Powder Technology 328 (2018) 215–226

Contents lists available at ScienceDirect

Powder Technology

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

Parametric study for MP-PIC simulation of bubbling fluidized beds with

Geldart A particles
Meiyan Feng a, Fei Li a,⁎, Wei Wang a,b, Jinghai Li a,b
a
State Key Laboratory of Multi-phase Complex Systems, Institute of Process Engineering, Chinese Academy of Sciences, Beijing 100190, PR China
b
University of Chinese Academy of Sciences, Beijing 100490, PR China

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

Article history: This is a parametric study for multi-phase particle in cell (MP-PIC) simulation of bubbling fluidized beds with
Received 17 August 2017 Geldart A particles using the open source MFIX program. The main parameters have been studied including
Received in revised form 7 December 2017 drag models, grid resolution and number of particles per parcel (PPP). And the calculated axial/radial solid
Accepted 13 January 2018
distribution and bed height are compared with the experimental data for validation. It is shown that the drag
Available online xxxx
model can significantly affect the calculation results of bubbling fluidized bed with Geldart A particles. Specifical-
Keywords:
ly, the Energy Minimization Multi-scale (EMMS) bubbling drag model can predict right bubbling phenomenon
Bubbling fluidized bed and also improve the accuracy compared to the homogeneous drag model. Bubble analysis shows that there
Multiphase flow exists a stable average bubble diameter when the bed becomes stable. The average bubble circularities are
MP-PIC about 0.5 for the two bubbling bed studied in this work, even though they have different average bubble
EMMS diameter. Parameter analysis shows that the accuracy of the calculated results improves with decreasing grid
Drag model size or PPP. There exists a threshold value for grid size/PPP, below which, grid/PPP independent result can be
obtained. The PPP plays the similar role to grid resolution in MP-PIC simulation.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction the bubbles lead to heterogeneous solid distributions in the beds. So it

MP-PIC method was proposed by Andrews and O'Rourke [1]
originally in 1996. For the fluid phase, it is described with Eulerian
method; while for the particle phase, the method similar to the
coarse-grained discrete phase model (DPM) approach was used. In
this method, “parcel” is on behalf of a group of particles with the same
size, speed and density, and the Newton's laws are used to track each
“parcel” instead of single particles. So the calculation loading is greatly
reduced. This is a great advantage for the MP-PIC method, especially
in the calculation of large-scale systems. The method does not track
interparticle collision directly, but take into account this effect through
solid-phase normal stress [2, 3]. All these attractive advantages make
MP-PIC suitable for simulation of large scale gas-solid flow applications
[4]. MP-PIC method was used in 2D/3D gas-solid two phase flow simu-
lation by Snider [2] for the first time. And after that, MP-PIC was widely
used in fluidization simulations, for example, bubbling fluidized bed
[5–7], fast fluidized bed [8–10] and a lot other industrial fluidized bed
applications.
Drag model is an important parameter in simulation of bubbling
fluidized beds [11–16]. Bubbling fluidized bed flows are characterized
with gas bubbles which are also known as meso-scale structures, and
⁎ Corresponding author.
E-mail address: lifei@ipe.ac.cn (F. Li).
is necessary to account for the effects of bubbles in bubbling bed simu-
lations. Currently, there are mainly two types of drag models, namely,
the homogeneous drag model [13, 14, 17–19] and the heterogeneous
drag model [20–26]. Generally the heterogeneous drag models account
for heterogeneous solid distribution due to meso-scale structure while
homogeneous drag models assume homogeneous solid distribution.
Liang et al. [7] simulated a pseudo-2D bubbling fluidized bed with
Geldart B particles by using Barracuda® (a software for MP-PIC meth-
od), and the results showed that the currently available adjusting mea-
sures (say, time step, mesh size and parcel number, drag model, solid
volume fraction at close packing, gas turbulence model), had little ef-
fects in the simulation results except for the near-wall mesh refinement.
They had just tested the homogeneous drag models [17, 19, 27, 28].
Pugsley et al. [6] investigated the capability of Barracuda® for simulat-
ing bubbling bed flows with Geldart A particles, and the results of
bubble properties showed acceptable accuracy without modifying the
homogeneous drag force or other constitutive relationships in the
model. It should also be noted that McKeen & Pugsley [29] showed
that the homogeneous drag force coefficient had to be multiplied by a
scaling factor less than 1 to obtain the reasonable bed expansion height
in simulating bubbling fluidized bed with two-fluid model. Chen et al.
[10] simulated CFB riser with Geldart A and Geldart B particles by
using Barracuda®. The results indicated that the Wen-Yu/Ergun drag
model overestimated the momentum transfer. All the above reports

https://doi.org/10.1016/j.powtec.2018.01.024
0032-5910/© 2018 Elsevier B.V. All rights reserved.
216 M. Feng et al. / Powder Technology 328 (2018) 215–226

show that the simulations results calculated with homogeneous drag

model arouse controversy in MP-PIC method. Because there exist
heterogeneous structures in fluidizations, it is necessary to account for
the heterogeneous drag model in MP-PIC method in simulating
bubbling fluidized beds [30, 31].
EMMS drag model is one of the heterogeneous drag models. It was
developed based on meso-scale structure analysis [20, 32, 33]. In this
model, the non-uniform flow is separated into different scales. In
details, the gas-solid interaction is decomposed into the interactions
between single particles and fluid (micro-scale effect) in both the
interior of particle-rich dense phase and the particle-lean dilute phase,
and the interaction between dilute phase and dense phase (meso-
scale effect). Finally, the drag forces at different scales are summed to
obtain the heterogeneous drag force. Recently, a structure-dependent
energy dissipation analysis [34] helped better understanding of the
EMMS stability condition and drag.
The EMMS drag model was widely used in two-fluid model (TFM) to
simulate fluidizations [35–39]. Li et al. [40] introduced the EMMS drag
Fig. 1. HD calculated with EMMS/bubbling drag model for two bubbling fluidized beds
in MP-PIC method and proved that MP-PIC combined with the EMMS (bubbling bed 1: Ug = 0.1 m/s, ρg = 1.225 kg/m3, μ = 1.8 ∗ 10−5 Pa·s, ρp = 1500 kg/m3,
drag model can successfully simulate the flow behavior in CFB risers. dp = 7.5 ∗ 10−5 m; bubbling bed 2: Ug = 0.06 m/s, ρg = 1.225 kg/m3, μ = 1.7894 ∗ 10−5
It can predict not only the macroscopically bottom-dense and top- Pa·s, ρp = 1780 kg/m3, dp = 6.5 ∗ 10−5 m).
dilute axial solid distribution as well as the so-called core-annular radial
distribution, but also the meso-scale phenomena of particle aggrega-
!
tion. However, the applicability of the EMMS drag in simulation of dU p ∇p ! βp
!
!  ! 
bubbling fluidized beds has not been studied yet. ¼− þ g þ U g x p − U p þ Fc ð4Þ
dt ρs ρs
In addition, some basic issues in MP-PIC calculations, such as the
impacts of the grid resolution and the number of particles per parcel
(PPP) on the simulation are seldom studied in the literature. In this Here, Fc represents the interaction force between particles. For
paper, the factors affecting bubbling fluidized bed simulations including traditional Discrete Element Methods (DEM), the collisions between
drag models, grid resolution and PPP will be studied in the frame of MP- particles need to be tracked to calculate Fc, which requires that the
PIC method. Two of the drag models including the homogeneous drag time step of equation for particles is small enough and needs real-time
[13] and the heterogeneous EMMS drag are used. The simulation results information to calculate the distance between a pair of particles in
are compared with the experimental data for validation. order to determine whether they collide or not. So the DEM method
requires a lot of computing resources and time. For the gas-solid flow
2. Model description of industrial installations or large scale laboratory device, the DEM
method is almost impossible to be used.
2.1. MP-PIC method

MP-PIC is a typical Euler-Lagrange method. The Navier-Stokes (N-S) outlet

equations are used to describe the movement of the gas phase, while
Newton's second law is used to describe the motion of particle phase.
The typical gas-solid equations are listed below [2].
For the gas phase, the continuity and the momentum equations are
as follows:
W
∂

*

εg ρg þ ∇  εg ρg u g ¼ 0 ð1Þ
∂t Bubbling bed 1 [27]:
H=1 m
∂
*

* *
 *
ε g ρg u g þ ∇  εg ρg u g u g ¼ −εg ∇p þ ∇  τg þ ε g ρg g ð2Þ W=0.14 m
∂t
H

X nT
V p
!
!  ! 
− np β U g x p −U p
p¼1
Vc p Bubbling bed 2 [41]:
H=2.464 m
Here, the subscript g represents the gas phase; εg is the gas volume W=0.267 m
!
fraction; ρg is the gas density; u g is the gas phase velocity at cell center;
p is the gas pressure; τg is the gas phase stress tensor; nT is the parcel
number in fluid cell; np is the number of particles per parcel; Vp is the
particle volume and Vc is the fluid cell volume. βp is the drag coefficient;
! ! !
U g ð x p Þ is the gas velocity at parcel location; x p is the parcel location;
!
U p is the parcel velocity.
For the particle phase, the equations of motion are as follows:
inlet
!
dxp !
¼ Up ð3Þ
dt Fig. 2. 2D geometry of the bubbling beds (H: reactor height; W: reactor width).
 M. Feng et al. / Powder Technology 328 (2018) 215–226 217

Table 1 in MP-PIC method thus a larger time step can be used. In addition, the
Summary of parameters used in the simulation. search process is not required in MP-PIC method, which can reduce
Bubbling bed 1 Bubbling bed 2 the amount of particle computation massively. The parcels are tracked
[27] [41] in MP-PIC method instead of single particles, which can further reduce
Reactor height, H (m) 1 2.464 the amount of calculation. Above all, these three aspects reduce the
Reactor width, W (m) 0.14 0.267 amount of calculation greatly, thus not only improving computational
Initial packing height, H0 (m) 0.5 1.2 efficiency, but also making it possible to simulate gas-solid flows in
Initial bed voidage 0.45 0.44
industrial equipments [40].
Average particle diameter, dp (m) 7.5 ∗ 10−5 5.18 ∗ 10−5
Particle density, ρp (kg/m3) 1500 1780
Solid volume fraction at close pack, εs,max 0.55 0.56 2.2. Drag model
Gas density, ρg (kg/m3) 1.225 1.225
Inlet gas velocity, Ug (m/s) 0.1 0.06 There are two drag models compared in this paper. One is a typical
Gas viscosity, μ (Pa·s) 1.8 ∗ 10−5 1.7894 ∗ 10−5
Restitution coefficient, ew 0.9 0.9
homogeneous drag model; the other accounting for inhomogeneous
Acceleration of gravity, (m/s2) 9.8 9.8 structure is EMMS/bubbling drag model. It should be mentioned that
Time step, Δt (s) 1 ∗ 10−4 5 ∗ 10−4 there are lots of homogeneous drag models reported previously, how-
Particle pressure model parameter, ps∗ (Pa) 100 100 ever those models are similar in trend with very limited differing in
Particle pressure model parameter, α 2.5 2.5
quantity [41, 42]. So only a commonly used homogenous drag model
is adopted.
The homogeneous drag model [13] used is an empirical correlation.
Table 2 This model does not consider the flow structure.
Numerical parameters used in bubbling beds simulations to test different drag models.
8 !
 ! 
Grid size/mm2 >  ! 
Geometry PPP Drag model >
> 3 ε s ε g ρg  U g x p − U p 
>
> C ε−2:7 εs b 0:2
Case 1 Bubbling bed 1 2∗2 186 Homogeneous <4 D0
dp
 g 
Case 2 EMMS/bubbling β¼ ! ! !  ð7Þ
>
> ε2s μ 7 εs ρg  U g x p − U p 
Case 3 Bubbling bed 2 3.34 ∗ 6.16 100 Homogeneous >
>
Case 4 EMMS/bubbling >
: 150 2
þ εs ≥ 0:2
εg d 4 dp
p

8

>
< 24 1 þ 0:15R0:687
Particle stress gradient force is used to characterize the interaction e
C D0 ¼ Re b 1000 ð8Þ
force between particles in MP-PIC method, which is expressed as: > Re
:
0:44 Re ≥ 1000
∇ps
Fc ¼ − ð5Þ !
 ! 
ε s ρs  ! 
εg ρg dp  U g x p − U p 
Re ¼ ð9Þ
P s ε αs μg
Ps ¼ ð6Þ
εs; max −ε s
Here, β is the interphase momentum transfer coefficient; μ is the
Here, ps represents particle pressure. p∗s, α are model parameters. viscosity; CD0 is the standard drag coefficient of a single particle; Re is
εs, max is the solid volume fraction at close pack. εs in Eq. (5) and the particle Reynolds number.
Eq. (6) is the solid volume fraction. Fc is only a function of εs. There- EMMS/bubbling drag model [36, 43–45] is applicable to simulate
fore, collisions between particles do not have to be tracked accurately bubbling bed based on the two-fluid model. In this model, HD = β/βw

Case1:Homogeneous Case2:EMMS/ Case3:Homogeneous Case4:EMMS/

bubbling bubbling
(a) (b)

Fig. 3. Particle distribution calculated with different drag models instantaneously. (a) Bubbling bed 1 at time = 20 s. (b) Bubbling bed 2 at time = 30 s.
218 M. Feng et al. / Powder Technology 328 (2018) 215–226

is known as the heterogeneity index [26], and it needs to be calculated Though the both drag models predicted bubbling bed fluidization, the
according to different operating conditions. particle distributions calculated with homogeneous drag model exhibit
The calculated HD for two bubbling fluidized beds are shown in some turbulent bed characteristics, e.g. the irregular bubble and cluster
Fig. 1. And their fitting functions are given as Eqs. (10) and (11), respec- shapes [48, 49], while EMMS/bubbling drag model predicts a much
tively. For detailed introduction to the EMMS/bubbling drag model, it is more typical bubbling bed. To quantitatively compare the calculated
referred to literature [45]. solid concentration distributions with experimental measurements,
8 Fig. 4 shows the time-averaged (10–20 s for bubbling bed 1 and 15–
> 1 ; εg ≤0:4 30 s for bubbling bed 2) axial distributions of solid concentration with
>
>
> −222:88267 þ 1115:14532ε g −1388:59658ε 2g
> ; 0:4bεg ≤0:41199
>
> different drag models, where the dashed line is the experimentally
>
> 181:40795−1120:39114ε g þ 2310:51077ε 2g
<
measured bed height. For bubbling bed 1, the simulation results
HD ¼ −1589:35592ε 3g ; 0:41199bεg ≤0:47196
>
> shows that both homogeneous and EMMS/bubbling drag models over
>
> −0:05576 þ 0:08337ε g þ 0:97598ε2g ; 0:47196bεg ≤0:72984
>
>
>
> −0:75971 þ 1:75842ε g ; 0:72984bε g ≤1 predict bed height. However, the bed height calculated with EMMS/
>
:
1 ; εg N1 bubbling drag model is closer to the experimental value than that
ð10Þ with homogeneous drag model. This is because EMMS/bubbling drag
model has considered the effects of meso-scale structures within the
8 fluidized bed. The simulation using the homogeneous drag model
> 1 ; ε g ≤ 0:4
>
>
>
> −132:66377 þ 669:1222εg −837:40663ε2g ; 0:4bε g ≤0:41799 exhibited larger bed heights due probably to larger bubbles and larger
>
>
>
> 49:12504−210:60763ε g þ 226:7762ε2g ; 0:41799bεg ≤ 0:46597 number of bubbles. For bubbling bed 2, the axial distribution of voidage
<
HD ¼ −0:07537 þ 0:15139ε g þ 1:07059ε 2g ; 0:46597bεg ≤ 0:69385 calculated with EMMS/bubbling is much closer to the experimental data
>
>
>
> 0:67911
>
> 1:0446− ; 0:69385bεg ≤ 1
>
> εg −0:78105
>
: 1 þ e 0:08531
1 ; ε g N1
ð11Þ

3. Model set-up and parameters

In this paper, there are two bubbling beds with Geldart A particles to
be simulated. One is the bubbling bed of fluid catalytic cracking (FCC)
catalyst used in McKeen & Pugsley [29], and it is called bubbling bed 1.
The other bubbling bed used by Zhu et al. [46] in their experiments is
called bubbling bed 2 in this paper. Both bubbling beds have similar ge-
ometry but with different dimensions. The geometry is shown in Fig. 2
and the specific parameters are shown in Table 1. The corresponding
HD of EMMS/bubbling drag model is also calculated according to Table 1.
The grid spacing of 3.34 mm ∗ 6.16 mm (80 ∗ 400 = 32,000) is
adopted for bubbling bed 2, which is sufficient to obtain the gird inde-
pendent results as reported in literature [47] for the two-dimensional
simulation.

4. Results and discussion

4.1. Effects of drag models

To compare the effects of different drag models on the simulation

results of bubbling bed, both bubbling bed 1 and bubbling bed 2 were
simulated with the two drag models as introduced in Section 2.2. The
numerical parameters [47] used in the simulations are shown in Table 2.
The particle distributions calculated with different drag models are
shown in Fig. 3. Fig. 3(a) shows the results for the bubbling bed 1. In
general, the results calculated with homogeneous drag model and
with the EMMS/bubbling drag all show bubbling bed characteristics: a
free board in the upper region and a dense lower region that is undergo-
ing bubbling. For the case 1, small bubbles formed in the bottom of bed
and gradually merged into larger ones. Large bubbles mainly rise in the
middle of bed and break up at bed surface. These large bubbles are
irregular in shape. For case 2, the calculated bed height is apparently
lower than the case 1 due to the reduced drag force with the EMMS/
bubbling drag model. Different from the case 1, the larger bubbles rise
up near the side wall alternatively one by one. And the bubble shapes
are more regular than in case 1. Fig. 3(b) shows the particle distributions
for the bubbling bed 2. The results are much similar to that of the
bubbling bed 1. As can be seen in Fig. 3(b), the calculated bed height
and the emulsion phase voidage are much lower with EMMS/bubbling
drag model as compared to with homogeneous drag model. There are Fig. 4. Axial distribution of solid concentration calculated with different drag models.
more distinct bubbles and the bubbles are more regularly in case 4. (a) Bubbling bed 1. (b) Bubbling bed 2.
 M. Feng et al. / Powder Technology 328 (2018) 215–226 219

than in the other case. While with the homogeneous drag model, the
solid concentration is apparently smaller than the experimental data.
Fig. 5 compares the time-averaged (15–30 s) radial solid concentra-
tion profiles with the experimental measurement at different heights in
bubbling bed 2. Generally, the figures show that the measured solid
concentrations are uniformly distributed along the radial direction.
With the homogeneous drag model, the calculated solid concentrations
are under predicted and the solid concentrations near the wall are
higher than the solid concentration in the center of the bed. While
with the EMMS/bubbling drag model, the calculated solid concentra-
tions are much uniformly distributed. And the value is closer to
experimental measurement than that calculated with homogeneous
drag model. Specifically, the solid concentration in h = 1.1 m fluctuates
along the radial direction. This is because there are large bubbles
continuously generated in this part of the bed.

4.2. Bubble behavior analysis

In order to further understand the bubble behavior in bubbling beds,

the results images were saved every 0.1 s for bubble characteristic
analysis (including bubble diameter and bubble circularity). Since
simulation results with EMMS bubbling drag model are closer to
experimental measurement, only the results with the EMMS bubbling
drag model were used here. At first, these images were binarized with
the open-source software “ImageJ” [50, 51]. Then bubbles were
identified with by detecting the closed edges. The bubble identification
process is shown in Fig. 6. At the same time, the areas of bubbles, which
(a) (b) (c)
Fig. 6. Identification of bubbles from simulation results (a) original calculation results.
(b) Binarized image by ImageJ. (c) Identified bubbles.
were larger than 20 pixels (corresponding to bubble diameter of 0.55
cm. This value is sufficient to characterize bubbles conditions in our opin-
ion) were exported for further analysis. Then according to the pixels

H = 0.4 m H = 0.6 m

H = 0.8 m H = 1.1 m
Fig. 5. Comparison of time-average radial solid concentration profiles at different heights in bubbling bed 2.
220 M. Feng et al. / Powder Technology 328 (2018) 215–226

(a) areas, the real areas of bubbles were calculated. Further, the diameters of
0.028
bubbles were calculated based on their areas by assuming they were
round bubbles. Fig. 7 shows the evolution process of the averaged bubble
0.024
diameter in the flow fields of bubbling bed 1 and 2. Fig. 8 shows the bub-
ble diameter distributions in stable state of bubbling bed 1 and 2.
0.020
Fig. 7(a) shows that the bubble diameter tends to be stable after
db (m)

0.016 about 5 s simulation in bubbling bed 1. The flow field also shows the
same trend. That's the reason why the data of 10–20 s is collected for
0.012 averaging in bubbling bed 1. The figure shows that there are only small
bubbles initially, and then bubbles grow bigger with time. After 5 s, bub-
0.008 ble diameter begins to fluctuate around an average bubble diameter of
0 5 10 15 20 0.022 m (293 times the particle diameter) with the standard deviation
Time (s)
of 0.002 m. Fig. 7(b) shows the similar process just like in bubbling bed
2, while the bubble diameter and the flow filed tend to be stable after
(b) about 8 s simulation in bubbling bed 2. So the data of 15–30 s is collected
0.07 for averaging in bubbling bed 2. The averaged bubble diameter is 0.039 m
0.06 (757 times the particle diameter) with the standard deviation of 0.006 m.
Fig. 8 shows the bubble diameter distributions in steady state for
0.05 bubbling bed 1 and 2. Generally, the percentage of bubble decreases
with the increasing of bubble diameter. Which means that small bub-
db (m)

0.04 bles prevails in the bubbling beds. However, there are still secondary
peaks for the both bubbling beds, i.e. 0.07 m for bubbling bed 1 and
0.03
0.09 m for bubbling bed 2. The appearance of the second peak shows
0.02 that there exists a second characteristic bubble diameter in bubbling
fluidized beds [52].
0.01 Circularity in “ImageJ” is defined as [50]:
0 5 10 15 20 25 30
Time (s)

Ab
Fig. 7. Evolution processes of the averaged bubbles diameters of (a) bubbling bed 1 with Circularity ¼ 4π  ð12Þ
EMMS/bubbling drag model. (b) Bubbling bed 2 with EMMS/bubbling drag model. P 2b

(a)
0.75

0.70

0.65
Circularity

0.60

0.55

0.50

0.45

0.40
0 5 10 15 20
Time (s)

(b)
1.0

0.8
Circularity

0.6

0.4

0.2
0 5 10 15 20 25 30
Time (s)

Fig. 8. Bubble diameter distributions in steady state of (a) bubbling bed 1 with EMMS/ Fig. 9. Evolution processes of the averaged circularity of bubble of (a) bubbling bed 1 with
bubbling drag model, (b) bubbling bed 2 with EMMS/bubbling drag model. EMMS/bubbling drag model. (b) Bubbling bed 2 with EMMS/bubbling drag model.
 M. Feng et al. / Powder Technology 328 (2018) 215–226
Fig. 10. Circularity distributions of bubble in steady state of (a) bubbling bed 1 with
EMMS/bubbling drag model, (b) bubbling bed 2 with EMMS/bubbling drag model.
where Ab is the bubble area and Pb is the bubble perimeter. The circular-
ity value of 1.0 indicates a perfect round bubble. As the value approaches
0.0, it indicates an increasingly elongated bubble shape. Fig. 9(a) shows
that the circularity of bubble tends to be stable after about 5 s simulation
in bubbling bed 1. This trend is similar with the bubble diameter
14s 14.1s 14.2s 14.3s
Fig. 11. The bubble coalescence process (bubbling bed 1).
221
evolution process shown in Fig. 7(a). It can be seen that the circularity
is bigger at the beginning of simulation. It is probably because the
bubbles are very small in this stage so that the bubbles are much more
rigid that the round shape can be maintained. Then the circularity de-
creases with time, and after about 5 s, the circularity begins to fluctuate
around an average value of 0.496 with the standard deviation of 0.039.
Fig. 9(b) shows that the circularity fluctuates at a very wide range at the
beginning of simulation in fluidized bed 2. May be it is because of only a
few small bubbles appeared at the beginning. And there are only 1–4
bubbles identified in each snapshot before 4.3 s. While after 4.3 s,
there are 10–56 bubbles identified in each snapshot. Small amount of
bubbles can lead to the strong fluctuation in circularity. Generally, the
bubble circularity and the flow filed tend to be stable after about 8 s
simulation in bubbling bed 2. The averaged circularity of bubble
diameters is 0.478 with the standard deviation of 0.040.
Fig. 10 shows the bubble circularity distributions in steady state for
bubbling bed 1 and 2. Generally, the both cases show Maxwellian like
distributions with the most probable bubble circularity of about 0.4.
This value is not consistent with the averaged circularity. Because the
circularity distributions are not axisymetrical about 0.4. But the
circularity of most bubbles are higher than 0.4, that the averaged bubble
circularity is higher than 0.4. It is worth noting that no matter in
bubbling bed 1 or in bubbling bed 2, the bubble circularity is about
0.5. That means the bubbles are not perfect round. This may be a reason
for the inaccurate calculations. Since most bubble drag models are
developed under the assumption that bubbles are round. So the drag
model accounting for imperfect round bubble will be need in the future.
Meanwhile, bubble coalescence behavior was also observed for
bubbling bed 1 by a series of snapshots at continuous instants as
shown in Fig. 11. Several bubble coalescence events were captured
with the involved bubbles were marked by dashed circles. Generally,
two small bubbles coalesced to form a large one. Before coalescence,
the two bubbles moved closer and then they began to coalesce. In the
initial stage of coalescence, there was an irregular interface structure
for the coalescing bubble as indicated in the snapshots at 14.1 s, 14.5 s,
14.6 s. When the coalescence was completed, the interface structure
disappeared to form a regular bubble. Almost all of the large bubbles
were created by coalescence in this case. A whole coalescence process
takes 0.1–0.2 s approximately.
4.3. Effects of grid size
In this section, the effects of different grid resolutions will be studied.
Only the bubbling bed 1 would be simulated and the number of
14.4s 14.5s 14.6s 14.7s
222 M. Feng et al. / Powder Technology 328 (2018) 215–226
Table 3
Numerical parameters used in bubbling bed 1 simulation to test the effects of grid size.
Geometry Grid size/mm2 PPP Drag model
Case 1, 7–10 Bubbling bed 1 2 ∗ 2, 1 ∗ 1, 3 ∗ 3, 5 ∗ 5 186 Homogeneous
Case 2, 11–14 2 ∗ 2, 1 ∗ 1, 3 ∗ 3, 5 ∗ 5 EMMS/bubbling
particles per parcel (PPP) is fixed to 186 (below which, the simulation
results do not change apparently as shown in Section 4.4) in this section.
All the cases tested in this section are listed in Table 3.
The calculation results with different grid resolutions for bubbling bed
1 are shown in Figs. 12 and 13. Fig. 12(a) and (b) show the contours of
voidage distribution with homogeneous drag model and EMMS/bubbling
drag model respectively. Generally, the simulation results are much
similar to those shown in Fig. 3(a). EMMS/bubbling drag model predicts
bubbles with regular shapes while homogeneous drag model predicts
bubbles with irregular shapes. With the minimum grid size, 1 mm for
the present study, more details can be captured for both models. When
the grid size increases to a certain extent, 5 mm for bubbling bed 1, the
homogeneous drag model predicts an over expended bed. And further in-
crease of the grid size to 10 mm, the homogeneous drag model fails to
predict the reasonable particle distribution and the particles are gathered
at the top of the fluidized bed as shown in Fig. 12(a). This is because the
wall boundary condition is applied for particle phase that the particles
cannot go out of the bed in MFIX. On the other hand, the EMMS/bubbling
drag model predicts a more reliable flow state.
Fig. 13 shows the time-averaged (10–20 s) axial distribution of solid
concentration for bubbling bed 1 with different grid size. Except for the
10 mm cases with homogeneous drag model, all the calculated solid
concentration curves show the “C” type distributions. Generally,
homogeneous drag model predicts much higher bed height with
much lower solid concentration (about 0.32) of the dense phase bed;
10mm 5mm 3mm
(a)
10mm 5mm 3mm
(b)
Fig. 12. Particle distribution for bubbling bed 1 calculated with different grid sizes at time = 20 s (PPP = 186). (a) With homogeneous drag model. (b) With EMMS/bubbling drag model.
EMMS/bubbling drag model predicts the opposite result, lower bed
height with higher solid concentration (about 0.38). It should also be
noted that the bubbling bed height changes little when the grid size is
less than 3 mm for both the drag models. So the grid size of 2 mm is
an appropriate value for bubbling bed 1 and hence used in later
computation.
Fig. 14 quantitatively compares the calculated bed heights with dif-
ferent grid sizes. The bed height “H” in Fig. 13 is calculated from Fig. 13
at a position where the solid concentration equals to half of the aver-
aged dense bed solid concentration [29]. Fig. 14 shows that: 1) No mat-
ter which drag model (homogeneous drag model or EMMS/bubbling
drag model) is used, the calculated bed height decreases with the de-
crease of the grid size. Until at the threshold grid size value, the bed
height reaches a stable value and does not change apparently with fur-
ther decrease of the grid size. 2) With homogeneous drag model, there
is a threshold between 3 mm and 5 mm in grid size, below which the
calculated bed heights are independent of the grid size. Meanwhile,
there is a threshold between 5 mm and 10 mm in grid size with
EMMS/bubbling drag model. It means that grid-independent solution
can be obtained in coarse grid size with EMMS/bubbling drag model.
For large-scale industrial simulation, this will greatly reduce the
computation time. 3) The bubbling bed heights calculated with
EMMS/bubbling drag model are less than that calculated with the ho-
mogeneous drag force. The bed heights predicted by both drag models
are higher than the experimental measurement, while the heights
predict with EMMS/bubbling drag model are much closer to the exper-
imental data. These results further proved that EMMS/bubbling drag
model is more suitable for the simulation of bubbling bed 1.
4.4. Effects of number of particles per parcel (PPP)
In order to investigate how PPP affects the calculation results, five
cases with different PPP (747, 373, 186, 93, 23, 10) have been simulated
2mm 1mm
2mm 1mm
 M. Feng et al. / Powder Technology 328 (2018) 215–226
Fig. 13. Axial distribution of solid concentration for bubbling bed 1 calculated with
different grid sizes (PPP = 186). (a) With homogeneous drag model. (b) With EMMS/
bubbling drag model.
for bubbling bed 1. Two different drag models were used in this section.
For all these cases, the grid sizes were kept constant. Specifically, 2 mm
grid is used. All the cases tested in this section are listed in Table 4.
Fig. 15(a) and (b) shows the snapshots of voidage distribution
with homogeneous drag model and EMMS/bubbling drag model at
time = 20 s. Even if PPP is changed, the bed behaves like turbulent
fluidization rather than bubbling fluidization with homogeneous
drag model. More details of the flow field can be captured with
the decrease of PPP with EMMS/bubbling drag model. For example,
the shapes of bubbles and the bubble emulsion interfaces are
clearer with the decrease of PPP. Over all, no matter which drag
model (homogeneous drag model or EMMS/bubbling drag model)
is used, the flow states change a little with the decrease of PPP
apparently.
Fig. 16(a) and (b) shows the time averaged (10–20 s) axial
distribution of solid concentration for bubbling bed 1 calculated with
homogeneous drag model and EMMS/bubbling bed model separately.
All the curves have a similar “C” type shape. The solid concentration is
223
Fig. 14. Comparison of bed heights calculated with different grid sizes (bubbling bed 1,
ppp = 186).
low at the bottom and surface of the bed. And in between, the solid
concentration in the dense region is about 0.34 approximately with
homogeneous drag model, while the solid concentration in dense
region is about 0.38 with EMMS/bubbling drag model. The axial solid
concentration distribution seems not sensitive to PPP, because the
curves do not change apparently with PPP.
Fig. 17 further compares the bed heights calculated with different
PPP. It shows that: 1) The calculated bed heights with homogeneous
drag model are bigger than those with EMMS/bubbling drag model for
all the PPP. And all the calculated bed heights are higher than the
experimental value. But the EMMS/bubbling drag predicted values are
much closer to the experimented measurement. 2) For both drag
models, the calculated bed height decreases with the decrease of PPP.
There also exists a threshold value for PPP, below which the calculated
bed height no longer changes with PPP. The figure shows that this
value locates between 186 and 373 for both drag models.
4.5. Discussion
The above calculation results show that the EMMS/bubbling drag
model can improve the accuracy of the simulation. However there are
still clear errors. There maybe two reasons mainly.
4.5.1. Drag model
The drag models used in this paper are established based on cell
averaged parameters in Eulerian-Eulerian framework. MP-PIC method
is a particle tracking method belonging to the Eulerian-Lagrangian
framework. Each parcel has its own speed, position and solid concentra-
tion. So the drag model needs to account for the discrete characteristics
of MP-PIC methods for further improvements.
4.5.2. Stress model
Particle volume concentration are very large in bubbling bed, thus
particle-particle interaction force (solid stress) play a great role in
particle movement. The current solid stress model does not account
for the impact of meso-scale structures (bubble), which may also affect
the simulation results. The solid stress model based on meso-scale
structures should be developed in future work [53].
224 M. Feng et al. / Powder Technology 328 (2018) 215–226

Table 4
Numerical parameters used in bubbling bed 1 simulation to test the effects of PPP.

Geometry Grid size/mm2 PPP Drag model

Case 1, 15–19 Bubbling bed 1 2∗2 186, 747, 373, 93, 23, 10 Homogeneous
Case 2, 20–24 EMMS/bubbling

5. Conclusion model, much coarser grids can be used to obtain reasonable result,
with EMMS/bubbling drag model. For large-scale industrial simula-
Two bubbling fluidized beds with Geldart A particles [29, 46] were tion, this will greatly reduce the computation time.
simulated with MP-PIC method by using both EMMS/bubbling drag 4) With the same grid size, decreasing PPP can improve the accuracy of
model and traditional homogeneous drag model. Different grid size the simulation result. But the axial solid concentration distribution
and different number of particles per parcel were tested to study their seems not sensitive to PPP in the present work.
effects on simulation results. The main conclusions are as follows:
It should be also noted that, there are some discrepancy between the
1) The EMMS/bubbling drag model can predicts more accurate radial/ simulation results and the experimental measurements even though
axial distribution of solid concentration and more reasonable flow the EMMS/bubbling drag have been used. So in future works, the drag
state than homogeneous drag model. Because meso-scale structures model needs to consider the characteristics of MP-PIC methods for
within the fluidized bed has been considered in the EMMS/bubbling further improvements. And the solid stress model based on meso-
drag model. But they all over predicted bed height in calculations of scale structures [53] should also be developed.
bubbling bed 1.
2) The EMMS/bubbling drag model calculated average bubble diameter Nomenclature
increases with simulation time at the initial stage, and then it Ab bubble area (m2)
becomes stable and fluctuates around the average value. No matter CD0 standard drag coefficient of single particle
in bubbling bed 1 or in bubbling bed 2, the averaged bubble circular- dp averaged particle diameter (m)
ity is about 0.5. That may be one reason for calculation errors. Since ew restitution coefficient
most bubble drag models were developed under the assumption Fc interaction force between particles (N)
that bubbles were perfectly round. So bubble drag model accounting np number of particles per parcel
for imperfectly round bubble effect may be a prospective way to nT parcel number in fluid cell
improve simulation accuracy. p gas pressure (Pa)
3) There exists a threshold value of grid size, below which the simula- Pb bubble perimeter (m)
tion result is independent of grid size. With the EMMS/bubbling ps particle pressure (Pa)
drag model, the threshold of grid size is bigger than that with the p∗s particle pressure model parameter (Pa)
homogeneous drag model. Compared with homogeneous drag Re Reynolds number

PPP = 747 373 186 93 23 10

(a)

PPP = 747 373 186 93 23 10

(b)

Fig. 15. Particle distribution for bubbling bed 1 calculated with different PPP at time = 20 s (grid size = 2 mm). (a) With homogeneous drag model. (b) With EMMS/bubbling drag model.
 M. Feng et al. / Powder Technology 328 (2018) 215–226 225

Fig. 17. Comparison of bed height with the change of PPP with different drag model.

εs, max solid volume fraction at max. packing

μ gas viscosity (Pa·s)
ρg gas density (kg/m3)
ρp particle density (kg/m3)
τg gas phase stress tensor (Pa)

Subscripts
g gas phase
s solid phase

Acknowledgment

This work was financially supported by the National Key R&D

Program of China (No. 2017YFB0602701-04) and the National Natural
Science Foundation of China (Nos. 91334204 and 21625605). Meiyan
Feng also thanks her colleague Yanpei Chen for her help in using
“ImageJ”.

References
Fig. 16. Axial distribution of solid concentration for bubbling bed 1 calculated with
different PPP (grid size = 2 mm). (a) With homogeneous drag model. (b) With EMMS/ [1] M.J. Andrews, P.J. O'Rourke, The multiphase particle-in-cell (MP-PIC) method for
bubbling drag model. dense particulate flows, Int. J. Multiphase Flow 22 (2) (1996) 379–402.
[2] D.M. Snider, An incompressible three-dimensional multiphase particle-in-cell
model for dense particle flows, J. Comput. Phys. 170 (2) (2001) 523–549.
[3] S. Benyahia, J.E. Galvin, Estimation of numerical errors related to some basic as-
Δt time step sumptions in discrete particle methods, Ind. Eng. Chem. Res. 49 (2010)
! 10588–10605.
ug gas phase velocity at cell center (m/s) [4] A. Stroh, F. Alobaid, M.T. Hasenzahl, J. Hilz, J. Ströhle, B. Epple, Comparison of three
! ! different CFD methods for dense fluidized beds and validation by a cold flow exper-
U gð x pÞ gas velocity at parcel location (m/s) iment, Particuology 29 (Supplement C) (2016) 34–47.
Ug gas inlet velocity (m/s) [5] D. Snider, S. Banerjee, Heterogeneous gas chemistry in the CPFD Eulerian–
! Lagrangian numerical scheme (ozone decomposition), Powder Technol. 199 (1)
Up parcel velocity (m/s) (2010) 100–106.
Vp particle volume (m3) [6] S. Karimipour, T. Pugsley, Application of the particle in cell approach for the simula-
Vc fluid cell volume (m3) tion of bubbling fluidized beds of Geldart A particles, Powder Technol. 220 (2012)
63–69.
W reactor width (m) [7] Y. Liang, Y. Zhang, T. Li, C. Lu, A critical validation study on CPFD model in simulating
! gas–solid bubbling fluidized beds, Powder Technol. 263 (2014) 121–134.
xp parcel location (m)
[8] A. Abbasi, P.E. Ege, H.I. de Lasa, CPFD simulation of a fast fluidized bed steam coal
gasifier feeding section, Chem. Eng. J. 174 (1) (2011) 341–350.
[9] C. Chen, F. Li, H. Qi, Modeling of the flue gas desulfurization in a CFB riser using the
Greek letters Eulerian approach with heterogeneous drag coefficient, Chem. Eng. Sci. 69 (1)
α particle pressure model parameter (2012) 659–668.
[10] C. Chen, J. Werther, S. Heinrich, H.-Y. Qi, E.-U. Hartge, CPFD simulation of circulating
β drag force coefficient (kg/m3 s) fluidized bed risers, Powder Technol. 235 (2013) 238–247.
βp drag force coefficient for particle (kg/m3 s) [11] H. Arastoopour, P. Pakdel, M. Adewumi, Hydrodynamic analysis of dilute gas—solids
βw Wen & Yu drag force coefficient (kg/m3 s) flow in a vertical pipe, Powder Technol. 62 (2) (1990) 163–170.
[12] T. Tanaka, T. Kawaguchi, Y. Tsuji, Discrete particle simulation of flow patterns in
εg gas volume fraction two-dimensional gas fluidized beds, Int. J. Mod. Phys. B 07 (09n10) (1993)
εs solid volume fraction 1889–1898.
226 M. Feng et al. / Powder Technology 328 (2018) 215–226
[13] D. Gidaspow, Hydrodynamics of fluidization and heat transfer: supercomputer
modeling, Appl. Mech. Rev. 39 (1986) 1–23.
[14] M. Syamlal, T. O'Brien, Simulation of granular layer inversion in liquid fluidized beds,
Int. J. Multiphase Flow 14 (4) (1988) 473–481.
[15] B.H. Xu, A.B. Yu, Numerical simulation of the gas-solid flow in a fluidized bed by
combining discrete particle method with computational fluid dynamics, Chem.
Eng. Sci. 52 (16) (1997) 2785–2809.
[16] M. Lungu, Y. Zhou, J. Wang, Y. Yang, A CFD study of a bi-disperse gas–solid fluidized
bed: effect of the EMMS sub grid drag correction, Powder Technol. 280 (Supplement
C) (2015) 154–172.
[17] C.Y. Wen, Y.H. Yu, Mechanics of fluidization, Chem. Eng. Prog. Symp. Ser. 62 (62)
(1966) 100–111.
[18] D.Z. Zhang, W.B. VanderHeyden, High-resolution three-dimensional numerical sim-
ulation of a circulating fluidized bed, Powder Technol. 116 (2) (2001) 133–141.
[19] D. Gidaspow, Multiphase Flow and Fluidization: Continuum and Kinetic Theory De-
scription, Academic Press, Boston, 1994.
[20] J. Li, Y. Tung, M. Kwauk, Method of energy minimization in multi-scale modeling of
particle-fluid two-phase flow, in: Basu P. Large, J. François (Eds.), Circulating Fluid-
ized Bed Technology, Pergamon 1988, pp. 89–103.
[21] T. O'Brien, M. Syamlal, Particle cluster effects in the numerical simulation of a circu-
lating fluidized bed, CFB IV, New York, 1993.
[22] H. Xiao, H. Qi, C. You, X. Xu, Theoretical model of drag between gas and solid phase,
J. Chem. Ind. Eng. (China) 54 (3) (2003) 311–315.
[23] N. Yang, W. Wang, W. Ge, J. Li, Choosing structure-dependent drag coefficient in
modeling gas-solid two-phase flow, China Particuology 1 (1) (2003) 38–41.
[24] S. Tran-Cong, M. Gay, E.E. Michaelides, Drag coefficients of irregularly shaped parti-
cles, Powder Technol. 139 (1) (2004) 21–32.
[25] R. Beetstra, M.A. Van der Hoef, J.A.M. Kuipers, Drag force of intermediate Reynolds
number flow past mono- and bidisperse arrays of spheres, AICHE J. 53 (2) (2007)
489–501.
[26] W. Wang, J. Li, 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) (2007) 208–231.
[27] J.F. Richardson, Incipientfluidization and particulate systems, in: J.F. Davidson, D.
Harrison (Eds.), Fluidization, Academic Press, London & New York 1971, pp. 25–64.
[28] R. Turton, O. Levenspiel, A short note on the drag correlation for spheres, Powder
Technol. 47 (1986) 83–86.
[29] T. McKeen, T. Pugsley, Simulation and experimental validation of a freely bubbling
bed of FCC catalyst, Powder Technol. 129 (1–3) (2003) 139–152.
[30] Z. Shi, W. Wang, J. Li, A bubble-based EMMS model for gas–solid bubbling fluidiza-
tion, Chem. Eng. Sci. 66 (22) (2011) 5541–5555.
[31] S. Vashisth, A.H. Ahmadi Motlagh, S. Tebianian, M. Salcudean, J.R. Grace, Comparison
of numerical approaches to model FCC particles in gas–solid bubbling fluidized bed,
Chem. Eng. Sci. 134 (2015) 269–286.
[32] H. Qi, F. Li, B. Xi, C. You, Modeling of drag with the Eulerian approach and EMMS the-
ory for heterogeneous dense gas–solid two-phase flow, Chem. Eng. Sci. 62 (6)
(2007) 1670–1681.
[33] J. Wang, W. Ge, J. Li, 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 (6) (2008) 1553–1571.
[34] Y. Tian, J. Geng, W. Wang, Structure-dependent analysis of energy dissipation in gas-
solid flows: beyond nonequilibrium thermodynamics, Chem. Eng. Sci. 171 (2017)
271–281.
[35] M. Zeneli, A. Nikolopoulos, N. Nikolopoulos, P. Grammelis, E. Kakaras, Application of
an advanced coupled EMMS-TFM model to a pilot scale CFB carbonator, Chem. Eng.
Sci. 138 (2015) 482–498.
[36] B. Lu, W. Wang, J. Li, Searching for a mesh-independent sub-grid model for CFD sim-
ulation of gas–solid riser flows, Chem. Eng. Sci. 64 (15) (2009) 3437–3447.
[37] A. Ullah, K. Hong, S. Chilton, W. Nimmo, Bubble-based EMMS mixture model applied
to turbulent fluidization, Powder Technol. 281 (2015) 129–137.
[38] B. Lu, W. Wang, J. Li, Eulerian simulation of gas–solid flows with particles of Geldart
groups A, B and D using EMMS-based meso-scale model, Chem. Eng. Sci. 66 (20)
(2011) 4624–4635.
[39] Y. Mei, M. Zhao, B. Lu, S. Chen, W. Wang, Numerical comparison of two modes of
gas-solid riser operation: fluid catalytic cracking vs CFB combustor, Particuology
31 (Supplement C) (2017) 42–48.
[40] F. Li, F. Song, S. Benyahia, W. Wang, J. Li, MP-PIC simulation of CFB riser with EMMS-
based drag model, Chem. Eng. Sci. 82 (2012) 104–113.
[41] X. Wang, C. You, Evaluation of drag force on a nonuniform particle distribution with
a meshless method, Particuology 9 (3) (2011) 288–297.
[42] F. Li, Investigations on the Turbulent Gas-Solid Two-phase Interactions in Fluidized
Desulfurization Process(Doctor Dissertation) Tsinghua University, Beijing, 2009.
[43] N. Yang, W. Wang, W. Ge, J. Li, CFD simulation of concurrent-up gas–solid flow in
circulating fluidized beds with structure-dependent drag coefficient, Chem. Eng. J.
96 (1–3) (2003) 71–80.
[44] K. Hong, A Two-phase Structure-dependent Multi-fluid Model and Its Application in
Simulation of Gas-Solid Fluidization(Doctor Dissertation) University of Chinese
Academy of Sciences, Beijing, 2013.
[45] K. Hong, W. Wang, Q. Zhou, J. Wang, J. Li, An EMMS-based multi-fluid model (EFM)
for heterogeneous gas–solid riser flows: part I. Formulation of structure-dependent
conservation equations, Chem. Eng. Sci. 75 (2012) 376–389.
[46] H. Zhu, J. Zhu, G. Li, F. Li, Detailed measurements of flow structure inside a dense
gas–solids fluidized bed, Powder Technol. 180 (3) (2008) 339–349.
[47] X. Lv, H. Li, Q. Zhu, Simulation of gas–solid flow in 2D/3D bubbling fluidized beds by
combining the two-fluid model with structure-based drag model, Chem. Eng. J. 236
(2014) 149–157.
[48] J. Chang, J. Zhao, K. Zhang, J. Gao, Hydrodynamic modeling of an industrial turbulent
fluidized bed reactor with FCC particles, Powder Technol. 304 (Supplement C)
(2016) 134–142.
[49] R.H. Venderbosch, The Role of Clusters in Gas-Solids Reactors(Ph. D. dissertation)
Twente University, Netherlands, 1998.
[50] T. Ferreira, W. Rasband, ImageJ User Guide, 2012.
[51] J. Ma, D. Liu, X. Chen, Bubbling behavior of cohesive particles in a two-dimensional
fluidized bed with immersed tubes, Particuology 31 (Supplement C) (2017)
152–160.
[52] J. Chen, N. Yang, W. Ge, J. Li, Computational fluid dynamics simulation of regime
transition in bubble columns incorporating the dual-bubble-size model, Ind. Eng.
Chem. Res. 48 (17) (2009) 8172–8179.
[53] G. Xu, J. Li, Multi-scale interfacial stresses in heterogeneous particle-fluid systems,
Chem. Eng. Sci. 53 (18) (1998) 3335–3339.