 1þ 10ð0:328es Þ29:7
b ¼ 0:006 þ 0:0917  ees =0:06266 þ 0:337  ees =0:01458
8
>
> 6:13  29:86  es þ 36:04  e2s 0:32  es  0:375
>
> 3:81 þ 68:52  es  426:97  e2 þ 1191:3  e3  1256:2  e4
< s s s 0:2  es \ 0:32
c ¼ 0:12 þ 4:8  es  48:93  e2s þ 239:2  e3s  434:4  e4s 0:015  es \ 0:2
>
>
>
> 0:18 þ 21:33  es 0:0078  es \ 0:015
:
0 es \ 0:0078 or es [ 0:375

In the CFD simulation of gas–solid flow in fluidized bed, programming the

calculation formula in Table 3.1 and writing it into User Defined Function (UDF),
QC-EMMS drag model can be easily incorporated into the Eulerian–Eulerian
two-fluid approach for calculation.

3.5 System-Scale Validation of Drag Model

Substitute the QC-EMMS heterogeneous drag model into the two-fluid approach,
and perform numerical simulation on gas–solid flow in Circulating Fluidized Bed
(CFB). Compare the result with experimental data to test the model accuracy. This
is the system-scale test of the model.

3.5.1 Geometric Model and Computational Conditions

Figure 3.14 shows the simplified two-dimensional (2D) geometric model of CFB
riser.
The gas–solid flow state of CFB is decided by three parameters. Operating gas
velocity Ug, solids mass flux Gs, and bed inventory Iinv. Given two of these
parameters, the other can be calculated. The computational accuracy is principally
decided by the drag model. Here, Ug and Iinv are given to find out Gs.
Before calculation, a certain amount of particles have accumulated on the bot-
tom. The initial solids accumulation height and concentration is respectively H0 and
εs0. Air is uniformly blown in from bottom at Ug, driving particles move upwards
and flowing out of the bed. There are two pressure outlets at both sides of the top
bed.
To ensure the constant particle inventory in bed, the outlet particles are returned
from time-to-time back into the bed from the return inlets at both sides of the
bottom. Returning velocity of particles are set to be the outlet flow rate divided by
3.5 System-Scale Validation of Drag Model 71

Fig. 3.14 CFB riser

140 350
Outlet

90

10500

Circulating solids
returning inlet
350 140
H0
h
x
Gas inlet

inlet area, particle density, and inlet particle concentration (εs,in = 0.3). This is
programmed and written by UDF.
Gas is exposed to slipless boundary conditions at the sidewall. For particles,
partial slip is assumed at the sidewall. The specular reflection coefficient is taken to
be 0.0001 [14]. Other settings are the same as the literature [15].
Table 3.2 lists the main computational parameters including CFB size, gas, and
particle properties. For condition 2, under the same operating gas velocity, three
different initial accumulation heights, i.e., bed inventories, are used to investigate
the ability of drag model to predict “choking” state.

3.5.2 Grid-Size Effects

With given gas velocity and bed inventory, particle circulating mass flux Gs is
calculated and compared with experimental data under various grid numbers. The
largest cell number is 16 times of the smallest one.
Table 3.3 indicates that, the computational accuracy of QC-EMMS model is
significantly higher. The results predicted by QC-EMMS model are always closer to
the experimental value than other two models under the same grid number.
The traditional uniform Gidaspow drag model greatly overestimates the particle
mass flux which is far larger than the experimental data. Comparing to Gidaspow
model, the heterogeneous drag model EMMS/matrix, which is also based on the
EMMS theory, can effectively improve computational accuracy. However, even
72 3 Cluster Model and Heterogeneous Drag Model

Table 3.2 Computational conditions

Parameter Condition 1 Condition 2
Operating gas velocity Ug (m/s−1) 1.52 2.1
Initial accumulation height H0(m) 3.09 2.04, 2.87, 3.23
Initial accumulation solids concentration εs0(-) 0.3
External particle circulating mass flux Gs (kg/m2s) To be calculated
Particle density ρp (kg/m3) 930
Particle diameter dp (μm) 54
Minimum fluidization solids concentration εsmf(-) 0.5
Gas density ρg (kg/m3) 1.225
Gas kinematic viscosity μg (Pa s) 1.789 × 10−5
Outlet pressure (Pa) 1.013 × 105
Gas-wall Slipless
Particle-wall Partial slip
Particle-wall specular reflection coefficient φ 0.0001
Particle elastic collision coefficient es 0.9
Time step Δt(s) 5 × 10−4

Table 3.3 Particle mass fluxes under different drag model and grid number
Grid number Experimental value Calculation results from different drag models Gs
(kg/m2s)
Gidaspow [15] EMMS/matrix [15] QC-EMMS
20 × 150 = 3000 14.3 161.31 73.77 22.5
40 × 300 = 12,000 167.16 28.04 18.6
60 × 450 = 25,000 169.00 19.73 17.7
80 × 600 = 48,000 173.33 19.04 17.1

under a fine grid 80 × 600, there is still a 33.15 % deviation. Comparatively,

QC-EMMS drag model is not only much more accurate than the uniform drag
model, but also is more accurate than EMMS/matrix model. For coarse grid
(20 × 150), for example, the computational error of the uniform model is as high as
1028.04 %, that of the EMMS/matrix model is 415.87 % while that of QC-EMMS
model is 57.3 %.
Figure 3.15 indicates that, QC-EMMS model is relatively grid-independent
comparing to EMMS/Matrix model. The QC-EMMS model results under different
gird numbers are not significantly divergent. However, the Gs value from
EMMS/matrix model reduces greatly with the increasing grid number. Only when
the grid is fine enough, does the computational result come to stabilize.
This leads to an important deduction, that an accurate mathematical model will
be grid-independent. Hence, a coarse grid can be employed for simulation to
3.5 System-Scale Validation of Drag Model 73

Fig. 3.15 Grid effects of 80

different drag models (under

Particle mas flux Gs (kg/m s)

condition 1)

2
EMMS/matrix
60

40
QC-EMMS

20

Experimental data

0
4 4 4 4 4
0 1x10 2x10 3x10 4x10 5x10
Grid number (-)

accelerate calculating process. On the contrary, an inaccurate model is more

grid-correlative and needs a quite fine grid for simulation.

3.5.3 Results Analysis

Simulation results under the same grid number (40 × 300 = 12,000) are discussed in
this section.

3.5.3.1 Distribution of Solids Concentration

Figure 3.16a shows the distribution of solids concentration in fluidized bed at

various times after flow stability. Obviously, particles are heterogeneously dis-
tributed. Many particles gather at the bed bottom, forming clusters with fairly high
local solids concentration. As time goes by, clusters keep forming and breaking
up. Particle-wall interaction gives rise to clustered matters and biggish flake-like
clusters near the wall. At the same time, particles separated from the tip of the
flaked matters and form belted matters. Experiments [16] have also observed such
phenomenon concurring with the particle clustering (Fig. 1.1).
Figure 3.16b shows the time-averaged solid concentration derived by the
instantaneous flow field data after flow stability (15–30 s). It is a typical hetero-
geneous distribution that is axially dilute at top, dense at bottom and radially dilute
in center, dense at side region. The radial heterogeneous structure is obvious in the
bottom region. In the upper bed region, the radial heterogeneous structure is also
detected when color reduction is narrowed.
74 3 Cluster Model and Heterogeneous Drag Model

(a) (b)

Fig. 3.16 Solids concentration distribution in CFB risers (under condition 1). a Instantaneous
results, b Time-averaged result (15–30 s)

Figure 3.17 shows the axial distribution curve of cross-sectional average solids
concentration. Comparing the computational results from EMMS/matrix model
[15] and from QC-EMMS model, the latter reproduces the S-shaped experimental
data. In the bottom dense region (h = 0–3 m), particles are densely accumulated
with higher concentration. Up into the intermediate transition region (h = 3–6 m),
solids concentration tapers. In the top dilute region (h > 6 m), solids concentration
is smaller and hardly varies with the height at all. The EMMS/matrix, fails to
present this S-shaped, although it gives the particle accumulation at the bottom.
From 4 m height upwards, solids concentration reduces linearly, and is much larger
than the experimental data. Thus, QC-EMMS drag model successfully predicts the
“S-shaped” distribution of axial solids concentration. This proves the model
accuracy.
3.5 System-Scale Validation of Drag Model 75

Fig. 3.17 Axial distribution

of solids concentration (under Experiment data
condition 1) 10 EMMS/matrix
QC-EMMS

Height h (m)
6

0
0.0 0.1 0.2 0.3
Cross-section averaged solids concentration εs,av (-)

Figure 3.18 gives the radial distribution of time-averaged solids concentration at

three heights. Radially, solids concentration at the center is smaller than that at the
sides. At height h = 2 m, the radial heterogeneous structure is quite significant since
the solids concentration at the side is even higher than 0.3. As height increases,
radial heterogeneity diminishes. This is because, while particles are carried upward
by gas flow, some of them fall down along the wall to the bed bottom. That is,
interparticle circulation takes place and causes the solids concentration larger at the
side of the bed bottom. Radial distribution of particle velocity which will be stated
later also indicates this interparticle circulation.

3.5.3.2 Radial Distribution of Particle Velocity and Slip Velocity

Figure 3.19a shows the radial distribution of particle velocity. In the bottom region
of CFB riser (h = 2, 3 m), particles at the bed center rise at a fairly high velocity
while, the particle velocity is smaller than 0, suggesting downward movement near
the wall. This indicates the existence of backmix and interparticle circulation. In the
upper part of CFB riser (h = 4 m), all the particles are carried upward by gas flow,
and the radial distribution of particle velocity is relatively uniform.
Figure 3.19b shows the radial distribution of local slip velocity derived by
subtracting gas and solid velocities at the same position. At all heights, the slip
76 3 Cluster Model and Heterogeneous Drag Model

Fig. 3.18 Radial distribution 0.4

of solids concentration (under

Solids concentration εs (-)

condition 1)
h=2 m
0.3

0.2

3m
0.1
4m
0.0
-1.0 -0.5 0.0 0.5 1.0
Radial position r/R (-)

velocity at bed center is higher than that at near-wall region. This indicates that
particles at bed center are more likely to be carried upward by gas flow. As height
increases, the slip velocity gradually diminishes, and the radial distribution curve
becomes milder. This is because particles speed under the action of gas flow and
then gradually come into the fully developed region at the top where the particle
velocity difference becomes very little.
Figures 3.18 and 3.19b show the spatial distributions of particle concentration
and slip velocity respectively. In order to describe the relation between these two
flow parameters, solids concentration is used as the X-axis and slip velocity at the
same axial and radial position is used as the Y-axis. After the coordinate trans-
formation, the relationship of slip velocity and solids concentration is obtained as
shown in Fig. 3.20.
As shown in Fig. 3.20, local slip velocity presents a unimodal profile with solids
concentration. The peak locates nearby the solids concentration of 0.15. From the
peak toward both sides, local slip velocity gradually diminishes. This concurs with
the experimental results in Fig. 3.4.
In Fig. 3.20, the slip velocity at the same solids concentration is obviously
dispersed. This suggests that, in addition to solids concentration, slip velocity is
also subject to other factors, which typically include bed height and radial position.
For the same local solids concentration, with the increase of bed height, the slip
velocity diminishes, especially in the fully developed region at the bed top.
3.5 System-Scale Validation of Drag Model 77

(a) (b)
8 2.5

Local slip velocity uslip (m/s)

Particle velocity us (m/s)

h=2 m h=2 m
4 2.0
3
1.5 3
0 4
1.0
-4
0.5 4
-8 0.0
-1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0
Radial position r/R (-) Radial position r/R (-)

Fig. 3.19 Radial distribution of velocity (under condition 1). a Particle velocity, b Gas–solid slip
velocity

Fig. 3.20 Local slip velocity 3.0

versus solids concentration
h=2 m
Local slip velocityuslip (m/s)

(under condition 1) 2.5

2<h≤3.8
3.8<h≤ 3.9
2.0
h>3.9
1.5

1.0

0.5

0.0
0.0 0.1 0.2 0.3 0.4
Solids concentration ε s (-)

3.5.4 Analysis on Local Heterogeneity

Instantaneous numerical simulation can monitor the variation of particle volume

concentration with time at different axial and radial positions in CFB. Figure 3.21
presents the instantaneous distribution of solids concentration within 10–30 s at
three radial positions on the cross section in the bottom dense region (h = 2 m),
transition region (h = 3.85 m), and dilute region (h = 4.2 m).
In the upper dilute region of CFB, the monitoring signals at all the three radial
positions always fluctuate nearby a very small solids concentration. Spikes appear
at very few moments where solids concentration increases abruptly and local par-
ticle clustering takes place. In the intermediate transition region, the monitoring
signals at all the three radial positions fluctuate widely and repeatedly. Especially at
78 3 Cluster Model and Heterogeneous Drag Model

r/R=0 r/R=0.5 r/R=1

0.5
εs=0.027 εs=0.017 εs=0.028
0.4
εs (-)

0.3

0.2
h=4.2 m
0.1

0.0
εs=0.12 εs=0.084 εs=0.18
0.4
εs (-)

0.3

0.2
h=3.85 m
0.1

0.0
εs=0.14 εs=0.24 εs=0.29
0.4
εs (-)

0.3

0.2
h=2 m
0.1

0.0
10 15 20 25 10 15 20 25 10 15 20 25 30
Time (s) Time (s) Time (s)

Fig. 3.21 Instantaneous change of local solids concentration (under condition 1)

bed side (r/R = 1), the fluctuation rate reaches its maximum and the maximum
solids concentration can be as large as 0.4. This suggests that dense clusters and rare
bubbles keep forming and breaking up, gas–solid is substantially mixed and gas–
solid interaction is quite intense. For the bottom dense region, at bed center
(r/R = 0), solids concentration fluctuates nearby 0.24. At the wall (r/R = 1), solids
concentration further increases, fluctuating nearby 0.29 and the fluctuating rate
reduces. That gives rise to a radially dilute-center, dense-side heterogeneous
structure.
Figure 3.22 shows the probability density distribution (PDD) from statistical
analysis of the time-series solids concentration in Fig. 3.21. The X-axis denotes
solids concentration coordinate. Y-axis denotes the number of occurrences of the
particle concentration, i.e., probability density.
On the top dilute region of CFB riser, the probability densities at all the three
radial positions are unimodal with the peak locating at small solids concentration.
This suggests a locally uniform single-phase structure which is particle dilute phase.
In the intermediate transition region, the probability densities at all the three radial
positions are bimodal, especially at the side (r/R = 1). This signifies a local het-
erogeneous structure where dense and dilute phases coexist. The characteristic
solids concentrations of dilute and dense phases are the solids concentrations
corresponding to the bimodal positions. At the side, for example, the dense
3.5 System-Scale Validation of Drag Model 79

r/R=0 r/R=0.5 r/R=1

250

200
PDD (-)

150

100
h=4.2 m
50

250

200
PDD (-)

150

100
h=3.85 m
50

250

200
PDD (-)

150

100
h=2 m
50

0
0.0 0.1 0.2 0.3 0.4 0.0
0.0
0.5 0.1 0.2 0.3 0.4 0.0
0.5
0.0 0.1 0.2 0.3 0.4 0.5
ε s (-) ε s (-) εs (-)

Fig. 3.22 Probability density distribution of solids concentration (under condition 1)

concentration is 0.35 while the dilute concentration is 0.024. For the bottom dense
region, the probability density is unimodal, and its peak locates at a large solids
concentration. This suggests local dense clusters as the principal form. From the
bed center to the sides, the solids concentration corresponding to the peak gradually
increases, suggesting that particle clustering becomes more significant and gives
rise to a radially dilute-center, dense-side heterogeneous structure.
The standard deviation σs of solids concentration, which describes the fluctuation
level with time, is often used to define flow domain transformation [17, 18]. It is
also used to describe local heterogeneity [19].
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
u 1 X N  2
rs ¼ t es;i  es ð3:13Þ
N  1 i¼1

1X N
es ¼ es;i ð3:14Þ
N i¼1

Here, εs,i is the instantaneous solids concentration; εs is the time-averaged solids

concentration; N is the number of samples.
80 3 Cluster Model and Heterogeneous Drag Model

When solids concentration fluctuates between the minimum fluidization con-

centration εsmf and the minimum 0, σs reaches its maximum,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rs;max ¼ es  ðesmf  es Þ ð3:15Þ

According to the instantaneous fluctuation of solids concentration, define local

heterogeneity γ to be the σs to σs,max ratio [19],
rs rs
c¼ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð3:16Þ
rs;max es  ðesmf  es Þ

Analyzing the instantaneous solids concentration fluctuation at different positions,

the time-averaged solids concentration and local heterogeneity are obtained. Using
the former as the X-axis and the latter as the Y-axis, the result is shown in Fig. 3.23.
Figure 3.23 indicates a unimodal local heterogeneity with the peak located
nearby local particle concentration of 0.15. From the peak toward both sides, local
heterogeneity diminishes and tends toward 0 at the εs → 0 dilute end and the
εs > 0.3 dense end. This agrees with the distribution of local slip velocity with solids
concentration shown in Fig. 3.20. This suggests that local slip velocity is a signature
of local heterogeneity, which concurs with our analysis in Sect. 3.2.
Nearby the solids concentration of 0.15, the instantaneous solids concentration
fluctuates widely. Dense clusters and dilute bubbles keep forming and breaking.
High flow heterogeneity leads to considerable local slip velocity. In the dilute space
where the local time-averaged solids concentration tends toward 0, the instanta-
neous solids concentration fluctuates modestly. Particles are almost uniformly
scattered. The local slip velocity is also fairly small. In the dense space where the
local time-averaged solids concentration is larger than 0.3, the instantaneous solids
concentration stabilizes at a fairly large value. The fluctuation rate is small. Large

Fig. 3.23 Local 0.8

heterogeneity (under h≤2 m
condition 1) 2<h≤3.8
Local heterogeneity γ (-)

0.6 3.8<h≤3.9
h>3.9

0.4

0.2

0.0
0.0 0.1 0.2 0.3 0.4
Solids concentration εs (-)
3.5 System-Scale Validation of Drag Model 81

number of particles exists uniformly. The slip velocity is hence also modest. The
relation of this instantaneous fluctuation with time-averaged solids concentration
can also be observed from Fig. 3.21.
In conclusion, QC-EMMS drag model can calculate the accurate particle cir-
culating flux, effectively minimize grid correlation, and successfully predict the
axial and radial heterogeneity of time-averaged flow parameters. The calculated
variation of local slip velocity and heterogeneity conforms to the experimental
results and our previous physical judgment.

3.5.5 Prediction of “Choking” State

Under a given operating gas velocity Ug, particle circulating mass flux Gs normally
increases with increasing bed inventory Iinv. However, when “choking” occurs,
under a given Ug, Gs does not change with Iinv, and the only change is the height of
the bottom dense region in bed. Whether the “choking” state in CFB is accurately
predicted is believed to be one of the criteria that measure the accuracy of a drag
model [20, 21].
CBF experiment (under condition 2) indicates that, under Ug = 2.1 m/s, when the
initial accumulation height increases from 2.04–3.23 m, the particle circulating
mass flux stays at a constant 24.1 kg/m2s, i.e., the CFB is at a “choking” state
[22, 23].
QC-EMMS model is used to simulate the gas–solid fluidization of the three bed
inventories, so as to test the model ability to predict this “choking” state.
Table 3.4 shows that, under the same operating gas velocity, particle mass flux
hardly changes with bed inventory. The computational error of QC-EMMS model
is persistently below 10 %.
Under the “choking” state, with increasing bed inventory, particle mass flux out
of bed virtually stays unchanged, while solids accumulation at bed bottom
increases. The axial solids concentration curve gradually changes from exponential
to an “S-shaped.”
The axial solids concentration distribution in Fig. 3.24 right presents this
“S-shaped” feature and the computational results agree well with the experimental
data. At the initial accumulation height of 2.04 m, it is an exponential distribution.
When the initial height increases to 2.87 m, it is “S-shaped.” When the initial height

Table 3.4 Resulting particle mass flux in circulation Gs (kg/m2s)

Initial accumulation height H0 Experiment Calculation Relative error
(m) value value (%)
2.04 24.1 22.6 −6.2
2.87 22.2 −7.8
3.23 22.7 −5.8
82 3 Cluster Model and Heterogeneous Drag Model

10 Experimental data 10 Experimental data

10 Experimental data
Calculation result Calculation result
Calculation result
8 8 H0=2.87 m 8 H0=3.23 m

h (m)
Height h (m)

Height h (m)
H0=2.04 m
6 6 6

Height
4 4 4

2 2 2

0 0 0
0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3
Cross-section averaged solids concentration ε (-) Cross-section averaged solids concentration εs (-) Cross-section averaged solids concentration ε
s,av (-)
s,av

Fig. 3.24 Axial distribution of solids concentration (under condition 2)

is further increased to 3.23 m, the height of the bottom dense region of this
“S-shaped” is increased. This verifies that QC-EMMS model can predict the
“choking” state accurately.

3.6 Summary

1. Through local slip velocity and O-S heterogeneous drag, the variation of local
flow heterogeneity with solids concentration εs is clarified. It is a unimodal
distribution, with the peak located at εs = 0.1–0.15, and the minimum located at
the εs → 0 dilute end and the εs → εsmf dense end.
2. Provide a heterogeneous region in the cluster density—local solids concentra-
tion correlation εsc(εs). The heterogeneous region corresponds to the range of
cluster density, locating between a 45° straight line signifying uniform state and
a line representing the maximum accumulation state. This 45° straight line
identifies the uniform-heterogeneous borderline.
3. Based on the understanding of local heterogeneity, the εsc(εs) curve profile is
determined. That is a unimodal distribution with solids concentration tending
toward the 45° line at the dilute and dense ends. A cluster density model on this
basis is established and verified by experiment. That is, the model passes the
“mesoscale” validation.
4. Based on the cluster density model, a local heterogeneous index describing the
level of particle clustering is defined. The cluster size equation in the EMMS
theory is corrected by this index so as to agree with physical realities.
5. Using the cluster density model and size correction equation, the basic equations
of the EMMS theory are supplemented and modified. QC-EMMS drag model is
established and agrees with the O-S model, reflecting the essence of heteroge-
neous drag reduction. That is, the model passes the “grid-scale” validation.
6. Mathematical expression of the QC-EMMS drag model is concluded. It enables
easy coupling with the two-fluid approach and simulation of gas–solid flow
3.6 Summary 83

processes in CFB. Comparing with uniform flow models and previous hetero-
geneous drag models, the new model has higher computational accuracy and
lower grid correlation. It can predict flow heterogeneity successfully and reflect
the unimodal distribution of local heterogeneity with solids concentration. It also
successfully predicts the “choking” state. That is, the model passes the
“system-scale” validation.

References

1. Qi HY (1997) Euler/Euler Simulation der Fluiddynamik Zirkulierender Wirbelschichten.

Aachen, Germany: Verlag Mainz, Wissenschaftsverlag. ISBN3-89653-224-3
2. Sankar SR, Smith TN (1986) Slip velocities in pneumatic transport-part I. Powder Technol
47:167–177
3. Wei W, Kun H, Bona L, Nan Z, Jinghai L (2013) Fluidized bed simulation:
structure-dependent multiscale CFD. CIESC J 64(1):95–106 (In Chinese)
4. Subbarao D (2010) A model for cluster size in risers. Powder Technol 199(1):48–54
5. Yang YL, Jin Y, Yu ZQ et al (1993) Local slip behavior in the circulating fluidized bed.
AIChE Symp Ser 89:81–90
6. Yang YL, Jin Y, Yu ZQ et al (1992) Investigation on slip velocity distributions in the riser of
dilute circulating fluidized-bed. Powder Technol 73(1):67–73
7. Tanner H, Li JH, Reh L (1994) Radial profiles of slip velocity between gas and solids in
circulating fluidized beds. AIChE Symp Ser 90(301):105–113
8. Yang Y (1991) Experimental and theoretical studies on hydrodynamics in cocurrent upward
and downward circulating fluidized Bes, Dissertation, Tsinghua University,Beijing, China (in
Chinese)
9. O’Brien TJ, Syamlal M (1993) Particle cluster effects in the numerical simulation of a
circulating fluidized bed. In: Avidan AA (ed) Preprint volume for CFB-IV. AIChE, New York,
pp 430–435
10. Harris AT, Davidson JF, Thorpe RB (2002) The prediction of particle cluster properties in the
near wall region of a vertical riser. Powder Technol 127(2):128–143
11. Wang W, Li JH (2007) Simulation of gas-solid two-phase flow by a multi-scale CFD approach
—extension of the EMMS model to the sub-grid level. Chem Eng Sci 62(1–2):208–231
12. Fei Li (2009) Investigations on the turbulent gas-solid two-phase interactions in fluidized
desulfurization process (Doctoral Dissertation). Tsinghua University, Beijing
13. Fei L, Cheng C, JinSheng W, Haiying Q (2011) QL-EMMS drag model and its revision for
fluidized dense gas-solid two-phase flow. J Eng Thermophys 1:75–79
14. Benyahia S (2012) Analysis of model parameters affecting the pressure profile in a circulating
fluidized bed. AIChE J 58(2):427–439
15. Lu BN, Wang W, Li JH (2009) Searching for a mesh-independent sub-grid model for CFD
simulation of gas-solid riser flows. Chem Eng Sci 64(15):3437–3447
16. Bi X, Zhu J, Jin Y, Yu Z (1993) Forms of particle aggregation in CFB. In: Proceedings of the
6th Chinese conference on fluidization. Huazhong University of Science and Technology,
Wuhan, pp 162–167 (In Chinese)
17. Bi HT, Grace JR (1995) Effect of measurement method on the velocities used to demarcate the
onset of turbulent fluidization. Chem Eng J Biochem Eng J 57(3):261–271
18. Zhu HY, Zhu J (2008) Characterization of fluidization behavior in the bottom region of CFB
risers. Chem Eng J 141(1–3):169–179
19. Brereton CMH, Grace JR (1993) Microstructural aspects of the behavior of circulating
fluidized beds. Chem Eng Sci 48(14):2565–2572
84 3 Cluster Model and Heterogeneous Drag Model

20. Ge W, Wang W, Yang N (2011) Meso-scale oriented simulation towards virtual process
engineering (VPE)-The EMMS Paradigm. Chem Eng Sci 66:4426–4458
21. Changjian C, Wei G (2008) Further Analysis on the choking criteria in the EMMS model.
Process Chem 4:620–624 (In Chinese)
22. Xu GW, Gao SQ (2003) Necessary parameters for specifying the hydrodynamics of
circulating fluidized bed risers-a review and reiteration. Powder Technol 137:63–76
23. Li J, Tung Y, Kwauk M (1988) Axial voidage profiles of fast fluidized beds in different
operating regions. In: Basu P, Large JF (eds) Circulating fluidized bed technology, vol. II.
Pergamon Press, Oxford, pp 193–203
Chapter 4
Condition Universality of Heterogeneous
Drag Model

4.1 General

Whether a mathematical model is theoretically valuable lies in whether it is uni-

versal. As far as the drag model describing heterogeneous gas–solid two-phase flow
is concerned, universality means its ability to provide accurate results for normal
particle types (Geldard A, B) under all operating conditions (i.e., “condition”,
typically decided by empty bed gas velocity Ug and particle mass flux Gs).
The O-S drag model used here as a test reference is based on experiment, but its
heterogeneous correction parameters only apply to two Gs values with no infor-
mation relating to Ug. Hence, its condition universality is uncertain. Studies have
indicated big errors when it is used for numerical simulation of fluidized bed,
despite the similarities of the Gs value of the simulation subject to that required by
the mode [1].
In the previous chapter, we established a cluster density and size model that
represents the maximum heterogeneity and a corresponding heterogeneous drag
model QC-EMMS. It simulates fluidized flow accurately and marks great
improvement from existing models. The only problem, however, is that its con-
dition universality is not solved.
For this reason, this chapter will continue to analyze flow heterogeneity so as to
examine the following problems. (a) Variation and quantitative representation of
system-scale flow heterogeneity with operating conditions. (b) Correlation of
mesoscale cluster model and drag model with overall (or system) heterogeneity.
(c) Condition universality testing of model.

© Springer-Verlag Berlin Heidelberg 2016 85

C. Chen, Investigations on Mesoscale Structure in Gas–Solid Fluidization
and Heterogeneous Drag Model, Springer Theses,
DOI 10.1007/978-3-662-48373-2_4
86 4 Condition Universality of Heterogeneous Drag Model

4.2 Evolution and Representation of Fluidization State

Gas–solid fluidization state is decided by a number of factors including operating

gas velocity Ug, particle mass flux Gs, bed inventory Iinv, gas–solid properties, and
the fluidized bed size, which contribute to a variety of gas–solid fluidized flow
patterns as shown in Fig. 4.1.
When Ug is increased, a particle fluid system generally experiences fixed bed,
particulate fluidization, bubbling fluidization, slugging, turbulent fluidization, fast
fluidization, and dilute phase pneumatic transport. Some of these patterns do not
necessarily appear subject to particle properties and the structure of the fluidized
bed. Under a very small Ug, gas flow is too weak to suspend particles and a fixed
bed state appears. When Ug equals to the minimum fluidization velocity Umf,
particle bed starts to expand uniformly leading to particulate fluidization. As Ug
continues to increase, the system will form a number of fluidization states, bubbling
fluidization characterized by “emulsion phase and dispersed bubbles of compacted
particles,” turbulent fluidization mixed with a lot of bubbles and clusters, fast
fluidization characterized by “dispersed clusters and compacted gas flow.” And
finally, reach the dilute pneumatic transport when Ug reaches its critical value Upt.
From fixed bed to turbulent fluidization, particles move up and down locally
under the action of gas flow and will not be carried away from the bed. This is often
termed “slow fluidization.” When entering fast fluidization, particle entrainment
increases significantly and, if no fresh particles are recharged, particles in bed will
soon be blown out. So a gas–solid separator and a downer are used to recover
particles carried out by gas flow and send them back into the riser to form a
circulating loop. Gs is an important parameter representing particle circulating net
flux.

Increase Ug

Fixed bed Particulate bed Bubbling bed Slugging Turbulent bed Fast bed Pneumatic transport

Fig. 4.1 Gas–solid fluidized flow pattern

4.2 Evolution and Representation of Fluidization State 87

Ug and Gs are counteractive operating parameters. The former signifies energy

transported into the fluidization system or suspended transport energy to particles.
The latter signifies the system energy consumed by particle flow. At a constant Ug,
Gs increases with the feeding rate. This inevitably increases the consumption of
suspended transport energy and causes the particle fluid system to transform from
single particles to clusters and emulsion phase, which is precisely opposite to
increasing Ug.
In order to relate observed flow patterns to operating conditions and allow
quantitative representation, people have proposed a variety of fluidized phase dia-
grams, such as the “pressure drop gradient—operating gas velocity” phase diagram,
“bed voidage—operating gas velocity” phase diagram, “Ug-Gs” phase diagram, and
the “Re*-Ar” phase diagram. But how to represent these many flow patterns with
consistent and relatively simple parameters, i.e., establish a general fluidized phase
diagram, has been a challenge to the fluidization research community.
Next section focuses on how to represent flow patterns in a quantitative manner
by analyzing the last two-phase diagrams.

4.2.1 “Ug-Gs” Fluidized Phase Diagram

So far the most commonly used fluidized phase diagrams divides the boundaries
between different flow patterns with “Ug-Gs” curve, using operating gas velocity as
the X-axis and particle circulating mass flux as the Y-axis.
Figure 4.2 shows a typical “Ug-Gs” fluidized phase diagrams. At the boundaries
between different flow patterns, there is a specific correlation between Ug and Gs.
However, different researchers have different understandings on this correlation,
and controversies typically focus on the understanding of the fast fluidization
boundary. That is, different definitions are given to critical velocities UTF and UFD.
As to the fast fluidization boundary, there are two main opinions so far. One
assumes that fast fluidization is between type A and type C choking [3, 4], and its
operating gas velocity is larger than significant entrainment velocity Use. The other
argues that, instead of simply using choking as the fast fluidization boundary [5],
pressure drop gradient or particle entrainment method must be used to determine
fluidization boundaries. The more frequently used method is pressure drop gradient,
which determines boundaries between different flow patterns by comparing the
variation of the pressure drop gradient in the upper and lower parts of bed with
operating gas velocity.
As shown in Fig. 4.3a, at a constant particle flux Gs = Gs1, when Ug is increased,
the pressure gradient in the lower part of bed dPL gradually diminishes. When the
pressure gradient in the upper part of bed dPU gradually increases, that means more
particles are carried up by gas flow. When the pressure gradient in the upper part
reaches its maximum, particles are being carried out of bed and fast fluidization
starts. The corresponding critical gas velocity is UTF. When UTF is further
increased, the average void age in the fast fluidized bed increases. The pressure
88 4 Condition Universality of Heterogeneous Drag Model

UFD
Gs ((kg/m2s))

UTF

UPT

ε av

Ug (m/s)

Fig. 4.2 “Ug-Gs” fluidization phase diagram (broken lines represent average bed voidage in fast
CFB, ρg = 1.21 kg/m3, μg = 1.81 × 10−5 Pa s, ρp = 1545 kg/m3, dp = 54 μm, D = 0.186 m)
Reproduced from Ref. [2] by permission of John Wiley & Sons Ltd

gradient in both the upper and lower part reduces, and the two gradients become
closer to each other. When the pressure gradient in the upper part precisely equals
to that in the lower part, this marks the start of dense pneumatic transport, and the
corresponding critical gas velocity is UFD. When Ug is increased again, the pressure
gradients in the upper and lower part remain equal to and reduce together. But when
dilute pneumatic transport starts, friction between the particle fluid system and the
bed wall increases significantly, causing the pressure gradient to rise again. The
minimum pressure gradient corresponds to gas velocity UPT, marking the boundary
between dense and dilute pneumatic transport.
Figure 4.3b shows that, the critical velocity at flow pattern boundaries is closely
related to particle mass flux. The fast fluidization region between UTF and UFD
reduces with Gs. When Gs is reduced to the initial transport flux Gstr, no fast
fluidization takes place anymore. As shown in Fig. 4.3c, at a constant Gs = Gstr, an
increased Ug leads to a direct change from slow fluidization to pneumatic transport.
Hence, for a constant particle circulating mass flux, fast fluidization boundary is
defined as Ug > Utr, Gs > Gtr, and UTF < Ug < UFD. Then, according to experimental
data, people find an empirical formula calculating Utr, UTF, and UFD as shown in
Eqs. (4.1)–(4.6) below.
Literature [6]:
!0:288  
UTF Gs D qp  qg D 0:69 0:2
pffiffiffiffiffiffi ¼ 1:463 Ret ð4:1Þ
gD lg qg dp
4.2 Evolution and Representation of Fluidization State 89

Fig. 4.3 Fluidization phase (a)

diagram expressed by
pressure drop gradient. dpL Gs=Gs1
Reproduced from Ref. [6] by
permission of John Wiley &

dp / dh
Sons Ltd

dpU

(b) Bubble/Turbulent Circulating fluidized bed

Fluidization UTF UFD UPT
Gs (kg/m2s)

Gs1

Dilute phase
Gstr transport

(c) Utr UPT

Gs=Gstr
dp / dh

Ug (m/s)

!0:442  
UFD Gs D qp  qg D 0:96 0:344
pffiffiffiffiffiffi ¼ 0:684 Ret
gD lg qg dp
where; ð4:2Þ
qg dp
Ret ¼ U
lg t
90 4 Condition Universality of Heterogeneous Drag Model

Literature [7]:
!0:345
Gs
ReTF ¼ 0:395 Ar0:572 ð4:3Þ
qg Ut
!0:359
Gs
ReFD ¼ 0:440 Ar0:563 ð4:4Þ
qg Ut

Literature [8]:
!0:03
Gs
ReTF ¼ 0:541 Ar 0:47
qg Ut
where; ð4:5Þ
!
Gs
63  Ar  927 and 2  254
qg Ut
!0:0043
Gs
ReFD ¼ 0:572 Ar 0:48
qg Ut
where; ð4:6Þ
!
Gs
63  Ar  10897 and 2  97
qg Ut

As discussed above, empirical formulas vary appreciably among different lit-

eratures and are closely related to fluid—particle properties (ρg, μg, Ar), particle
circulating mass flux (Gs), and fluidized bed size (D). In other words, the fluidized
state of a particle—fluid system must be jointly represented by a number of
parameters including Ug, Gs, Ar, and Ut. In order to represent different flow patterns
with relatively simple and consistent parameters, people further proposed a
dimensionless “Re*-Ar” phase diagram.

4.2.2 “Re*-Ar” Fluidization Phase Diagram

A phase diagram was established first using dimensionless particle size d*p as the X-
axis and dimensionless operating gas velocity U* as the Y-axis [9],
4.2 Evolution and Representation of Fluidization State 91

!1=3
qg ðqp  qg Þdp3 g
dp ¼ Ar 1=3
¼ ð4:7Þ
l2g
" #
q d p e G
U ¼
g g s
Ug   =Ar1=3 ð4:8Þ
lg 1  eg q p

The merit of this phase diagram is a broader range of application, since

dimensionless parameters can eliminate divergences among different gas/solid
properties and particle circulating mass fluxes.
Later, it was discovered that the use of Ar number, which represents the prop-
erties of both gas and solid, and Re* which reflects the overall velocity slip between
these two phases, helps better represent the flow pattern regions under different
properties [10],

qg ðqp  qg Þdp3 g
Ar ¼ ð4:9Þ
l2g
" #
q d p e G
Re ¼
g g s
Ug   ð4:10Þ
lg 1  eg qp

This gave birth to a “Re*-Ar” general fluidization phase diagram as shown in

Fig. 4.4.
Re*

Ar

Fig. 4.4 “Re*-Ar” fluidization phase diagram. Reprinted from Ref. [11]. Copyright 2011, with
permission from Elsevier
92 4 Condition Universality of Heterogeneous Drag Model

Figure 4.4 shows that, for Geldart type A particles as an example, at a constant
Ar number, as Re* increases, particulate fluidization, bubbling fluidization, slug-
ging, turbulent fluidization, fast fluidization, and pneumatic transport take place in
turn. For type B particles, no particulate fluidization and turbulent fluidization exist.
For type D particles, “plug flow” and slugging easily take place.
The “Re*-Ar” phase diagram, however, has some defects.
First, its division of slugging fluidization for Geldart type A particles is not so
logical. The typical indication of slugging is the presence of gas plugs, i.e., bubbles
gradually increase close to the bed area while they ascend. This phenomenon
generally does not happen except in a fairly slender (large length-to-diameter ratio)
fluidized bed or when coarse and heavy particles (Geldart type B or D) are used. It
rarely takes place for Geldart type A particles. Analysis of the statistical data on the
starting slugging velocity UMS and type C choking rate UCC also indicates that only
two experimental data are within the category of type A particles while the rest are
all type B or D particles. This suggests that, within Geldart type A particles, curves
MS and CC are not so reliable and the slenderness ratio of the fluidized bed system
has to be accounted for.
Second, it is not so reasonable to use curve AC representing the type A choking
rate as the borderline for fast fluidization. Figure 4.4 assumes that one of the
important indicators of fast fluidization is the existence of particle accumulation at
bed bottom that presents an “S-shaped” axial solids concentration profile that is
dilute at top and dense at bottom. Consequently, Type A choking rate UAC is used
as the initial velocity of fast fluidization. In reality, more and more researchers find
that “exponential” axial solids concentration distribution also falls under the fast
fluidization category [12]. Hence, as discussed in Fig. 4.3 of the previous subsec-
tion, the transformation rate from turbulent fluidization toward fast fluidization
should be the gas velocity when the pressure gradient in the upper part of the bed
dPU reaches its maximum. Type A choking rate divides fast fluidization into two
regions. One is S-shaped axial particle concentration distribution that is dilute at
top, dense at bottom appears in bed (UTF < Ug < UAC). The other is exponential
axial particle concentration that appears in (UAC < Ug < UMP).
Besides, the assumption that, at a constant Ar, Re* continuous to increase from
conventional slow fluidization to fast fluidization and to pneumatic transport, is
unreasonable. Re* denotes the overall gas–solid slip velocity. The overall slip
velocity is modest and approximates to single particle terminal deposition velocity
in a dilute pneumatic transport region [13]. Hence, from fast fluidization to dilute
pneumatic transport, Re* should reduce other than increase.
Despite these shortfalls, the “Re*-Ar” phase diagram offers an important clue for
representing fluidization conditions. At a constant gas–solid property, representing
Ar number, Re* can solely determine various flow states, or “conditions.”
As discussed in the previous chapter, local slip velocity—local heterogeneity—
mesoscale structure is a trinity. A given element is the key to identify the remaining
two. Analogously, on the system scale, there also exists such trinity, i.e., the overall
slip velocity can reflect the overall heterogeneity. Considering that, Re* represents
the overall slip velocity, “Re*-Ar” phase diagram can be used to represent operating
4.2 Evolution and Representation of Fluidization State 93

conditions in a quantitative manner. In this way, the overall heterogeneity can be

easily connected to operating conditions.

4.3 Overall Heterogeneity Analysis

From phase diagram analysis above, Re* number representing gas–solid slip
velocity can be used to determine the flow state under specific properties and
system conditions. This implies that overall interphase slip velocity is a major
feature of a fluidization system and corresponds to flow state, which is either
uniform or heterogeneous. Obviously, there should be a relationship between
overall slip velocity and the overall flow heterogeneity.

4.3.1 Overall Slip Velocity

Figure 4.5 gives a qualitative description of the relation between overall slip
velocity and bed average solids concentration and identifies different flow states.
This chart does not include slugging state and therefore only applies to horizontally
large fluidization systems dealing with Geldar A particles. Anyway, this can already
give a qualitative explanation.

Slip
velocity

Ut
Umf
Cross-sectional average solids concentration εsmf

Fig. 4.5 Overall slip velocity versus average solids concentration in bed. Reprinted from Ref.
[14]. Copyright 1979, with permission from Elsevier
94 4 Condition Universality of Heterogeneous Drag Model

As shown in Fig. 4.5, the system-scale interphase slip velocity also presents a
“unimodal” profile, similar to the relation between local slip velocity and local
solids concentration in Fig. 3.4. The overall slip velocity reaches its maximum in
the fast fluidization region, suggesting that, this is caused by the drag reduction in
connection with the existence of mesoscale clustering structure. Undoubtedly, this
is exactly a signature of flow heterogeneous.
From the fast fluidization region toward the pneumatic transport region at the left
or the slow fluidization region on the right, the overall slip velocity diminishes
gradually. This suggests degraded particle clustering and flow heterogeneity. In the
dilute pneumatic transport region where the bed average solids concentration is
modest, particles are uniformly carried up and transported out of bed, and the slip
velocity approximates to particle terminal velocity Ut. When the solids concen-
tration in bed reaches the minimum fluidization concentration εsmf, when particles
are just being blown up by gas flow, they are almost uniformly expanded and the
slip velocity is modest, i.e., it is the minimum fluidization velocity Umf. Hence, the
bed minimum overall slip velocity is Ut or Umf. Besides, the maximum slip velocity
is also related to particle circulating mass flux. At the same bed average solids
concentration, a larger Gs means a larger slip velocity.

4.3.2 Relationship of Overall Heterogeneity and Condition

As analyzed above, from particulate fluidization, bubbling fluidization, turbulent

fluidization to fast fluidization, the overall heterogeneous is increased, causing an
increase in the overall slip velocity. The overall gas-slip Reynolds number Re* is
consequently larger. From fast fluidization to pneumatic transport, the overall
heterogeneity is reduced, causing a reduction in the overall slip velocity. The Re*
value is gradually smaller.
Figure 4.6 gives a better presentation of the variation of Re* and overall het-
erogeneity with operating conditions.
As shown in the Fig. 4.6, at the small and large ends of operating gas velocity,
flow tends toward uniform and Re* is modest. In the intermediate region, hetero-
geneity is higher and Re* is larger. In other words, both Re* number and overall
heterogeneity presents a “small-ended, large-center” unimodal profile with flow
pattern.

Umf Umb Uc U TF UAC UMP

Fixed Particulate Bubble Turbulent S-shpae fast Exponential fast
bed flow fluidization fluidization fluidization fluidization transport

Uniform Increasing heterogeneity Uniform

Re* increases

Fig. 4.6 Fluidization pattern conversion

4.3 Overall Heterogeneity Analysis 95

Continuously increasing Ug converts compactly accumulated particles into

emulsion phase, then breaking up them into clusters and then into single particles as
shown in Fig. 4.1. At both “ends” where Ug is very small or very large, i.e., during
particulate fluidization and dilute pneumatic transport, flow presents a uniform
profile. Flow heterogeneity only occurs in the intermediate part, and there definitely
exists one maximum heterogeneous state.
In summary, gas–solid slip Reynolds number Re* is a bridge lining operating
conditions to overall heterogeneity. It not only represents the fluidization condi-
tions, but is also positively related to overall heterogeneity.

4.4 Condition Universality of Model

To make the heterogeneous drag model to be condition universal, it is important to

examine the condition universality of a mesoscale structure, especially a cluster
density model. This involves two aspects. One is how cluster density varies with
operating conditions. The other is how to incorporate the effects into the model.

4.4.1 Relationship of Cluster Density Versus Fluidization

Heterogeneity

The previous chapter has defined the heterogeneous region and the range of particle
density using a 45° straight line that represents uniform state and a “εsc = εsmf” level
line. The cluster density curve has been characterized as “unimodal distribution,
with the ends tending toward the 45° line.” However, it only gave one εsc(εs) curve,
the maximum of which just reaches the upper boundary, i.e., the minimum flui-
dization particle concentration εsmf.
In reality, in the heterogeneous region above the 45° line, there can be more than
one εsc(εs) curve characterizing this “unimodal” profile. The extent to which the
curve departs from 45° line indicates the level of overall flow heterogeneity. Hence,
the variation of flow conditions is actually reflected on the cluster density curve’s
moving nearer to or farther away from the 45° line by virtue of the key signature of
overall heterogeneity, as shown in Fig. 4.7.
As shown in Fig. 4.7, the nearer the cluster density curve is to the 45° line, the
smaller the overall heterogeneity of the fluidization system. The εsc(εs) curve given
in the previous chapter (the uppermost curve in Fig. 4.7) actually stands for the
maximum overall heterogeneity.
This assumption is confirmed by the experimental result of cluster density.
As shown in Fig. 4.8, all the three groups of experimental data are located in the
heterogeneous region above the 45° line. All present a unimodal profile, with the
peak locating at εs = 0.15, and tend toward the 45° line at the εs → 0 dilute end and
96 4 Condition Universality of Heterogeneous Drag Model

Fig. 4.7 Cluster density 0.6

versus overall heterogeneity εsmf
0.5

(-)
Cluster density εsc
0.4

0.3
Uniform state
0.2

Decreasing overall heterogeneity

0.1

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration ε (-)
s

the εs → 0.6 dense end. As the fluidization condition varies, so does the extent to
which the three groups depart from the 45° line. Note that, the experimental data in
the first two charts are not from only one fluidization condition. The former is an
extensive integration of the literature data, while the latter is the experimental result
from the same fluidized bed system under different operating conditions.
In summary, fluidization condition is represented by the extent to which cluster
density curve departs from the uniform state 45° line. Overall heterogeneity fully
correspond conditions to and couples it with the cluster density model. Next section
will give a quantitative representation of this coupling to ensure the model con-
dition universality.

4.4.2 Mathematic Description of Model Universality

To quantify the effects of operating condition, i.e., the extent to which cluster
density curve departs from the 45° line, a coefficient Ψ is introduced into the
original cluster density formula (Eq. 3.7) and get,

30:35e2s
esc ¼ es þ W 2 ðesmf  es Þ ð4:11Þ
e15:37ðes þ0:03Þ  0:96

According to the analysis in the previous section, Ψ actually represents the

overall heterogeneity of a fluidization system and is therefore referred to as overall
heterogeneous index. It changes between 0 and 1. When Ψ = 1, the overall flow
heterogeneity reaches its maximum, and the maximum εsc is at the upper boundary
of the heterogeneous region. The smaller the Ψ value, the closer the εsc(εs) is to the
45° line. When Ψ = 0, the εsc(εs) curve is the 45° line indicating that the entire
system is under uniform fluidization state.
4.4 Condition Universality of Model 97

Fig. 4.8 Cluster densities (a) 0.6

under various operating
conditions. a Experimental

Cluster density εsc(-)

0.5
validation 1. b Experimental
validation 2. c Experimental
0.4
validation 3
0.3
Uniform state
0.2

0.1
[52]
Experimental data
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration ε (-)
s
(b)
0.6

0.5
(-)
Cluster density εsc

0.4

0.3

0.2 Uniform state

0.1
[48]
Experimental data
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)
(c) 0.6

0.5
(-)
Cluster density εsc

0.4

0.3

0.2
Uniform state
0.1
[51]
Experimental data
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)
98 4 Condition Universality of Heterogeneous Drag Model

Table 4.1 Experimental Ar Ug Gs εs Re* Ψ

conditions and the (–) (m/s) (kg/m2 s)
corresponding Ψ values
16.4 5.5 201 0.043 11.55 0.5
8.1 202 0.02 6.89 0.2
32.5 0.91 0 0.25 4.86 0.15
1.45 1.14 0.15 7.42 0.4

Establishing the mathematical correlation between overall heterogeneous index

Ψ and condition parameter Re*, the condition universality of the cluster density
model is then achieved.
According to the experimental data of cluster density, the overall heterogeneous
index Ψ under four groups of specified conditions (Re*), are as listed in Table 4.1.
Referring to the “Ar-Re*” phase diagram, when the real Re* number equals to
the Ret number based on single particle terminal velocity, the flow state tends
toward uniform, hence Ψ = 0.
According to the calculation formula of Ret [11],

1
Ret ¼ Ar Ar\32:9
18 ð4:12Þ
Ret ¼ 0:153Ar0:71 32:9\Ar\106;520

In Table 4.1, the Ret values corresponding to the two Ar values are 0.9 and 0.8.
In fast fluidization, the axial distribution of cross-sectional average solids con-
centration is an “exponential” or “S-shaped” structure that is dense at bottom, dilute
at top, while the radial particle concentration is a “ring-nucleus” structure that is
dense at side, dilute at center. In an axially “S-shaped” fast fluidized bed, the bed is
obviously divided into a dilute upper part and a dense lower part. When the
operating gas velocity is increased, more and more bottom particles are carried up
by gas flow, causing the bottom solids concentration to reduce and the axial solids
concentration to present an “exponential” profile. When the gas velocity is further
increased, the axially dilute-top, dense-bottom heterogeneous structure gradually
disappears, and relatively uniform pneumatic transport starts. That explains why at
the boundary between fast fluidization and pneumatic transport where axial heter-
ogeneous distribution is going to appear, Re* reaches its maximum and conse-
quently Ψ reaches its maximum 1. This boundary corresponds to curve MP in the
“Ar–Re*” phase diagram in Fig. 4.4.
According to the calculation formula of ReMP [11],

ReMP ¼ 14:9Ar0:27 ð4:13Þ

Here, the ReMP values corresponding to the two Ar values in Table 4.1 are 31.7
and 38.02.
4.4 Condition Universality of Model 99

Fig. 4.9 The relationship 1.2

between Ψ and Re*
1.0

0.8

Ψ (-)
0.6

0.4

0.2

0.0
0 5 10 15 20 25 30 35 40
Re* (-)

Therefore, within Ret < Re* < ReMP, as Re* increases, the overall heterogeneous
index Ψ gradually increases from 0 to 1. Neglecting the difference of property
parameter Ar and fitting the eight groups of (Re*, Ψ) data, the resulting curve is
shown in Fig. 4.9.
The fitting curve in Fig. 4.9 presents a Boltzman curve. The overall heteroge-
neous index Ψ increases monotonously with Re* and gradually tends toward its
maximum 1. Then, it can be described by,

1:3
W¼1  ð4:14Þ
1þ eðRe 8:5Þ=5:9

Equations (4.14) and (4.11) together make up a complete cluster density model
that reflects the cluster density results in different flow states under different
operating conditions.
Cluster size equation (Eq. 3.10) also contains the overall heterogeneous index Ψ,
thus, it also reflects the condition effects. This is because cluster size is corrected by
local heterogeneous index Φ which is related to heterogeneous index Ψ. According
to the definition of Φ (Eq. 3.8) and cluster density formula (Eq. 4.11), Φ is
determined by Ψ and ɛs.

30:35e2s
U¼W 2 ð4:15Þ
e15:37ðes þ0:03Þ  0:96

Equations (3.10), (4.14), and (4.15) together make up a complete cluster size
model that indicates the cluster size results under different conditions.
Therefore, when this condition-universal mesoscale structure (cluster density
and size) model is used, within a specific range of particle properties, the
QC-EMMS drag model is also condition universal.
100 4 Condition Universality of Heterogeneous Drag Model

4.5 Universal Validation

Like the previous chapter, the universality validation here is also conducted on grid
and system scales.
Grid-scale validation directly compares the QC-EMMS model with the O-S drag
model. System-scale test calculates a number of fluidization conditions with the
universalized QC-EMMS drag model and compares the result with the experimental
data.

4.5.1 Grid-Scale Validation

Table 4.2 lists the overall heterogeneous index Ψ under the O-S model conditions
from Eq. (4.14).
The Ψ values under particle mass fluxes of 98 and 147 kg/m2 s are 0.8 and 0.96.
Ψ increases with increasing Gs, suggesting higher flow heterogeneity and more
significant particle clustering. This is echoed by the O-S drag function curve. That
is, under high Gs, the drag function reduces at a greater rate.
Figure 4.10 shows the cluster characteristic parameters and drag function derived
with the overall heterogeneous index Ψ in Table 4.2, using the mesoscale structure
model and the QC-EMMS drag model.
Compared to the QL-EMMS model results as shown in Figs. 2.2 and 2.3 which
cannot reflect the condition effects, QC-EMMS drag model is well condition uni-
versal. In Fig. 4.10, when (Ug, Gs) increases from (3.7, 98) to (4.3, 147), both
cluster density and size increase, leading to the significant reduction of drag
function. This indicates that the increased Gs results in more significant particle
clustering and higher overall heterogeneity.
In Fig. 4.10c, compared with the O-S curve, QC-EMMS model not only reflects
the essence of drag function distribution relative to solids concentration, but also its
variation with condition. The drag function curve is a “concave” ascending profile
and descends with the increase of Gs. However, as discussed in the previous
chapter, as slip velocity differs, the QC-EMMS drag function is quantitatively
different from the O-S drag function.
To further validate the accuracy and universality of QC-EMMS drag model, it is
applied for fluidized bed simulation under various conditions. That is a system-scale
validation.

Table 4.2 Ψ values Ug (m/s) Gs (kg/m2 s) Re* Ψ

corresponding to O-S
experimental condition 3.7 98 18.5 0.8
4.3 147 21.4 0.96
4.5 Universal Validation 101

Fig. 4.10 QC-EMMS model (a) 0.6

results under various (Ug, Gs) (4.3, 147)
(ρg = 1.144 kg/m3, 0.5

(-)
μg = 1.848 × 10−5 Pa s,

Cluster density εsc

ρp = 1714 kg/m3, dp = 76 μm, 0.4 (3.7, 98)
εsmf = 0.6). a Cluster density.
b Dimensionless cluster size. 0.3
c Drag model Uniform state
0.2

0.1

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration ε s (-)
(b) Dimensionless cluster size 120

90
dcl/dp (-)

(4.3, 147)
60

30 (3.7, 98)

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)
(c) 10
6
β (Ns/m )
4

5 Uniform state
10
4
10
QC-EMMS
3
Drag function

10
2
10 2
Ug(m/s) Gs(kg/m s)
10
1 3.7 98
O-S 4.3 147
0
10
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Solids concentration εs (-)

4.5.2 System-ScaleValidation

To examine the condition universality of a drag model, it is necessary to select a

range of representative flow states and operating conditions. For this purpose, this
section selects five conditions, low-flux bed with Geldart A particles (case 1),
high-flux bed with A particles (case 2), two-dimensional (2D) bed with Geldart B
particles (case 3), and three-dimensional (3D) bed with Geldart B particles (case 4).
102 4 Condition Universality of Heterogeneous Drag Model

Fig. 4.11 Distribution of

computational conditions

UFD

Gs (kg/m2s)
UTF

UPT

ε av

Ug (m/s)

Among them, Case 2 also contains two different operating gas velocities and par-
ticle mass fluxes.
These conditions are identified by solid circles in the “Ug-Gs” fluidized phase
diagram, as shown in Fig. 4.11. These conditions are widely different from each
other and encompass the majority of the diagram. Numerical simulation of fluidized
bed under these conditions, therefore, can effectively validate the condition uni-
versality of QC-EMMS drag model.
Low-flux and high-flux bed using Geldart A particles (Case 1 and 2)
Geldart A particles are more likely to form clusters under gas flow, resulting in high
flow heterogeneity. Uniform drag models cannot predict the heterogeneous fluidized
beds using Geldart A particles [15–18]. Therefore, numerical simulations of Geldart A
particle fluidized bed are first conducted and validated by experimental data.
1. Geometric model and computational conditions
The 2D simplified geometric model of fluidized bed reactor is shown in
Fig. 4.12, the two-dimensional (2D) simplified geometric model of fluidized bed
reactor. At a constant operating gas velocity and a given initial number of
particles, gas flow carries the particles up and out of the pressure outlet at the
top. All the outgoing particles are returned into the bed from the bottom to
ensure a constant particle inventory in bed. That is, particle circulating mass flux
is calculated at a given gas velocity and bed inventory.
Table 4.3 lists the computational conditions, assuming the same gas–solid wall
conditions as those of the case in Chap. 3. Gs is the value to be solved. The
experiment-measured Gs for case 1 equals to 106 kg/m2 s. Case 2 includes two
different operating velocities 5.5 and 8.1 m/s. The experiment-measured Gs
values are fairly high, that is 201 and 202 kg/m2 s, respectively.

2. Result analysis for Case 1

Figure 4.13 shows the variation of particle mass flux with time. The calculated
Gs is almost 0 before 5 s. This is because particles are initially accumulated on
4.5 Universal Validation 103

Fig. 4.12 Two-dimensional (a) (b) Outlet

simplified geometric model

100
for CFB riser (Geldart A
particle). a Case 1. b Case 2
Outlet
100

254

15100
10850

Circulating solids
Circulating solids

inlet
H0
100

100 100
H0

inlet
h
x h
x
Gas inlet Gas inlet

Table 4.3 Computational conditions

Parameter Case 1 Case 2a Case
2b
Operating gas velocity Ug (m/s) 3.8 5.5 8.1
Initial accumulation height H0 (m) 2.97 2.15 1.02
Overall heterogeneous index Ψ 0.9 0.3 0.5
Gas density ρg (kg/m3) 1.212 1.225
Gas kinematic viscosity μg (Pa s) 1.908 × 10−5 1.789 × 10−5
Particle density ρp (kg/m3) 1380 1500
Particle diameter dp (μm) 65 67
Minimum fluidization particle concentration 0.5 0.58
εsmf (–)
Number of cells 30 × 154 30 × 200
Initial accumulation concentration εs0 (–) 0.3
Particle mass flux in circulation Gs (kg/m2 s) To be
calculated
Outlet pressure (Pa) 1.013 × 105
Gas-wall interaction Slipless
Particle-wall interaction Partial slip
Particle-wall specular reflection coefficient φ 0.0001
Particle elastic collision coefficient es 0.9
Time step Δt (s) 5 × 10−4
104 4 Condition Universality of Heterogeneous Drag Model

Fig. 4.13 Instantaneous 400

particle mass flux

Particle mass flux Gs (kg/m s)

350

2
300

250

200

150

100

50 Experimental data

0
0 5 10 15 20 25 30 35 40
Time (s)

the bed bottom and, though carried up by gas flow, are yet to reach the outlet at
the top. After 5 s, Gs fluctuates widely with time and starts to stabilize after 10 s,
fluctuating around the average value.
Table 4.4 compares the QC-EMMS simulation result with that of different drag
models for the same case, including Gidaspow, EMMS/matrix [19], QL-EMMS
drag model [20]. Gidaspow drag model greatly overestimates the particle mass
flux since it cannot address clustering effects. Comparatively, the EMMS-based
drag models are able to effectively reduce computational error. QC-EMMS
model provides the highest computational accuracy with relative error of merely
2.85 %. This confirms that QC-EMMS drag model applies to numerical simu-
lation of Geldart A particle fluidized bed and its computational accuracy is
higher than other EMMS models.
Figure 4.14 shows the axial distribution of solids concentration. The compu-
tational result from the QC-EMMS model agrees well with the experimental
result. As the bed height increases, the cross-sectional average solids concen-
tration gradually reduces, presenting the “dilute at top, dense at bottom” het-
erogeneous distribution in bed. The result from the Gidaspow drag model [21],
on the other side, indicates an almost uniform solids distribution along the
height of bed. This is obviously attributable to the inability of a uniform drag
model to address mesoscale particle clustering effect.

3. Result analysis for Case 2

The applicability of QC-EMMS model to high-flux bed (Gs > 200 kg/m2 s) is
tested through Case 2. As shown in Table 4.5, under both two operating gas

Table 4.4 Particle mass fluxes predicted by different drag models with relative errors
Experimental Gidaspow EMMS/matrix QL-EMMS QC-EMMS
value
Particle mass flux 106 444 198 133 109
Gs (kg/m2 s)
Relative error (%) – 319 86.6 25.5 2.85
4.5 Universal Validation 105

Fig. 4.14 Axial distribution 12

of cross-sectional solids Experimental data
concentration Calculation result
10

Height h (m)
6

0
0.0 0.1 0.2 0.3
Cross-section averaged solids
concentration εs,av (-)

Table 4.5 Calculated particle circulating mass flux Gs for Case 2

Operating gas velocity Calculated value Experimental value Error
(m/s) (kg/m2 s) (kg/m2 s) (%)
5.5 217.3 201 8.1
8.1 192.2 202 4.9

velocities (Ug = 5.5, 8.1 m/s), the computational error of particle mass flux is
smaller than 10 %.
Figure 4.15 compares the computational and experimental results of the axial
distribution of solids concentration. As shown in the figure, the axial distribution
of cross-sectional solids concentration under both conditions well agrees with
the experimental data. As the bed height increases, εs gradually reduces expo-
nentially, presenting the “dilute-top, dense-bottom” overall heterogeneous pro-
file in fluidized bed.
The particle mass fluxes under the two conditions are close. When the apparent
operating gas velocity is reduced from 8.1 to 5.5 m/s, the cross-sectional solids
concentration εs,av at any bed height shows an increase. That is because when
gas velocity reduces, the gas flow carrying effect diminishes, and the overall
solids concentration in bed increases. This confirms that QC-EMMS drag model
can predict the accurate heterogeneous flow characteristics in high-flux bed.
Fluidized bed using Geldart B particles (Case 3 and 4)
For fluidized bed using Geldart B particles, the flow is not that heterogeneous
because large-size particles are difficult to form clusters. Although some studies
106 4 Condition Universality of Heterogeneous Drag Model

Fig. 4.15 Axial distribution 16

of cross-sectional solids Experimental data
concentration 14 Calculation result

12

Height h (m)
10

6
Ug/Gs
4 5.5/201
8.1/202

0
0.00 0.03 0.06 0.09 0.12
Cross-section averaged solids
concentration εs,av (-)

have demonstrated that a uniform drag model can achieve fairly high computational
accuracy, some others maintain that only by introducing a heterogeneous drag
model is it possible to make accurate numerical simulation [22].
This section will test the applicability of QC-EMMS drag model to the numerical
simulation of gas–solid fluidized bed using Geldart B particles. The two cases
involve 2D and 3D simulation, respectively.
1. Geometric model and computational condition
Figure 4.16 shows a simplified geometric model of fluidized bed reactor. Similar
to the cases above, particle mass flux is calculated at a given operating gas
velocity and bed inventory.
Table 4.6 lists the gas–solid properties and computational conditions. Gas–solid
wall conditions are the same as those used previously. Particle circulating mass
flux Gs is the value to be predicted.
The experimental measured Gs values in Case 3 and Case 4 are 151.6 and
30 kg/m2 s, respectively.
2. Results analysis for Case 3
Figure 4.17 shows the calculated instantaneous particle circulating mass flux.
Obviously, Gs fluctuates widely with time and starts to stabilize after 5 s,
fluctuating around the average value of 165.9 kg/m2 s. This implies that the gas–
solid two-phase flow state in bed reaches is stable state. The calculated average
165.9 kg/m2 s is only 9.4 % divergent from the experimental value of
151.6 kg/m2 s, confirming the accuracy of QC-EMMS model.
Figure 4.18 shows the instantaneous distribution of solids concentration within
10–20 s after flow stability. The simulation result discloses the mesoscopic flow
details in fluidized bed and gives a real picture of the experimentally photo-
graphed particle clustering in Fig. 1.1. Particle clusters with fairly high local
particle concentration are detected in bed. The clusters come in different shapes
(flocs, belts, clusters, and flakes) that change continuously with spatial position
4.5 Universal Validation 107

(a) (b)

411 700
Outlet

411
8500

Circulating solids
inlet
342 411
H0

h
x
Gas inlet

Fig. 4.16 Simplified geometric model for CFB riser (Geldart B particles). a Case 3. b Case 4

Table 4.6 Gas–solid properties and computational conditions

Parameter Case 3 Case 4
Operating gas velocity Ug (m/s) 7.6 3
Initial accumulation height H0 (m) 2.16 1.3
Overall heterogeneous index Ψ 0.95 0.82
Initial accumulation concentration εs0 (–) 0.3
Gas density ρg (kg/m3) 1.225 1.18
Gas kinematic viscosity μg (Pa s) 1.789 × 10−5 1.5 × 10−5
Particle density ρp (kg/m3) 2500 2600
Particle diameter dp (μm) 300 140
Grid number 40 × 200 28 × 15 × 150
solids concentration in critical fluidization εsmf (–) 0.5
Outlet pressure (Pa) 1.013 × 105
Time step Δt (s) 5 × 10−4

and keep breaking up and forming over time. At the bed bottom, especially near
the wall surface, particles are noticeably clustered and the cluster density
(internal particle concentration) is fairly high. Up the bed height, cluster density
reduces, attributable to the reduced local solids concentration. Besides that, up
the bed height, cluster axial size increases but radial size reduces, looking like a
108 4 Condition Universality of Heterogeneous Drag Model

Fig. 4.17 Instantaneous 1000

Particle mass flux Gs (kg/m s)

particle mass flux

2
800

600
Time-avreaged Calculation result: 165.9
400

200

0
Experimental data:151.6

-200
0 5 10 15 20
Time (s)

Fig. 4.18 Instantaneous

solids concentration

strip. Previous experiments only measured the axial size and thus cannot give
the real size of clusters. In reality, it is the windward size of clusters (mainly
decided by the radial size) that makes sense to gas–solid interactive drag force.
This understanding is very important for developing cluster size models and
heterogeneous drag models.
4.5 Universal Validation 109

Fig. 4.19 Time-averaged

solids concentration (10–20 s)

Figure 4.19 shows the time-averaged solids concentrations after flow stability.
Axially, a “dilute-top, dense-bottom” heterogeneous profile is observed. In the
bottom dense region, a radially dilute-center, dense-side “ring-nucleus” structure
appears. In the upper dilute region, similar radial heterogeneous structure is also
detected within a smaller color range.
Figure 4.20 compares the computational and experimental results of the axial
distribution of solids concentration quantitatively. The computational result
agrees well with the experimental data. Solids concentration at the bottom is
larger and gradually diminishes up the bed height.
Figure 4.21 compares the computational and experimental results of the radial
distribution of solids concentration at two axial heights. These two have the
same variation trend and quantitatively agree well with each other. The particle
velocity at the bed center is larger. This is because particles at the center are
more dilute and the gas flow carrying effect is greater. The particle velocity at
the side is smaller and negative. This suggests that particle moves up and down
along the wall and particle backmix and internal circulation exist in the fluidized
bed. It is right this intraparticle circulation that causes mesoscale structure
110 4 Condition Universality of Heterogeneous Drag Model

Fig. 4.20 Axial distribution 8

of cross-sectional solids Experimental data
concentration Calculation result

Height h (m)
4

0
0.0 0.1 0.2 0.3
Cross-section averaged solids
concentration εs,av (-)

(a) (b)
12 12
us (m/s)
us (m/s)

h=1.54 m h=4.209 m
8 8

4 4
Particle velocity

Particle velocity

0 0

-4 -4

-8 -8
-1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0
Radial position r/R (-) Raidal position r/R (-)

Fig. 4.21 Radial distribution of particle velocity. a h = 1.54 m. b h = 4.209 m

clusters to form locally, which eventually account for a dilute-center, dense-side

“ring-nucleus” structure in the radial direction.
In summary, QC-EMMS drag model not only calculates particle circulating
mass flux accurately, but also discloses the real mesoscopic flow details and
accurately predicts the local and overall heterogeneity inside the fluidized bed
using Geldart B particles.
3. Result analysis for Case 4
The applicability of QC-EMMS drag model to simulate 3D fluidized bed is
tested through Case 4.
4.5 Universal Validation 111

Fig. 4.22 Instantaneous 300

particle mass flux

Particle mass flux Gs (kg/m s)

2
250
Drag model : Wen-Yu
200

150

100

50 QC-EMMS

Experimental data
0
0 10 20 30 40 50
Time (s)

Figure 4.22 shows the calculated instantaneous particle mass flux Gs using the
uniform flow Wen-Yu drag model and the heterogeneous QC-EMMS drag
model. The Gs from the Wen-Yu model starts to stabilize after 10 s and the
time-averaged value after stability is 126.4 kg/m2 s. The Gs from the QC-EMMS
model starts to stabilize after 5 s and the time-averaged value is 40.4 kg/m2 s,
which agrees well with the experimental result. Therefore, for Geldart B particle
fluidized bed, even when using 3D simulation, the uniform drag model still
overestimates the drag, leading to excess gas flow carrying capacity. The
QC-EMMS drag model can effectively reflects the drag reduction, accurately
predicts particle mass flux, and significantly improves computational accuracy.
Figure 4.23 compares the axial distribution of solids concentration predicted by
the Wen-Yu and the QC-EMMS drag model and also the experimental

Fig. 4.23 Axial distribution 8

of cross-sectional solids Experimental data
concentration Wen-Yu
QC-EMMS
6
Height h (m)

0
0.0 0.1 0.2 0.3
Cross-section averaged solids
concentration εs,av (-)
112 4 Condition Universality of Heterogeneous Drag Model

Fig. 4.24 Solids concentrations on cross sections of different heights

measurement. For the Wen-Yu drag model, the cross-sectional average solids
concentration is almost on a vertical line along the bed height, i.e., particle
concentration axially shows a uniform distribution. The result from the
QC-EMMS drag model almost fully agrees with the experimental data and is
more accurate than the previous 2D simulation. The solids concentration at the
bottom is larger, with the maximum close to 0.3. As the axial bed height
increases, solids concentration gradually reduces. This confirms that the
QC-EMMS drag model can accurately predict the axially dilute-top,
dense-bottom heterogeneous characteristics in gas–solid fluidized bed and 3D
simulation helps improve the computational accuracy.
Figure 4.24 shows the cross-sectional solids concentration distribution at three
different heights. The solids concentration at the bed center is smaller and that
near the side is larger, presenting an obvious “ring-nucleus” heterogeneous
structure. As the bed height increases, from 0.2 to 1.2 m, the cross-sectional
solids concentration reduces on the whole, presenting a dilute-top, dense-bottom
axial heterogeneous structure.
Figure 4.25 shows the radial distribution of solids concentration at different
heights. Solids concentration gradually increases from center toward sides,
corresponding to Fig. 4.24. As the bed height increases, solids concentration
reduces on the whole and the radial distribution curve becomes milder. This is
because particles in fluidized bed are carried up by gas flow, and some of them
start to move down along the wall after coming into the upper region, giving
rising to intraparticle circulation. The downcoming particles accumulate com-
pactly near the wall and contribute to increased particle concentration.
Therefore, the QC-EMMS drag model applies to 3D fluidized bed simulation to
obtain accurate particle mass flux and heterogeneous distribution profile.
4.6 Summary 113

Fig. 4.25 Radial distribution 0.5

of solids concentration
h=0.2 m

Solids concentration εs (-)

0.4

0.3
0.8
0.2

0.1
1.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0
Position in x direction (m)

4.6 Summary

1. It is discovered that, as fluidized gas velocity increases, the system undergoes

bubbling flow—fast flow—pneumatic transport flow, which corresponds to the
unimodal distribution of overall slip velocity featuring an increase followed by a
reduction.
2. Through fluidization phase diagram and flow heterogeneity analysis, it is dis-
covered that the overall heterogeneity of different flow states under different
operating conditions is also unimodal and corresponds to overall slip velocity.
The overall slip Reynolds number Re* can link condition to overall
heterogeneity.
3. Through fluidization phase diagram and flow heterogeneity analysis, it is dis-
covered that the overall heterogeneity of different flow states under different
operating conditions is also unimodal and corresponds to overall slip velocity.
The overall slip Reynolds number Re* can link condition to overall
heterogeneity.
4. Grid-scale and system-scale model testing indicates that the variation with
condition from the QC-EMMS model agrees with the O-S model. The
QC-EMMS model can accurately simulate greatly different fluidized under
various conditions, with computational error of particle circulating mass flux
always limited to 10 %. It can accurately predict dynamics of clusters forming
and breaking up with time, and the axial and radial heterogeneous distribution of
particle concentration and velocity.
114 4 Condition Universality of Heterogeneous Drag Model

References

1. Qi HY (1997) Euler/Euler Simulation der Fluiddynamik Zirkulierender Wirbelschichten.

Verlag Mainz, Wissenschaftsverlag, Aachen, Germany. ISBN3-89653-224-3
2. Bai D, Jin Y, Yu Z (1993) Flow regimes in circulating fluidized beds. Chem Eng Technol 16
(5):307–313
3. Bi HT, Grace JR (1995) Flow regime diagrams for gas-solid fluidization and upward transport.
Int J Multiph Flow 21(6):1229–1236
4. Yang WC (2004) “Choking” revisited. Ind Eng Chem Res 43(18):5496–5506
5. Wang XS, Rhodes MJ (2002) On the use of choking as a boundary for fast fluidization. Chem
Eng Commun 189(2):223–236
6. Bai D, Jin Y, Yu Z (1993) Flow regimes in circulating fluidized beds. Chem Eng Technol 16
(5):307–313
7. Namkung W, Kim SW, Kim SD (1999) Flow regimes and axial pressure profiles in a
circulating fluidized bed. Chem Eng J 72(3):245–252
8. Das M, Bandyopadhyay A, Meikap BC et al (2008) Axial voidage profiles and identification
of flow regimes in the riser of a circulating fluidized bed. Chem Eng J 145(2):249–258
9. Bi HT, Grace JR (1995) Flow regime diagrams for gas-solid fluidization and upward transport.
Int J Multiph Flow 21(6):1229–1236
10. Rabinovich E, Kalman H (2011) Flow regime diagram for vertical pneumatic conveying and
fluidized bed systems. Powder Technol 207(1):119–133
11. Rabinovich E, Kalman H (2011) Flow regime diagram for vertical pneumatic conveying and
fluidized bed systems. Powder Technol 207(1):119–133
12. Smolders K, Baeyens J (2001) Gas fluidized beds operating at high velocities: a critical review
of occurring regimes. Powder Technol 119(2):269–291
13. Matsen JM (1982) Mechanisms of choking and entrainment. Powder Technol 32(1):21–33
14. Yerushalmi J, Cankurt NT (1979) Further-studies of the regimes of fluidization. Powder
Technol 24(2):187–205
15. Wang J (2009) A review of Eulerian simulation of Geldart A particles in gas-fluidized beds.
Ind Eng Chem Res 48(12):5567–5577
16. Sundaresan S (2000) Modeling the hydrodynamics of multiphase flow reactors: current status
and challenges. AIChE J 46(6):1102–1105
17. Wang W, Li Y (2004) Simulation of the clustering phenomenon in a fast fluidized bed: the
importance of drag correlation. Chin J Chem Eng 12(3):335–341
18. Ullah A, Wang W, Li J (2013) Evaluation of drag models for cocurrent and countercurrent
gas–solid flows. Chem Eng Sci 92:89–104
19. Dong W, Wang W, Li JH (2008) A multiscale mass transfer model for gas-solid riser flows:
part II-sub-grid simulation of ozone decomposition. Chem Eng Sci 63(10):2811–2823
20. Li F (2009) Investigations on the turbulent gas-solid two-phase interactions in fluidized
desulfurization process. Doctoral dissertation, Tsinghua University, Beijing
21. Dong W, Wang W, Li JH (2008) A multiscale mass transfer model for gas-solid riser flows:
part II-sub-grid simulation of ozone decomposition. Chem Eng Sci 63(10):2811–2823
22. Wang JW, Ge W, Li JH (2008) Eulerian simulation of heterogeneous gas-solid flows in CFB
risers: EMMS-based sub-grid scale model with a revised cluster description. Chem Eng Sci
63:1553–1571
Chapter 5
Conclusions and Outlook

An investigation with flow heterogeneity as the principal line was conducted on the
formation and evolution of mesoscale structure in heterogeneous gas–solid
two-phase fluidization and its effects on drag force. A mesoscale structure mathe-
matical model was established, and the Energy Minimization Multi-scale (EMMS)
theory was refined. An accurate heterogeneous drag model with certain condition
universality was developed and proved to have the ability to simulate heteroge-
neous flow processes under various conditions accurately.

5.1 Research Findings

5.1.1 EMMS Theory and the Effects of Mesoscale Structure

on Drag

1. The existing EMMS drag models cannot reflect the essence of drag reduction
caused by heterogeneous flow. This is because they fail to give a logical
description of mesoscale structure (cluster) characteristics.
2. Clarified the fuzzy understanding on cluster variation in dense gas–solid
two-phase flow. As solids concentration increases, clusters first form and then
keep breaking up and eventually return to single particle state. The curve of
cluster size versus local solids concentration is unimodal and tends toward
single particle size at the dilute and dense ends.
3. Cluster density makes a material difference to drag force. The key to refining the
EMMS theory lies in the supplementation of a cluster density model and the
correction of the cluster size model.

5.1.2 Mesoscale Structure (Cluster) Model

1. Based on local slip velocity and O-S heterogeneous drag analysis, the variation
of local flow heterogeneity with solids concentration εs was identified. It is a

© Springer-Verlag Berlin Heidelberg 2016 115

C. Chen, Investigations on Mesoscale Structure in Gas–Solid Fluidization
and Heterogeneous Drag Model, Springer Theses,
DOI 10.1007/978-3-662-48373-2_5
116 5 Conclusions and Outlook

unimodal shape, with the peak locating at εs of 0.1–0.15, and tending toward the
minimum at the dilute end (εs → 0) and the dense end (εs → εsmf).
2. In the diagram of cluster density versus local solids concentration εsc(εs), the
range of cluster density was delineated. (a) A 45° straight line representing
uniform state was proposed to be the lower boundary of the εsc(εs) curve. On this
45° line, the εsc is constantly equals to the εs, suggesting uniform flow without
local particle clustering. (b) A “εsc = εsmf” level line representing the maximum
heterogeneous state is proposed to define the upper boundary of the εsc(εs), since
εsc can never exceed minimum fluidization concentration εsmf.
3. The εsc(εs) curve was determined to be unimodal. The εsc tends toward the 45°
line on the dilute and dense ends where the flow is uniform. A cluster density
model on this basis was established and validated through experimental data,
i.e., it passed a “mesoscale” validation.
4. Based on the cluster density model, a local heterogeneous index was established
to describe the level of particle clustering. The cluster size equation in the
EMMS theory is corrected with this index to accord with the physical realities.

5.1.3 Heterogeneous Drag Model

1. Based on the proposed mesoscale structure model, the EMMS theory was
refined and a more accurate heterogeneous drag model (QC-EMMS) was
developed. QC-EMMS model reflects the essence of heterogeneous drag, i.e., a
“concave” rise with solids concentration and tending toward uniform state at the
dilute and dense ends.
2. The mathematical expression of the heterogeneous drag model was concluded to
make it easily couple with the two-fluid approach for numerical simulation of
gas–solid flows. The simulation results of gas–solid fluidized bed agree well
with experimental results. Compared with uniform flow models and previous
heterogeneous drag models, QC-EMMS model is more accurate and more grid
independent. It successfully predicts local gas–solid slip velocity and flow
heterogeneity, and reproduces the unimodal distribution profile of heterogeneity
with solids concentration. It can also predict “choking” state.

5.1.4 Conditional Universality of Model

1. The overall heterogeneity of different flow states under different operating

conditions is found to be unimodal that is similar to the local heterogeneity. The
overall slip-velocity Reynolds number Re* only reflects the overall heteroge-
neity but also represents the operating condition.
5.1 Research Findings 117

2. The correlation between cluster density model and operating condition was
established. It is reflected by the variation of εsc(εs) curve in the heterogeneous
region. The variation was represented by the overall heterogeneous index Ψ.
The mathematical expression of Ψ and Re* was established to achieve the
condition universality of the mesoscale structure model, and the QC-EMMS
drag model within a specific range of particle properties.
3. The model condition universality was validated on grid-scale and system-scale.
The variation with condition from the QC-EMMS model agrees well with the
O-S model. The QC-EMMS model can accurately simulate greatly different
fluidized beds under various conditions. The computational errors of particle
circulating mass flux are always less than 10 %. The dynamics of cluster
forming and breaking up with time, and the axial and radial heterogeneous
distribution of solids concentration and velocity all agree with the experimental
measurements.

5.2 Research Features and Innovations

Research features of this study are using flow heterogeneity analysis as the principal
line, mesoscale structure (cluster) characterization as the core, refining basic theory,
establishing a theoretical model, and developing the simulation capability as the
goal, and mesoscale, grid-scale and system-scale experimental validation as the
guarantee.

Innovations
1. Fundamentally changed the previous fuzzy understanding on the main param-
eters of mesoscale structure, cluster density εsc, and cluster size dcl. The cluster
density is actually the main determinant for drag. The existing cluster size
definitions are chaotic and responsible for the misunderstanding that, cluster size
increases to infinite at extremely dense solids concentration. While in reality,
cluster size tends toward single particle size dp at dense end.
2. Cluster density and size is found to vary in a “unimodal” profile with local solids
concentration. With this discovery, a corresponding mesoscale structure model
and a new heterogeneous drag model was established.
3. It is discovered that on system-scale, gas–solid flow heterogeneity also varies in
a “unimodal” profile with flow state (condition) and can be quantitatively
described with gas–solid slip Re* number. On this basis, we established the
correlation between mesoscale structure model and system heterogeneity and
virtually achieved condition universality of the heterogeneous drag model.
Experimental validation on mesoscale, grid-scale, and system-scale proved that
this model accurately describes the basic rule of fluidized gas–solid interaction and
further refines the EMMS theory.
118 5 Conclusions and Outlook

5.3 Outlook

1. Effects of particle size. Existing models are yet to consider the particle size
distribution, whereas this makes a difference to mesoscale structure. This pos-
sibly contributes to the modest error in our numerical simulation. For this rea-
son, it is necessary to correct the uniform-sized mesoscale structure model for its
size distribution.
2. Cluster size model. Though we have corrected the original cluster size formula
in EMMS theory after qualitative analysis, we do not have substantial experi-
mental validation due to limitations of experimental techniques. Besides, the
assumption of the EMMS theory that cluster size is inversely related to sus-
pended transport energy only applies to limit flows, from the fast fluidization to
pneumatic transport transition. Hence, it would be necessary in the next step to
purposely design separate experiments that properly measure cluster size using
optical fiber, speed camera, and other technologies. Furthermore, cluster size
variation rule from fixed bed, bubbling fluidization, particulate fluidization to
fast fluidization needs investigation.
3. A general heterogeneous drag model. Existing models have to involve an
EMMS analysis on the definite flow conditions to be simulated before a formula
for heterogeneous drag correction factor Hd can be obtained. The Hd formula
obtained this way only applies to the present condition and is far from being a
general expression. Therefore, generalized research has to be conducted.