Advances in Engineering Software 42 (2011) 375–386

Contents lists available at ScienceDirect

Advances in Engineering Software

journal homepage: www.elsevier.com/locate/advengsoft

Adjustment of drag coefficient correlations in three dimensional

CFD simulation of gas–solid bubbling fluidized bed
Ehsan Esmaili, Nader Mahinpey ⇑
Dept. of Chemical and Petroleum Engineering, Schulich School of Engineering, University of Calgary, AB, Canada T2N 1N4

a r t i c l e i n f o a b s t r a c t

Article history: Fluidized beds have been widely used in power generation and in chemical, biochemical, and petroleum
Received 13 November 2009 industries. 3D simulation of commercial scale fluidized beds has been computationally impractical due to
Received in revised form 8 November 2010 the required memory and processor speeds. In this study, 3D Computational Fluid Dynamics simulation
Accepted 10 March 2011
of a gas–solid bubbling fluidized bed is performed to investigate the effect of using different inter-phase
Available online 9 April 2011
drag models. The drag correlations of Richardon and Zaki, Wen–Yu, Gibilaro, Gidaspow, Syamlal–O’Brien,
Arastoopour, RUC, Di Felice, Hill Koch Ladd, Zhang and Reese, and adjusted Syamlal are reviewed using a
Keywords:
multiphase Eulerian–Eulerian model to simulate the momentum transfer between phases. Furthermore,
Multiphase flow
Fluidized bed
a method has been proposed to adjust the Di Felice drag model in a three dimensional domain based on
Computational Fluid Dynamics the experimental value of minimum fluidization velocity as a calibration point. Comparisons are made
Inter phase drag model with both a 2D Cartesian simulation and experimental data. The experiments are performed on a Plexi-
Coefficient of restitution glas rectangular fluidized bed consisting of spherical glass beads and ambient air as the gas phase. Com-
Eulerian–Eulerian model parisons were made based on solid volume fractions, expansion height, and pressure drop inside the
fluidized bed at different superficial gas velocities. The results of the proposed drag model were found
to agree well with experimental data. The effect of restitution coefficient on three dimensional prediction
of bed height is also investigated and an optimum value of restitution coefficient for modeling fluidized
beds in a bubbling regime has been proposed. Finally sensitivity analysis is performed on the grid interval
size to obtain an optimum mesh size with the objective of accuracy and time efficiency.
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction studied the effect of implementing different drag models on simu-

Gas–solid fluidized bed reactors are used in many industrial
operations, such as energy production and petrochemical pro-
cesses. Some of the distinct advantages of gas–solid fluidized bed
reactors over other methods of gas–solid reactors are controlled
handling of solids, isothermal conditions due to good solids mixing
and the large thermal inertia of solids, and high heat flow and reac-
tion rates between gas and solids due to large gas-particle contact
area. Hence, the fluidized bed reactors are widely used in gasifica-
tion, combustion, catalytic cracking and various other chemical
and metallurgical processes. Two approaches are typically used
for CFD modeling of gas–solid fluidized beds. The first one is
Lagrangian–Eulerian modeling [1–6], which solves the equations
of motion individually for each particle and uses a continuous
interpenetrating model (Eulerian framework) for modeling the
gas phase. In large systems of particles, the Lagrangian–Eulerian
model requires powerful computational resources because of the
numbers of equations that are being solved. Bokkers et al. [5] have
⇑ Corresponding author. Fax: +1 403 284 4852.
E-mail address: nader.mahinpey@ucalgary.ca (N. Mahinpey).
lation of gas–solid fluidized bed using Discrete Particle Model
(DPM) which assume a Lagrangian–Eulerian model for the multi-
phase fluid flow. van Sint Annaland et al. [6] have also studied
the particle mixing and segregation rates in a bi-disperse freely
bubbling fluidized bed with a new multi-fluid model (MFM) based
on the kinetic theory of granular flow for multi-component sys-
tems. The second approach is Eulerian–Eulerian modeling [7–13],
which assumes that both phases can be considered as fluid and
also take the interpenetrating effect of each phase into consider-
ation by using drag models. Therefore, applying a proper drag
model in Eulerian–Eulerian modeling is of a great importance.
Many researchers have applied 2D Cartesian simulations to
model pseudo-2D beds [1,7,11,13]. Behjat et al. [11] applied a
two-dimensional CFD (Computational Fluid Dynamics) technique
to the fluidized bed in order to investigate the hydrodynamic and
the heat transfer phenomena. They concluded that the Eulerian–
Eulerian model is suitable for modeling industrial fluidized bed
reactors. Their results indicate that considering two solid phases,
particles with smaller diameters have lower volume fraction at
the bottom of the bed and higher volume fraction at the top of
the bed. They also showed that the gas temperature increases as
it moves upward in the reactor due to the heat of polymerization

0965-9978/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.advengsoft.2011.03.005
376 E. Esmaili, N. Mahinpey / Advances in Engineering Software 42 (2011) 375–386

Nomenclature

A constant in RUC-drag model (–) Greek letters

A constant in Syamlal–O0 Brien drag model (–) b angel of internal friction(°)
B constant in RUC-drag model (–) eg gas phase volume fraction (–)
B constant in Syamlal–O0 Brien drag model (–) es solid phase volume fraction (–)
Cn drag factor on multi-particle system (–) cHs dissipation of granular temperature (kg m1 s3)
ds diameter of solid particles (m) D change in variable, final–initial (–)
e restitution coefficient of solid phase (–) r the Dell operator (m(1))
F drag factor in HKL drag model (–) Hs granular temperature(m2 s2)
Fr friction factor from Johnson et al. frictional viscosity (–) ks bulk viscosity (kg m1 s1)
F0, F1, F2, F3 drag constants in the HKL drag function (–) lg gas viscosity (kg m1 s1)
g the gravitational acceleration (=9.81) (m s2) ls granular viscosity (kg m1 s1)
g0 the general radial distribution function (–) ls,col collisional viscosity (kg m1 s1)
I the unit tensor (–) ls,kin kinetic viscosity (kg m1 s1)
I2D the second invariant of the deviatoric stress tensor (–) ls,fric frictional viscosity (kg m1 s1)
Ksg drag factor of phase s in phase g (kg m3 s1) ldil dilute viscosity in Gidaspow kinetic viscosity model
kHs conductivity of granular temperature (kg m1 s1) (kg m1 s1)
n coefficient in the Richardson and Zaki drag correlation p the irrational number p (–)
(–) qg gas density (kg m3)
P pressure (Pa) qs solid density (kg m3)
Ps solids pressure (Pa) s the stress–strain tensor (Pa)
Ps,fric frictional pressure (Pa)
DP pressure drop (Pa) Subscripts
r  qs diffusive flux of fluctuating energy (kg m1 s3) col collisional
Re the Reynolds number (–) dil dilute
Rem the modified Reynolds number in the Richardson Zaki fr frictional
correlation (–) g gas or fluid phase
Res the particle Reynolds number (–) kin kinetic
t time (S) max maximum
Dt time interval (S) mf minimum fluidization condition
Umf minimum fluidization velocity (m s1) min minimum
us,i, us,j solid phase velocity in the i and j direction (m s1) q general phase q
~
V velocity (m s1) s solid phase
vr the relative velocity correlation (–)
w factor in the HKL drag correlation (–)

reaction leading to the higher temperatures at the top of the bed
[11]. Peiranoa et al. [14] investigated the importance of three
dimensionality in the Eulerian approach simulations of stationary
bubbling fluidized beds. The results of their simulations show that
two-dimensional simulations should be used with caution and only
for sensitivity analysis, whereas three-dimensional simulations are
able to reproduce both the statics (bed height and spatial distribu-
tion of particles) and the dynamics (power spectrum of pressure
fluctuations) of the bed. In addition, they assumed that the accurate
prediction of the drag force (the force exerted by the gas on a single
particle in a suspension) is of little importance when dealing with
bubbling beds. However, in the present study, it is found that using
a proper drag model can significantly increase the accuracy of
results in the 3D simulation of bubbling fluidized beds.
Cammarata et al. [8] compared the bubbling behavior predicted
by 2D and 3D simulations of a rectangular fluidized bed using the
commercial software ANSYS-CFX (a CFD software). The bed expan-
sion, bubble hold-up, and bubble size calculated from the 2D and
3D simulations were compared with the predictions obtained from
the Darton equation [15]. A more realistic model of physical behav-
ior for fluidization was obtained using 3D simulations. They also
indicated that 2D simulations can be used for sensitivity analyses.
Xie et al. [10] compared the results of 2D and 3D simulations of
slugging, bubbling, and turbulent gas–solid fluidized beds. They
also investigated the effect of using different coordinate systems.
Their results show that there is a significant difference between
2D and 3D simulations, and only 3D simulations can predict the
correct bed height and pressure spectra. Li et al. [12] conducted a
three-dimensional numerical simulation of a single horizontal
gas jet into a laboratory-scale cylindrical gas–solid fluidized bed.
They proposed a scaled drag model and implemented it into the
simulation of a fluidized bed of FCC (Fluid Catalytic Cracking) par-
ticles. They also obtained the jet penetration lengths for different
jet velocities and compared them with published experimental
data, as well as with predictions of empirical correlations. Zhang
et al. [16] suggested a mathematical model based on the two-fluid
theory to simulate both homogeneous fluidization of Geldart A
particles and bubbling fluidization of Geldart B particles in a
three-dimensional gas–solid fluidized bed. The usage of their mod-
el is easy since it does not include adjustable parameters. It is capa-
ble of predicting the fluidization behavior leading to similar results
as the more complex Eulerian–Eulerian models.
Li and Kuipers [17] studied the formation and evolution of flow
structures in dense gas-fluidized beds with ideal collisional parti-
cles (elastic and frictionless) by employing the discrete particle
method, with special focus on the effect of gas–particle interaction.
They have concluded that gas drag, or gas–solid interaction, plays a
very important role in the formation of heterogeneous flow struc-
tures in dense gas-fluidized beds with ideal and non-ideal particle–
particle collision systems. They discovered that the non-linearity of
gas drag has a ‘‘phase separation’’ function by accelerating particles
in the dense phase and decelerating particles in the dilute phase to
trigger the formation of non-homogeneous flow structures.
Goldschmidt et al. [13] investigated a two-dimensional multi-fluid
Eulerian CFD model to study the influence of the coefficient of
restitution on the hydrodynamics of a dense gas–solid fluidized
 E. Esmaili, N. Mahinpey / Advances in Engineering Software 42 (2011) 375–386
beds. They demonstrated that, in order to obtain reasonable bed
dynamics from fundamental hydrodynamic models, it is signifi-
cantly important to take the effect of energy dissipation due to
non-ideal particle–particle encounters into account.
A few works in the literature have investigated the effect of
using different drag models in 3D simulation of fluidized beds to
obtain an optimum drag model for simulation of bubbling gas–so-
lid fluidized beds. Therefore, the underlying objective of this study
is to present an optimum drag model to simulate the momentum
transfer between phases and to compare the results of 3D and
2D simulations of gas–solid bubbling fluidized beds. Furthermore,
a method has been proposed to adjust the Di Felice Drag Model
[18] based on the experimental value of minimum fluidization
velocity as the calibration point. The effect of restitution coefficient
on the three dimensional prediction of bed height is also investi-
gated and an optimum value of restitution coefficient for modeling
fluidized beds in bubbling regime has been proposed.
2. Experimental setup
Experiments were carried out in the Department of Chemical
and Biological Engineering at the University of British Columbia.
The fluid bed is a Plexiglas rectangular shape column consisting
of spherical glass beads with ambient air as the gas phase. The
column dimensions are 0.280 (m) in width, 1.2 (m) in length,
and 0.0254 (m) in depth. Ambient air is uniformly injected into
the column via a gas distributor which is a perforated plate with
a hole to plate cross sectional area ratio of approximately 1.2%.
Pressure drops were measured using three differential pressure
transducers located at the elevations of 0.03, 0.3 and 0.6 (m) above
Fig. 1. Geometry of 3D Plexiglas fluidized bed.
377
the gas distributor. Fig. 1 illustrates the shape of the column used
in this research, along with its dimensions and pressure transducer
locations. Spherical, non-porous glass beads, Geldart group B parti-
cles, with a particle size distribution of 250–300 (lm) and density
of 2500 (kg/m3) were used as the granular parts. The static bed
height is 0.4 (m) with a solid volume fraction of approximately
60%. Several experiments were conducted at steady-state bed
operations in order to calculate the void fraction and minimum
fluidization velocity. In order to estimate the minimum fluidization
velocity, measurements were carried out at increasing velocity
increments from fixed bed to high inlet velocity (0.6 (m/s)). From
the data obtained, minimum fluidization velocity is estimated as
Umf = 0.065 (m/s).
3. Hydrodynamic model
In this study the general model of multiphase flow based on
Eulerian–Eulerian approach has been derived. The model solves
sets of transport equation for momentum and continuity of each
phase and granular temperature for the solid phase. These sets of
equations are linked together through pressure and interphase
momentum transfer correlations (drag models). The solid phase
properties have been obtained using the kinetic theory of granular
flow.
3.1. Continuity equation
The continuity equation in absence of mass transfer between
phases is given for each phase by:
378 E. Esmaili, N. Mahinpey / Advances in Engineering Software 42 (2011) 375–386

   
@ 2
ðeg qg Þ þ r  ðeg qg ~V g Þ ¼ 0; ð1Þ ss ¼ es ls r~
V s þ ðr~ V sI;
V s ÞT þ es ks þ ls r  ~ ð9Þ
@t 3
@
ðes qs Þ þ r  ðes qs ~
V s Þ ¼ 0: ð2Þ where ks is the granular bulk viscosity that is the resistance of gran-
@t
ular particles to compression or expansion. The following model is
And the volume fraction constraint requires eg + es = 1. developed from the kinetic theory of granular flow by Lun et al. [19]
where e, q, and ~V are the volume fraction, the density and the for ks:
instantaneous velocity, respectively. By considering the mass
rffiffiffiffiffiffi
transfer between the phases, the term ðm _ gs  m
_ sg Þ would then be 4 Hs
added to the right hand side of the above equations, where, m _ is ks ¼ es qs ds ð1 þ eÞ ; ð10Þ
5 p
the rate of mass transfer between phases.
where ds is the particle diameter.
3.2. Gas phase momentum equation In the solid stress tensor equation ls is the granular shear vis-
cosity that consists of a collision term, a kinetic term, and a friction
Assuming no mass transfer between phases and no lift and vir- term:
tual mass forces, the conservation of momentum for the gas phase ls ¼ ls;col þ ls;kin þ ls;fric : ð11Þ
can be expressed as:
The collisional viscosity is a viscosity contribution due to collisions
@ between particles and has the highest contribution in the viscous
ðeg qg ~
V g Þ þ r  ðeg qg ~
Vg~ g  eg rP þ eg q g þ K sg ð~
VgÞ ¼ r  s g Vs  ~
V g Þ;
@t regime. The corresponding correlation is taken from the kinetic the-
ð3Þ ory of granular flow by Lun et al. [19].
rffiffiffiffiffiffi
where P is the pressure, g is the gravity and Ksg is the drag coeffi- 4 Hs
cient between the gas and the solid phase which will be explained ls;col ¼ es qs ds ð1 þ eÞ : ð12Þ
5 p
g is given by:
in detail in Section 3.5. The gas stress tensor s
  The kinetic viscosity is expressed by Gidaspow model [22,23] as:
  2
sg ¼ eg lg r~
V g þ ðr~ V gI:
V g ÞT þ eg kg þ lg r  ~ ð4Þ 2ldil

4
2
3 ls;kin ¼ 1 þ ð1 þ eÞes g 0 ; ð13Þ
g 0 ð1 þ eÞ 5

3.3. Solid phase momentum equation

ldil ¼ ðconstantÞ  ðbulk densityÞ  ðmean free pathÞ
Assuming no mass transfer between phases and no lift and vir-  ðosccillation velocityÞ
tual mass forces, the conservation of momentum for the solid pffiffiffiffi  pffiffiffiffiffiffi
phase can be expressed as: 5 p ds
ldil ¼ ðes qs Þ Hs : ð14Þ
96 es
@
ðes q s ~
V s Þ þ r  ðes qs ~
V s~ s  rP s þ es rP þ es q g þ K sg ð~
V sÞ ¼ r  s s Vs  ~V g Þ;
@t The Schaeffer expression [24] for the frictional viscosity can be
ð5Þ
written as
@ ~ ~ ~ ~ ~
ðes qs V s Þ þ rðes qs V s V s Þ ¼ r  ss  rP s þ es rP þ es qs g þ K sg ðV s  V g Þ;

@t Ps;fric sin b
ð6Þ ls;fric ¼ pffiffiffiffiffiffi ; ð15Þ
2 I2D
where Ps is the granular pressure, derived from the kinetic theory of where Ps,fric is the frictional pressure, the constant b = 28.5° [25] is
granular flow, and is composed of a kinetic term and a term due to the angel of internal friction and I2D is the second invariant of the
particle collisions. In the regions where the particle volume fraction deviatoric stress tensor which can be written as
es is lower than the maximum allowed fraction es,max, the solid pres-
sure is calculated independently and is used in the pressure gradi- 1
I2D ¼ ½ðDs11  Ds22 Þ2 þ ðDs22  Ds33 Þ2 þ ðDs33  Ds11 Þ2 
ent term rPs It can be expressed as (Lun et al. [19]): 6
þ D2s12 þ D2s23 þ D2s31 ; ð16Þ
Ps ¼ es qs Hs þ 2qs ð1 þ eÞe2s g 0 Hs ; ð7Þ
 
where Hs is the granular temperature; e is the restitution coeffi- 1 @us;i @us;j
Dsij ¼ þ : ð17Þ
cient of granular particles and g0 is the radial distribution function. 2 @xj @xi
Different values for the coefficient of restitution, from 0.73 to 1,
have been proposed in literature. In this study the effect of restitu- Johnson et al. [26] made a simple algebraic expression for the solid
tion coefficient on the simulation of bubbling fluidized bed has been pressure in the frictional region:
investigated in order to obtain an optimum value for the entire
range of study. The results are presented in Section 5.3. For the ra- ðes  es;min Þn
Ps;fr ¼ Fr ; ð18Þ
dial distribution function, g0, the following correlation has been ðes;max  es Þp
proposed by Ibdir and Arastoopour [20] and it is well related to Fr ¼ 0:1es : ð19Þ
the data from the molecular simulator by Alder and Wainwright
[21]. In which es,min = 0.5, n = 2, and p = 3 are all experimental based
" parameters.
 1 #1
3 es 3
g0 ¼ 1 : ð8Þ 3.4. Kinetic theory of granular flow (KTGF)
5 es;max
In momentum equation, s
s is the solid stress tensor and can be The transport equation for granular temperature of solid phase
written as: Hs can be written as:
 E. Esmaili, N. Mahinpey / Advances in Engineering Software 42 (2011) 375–386 379

  
3 @ where
ðes qs Hs Þ þ r  es qs ~
V s Hs s : r~
¼s V s  r  qs  cHs  3K sg Hs ;
2 @t
Ret ¼ mr;s Rets ; ð25Þ
ð20Þ
A þ 0:06B  Rets
where ss , qs and c are the solid stress tensor, flux of fluctuating mr;s ¼ ; ð26Þ
Hs 1 þ 0:06Rets
energy and collisional energy dissipation respectively. 0rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffiffiffi 12
qs can be written as: 23:04 þ 2:52 4Ar
 4:8
B 3 C
qs ¼ kHs rHs ; ð21Þ Rets ¼ B@
C ;
A ð27Þ
1:26
where kHs is the granular conductivity of granular temperature and
3
the corresponding correlation based on Gidapow model [22] is ðqs  qg Þds qg~
g
given by: Ar ¼ ; ð28Þ
lg
pffiffiffiffiffiffiffiffiffiffi  2 rffiffiffiffiffiffi
150ds qs Hs p 6 Hs Ret, is the Reynolds number under terminal settling conditions for
kHs ¼ 1 þ ð1 þ eÞeg g 0 þ 2ds qs e2s ð1 þ eÞg 0 :
384ð1 þ eÞg 0 5 p the multi particle system, vr,s is the terminal velocity, Rets is the
ð22Þ Reynolds number under terminal settling conditions for the single
The algebraic equation for the collisional energy dissipation, cHs , is particle and Ar is the Archimedes number.
derived by Lun et al. [19] as follow: Also, the parameter C1 in Syamlal–O’Brien equation needs to be
qffiffiffiffiffiffiffi adopted in order to guarantee the continuity of velocity voidage
12ð1  e2 Þg 0 correlation as follows [37]:
cHs ¼ pffiffiffiffi qs e2s H2s : ð23Þ
ds p
logðC 2 Þ
When the restitution coefficient, e goes to 1, the dissipation of the C 1 ¼ 1:28 þ : ð29Þ
granular temperature goes to zero. This means that the particles logð0:85Þ
are perfectly elastic [19]. Using the same concept, a method has been proposed to modify the
drag model presented by Di Felice [18]. At minimum fluidization
3.5. Drag models condition, neglecting the gas-wall friction and the solid stress trans-
mitted by particles, the momentum balance can be written as:
The drag force between the gas phase and the particles is one of Buoyancy Force = Drag Force
the dominant forces in a fluidized bed. Generally, drag coefficients,
Ksg, are obtained from two types of experimental data. The first type K sg ~
es ðqs  qg Þg ¼ jV s  ~
V g j: ð30Þ
is for the high value of the solid volume fractions or packed-bed eg
pressure drop data, such as the Ergun drag model [27]. These types
of correlations require a complementary drag model for low values Considering the fact that at minimum fluidization condition ~
Vs ¼ 0
of the solid volume fractions, like the Gidaspow drag model [22,23]. V g ¼ U experiment
and ~ mf , the Eq. (29) can be reduced to:
In the second class of data, the terminal velocity of particles in flu- K sg
idized or settling beds is employed to derive the drag model as a es;mf ðqs  qg Þg ¼ U experiment
mf : ð31Þ
eg;mf
function of void fraction and Reynolds number. An example for this
category is the Richardson and Zaki model [28]. Substituting the Di Felice drag correlation into Eq. (30) and utilizing
In this paper, eleven widely used drag models that have been the least square method as a non-linear optimization algorithm, the
reported in the literature are investigated for the modeling of a drag model parameters P and Q in Di Felice drag correlations will be
3D fluidized bed. The corresponding correlations for each drag modified for the system under study using experimental data at
model are summarized in Table 1. minimum fluidization condition U experiment
mf ¼ 0:065 ðm=sÞ. When
adjusting the drag models it should be kept in mind that the adjust-
3.5.1. Adjustment of drag coefficient ment should not alter the behavior of the drag correlation when
In all drag correlations, the drag force depends on the local rel- voidage approaches one. Most drag correlations are formulated
ative velocity between phases and the void fraction. However, in such that in that limit, the single sphere drag coefficient, CD, can
deriving such general empirical drag correlations some other fac- be recovered.
tors, such as particle size distribution and particle shape have
not been considered. Also, void fraction dependency is very diffi- 4. Numerical simulation
cult to be determined for any condition other than a packed bed
or infinite dilution (single particle). On the other hand, most Governing equations of mass and momentum conservation as
researchers have information on the minimum fluidization veloc- well as the granular temperature equation are solved using finite
ity of their own material. In this respect, Syamlal and O’Brien volume method employing the Phase-Coupled Semi Implicit Meth-
[37] introduced a method to modify their original drag law using od for Pressure Linked Equations (PC-SIMPLE) algorithm, which is
minimum fluidization velocity, commonly available experimental an extension of the SIMPLE algorithm to multiphase flow. A mul-
information for the specific material. ti-fluid Eulerian–Eulerian model, which considers the conservation
The parameter C2 in Syamlal–O’Brien drag equation is related to of mass and momentum for each phase, has been applied. The ki-
the minimum fluidization velocity through the velocity voidage netic theory of granular flow, which considers the conservation of
correlation and the terminal Reynolds number, Ret [37] and is solid fluctuation energy, was used for closure of the solid stress
changed until the following criterion is met: terms. The three-dimensional (3D) geometry has been meshed
( )
eg lg Minimize
using 336,000 structured rectangular cells. Volume fraction, den-
Objective function : U experiment
mf  Rets  ! 0; sity, and pressure are stored at the main grid points that are placed
qg ds
in the center of each control volume. A staggered grid arrangement
is used, and the velocity components are solved at the control vol-
eg lg
mg ¼ Rets  ¼ U experiment
mf ; ð24Þ ume surfaces. Fig. 2 shows a schematic view of the staggered grid
qg ds cells for velocity components and pressure.
380 E. Esmaili, N. Mahinpey / Advances in Engineering Software 42 (2011) 375–386

Table 1 ( pffiffiffiffiffiffiffiffiffiffi
2=es =40; 0:01 < es < 0:1
Summary of drag coefficient correlations. F1 ¼
0:11 þ 0:00051eð11:6es Þ ; es > 0:4
8  pffiffiffiffiffiffiffi 
1. Richardon and Zaki [28] (1954) >
< ð1  wÞ 1þ3 es =2þð135=64Þes2lnðes Þþ17:89 es
3 q eg es 3
1þ0:681es 11:03es þ15:41es
þ w½10es =ð1  es Þ3 ; es < 0:4
K sg ¼ 4dg v 2 C D j~
Vs  ~
Vgj F2 ¼
s r >
: 10es 3 ; es P 0:4
vr ¼ e n1
g
ð1es Þ
8 0:9351es þ 0:03667; es < 0:0953
>
> 4:65; Rem < 0:2 F3 ¼
< 4:4Re0:03 ; 0:2 > Rem < 1 0:0673 þ 0:212es þ 0:0232=ð1  es Þ5 ; es P 0:0953
m
n¼ w ¼ eð10ð0:4es Þ=es Þ
>
> 4:4Re0:1
m ; 1 > Rem < 500
:
2:4; Rem > 500
10. Zhang and Reese drag model [36] (2003)
Rem ¼ Re s 8 2
vr < 150 es lg þ 1:75 es qg ~ V r ; eg 6 0:8
q g d s j~
V s ~
Vg j K sg ¼ eg d2s ds
Res ¼ lg : 3 C es qg ~ 2:65
4 D ds
V r eg ; eg P 0:8
h i0:5
2. Wen–Yu drag model [29] (1966) ~ ~ ~ 2
V r ¼ ðV s  V g Þ þ 8Hs =p
3q e ð1eg Þ
K sg ¼ g g C D j~
4ds
Vs  ~
V g je2:65 g
pffiffiffiffiffiffiffi
C D ¼ ð0:28 þ 6= Res þ 21=Res Þ
0:687
C D ¼ eg24
Res ½1 þ 0:15ðeg Res Þ  qg ds ~
Vr
Res ¼ lg
q d s j~
V s ~
Vg j
Res ¼ g l
g

3. Gibilaro drag model [30] (1983, 1985)

h i
es q g ~ ~ 1:8
K sg ¼ 17:3
Res þ 0:336 ds jV s  V g jeg
qg eg ds j~
V s ~
Vg j
Res ¼ 2lg

4. Gidaspow drag model [31] (1986)

K sg ¼ ð1  usg ÞK Ergun
sg þ usg K WenYu
sg
e2s lg es q
K Ergun
sg ¼ 150 þ 1:75 ds g j~
Vs  ~
V g j; eg 6 0:8
eg d2s
WenYu 3 es q g ~ ~
K sg ¼ 4 C D d jV s  V g jeg2:65 ; eg P 0:8
s
(
24 0:687
CD ¼ eg Res ½1 þ 0:15ðeg Res Þ ; Res < 1000
0:44; Res P 1000
q g d s j~
V s ~
Vg j
Res ¼ lg
es Þ
usg ¼ Arctan½1501:75ð0:2
p þ 0:5

5. Syamlal–O0 Brien drag model [32] (1988)

3e e q
K sg ¼ s g 2 g C D j~
4ds v r
Vs  ~
Vgj
& ’2
4:8ffiffiffi
C D ¼ 0:63 þ p Re
vr
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
v r ¼ 12 ½A  0:06Re þ ð0:06ReÞ2 þ 0:12Reð2B  AÞ þ A2 
A ¼ e4:14
g
8 1:28
< C 2 eg ; eg < 0:85
B ¼ eCg 1 ; eg P 0:85
:
C 1 ¼ 2:65; C 2 ¼ 0:8
6. Arastoopour drag model [33] (1990)
l m
K sg ¼ 17:3
es q g ~ ~ 2:8 Fig. 2. Schematic view of staggered grid, volume fractions are stored at the main
Res þ 0:336  ds jV s  V g jeg
grid points (P) while the velocity components at control volume surfaces.
q d j~
V ~V j
Res ¼ g s l s g
g

7. RUC-drag model [34] (1994)

lg ð1eg Þ2 q ð1e Þ A pressure correction equation is built based on total volume
K sg ¼ A þ B g ds g j~ Vs  ~
Vgj
eg d2s
3
26:8eg continuity. Pressure and velocities are then corrected so as to sat-
A¼ 2 1
ð1eg Þ3 ð1ð1eg Þ3 Þð1ð1eg Þ3 Þ2
2
isfy the continuity constraint. A grid sensitivity analysis is per-
e2g
B¼ 2
formed using different mesh sizes and 5 mm mesh interval
ð1ð1eg Þ3 Þ2
8. Di Felice drag model [18] (1994)
spacing was chosen for all the simulation runs. The detailed results
eq
K sg ¼ 34 C D sds g j~
Vs  ~
V g jf ðes Þ for sensitivity analysis have been discussed in Section 5.4. Second-
f(es) = (1  es)x order upwind discretization scheme was used for discretizing the
h i
x = P – Q  exp ð1:5bÞ
2
governing equations. An adaptive time-stepping algorithm with
2
b = log(Res)
100 iterations per each time step and a minimum value of order
P = 3.7 and Q = 0.65 105 for the lower domain of time step was used to ensure a stable
9. Hill Koch Ladd drag correlation [35] (2001)
convergence. The adaptive determination of the time step size is
3q e ð1eg Þ based on the estimation of the truncation error associated with
K sg ¼ g g C D j~
4ds
Vs  ~
Vgj
12e2g
the time integration scheme (i.e., first-order implicit or second-
CD ¼ F
Res order implicit). If the truncation error is smaller than a specified
q d e j~
V ~
Res ¼ g s 2g l s g
V j tolerance, the size of the time step is increased; if the truncation
g
8 error is greater, the time step size is decreased. The convergence
< F ¼ 1 þ 3=8Res ;
> es 6 0:01 and Res 6 ðF 2  q
1Þ=ð3=8  F3Þ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi criteria for other residual components associated with the relative
F ¼ F 0 þ F 1 Re2s ; es P 0:01 and Res 6 F 3 þ F 23  4F 1 ðF 0  F 2 Þ=ð2F 1 Þ
>
: error between two successive iterations has been specified in the
F ¼ F 2 þ F 3 Res ; Otherwise
8  pffiffiffiffiffiffiffi  order of 105. A detailed study has been carried out on the effect
>
< ð1  wÞ 1þ3 es =2þð135=64Þe2s lnðes Þþ17:14 es
3 þ w½10es =ð1  es Þ3 ;
1þ0:681es 8:4es þ8:16es
0:01 < es < 0:4 of restitution coefficient and the results have been presented in
F0 ¼
>
: 10es
; es > 0:4 Section 5.3. Including the adjusted drag model cases, 12 different
ð1es Þ3
 E. Esmaili, N. Mahinpey / Advances in Engineering Software 42 (2011) 375–386 381

Table 2 lal–O’Brien model and the parameters P and Q to 5.2 and 0.31 for
Computational model parameters. the Di Felice model. The associated parameters of the models were
Parameter Value estimated by adopting the model to experimental data using
Particle density 2500 (kg/m3) non-linear parameter estimation analysis.
Gas density 1.225 (kg/m3) Fig. 3 shows a snapshot of solid volume fraction contours for the
Mean particle diameter 275 lm twelve drag models studied in this work at a superficial gas veloc-
Initial solid packing 0.6 ity of 0:21 ðm=sÞ and after 10 (s) real-time simulations. In this fig-
Superficial gas velocity 11.7, 21, 38, 46(cm/s)
Bed dimension 0.28 (m)  1.2 (m)  0.025 (m)
ure, comparison between all drag models and experimental
Static bed height 0.4 (m) snapshot has been made in terms of bed height and bubble size
Grid interval spacing 0.002 (m) and shape. It can be readily observed that the two adjusted models
Inlet boundary condition type Inlet – velocity (i.e., Di Felice adjusted model and Syamlal–O’Brien adjusted model
Outlet boundary condition type Pressure – outlet
[37]) show the best results simulating the bed height. The adjusted
Under-relaxation factors Pressure 0.6 Di Felice model is more accurate in the prediction of the bubble
Momentum 0.4
shapes and fluctuating behavior of the free surface of the bed. It
Volume fraction 0.3
Granular temperature 0.2 can be seen that the original Syamlal–O’Brien model [32] repre-
sents the lowest bed expansion and gas void fraction. This fact
could have been foreseen from the minimum fluidization velocity
drag models are studied in this work to simulate the momentum prediction by this model, which is almost six times larger than
transfer between the phases (Richardon and Zaki [28], Wen–Yu experimental data [38]. Expansion of the bed started with forma-
drag model [29], Gibilaro drag model [30], Gidaspow drag model tion of bubbles for all the models and eventually reached a statis-
[31], Syamlal–O’Brien drag model [32], Arastoopour drag model tically steady-state bed height. After this point, an unsteady
[33], RUC-drag model [34], Di Felice drag model [18], Hill Koch chaotic generation of bubbles was observed after almost 3 (s) of
Ladd drag correlation [35], Zhang and Reese drag model [36], ad- real-time simulation. Disregarding the two adjusted drag model,
justed Syamlal [37], adjusted Di Felice drag model). The drag mod- Fig. 3 shows that the original Di Felice [18] and Gibilaro [30] drag
els available in Fluent 6.3 suited for a fluidized bed simulation is models have produced better results in predicting the bed expan-
the Gidaspow model, the Syamlal–O0 Brien model and the Wen– sion among other drag models. The drag model proposed by
Yu drag model. For the other nine drag models, specific User De- Richardson–Zaki [28] has given the worst results with respect to
fined Functions (UDF) in C++ have been implemented and up- bubble shapes since it shows symmetry in contours of solid vol-
loaded into the software. FLUENT 6.3 on a 20 AMD/Opteron 64bit ume fraction after 10 (s) real time which is not reasonable. The rest
processor Sun Grid Microsystems workstation W2100Z with 4 GB of the models showed approximately the same range of bed expan-
RAM is employed to solve the governing equations. Computational sion. There also exists a more recent correlation which is based on
model parameters are listed in Table 2. extensive lattice Boltzmann simulations by van der Hoef et al. [39],
and Beetstra et al. [40]. They have proposed expressions for nor-
malized drag force for both mono-dispersed and poly-dispersed
5. Result and discussion
systems. Their results found to be in excellent agreement (devia-
tion smaller than 3%) with the simulation data of several models
CFD modeling has been performed using FLUENT 6.3. Simula-
proposed in literatures.
tions have been carried out on a 3D fluidized bed using a transient
Eulerian–Eulerian model. Several superficial gas velocities, 0.11,
0.21, 0.38, and 0:46 ðm=sÞ, that correspond to 1.6, 3.2, 5.8, and 5.1. Pressure drop
7Umf, respectively, have been studied. In the following section,
the simulation results have been compared with the experimental Fig. 4 shows the time average pressure drop inside the bed be-
data in order to validate the model. As previously discussed, sev- tween two specific elevations (i.e. 0.03 (m) and 0.3 (m) as demon-
eral drag models have been proposed in the literature to model strated in Fig. 1) for different studied cases and experimental
the momentum transfer between the phases. In the present work, results. In order to calculate the average pressure at each pressure
a complete study has been performed on all those drag models and sensor (i.e. y = 0.03 (m)), both spatial and time averaging have been
finally a method has been developed to adjust the drag model pro- applied. At first, the spatial averaging, which is the average value of
posed by Di Felice [18]. The adjustment of the two drag models pressure for all nodes in the plane of first pressure sensor (plane
with the methods discussed earlier was taken into account by opti- y = 0.03 (m)) has been utilized. Subsequently, the time averaging
mizing the parameters C1 and C2 to 11.772 and 0.182 for the Syam- of spatial-averaged pressure values in the period of 3–10 (s) real

(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) (l) (m)
Fig. 3. Contours of solid volume fraction (U = 0.21 (m/s) t = 10 (s)): (a) experiment; (b) Syamlal–O’Brien adjusted; (c) Syamlal–O’Brien; (d) Arastoopour; (e) Gibilaro; (f) Hill
Koch Ladd; (g) Zhang–Reese; (h) Richardson–Zaki; (i) RUC; (j) Di Felice adjusted; (k) Di Felice; (l) Wen–Yu and (m) Gidaspow.
382 E. Esmaili, N. Mahinpey / Advances in Engineering Software 42 (2011) 375–386

Fig. 4. Pressure drop inside the bed ðDP 1 ¼ P z¼0:03m  P z¼0:6m Þ.

time has been incorporated. As indicated in Fig. 4, the pressure
drop for all the models showed a declining trend with increase of
the superficial gas velocity, providing good qualitative agreement
with the experimental data. It can be seen that the adjusted Di
Felice model gives closer result to experimental data and for all
four superficial gas velocity shows significant improve in result
compared to original Di Felice model [18]. As shown in Fig. 4 the
adjusted model based on Syamlal–O’brien [37] study also gives
acceptable results especially in lower superficial gas velocities.
Fig. 5 shows a comparison between all drag models in predic-
tion of overall bed expansion ratio (i.e., DP2 as indicated in
Fig. 1) with respect to superficial gas velocity. It can be seen that
the overall pressure drop for all drag models does not change too
much by increasing the superficial gas velocity. This is in good
agreement with both theoretical and experimental predictions, ex-
cept for the highest velocity, where the deviation may be due to
the fact that at high velocities of gas, the elevation 0.6 m above
the distributor is actually inside the bed for the case of the adjusted
Syamlal–O’Brien drag model [37]. Hence, it, in fact, it represents
the pressure drop between two elevations inside the bed [38].
Fig. 6 shows a comparison between 2D vs. 3D simulation of flu-
idized bed using three different drag models (adjusted Di Felice
and Syamlal–O’Brien [37] drag model and original Di Felice [18]
drag model). It can be seen that the pressure drop for both 2D
and 3D simulation shows a declining trend by increasing the
superficial gas velocity which is in good qualitative agreement
with the experimental data. However, 3D simulations show their
superiority in predicting the pressure drop inside the bed com-
pared to 2D simulations. The reason can be the effect of participat-
ing governing equations of the z direction (depth of the bed) in
Navier Stokes equation of multiphase flow. It can be concluded that
although three-dimensional simulation takes more time and

Fig. 5. Overall Pressure drop ðDP 2 ¼ P z¼0:03m  P z¼0:6m Þ.

 E. Esmaili, N. Mahinpey / Advances in Engineering Software 42 (2011) 375–386 383

Fig. 6. 2D vs. 3D simulation of pressure drop inside the bed ðDP 1 ¼ P z¼0:03m  P z¼0:3m Þ.

computing processors than two-dimensional simulation, it gives
more accurate results when the models are compared with exper-
imental data.
5.2. Bed expansion ratio
The experimental data of the time-average bed expansion ratio
were compared with corresponding values predicted by different
drag models for various superficial gas velocities as depicted in
Fig. 7. For this series of simulations, a static bed height of
H0 = 0.4 (m) over a range of superficial velocities 11.7, 21, 38,
and 46 ðcm=s was used. All drag models demonstrate a consistent
increase in bed expansion with gas velocity and predict the bed
expansion reasonably well. Fig. 7 shows the considerable relative
increase in bed expansion as the fluidizing velocity increases; a
5% increase was obtained at 0.11 m/s, a 20% increase at 0.21 m/s,
42% at 0.38 m/s, and up to a 50% increase in bed height was
measured at 0.46 m/s, the highest fluidized velocity investigated.
It can be seen that using adjusted Di Felice drag model, the bed
expansion ratio can be predicted fairly accurately over a whole
range of superficial gas velocities compared to experimental data.
All the available drag correlations with the exception of two ad-
justed drag models (i.e., Di Felice and Syamlal–O’Brien [37]) at high
superficial gas velocity (0.46 m/s), underestimate the bed expan-
sion. The adjusted Syamlal–O’Brien drag model showed good
agreement with experimental results only up to a moderate range
of gas velocity. Fig. 7 shows that for higher superficial gas veloci-
ties, the adjusted Syamlal–O’Brien drag model comparatively over-
estimated the bed expansion ratio. Fig. 7 also reveals that even the
original Di Felice [18] drag model gives the best result for the pre-
diction of bed expansion ratio among all the other conventional
drag laws. This fact vindicates the claim that this drag model
was opted for adjustment based on minimum fluidization velocity
[38].

Fig. 7. Comparison of simulated bed expansion ratio with experimental data.

384 E. Esmaili, N. Mahinpey / Advances in Engineering Software 42 (2011) 375–386
(a) (b) (c) (d)
Fig. 8. Contours of solid volume fraction (t = 10 (s), Di Felice adjusted drag model):
(a) U = 0.117 (m/s); (b) U = 0.21 (m/s); (c) U = 0.38 (m/s) and (d) U = 0.46 (m/s).
Fig. 8 shows the predicted contours of solid volume fraction at
t = 10 (s) using adjusted Di Felice drag model for four different
superficial gas velocities. It can be easily seen that by increasing
the gas velocity the bigger bubbles will be generated inside the
bed and as a result the bed height will increase significantly. By
further increasing the superficial gas velocity, the hydrodynamic
regime of the fluid flow inside the bed will transfer from bubbling
regime to slugging regime.
5.3. Effect of restitution coefficient
The restitution coefficient, e specifies the coefficient of restitu-
tion for collisions between solid particles. The restitution coeffi-
cient compensates for the collisions to be inelastic. In a
completely elastic collision the restitution coefficient will be equal
to one. Fig. 9 shows a snapshot of solid volume fraction contours at
the superficial gas velocity of U = 0.21(m/s) and t = 10 (s) using ad-
Experiment
Fig. 9. Comparison between experiment and simulated bed height for various values of the coefficient of restitution (U = 0.21 (m/s) t = 10 (s), adjusted Di Felice).
Table 3
Grid size sensitivity results.
Mesh DP 1 DP 2 Mean solid volume Simulation time for
spacing (kPa) (kPa) fraction at z = 0.2 (m) 10 (s) real time (h)
(mm)
2 2.945 5.18 0.55 300
4 2.964 5.10 0.54 148
5 2.975 5.05 0.54 52
justed Di Felice drag model for seven different restitution coeffi-
cients proposed for simulation of fluidized beds in literature. A
comparison between different values of restitution coefficient
and experiment in terms of bed height and bubble size and shape
is shown in Fig. 9. It can be seen that as collisions become less ideal
(and more energy is dissipated due to inelastic collisions) particles
become closely packed in the densest regions of the bed, resulting
in sharper porosity contours and larger bubbles [13]. In this study
the value of e = 0.92 for the coefficient of restitution has been used
for the whole simulation which seems to be in good agreement
with experiment in terms of bubble shape and bed height.
5.4. Mesh size sensitivity analysis
Wang et al. [41] concluded that in order to obtain correct bed
expansion characteristics, the grid size should be of the order of
three particle diameters which requires smaller grid size and high-
er computer resources.
A mesh size sensitivity analysis has been carried out to study
the effect of grid size resolution on the results predicted by numer-
ical simulation. In this respect, the geometry of the fluidized bed
has been meshed using three distinctive grid intervals of 2, 4,
and 5 (mm) to simulate the hydrodynamic behavior of the bed.
The adjusted Di Felice drag model has been chosen for modeling
the momentum transfer between the phases in sensitivity analysis
simulations. All the simulations performed at superficial gas veloc-
ity of U ¼ 0:21 ðm=sÞ. Table 3 shows the predication of pressure
drop inside and across the bed, DP1 and DP2, respectively. Predic-
tion of time mean average solid volume fraction at bed elevation of
Z = 0.2 (m) also was checked. The time mean average was calcu-
lated on the real-time simulation interval of 2–10 s to ensure that
statistical steady state behavior inside the bed was attained [38]. It
can be easily observed that the results did not show any notewor-
thy dissimilarity in fluid dynamics behavior of the beds. Table 2
also compares the time required for 10 s of real-time simulation.
 E. Esmaili, N. Mahinpey / Advances in Engineering Software 42 (2011) 375–386
(a) (b) (c)
Fig. 10. Contours of solid volume fraction (U = 0.21 (m/s), t = 6 (s), Di Felice
adjusted drag model): (a) mesh size 2 mm; (b) mesh size 4 mm and (c) mesh size
5 mm.
It can be seen that the required time for simulating 10 s of 3D flu-
idized bed drastically increases from 52 h to almost 2 weeks for a
decrease in grid interval spacing from 5 to 2 mm, respectively.
Therefore, the mesh interval size of 5 mm has been chosen for
the rest of simulation to obtain reasonable time efficiency without
losing the accuracy of results. Fig. 10 also shows the contours of so-
lid volume fraction for three different mesh size resolutions. Here-
in, similarities of bed expansion and bubble shapes among the
simulations can be easily appreciated. The above results indicate
that the grid size spacing selected for simulation in this work
(i.e., 5 mm) was adequate for satisfactory prediction of the hydro-
dynamics in computational geometry.
6. Conclusion
Numerical simulation of a bubbling gas–solid fluidized bed
were performed in a three dimensional solution domain using
the Eulerian–Eulerian approach to investigate the effect of using
different drag correlations for modeling the momentum transfer
between phases. The drag models of Richardon and Zaki,
Wen–Yu, Gibilaro, Gidaspow, Syamlal–O’Brien, Arastoopour, RUC,
Di Felice, Hill Koch Ladd, Zhang and Reese, and adjusted Syamlal
were reviewed and a method proposed for adjusting the original
Di Felice darg model in a three dimensional domain based on the
experimental minimum fluidization conditions. In this respect,
FLUENT 6.3 was used to perform the calculations while the drag
correlations have been implemented in C++ and uploaded in
FLUENT as User Defined Functions (UDF).The results have been
compared to experimental data in terms of pressure drop and
bed expansion ratio. It is concluded that the adjusted Di Felice
model predicts the hydrodynamic behavior of fluidized bed more
accurately that all other drag models. The effect of using three-
dimensional analysis vs. two-dimensional simulation of fluidized
385
beds is also investigated. The results show that although
three-dimensional simulation takes more time and computing pro-
cessors than two-dimensional simulation, it gives more accurate
results when the models are compared with experimental data.
Finally, sensitivity analysis was carried out to investigate the effect
of using various restitution coefficients as well as the different grid
interval spacing on the results. Further modeling efforts are
required to study the influence of other parameters such as gas
distributors, and also, the effect of particle size distribution which
has been underestimated using the mean particle diameter. More-
over, new experimental studies should be carried out using recent
advancements in instrumentation engineering in order to resolve
the available experimental discrepancies reported in the literature
such as void fraction measurements, and bed expansion ratio.
References
[1] Hoomans BPB, Kuipers JAM, Briels WJ, Swaaij VWPM. Discrete particle
simulation of bubble and slug formulation in a two-dimensional gas-
fluidized bed: a hard-sphere approach. Chem Eng Sci 1996;51:99.
[2] Xu B, Yu A. Numerical simulation of the gas–solid flow in a fluidized bed by
combining discrete particle method with computational fluid dynamics. Chem
Eng Sci 1997;52:2785.
[3] Kafui KD, Thornton C, Adams MJ. Discrete particle-continuum fluid modeling
of gas–solid fluidized beds. Chem Eng Sci 2002;57:2395.
[4] Goldschmidt MJV, Beetstra R, Kuipers JAM. Hydrodynamic modeling of dense
gas-fluidised beds: comparison and validation of 3d discrete particle and
continuum models. Powder Technol 2004;142:23.
[5] Bokkers GA, van Sint Annaland M, Kuipers JAM. Mixing and segregation in a
bidisperse gas–solid fluidised bed: a numerical and experimental study.
Powder Technol 2004;140:176–86.
[6] van Sint Annaland M, Bokkers GA, Goldschmidt MJV, Olaofe OO, van der Hoef
MA, Kuipers JAM. Development of a multi-fluid model for poly-disperse dense
gas–solid fluidised beds, part II: segregation in binary particle mixtures. Chem
Eng Sci 2009;64:4237–46.
[7] van Wachem BGM, Schouterf JC, Krishnab R, van den Bleek CM. Eulerian
simulations of bubbling behavior in gas–solid fluidized beds. Comput Chem
Eng 1988;22(Suppl.):S299–306.
[8] Cammarata L, Lettieri P, Micale GDM, Colman D. 2d and 3d CFD simulations of
bubbling fluidized beds using Eulerian–Eulerian models. Int J Chem Reactor
Eng 2003;1. article A48.
[9] Sun J, Battaglia F. Hydrodynamic modeling of particle rotation for segregation
in bubbling gas-fluidized beds. Chem Eng Sci 2006;61:1470.
[10] Xie N, Battaglia F, Pannala S. Effects of using two-versus three-dimensional
computational modeling of fluidized beds: part I, hydrodynamics. Powder
Technol 2008;182.
[11] Behjat Y, Shahhosseini S, Hashemabadi SH. CFD modeling of hydrodynamic
and heat transfer in fluidized bed reactors. Int Commun Heat Mass Transfer
2008;35:357–68.
[12] Li T, Pougatch K, Salcudean M, Grecov D. Numerical simulation of horizontal
jet penetration in a three-dimensional fluidized bed. Powder Technol
2008;184:89–99.
[13] Goldschmidt MJV, Kuipers JAM, van Swaaij WPM. Hydrodynamic modeling of
dense gas-fluidized beds using the kinetic theory of granular flow: effect of
coefficient of restitution on bed dynamics. Chem Eng Sci 2001;56:571–8.
[14] Peiranoa E, Delloumea V, Lecknera B. Two- or three-dimensional simulations
of turbulent gas–solid flows applied to fluidization. Chem Eng Sci
2001;56:4787–99.
[15] Darton RC, LaNauZe RD, Davidson JF, Harrison D. Bubble growth due to
coalescence in fluidised beds. Trans Inst Chem Eng 1977;55:274.
[16] Zhang Kai, Brandani Stefano, Bi Jicheng, Jiang Jianchun. CFD simulation of
fluidization quality in the three-dimensional fluidized bed. Prog Nat Sci
2008;18:729–33.
[17] Li Jie, Kuipers JAM. On the origin of heterogeneous structure in dense gas–solid
flows. Chem Eng Sci 2005;60:1251–65.
[18] Di Felice R. The voidage functions for fluid–particle interaction system. Int J
Multiphase Flow 1994;20(1):153–9.
[19] Lun CKK, Savage SB, Je rey DJ, Chepurniy N. Kinetic theories for granular flow:
inelastic particles in Couette flow and slightly inelastic particles in a general
flow field. J Fluid Mech 1984;140:223–56.
[20] Ibdir H, Arastoopour H. Modeling of multi-type particle flow using kinetic
approach. AICHE J 2005.
[21] Alder BJ, Wainwright TE. Studies in molecular dynamics. II. Behavior of small
number of elastic spheres. J Chem Phys 1960;33:2363–82.
[22] Gidaspow D. Multiphase flow and fluidization-continuum and kinetic theory
descriptions. Boston: Academic Press; 1994.
[23] Ding J, Gidaspow D. A bubbling fluidization model using kinetic theory of
granular flow. AIChE J 1990;36(4):523–38.
[24] Schaeffer Ergun. Fluid flow through packed columns. Chem Eng Prog
1952;48(2):89–94.
386 E. Esmaili, N. Mahinpey / Advances in Engineering Software 42 (2011) 375–386
[25] Johnson PC, Jackson R. Frictional–collisional constitutive relations for granular
materials, with application to plane shearing. J Fluid Mech 1987;176:67–93.
[26] Johnson PC, Nott P, Jackson R. Frictional–collisional equations of motion for
particulate flows and their application to chutes. J Fluid Mech 1990;210:
501–35.
[27] Ergun S. Fluid flow through packed columns. Chem Eng Prog 1952;48(2):
89–94.
[28] Richardson JF, Zaki WN. Sedimentation and fluidization: part I. Trans Inst
Chem Eng 1954;32:35–53.
[29] Wen CY, Yu YH. Mechanics of fluidization. Chem Eng Prog Symp Ser 1966:
100–11.
[30] Gibilaro LG, Di Felice R, Waldram SP, Foscolo PU. Generalized friction factor
and drag coefficient correlations for fluid–particle interactions. Chem Eng Sci
1985;40:1817–23.
[31] Huilin Lu, Yurong He, Gidaspow Dimitri. Hydrodynamic modeling of binary
mixture in a gas bubbling fluidized bed using the kinetic theory of granular
flow. Chemical Engineering Science 2003;58:1197–205.
[32] Syamlal M, O’Brien TJ. Simulation of granular layer inversion in liquid fluidized
beds. Int J Multiphase Flow 1988;14(4):473–81.
[33] Arastoopour H, Pakdel P, Adewumi M. Hydrodynamic analysis of dilute gas–
solids flow in a vertical pipe. Powder Technol 1990;62(2):163–70.
[34] Du Plessis JP. Analytical quantification of coefficients in the Ergun equation for
fluid friction in a packed bed. Transport Porous Media 1994;16:189–207.
[35] Hill RJ, Koch DL, Ladd JC. Moderate Reynolds number flows in ordered and
random arrays of spheres. J Fluid Mech 2001;448:243–78.
[36] Zhang Y, Reese JM. The drag force in two fluid models of gas–solid flows. Chem
Eng Sci 2003;58(8):1641–4.
[37] Syamlal M, O’Brien TJ. Derivation of a drag coefficient from velocity–voidage
correlation. US Department of Energy, Office of Fossil; Energy, National Energy
Technology Laboratory, Morgantown, WV; 1987.
[38] Vejahati Farshid, Mahinpey Nader, Ellis Naoko, Nikoo Mehrdokht B. CFD
simulation of gas–solid bubbling fluidized bed: a new method for adjusting
drag law. Can J Chem Eng 2009;87:19–30.
[39] Van der Hoef MA, Beetstra R, Kuipers JAM. J Fluid Mech 2005;528:233–54;
Beetstra R, Van der Hoef MA, Kuipers JAM. AIChE J 2007;53:489–501.
[40] Beetstra R, van der Hoef MA, Kuipers JAM. Drag force of intermediate Reynolds
number flow past mono- and bidisperse arrays of spheres. AIChE J 2007;53:
489–501.
[41] Wang Junwu, van der Hoef MA, Kuipers JAM. 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 2009;64:622–5.