You are on page 1of 11

Chemical Engineering Journal 244 (2014) 3343

Contents lists available at ScienceDirect

Chemical Engineering Journal


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

Verication and validation of a coarse grain model of the DEM


in a bubbling uidized bed
Mikio Sakai a,, Minami Abe b, Yusuke Shigeto b, Shin Mizutani b, Hiroyuki Takahashi c, Axelle Vir d,e,
James R. Percival d, Jiansheng Xiang d, Christopher C. Pain d
a

Resilience Engineering Research Center, School of Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan
Department of Systems Innovation, School of Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan
Department of Nuclear Engineering and Management, School of Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan
d
Applied Modeling and Computation Group, Department of Earth Science and Engineering, Imperial College London, Prince Consort Road, London SW7 2BP, UK
e
Faculty of Aerospace Engineering, Delft University of Technology, 2629 HS Delft, The Netherlands
b
c

h i g h l i g h t s

g r a p h i c a l a b s t r a c t

 The coarse grain model was veried

in two-dimensional uidized beds so


far.
 In this study, the coarse grain model
is veried and validated in a 3D
uidized bed.
 Macroscopic behavior agreed
between the original system and the
coarse grain model.
 Adequacy of the coarse grain model is
shown through this study.
 The coarse grain model fabricates
application of the DEM in industrial
scale system.

a r t i c l e

i n f o

Article history:
Received 7 October 2013
Received in revised form 19 December 2013
Accepted 11 January 2014
Available online 18 January 2014
Keywords:
Discrete element method
Coarse grain model
Scaling law
Fluidized beds
DEMCFD method
Gassolid ow

a b s t r a c t
Dense granular ows are encountered in various engineering elds. The discrete element method (DEM)
is used extensively for the numerical simulation of these ow types. Improvements in computer specications have made it possible to apply the DEM in various systems. Although the DEM is well-established,
it has a fatal problem. The problem is that the number of calculated particles is substantially restricted
when the simulation needs to be nished within practical time using a single personal computer. On
the other hand, many industries require application of the DEM to large scale systems on a single personal computer. We therefore developed the DEM coarse grain model in our previous studies. In the
coarse grain model, a group of original particles is simulated using a large-sized particle termed a coarse
grain particle, where the total energy is modeled to agree between the coarse grain and original particles.
The coarse grain model therefore makes it feasible to perform large scale DEM simulations by using a
smaller number of particles than the actual number. The coarse grain model thus far has been veried
in two-dimensional uidized beds. In this study, adequacy of the coarse grain model is veried and validated in a three-dimensional uidized bed, where scaling ratio is low and high respectively. Consequently, the coarse grain model is shown to simulate the macroscopic behavior of solid particles in
three-dimensional dense gassolid ows.
2014 Elsevier B.V. All rights reserved.

Corresponding author. Tel./fax: +81 3 5841 6977.


E-mail address: mikio_sakai@n.t.u-tokyo.ac.jp (M. Sakai).
1385-8947/$ - see front matter 2014 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.cej.2014.01.029

34

M. Sakai et al. / Chemical Engineering Journal 244 (2014) 3343

Nomenclature
m
I
v
Ff
FC
Fg
p
V
T
k
e
uf
g
ds
Cd
Res

particle mass, kg
inertial moment of particle, kg m2
particle velocity, m/s
drag force, N
contact force, N
gravity, N
pressure, N/m2
particle volume, m3
torque, N m
stiffness, N/m
restitution coefcient,
uid velocity, m/s
gravitational acceleration m/s2
solid particle diameter, m
drag coefcient,
particle Reynolds number,

1. Introduction
The discrete element method (DEM) [1] is a Lagrangian approach, where individual solid particle motion is computed based
on Newtons second law of motion. In the DEM, the contact force
is modeled using springs, dashpots and a friction slider. The DEM
is used extensively to simulate dense granular ows such as in a
screw conveyer [2], ribbon mixer [3], ball mill [4], etc. The DEM
can also be applied to simulate complex phenomena related to
agglomeration because of cohesive forces [5,6], segregation in a
polydispersive system [7] or non-spherical particle behavior [8].
Recently, the DEM has been applied to more complex systems such
as soliduid interaction problems. In these applications, solid particle behavior can be simulated by coupling the DEM with computational uid dynamics (DEMCFD method) [9], where the local
volume average technique [10] is introduced into the governing
equations of the uid phase. The DEMCFD method has been used
to model a variety of dense gassolid ows such as in a bubbling
uidized bed [1113], air-and-screen cleaning device [14], die lling [15], etc. Very recently, new LagrangianLagrangian approaches have made it possible to simulate solidliquid ows
involving free surfaces, where the DEM is coupled with a Lagrangian CFD approach such as the Smoothed Particle Hydrodynamics
(SPH) [16] or the Moving Particle Semi-implicit (MPS) methods
[17]. The LagrangianLagrangian approach is called the DEM
MPS method [18,19] or the DEMSPH method [20]. These methods
make it possible to simulate complex soliduid coupling problems such as wet ball mills, where soliduid interactions and free
surface uid ows must be computed simultaneously.
Although the DEM is well-established, it has a fatal problem.
The problem is that number of the calculated particles is substantially restricted when the simulation needs to be nished in a practical time by a single PC. This is especially critical because many
DEM simulations are performed without using a supercomputer.
The calculation time therefore becomes extremely long when simulations are performed by a single personal computer using an
excessive number of particles. This problem cannot be solved even
using the latest multi-core processors [21,22]. Many industries require application of the DEM to large scale systems on a single PC.
We therefore developed a coarse grain model of the DEM in our
previous studies [2325]. This coarse grain model is based on a
scaling law and can simulate the group of original particles by
using a large-sized particle termed a coarse grain particle, where
the total energy is modeled to agree between that of the coarse
grain and original particles. Therefore, solid particle rotation and

Subscript
O
original particle,
CGM
coarse grain model,
Greek letters
x
angular velocity, rad/s
g
damping coefcient, N s/m
l
friction coefcient,
d
particle displacement, m
e
void fraction,
b
inter-phase momentum transfer coefcient, kg/m2s
qf
uid density, kg/m3
sf
viscous stress tensor, N/m2
lf
uid viscosity, Pa s

cohesive forces can be modeled by the coarse grain model. Displacement due to rotational motion or contact or cohesive forces
cannot be simulated by other DEM scaling law models [26,27]
and our coarse grain model is the only one which models these
essential phenomena. In previous studies, the coarse grain model
was veried in powder processes such as in a pneumatic conveying
system [23] and uidized beds [24,25]. In these studies, the coarse
grain model was shown to simulate the macroscopic ow properties in original particle systems, namely, minimum uidization
velocity, bubble velocity, bed height and bed pressure drop. Incidentally, the similar techniques were studied in other Lagrangian
methods (e.g., Refs. [28,29]).
Adequacy of the coarse grain model in the DEMCFD simulation
has been veried only in two-dimensional uidized beds [24,25].
In these verication studies, the coarse grain ratio, which was dened to be the ratio of calculated particle size to original one, was
set to be low, i.e., values of 2.0 and 3.0. This is because it took an
extremely long time to solve the original system when the coarse
grain ratio is high. In the simulations, the calculation time was affected by the number of calculated particles. When the coarse
grain model was applied, the number of original particles becomes
l3 times larger than that of the coarse grain particles. Thus, the
computation time of the original particle system is extremely long
when the coarse grain ratio is high. Moreover, the coarse grain
model has not previously been validated in DEMCFD simulation.
When the validity of the coarse grain model is proven in threedimensional DEMCFD simulation, design or investigation of operational conditions is expected to be efcient for industrial use.
Hence, the verication and validation of the coarse grain model

Fig. 1. Overview of the coarse grain model (Coarse grain ratio l = 2.0).

35

M. Sakai et al. / Chemical Engineering Journal 244 (2014) 3343

in a three-dimensional DEMCFD simulation are required for


industrial applications.
In this study, the coarse grain model is veried and validated
in three-dimensional uidized beds. As a rst step, in the verication and validation test, non-cohesive particles are simulated in a
three-dimensional uidized bed. The verication and validation
tests are performed in not only uidized state but also xed state.
In the verication test, the macroscopic ow properties of the uidized bed such as bubble velocity, bed height and pressure drop,
obtained from the coarse grain model are shown to agree with
that of the original particle systems, where a low scaling ratio is
applied. When the coarse grain ratio is high, calculation for the
original system becomes substantially impossible by using a single PC, because of the huge calculation cost. Hence, the adequacy
of the coarse grain model whose scaling ratio becomes high can
only be demonstrated by a validation test. Consequently, the
coarse grain model is shown to simulate the macroscopic properties of the large-scale particle system, where the calculations are
performed using a smaller number of particles than that in the actual system.

The contact force can be divided into normal and tangential


component. Both components of the force are functions of relative
position and relative velocity of the particles. Namely, the normal
direction is dened by center of mass between the particles. The
tangential direction is dened by the relation between the relative
velocity and normal component of relative velocity. The normal
component of the contact force is given by:
3

2.1. Solid phase

mCGM v_ CGM

F fCGM  V CGM rp
3

l F fO  l V O r p l

P
3

F CCGM F gCGM
P
3
F C O l F gO

where m, v, l, Ff, V, p, FC and Fg indicate solid mass, solid velocity,


coarse graining ratio, drag force, particle volume, pressure, contact
force and gravitational force, respectively. The subscripts CGM and
O refer to the coarse grain model and original particle. A strong
point of the coarse grain model is that original physical properties
can be utilized directly.

where kO, d and gO are the stiffness, the overlap and the damping
coefcient, respectively. As addressed above, the translational motion of the coarse grain particle is assumed to agree with that of
the original particles. Hence, the CGM subscript can be replaced
by O in this equation. In the coarse grain model, the normal component of the displacement dn is modeled by overlap as is the case
with the existing DEM. Moreover, the potential energy, i.e., elastic
energy, is agreed between the coarse grain particle and group of
the original particles by this modeling.
By the same manner as normal component, the tangential component of the contact force is given as:

2. Numerical modeling

Based on our previous studies [2325], we rst address the


coarse grain model for non-cohesive particles which was developed to model contact, drag and gravitational forces as the scaling
law model. As illustrated in Fig. 1, there are l3 original particles in
the coarse grain particle whose size is l times larger than the original particle. In this gure, the coarse grain ratio l was set to 2.0.
The translational motion of the coarse grain particle is assumed to
be the same as average of that of the original particles. Accordingly, the velocity and displacement of the coarse grain particle
are assumed to be averages of the original particles i.e.,
m0 m CGM and do dCGM . As far as the rotational motion is concerned, the original particles existing in the coarse grain particle
are assumed to rotate around each center of mass with equal
angular velocity, where the angular velocity is averaged, as illustrated in Fig. 1. The contact force acting on the coarse grain particle was estimated under the assumption that the total energy
of the coarse grain particle is consistent with that of the original
particles. In addition, when a binary collision of the coarse grain
particles occurs, binary collisions between each of the original
particles (i.e., l3 binary collisions) are assumed to occur simultaneously. The contact force acting on the coarse grain particle is
evaluated using springs, dash-pots and a friction slider, as for
the existing DEM. From the assumption of the translational motion of the coarse grain model, the drag force acting on a coarse
grain particle becomes l3 times larger than that of the original
particle. The gravitational force is modeled by balancing the
coarse grain particle with the group of original particles. As far
as the particle shape is concerned, a spherical model is applied
in the same way as existing DEMCFD simulations. Consequently,
the following relationship is obtained between the coarse grain
and original particles:

 nO
F C nCGM l kO dnCGM  gO v nCGM l kO dnO  gO v

F C tCGM

8
3
>
>
> l kO dtCGM  gO v tCGM
>
>
>
>
 tO jF C t j < lO jF C n j
< l3 kO dtO  gO v
CGM
CGM
3
>
>
l lO jF C nCGM jv tCGM =jv tCGM j
>
>
>
>
>
: l3 l jF jv
 tO =jv
 tO j jF C t j P lO jF C n j
O C nO
CGM
CGM

where lO is the coefcient of friction. The translational motion of


the coarse grain particle is assumed to agree with that of the original particles.
The drag force Ff acting on the coarse grain particle is written
by:

F fCGM

b
b
3
 O V O
uf  v CGM V CGM l
uf  v
1e
1e

where e, uf and b are the void fraction, the uid velocity and an
interphase momentum transfer coefcient, respectively. From
the assumption of the translational motion of the coarse grain
model, namely, agreement of kinetic energy between the coarse
grain and original particles, the drag force acting on a coarse
grain particle becomes l3 times larger than that of the original
particle.
The following relationship is obtained for the rotational motion,
namely, angular velocity and torque, in the original and coarse
grain particles:
3

x_ CGM

4
lrO  l F COt
T CGM r CGM  F C CGMt
l TO 1 
_O

x
5
5
l
ICGM
ICGM
l IO
l IO

Incidentally, Eq. (5) ensures that the rotational energy between


the coarse grain particle and group of original particles agrees.
2.2. Gas phase
The governing equations of the gas phase are the same as those
of the original DEMCFD method even for the coarse grain model.
According to the volume average technique [10], the governing
equations of the gas phase are the uid continuity and Navier
Stokes equations for an incompressible uid, given by:

@e
r  euf 0
@t

@eqf uf
r  eqf uf uf erp  f r  esf eqf g
@t

36

M. Sakai et al. / Chemical Engineering Journal 244 (2014) 3343


Table 3
Calculation conditions (xed state).

20mm

Case 2-1 Case 2-2 Case 2-3 Case 2-4


Solid phase
Coarse grain ratio,
Calculated particle size, lm
Original size, lm
Number of particles,

5mm

50mm
kg/m3
N/m

2500
10
0.9
0.3

kg/m3
Pa s

1.0
1.8  105

Table 2
Calculation conditions (uidized state).
Case 1-1 Case 1-2 Case 1-3 Case 1-4
Solid phase
Coarse grain ratio,
Calculated particle size, lm
Original size, lm
Number of particles,

1.0
150
150
540,000

3.0
450

67,500

20,000

1.0
450
450
20,000

by two approaches, namely, usage of modeling equations and direct numerical simulation [30]. When the local volume average
technique is applied, modeling equations are used in the evaluation of the drag force. A combination of the Ergun [31] and Wen
and Yu [32] correlations has often been employed in previous studies and this combination is therefore employed here. The drag force
of both dilute and dense gassolid ow can be precisely computed
by the model. A void fraction of 0.8 is adopted as the boundary between these two equations in the evaluation of b. The b is given by:

Table 1
Physical properties.

Gas phase
Gas density
Viscosity

2.0
300

Gas phase
Grid size (x, y and z direction), mm 1.0  1.0  1.0
Supercial velocity, m/s
0.015

Fig. 2. Schematic diagram of a 3D uidized bed.

Solid phase
Solid density
Stiffness
Coefcient of restitution
Coefcient of friction

1.0
150
150
540,000

2.0
300

3.0
450

67,500

20,000

1.0
450
450
20,000

8
2
< 150 1e l2f 1:751  e qf juf  v
e
ds
d
s

:3C
4

e1e
ds

qf juf  v je

2:65

e > 0:8

e 6 0:8

10

where lf, ds and Cd are the viscosity, particle diameter and drag
coefcient for an isolated particle, respectively. The drag coefcient
Cd depends on the particle Reynolds number (Res) and is given by:

8
0:687
24
>
< Res 1 0:15Res Res 6 1000
C d 0:44
Res > 1000
>
:

11

where Res is expressed as:

Res

Gas phase
Grid size (x, y and z direction), mm 1.0  1.0  1.0
Supercial velocity, m/s
0.07

juf  v jeqf ds

lf

12

2.3. Algorithm
where q f, f, sf and g are the uid density, drag force, molecular viscous stress tensor expressed by Newtons law of viscosity and gravitational acceleration, respectively. f is dened according to
Newtons third law of motion:

PNgrid
f

F fCGM
V grid

i1

where Vgrid and Ngrid are the volume of the CFD grid and the number
of solid particles located in the local grid, respectively. Ff is given by
Eq. (4).
The e is the ratio of the void volume to the CFD grid volume. The
value of e was estimated by counting all the solid particles existing
in the grid as:

1

PNCGM
grid
i1

V grid

V CGM
si

1

PNCGM
grid
i1

V grid

l V org
si

1

PNorg
grid
i1

V grid

V org
si

where Vs indicates the volume of the solid particles inside the CFD
grid. Eq. (9) indicates that estimation of the void fraction is regarded
to be equivalent between the original system and the coarse grain
model.
A proper drag model for the description of b becomes essential
in the DEMCFD method. The drag force can be evaluated mainly

(1) Solid phase


The solid particle motion, namely, the velocity, position, angular
velocity and angle, was updated using the leap-frog method.
(2) Gas phase
A semi-implicit nite volume method employing a staggered
grid was used for discretizing the incompressible NavierStokes
equation. A hybrid scheme which is a combination of a rst order
upwind scheme and a second order central scheme was used in the
convection term. In this study, the pressurevelocity coupling was
based on the fractional step method for incompressible uid ows.
3. Verication test
Eight simulation cases were performed in the verication. The
adequacy of the coarse grain model is veried by simulating the
original particle system in a three-dimensional uidized bed,
where the coarse grain ratio is set to 2.0 and 3.0. The effectiveness
of the coarse grain model is illustrated by comparing the simulation results from Cases 1-3 and 1-4, where an equivalent particle
size is used with and without the coarse grain model.

37

M. Sakai et al. / Chemical Engineering Journal 244 (2014) 3343

Case 1-1

Case 1-2

Case 1-3

Case 1-4

Fig. 3. Simulation results at quasi-steady state.

0.6

250
Case 1-1
Case 1-2
Case 1-3

200

0.4

Bed height (mm)

Pressure drop (kPa)

0.5

0.3
0.2

100

50

0.1
0
1.5

150

2.5

3.5

4.5

Time (s)
Fig. 4. Comparison of the pressure drop between the coarse grain model and the
original system in a uidized bed.

Coarse grain ratio


Fig. 5. Comparison of the bed height between the coarse grain model and the
original system.

3.1. Calculation conditions


Fig. 2 shows a schematic diagram of the calculation domain. The
domain was discretized with 1 mm sized uniform grids in the x, y
and z directions. Initially, the spherical particles were packed randomly. The grid size was set to be sufciently larger than the particle diameter. The number of grids was 20, 50 and 5, in the x, y and
z direction, respectively. Gas was injected from the base, where the
supercial velocity was set to be 0.070 m/s and 0.015 m/s for uidized and xed state, respectively.
The physical properties of the particles and gas are illustrated
in Table 1. The solid particle was a glass bead with density of
2500 kg/m3. The value of the spring constant was set to be softer
than that estimated for the actual material, according to the previous studies [9,2325]. Coefcients of restitution and friction
were set to be 0.9 and 0.3, respectively. The gas density and viscosity were set to be 1.0 kg/m3 and 1.8  105 Pa s, respectively.

Four kinds of simulations were performed in the verication


tests for uidized state. Table 2 shows the calculation conditions.
Case 1-1 is the original particle system, where the number of solid
particles was 540,000. The coarse grain model was applied in Cases
1-2 and 1-3 where the coarse grain ratio was set to 2.0 and 3.0 and
the number of calculated particles became 67,500 and 20,000. The
number of solid particles became smaller than the actual ones,
since the coarse grain particle represented the group of original
particles. Case 1-4 was performed to investigate the effectiveness
of the coarse grain model, where the calculated particle size was
equal to that in Case 1-3, though the coarse grain model was not
applied.
Four kinds of simulations were performed in the verication
tests for xed state as shown in Table 3. The verication was performed by the same manner as that of uidized state. Case 2-1 is an
original particle system. The coarse grain model was applied in

38

M. Sakai et al. / Chemical Engineering Journal 244 (2014) 3343

Group 1

Group 2

Group 3
Case 1-1

Case 1-2

Case 1-3

Fig. 6. Appearance of different size of bubbles at quasi-steady state.

3.2. Results and discussion


3.2.1. Fluidized state
Fig. 3 shows a typical snapshot obtained from the simulation
results of the uidized state at quasi-steady state. Splashes and
bubbles were observed in the original particle system (Case 1-1)
and the coarse grain model (Case 1-2 and Case 1-3). It appears that
the coarse grain model can simulate the macroscopic ow
properties of the uidized bed. The solid particles hardly moved
in Case 1-4. These results indicate that we could not obtain the
expected results by using simply large particles in a large-scale
DEM simulation. Hence, a scaling law solution such as the coarse
grain model is needed to simulate macroscopic solid phase behavior using large-scale particles. We now compare simulation results
between the original particle system and the coarse grain model.
The bed height and pressure drop obtained from Cases 1-1 to
1-3 are shown in Figs. 4 and 5. These were obtained at the quasisteady state. The bed height and pressure drop were equivalent
between the coarse grain model and original particle system.
Fig. 6 designates the iso-surface with value of 0.85. A variety of
bubble sizes was observed, as shown in this gure. In the present
study, we classied bubbles into three groups by size to verify
the bubble velocity precisely. Groups 1, 2 and 3 are dened to contain bubbles of diameter less than 10 mm, of 10 mm to 20 mm and
more than 20 mm respectively. Fig. 7 designates the velocities in
each group, where velocity was measured at 10 mm height from

the bottom of the bed. As the bubble size increased, so did the
velocity. The velocity can be regarded to be equivalent between
the original particle system and the coarse grain model. Hence,
the macroscopic ow properties of the uidized bed obtained from
the coarse grain model can be regarded to be equivalent to the
original particle system. In other words, the coarse grain model
could simulate the original particle system by using a smaller
number of particles than actual one. This is because the drag force,
contact force and gravity were successfully simulated by the coarse
grain model through the scaling law.
The optimal value or maximum limit of the coarse grain ratio
depends on both the CFD grid size and the original particle

0.4
0.35

Bubble velocity (m/s)

Cases 2-2 and 2-3 where the coarse grain ratio was set to 2.0 and
3.0, respectively. Case 2-4 was performed to investigate the effectiveness of the coarse grain model.

Case 1-1
Case 1-2
Case 1-3

0.3
0.25
0.2
0.15
0.1
0.05
0
0

Group
Fig. 7. Bubble velocity in each group.

39

M. Sakai et al. / Chemical Engineering Journal 244 (2014) 3343

Case 2-1

Case 2-2

Case 2-3

Case 2-4
Fig. 8. Simulation results in a xed bed.

0.6

0.4
0.3

200mm

Pressure drop (kPa)

50mm

Case 2-1
Case 2-2
Case 2-3

0.5

0.2
0.1
0
0.8

0.85

0.9

0.95

Time (s)
Fig. 9. Comparison of the pressure drop between the coarse grain model and the
original system in a xed bed.

diameter. The grid size is generally decided based upon the uid
ow characteristics. Moreover, in the DEMCFD method, the grid
size should be larger than the calculated particle size according
to the local volume average technique [10]. On the other hand,
the coarse grain ratio should be set to be as large as possible from
a viewpoint of efciency. As a matter of course, the coarse grain ratio might set to be much higher when the original particle size is
small. Hence, the optimal value or maximum limit of the coarse
grain ratio is inuenced by the phenomena in the system and cannot be decided universally.

3.2.2. Fixed state


We then performed verication tests in a xed bed. Fig. 8 shows
a typical snapshot obtained from the simulation results of the xed
state (from Cases 2-1 to 2-4). The beds were never moved in all the
cases. The bed height became around 16 mm in all the cases. The

20mm
Fig. 10. Schematic diagram of a uidized bed.

pressure drop obtained from Cases 2-1 to 2-3 is shown in Fig. 9.


As illustrated in the gure, the pressure drop was equivalent between the coarse grain model and original particle system. On
the other hand, the pressure drop obtained from Case 2-4 were extremely different, namely, approximately 19 Pa. Thus, the pressure
drop could not be simulated accurately by using simply large-sized
particles without the coarse grain model. Consequently, adequacy
of the coarse grain model was proved in the xed state.
4. Validation test
Verication of the coarse grain model may be difcult when the
coarse grain ratio increases. This is because the calculation time for
the original particle system may become extremely long. In this
study, the simulation and experimental results are compared.
The adequacy of the coarse grain model with high coarse grain ratio is shown through the validation tests.

40

M. Sakai et al. / Chemical Engineering Journal 244 (2014) 3343

Table 4
Conditions of the validation.
Case 3-E
Experiment

Case 3-S
Simulation

Case 4-E
Experiment

Case 4-S
Simulation

Solid phase: glass beads


Original particle size, lm
Coarse grain ratio,
Calculated particle size, lm
Number of particles,
Weight of powder bed, kg

150

28,800,000
0.127

5.0
750
230,400

150

28,800,000
0.127

5.0
750
230,400

Gas phase: air


Grid size (x, y and z direction), mm
Supercial velocity, m/s

0.070

2.5  2.5  2.5

0.015

2.5  2.5  2.5

Porous media
Flow meter
Control valve
Accumulation tank
Compressor

(a) Experiment

Fig. 11. Schematic diagram of the experimental device.

4.1. Calculation conditions


Fig. 10 shows a schematic diagram of the calculation domain
which was 50 mm wide, 20 mm long and 200 mm high. The domain was discretized by structural grids of 2.5 mm in the x, y
and z directions. The grid size was chosen to be large relative to
the particle diameter, where the uid ow characteristics could
be simulated accurately. The number of grids was 20, 80 and 8,
in the x, y and z directions, respectively. The number of particles
was set to be 230,400, which corresponded to the actual mass of
powder used in the experiment. Initially, the spherical particles
were randomly packed. Air was injected from the base, where
the supercial velocity was set to be 0.070 m/s and 0.015 m/s for
uidized and xed states, as is the case with the verication tests.
The properties of the particles and gas were the same as those
in the verication. The density, stiffness, coefcient of restitution
and coefcient of friction were set to be 2500 kg/m3, 10 N/m, 0.9
and 0.3, respectively. The gas density and viscosity were set to be
1.0 kg/m3 and 1.8  105 Pa s, respectively. The calculation conditions are summarized in Table 4 for Case 3-S and Case 4-S.

4.2. Experimental conditions


Fig. 11 shows a schematic diagram of the experimental device.
The container inner size was the same as the calculation domain,
i.e., 50 mm wide, 20 mm long and 200 mm high. Glass beads with
average diameter of approximately 150 lm (UB-810L, UNION Co.

(b) Simulation
Fig. 12. Particle spatial distribution at quasi-steady state.

Ltd., Osaka, Japan) were inserted randomly. The mass of the powder bed was estimated to be 0.127 kg. The number of the particles
was estimated from the mass of the powder bed. Gas was injected
at the base. The supercial velocity was set to be 0.070 m/s and
0.015 m/s for uidized and xed states. The supercial velocity
and pressure were measured using a ow meter (FD-A10, KEYENCE
Corporation, Osaka, Japan) and pressure gauge (AP-C35, KEYENCE
Corporation, Osaka, Japan), respectively. The supercial velocity
and pressure were recorded using a data logger (NR-500 series,
KEYENCE Corporation, Osaka, Japan). The experimental conditions
are listed in Table 4 for Case 3-E and Case 4-E.

41

M. Sakai et al. / Chemical Engineering Journal 244 (2014) 3343

2.5

2.5
Simulation
Experiment

Simulation
Experiment

Pressure drop (kPa)

Pressure drop (kPa)

1.5

0.5

1.5

0.5

0
1

1.5

2.5

3.5

4.5

2.2

Time (s)

Table 5
Comparison of the bed height between the simulation and experiment.
Simulation

Experiment

105
0.31
7.8

97.4
0.31

(a) Experiment

(b) Simulation
Fig. 14. Particle spatial distribution in a xed bed.

2.6

2.8

Time (s)

Fig. 13. Comparison of the pressure drop between the calculation and experimental
result in a uidized bed.

Bed height (mm)


Standard deviation (mm)
Error (%)

2.4

Fig. 15. Comparison of the pressure drop between the coarse grain model and the
original system in a xed bed.

4.3. Results and discussion


4.3.1. Fluidized state
Fig. 12 illustrates typical snapshots obtained from the simulation and experiment results of the uidized state at quasi-steady
state. In both the simulation and experiment, splashes and bubbles
were observed in the quasi-steady state. The macroscopic ow
properties agreed qualitatively between the simulation and
experiment.
The pressure drop and bed height were measured in the experiment and simulation results for quantitative comparison. Unlike
the above verication test, the calculation results related to the
bubble could not be obtained due to the dense granular ow.
Fig. 13 shows a comparison of the pressure drop between the simulation and experiment. Although uctuations were observed in
the simulation, the recorded pressure became smooth in the experiment. This is because the time resolution of the pressure gauge is
lower than that of the calculation output. The pressure drop agreed
quantitatively between the experiment and simulation. The bed
height was measured by a visual observation to compare the calculation and experiment results fairly. The simulated bed height had
an error of 7.8% against the experimental height (see Table 5)
which is an allowable margin of error from an engineering viewpoint. The coarse grain model can therefore simulate bulk state
of the bed by setting a higher coarse grain ratio. This is because
the particleparticle and particleuid interactions were modeled
precisely in the coarse grain model. This result further supports the
adequacy of the coarse grain model. Moreover, the results imply
that the coarse grain model might simulate phenomena related
to bulk states, such as bubbling, channeling, slugging, etc. Consequently, the coarse grain model could compute the original particle
system using a much smaller number of particles than actual one.
This fact can be understood by comparing Case 1-1 with Case 3-S.
This is because the coarse grain model can enlarge the calculation
domain size, though number of calculated particles was similar in
these cases.
As mentioned above, the coarse grain ratio was restricted by the
grid size in the DEMCFD method. The maximum limit of the
coarse grain ratio depends on the CFD grid size. The grid size
should be larger than the calculated particle size according to the
local volume average technique [10]. The bulk state of the uidized
bed could be simulated by the existing grid size. This implies that
not only the coarse grain ratio but also the CFD grid size were
suitable.

42

M. Sakai et al. / Chemical Engineering Journal 244 (2014) 3343

Incidentally, in previous studies (for example, Ref. [33]), good


results were obtained by EulerianEulerian simulation in largescale systems, though the CFD grid resolution was similar to that
in this study. The EulerianEulerian simulation can be performed
without limits on the number of the calculated particles in contrast
to the DEMCFD method. This fact indicates that the DEMCFD
method can be applied to industrial systems when the fatal problem related to the number of calculated particles is solved. Thus,
the coarse grain model becomes an important approach from the
viewpoint of industrial application of the DEM.
4.3.2. Fixed state
By the same manner as the verication test, a validation test
was performed in a xed bed. Fig. 14 shows a typical snapshot obtained from the simulation and experiment results of the xed
state (from Cases 4-S and 4-E). The beds did not move in these
cases. The bed height became around 84 mm in the simulation
and experiment. The pressure drop obtained from Cases 4-S and
4-E is shown in Fig. 15. The pressure drop of xed bed reached
steady state earlier than that of the uidized state. As illustrated
in the gure, the pressure drop was equivalent between the coarse
grain model and original particle system. Consequently, adequacy
of the coarse grain model was proved in the xed state.
5. Conclusions
Verication and validation tests of the coarse grain model were
performed to determine its adequacy in a three-dimensional uidized bed. The adequacy of the coarse grain model was proven in
uidized and xed state through this study. The coarse grain model
makes the application of the DEM feasible in an industrial scale
system. The following new results were obtained through this
study.
(1) The adequacy of the coarse grain model was veried in the
simulation of three-dimensional uidized beds, where the
coarse grain ratio was set to be low, i.e., at 2.0 and 3.0. The
verication tests were performed in both uidized and xed
beds. In uidized state, the appearance of features such as
splashes and bubbles in the uidized bed agreed qualitatively between the original particle system and the coarse
grain model. Besides, the macroscopic ow properties such
as bed height, pressure drop and bubble velocity agreed
quantitatively between the original system and the coarse
grain model. Also, as far as the xed bed is concerned, the
pressure drop was in good agreement between the original
system and coarse grain model.
(2) When the coarse grain ratio is high, calculation of the original system becomes substantially impossible using a single
PC, because the calculation cannot be completed within a
practical time. Hence, the adequacy of the coarse grain
model with high coarse grain ratio can only be shown efciently by experiment. Hence, we performed a validation
test, where the coarse grain ratio was set to be high, i.e., at
5.0. The verication tests were performed in uidized and
xed beds. In uidized state the appearance of the bed
agreed qualitatively between the original particle system
and the coarse grain model. The macroscopic ow properties
such as the pressure drop and bed height agreed quantitatively between the original system and the coarse grain
model. Also, as far as the xed bed is concerned, the pressure
drop was in good agreement between the original system
and coarse grain model. Consequently, the coarse grain
model is shown to be applied to an actual three-dimensional
uidized bed system.

Hereafter, the coarse grain model is going to be applied to further large-scale or industrial systems.
Acknowledgements
The authors gratefully acknowledge nancial support provided
by the TOYOTA Motor Corporation and Grants for Excellent Graduate Schools from the Japan Society for the Promotion of Science
(JSPS), Japan.
References
[1] P.A. Cundall, O.D.L. Strack, A discrete numerical model for granular assembles,
Geotechnique 29 (1979) 4765.
[2] Y. Shigeto, M. Sakai, Arbitrary-shaped wall boundary modeling based on
signed distance functions for granular ow simulations, Chem. Eng. J. 231
(2013) 464476.
[3] M. Robinson, P.W. Cleary, Flow and mixing performance in helical ribbon
mixers, Chem. Eng. Sci. 84 (2012) 382398.
[4] M. Sakai, K. Shibata, S. Koshizuka, Effect of nuclear fuel particle movement on
nuclear criticality in a rotating cylindrical vessel, J. Nucl. Sci. Technol. 42
(2005) 267274.
[5] K. Washino, H.S. Tan, M.J. Hounslow, A.D. Salman, A new capillary force model
implemented in micro-scale CFDDEM coupling for wet granulation, Chem.
Eng. Sci. 93 (2013) 197205.
[6] P.Y. Liu, R.Y. Yang, A.B. Yu, Dynamics of wet particles in rotating drums: effect
of liquid surface tension, Phys. Fluids 23 (2011) 013304.
[7] M. Marigo, M. Davies, T. Leadbeater, D.L. Cairns, A. Ingram, E.H. Stitt,
Application of Positron Emission Particle Tracking (PEPT) to validate a
Discrete Element Method (DEM) model of granular ow and mixing in the
Turbula mixer, Int. J. Pharm. 446 (2013) 4658.
[8] Y. Guo, C. Wassgren, B. Hancock, W. Ketterhagen, J. Curtis, Granular shear ows
of at disks and elongated rods without and with friction, Phys. Fluids 25
(2013) 063304.
[9] Y. Tsuji, T. Kawaguchi, T. Tanaka, Discrete particle simulation of two
dimensional uidized bed, Powder Technol. 77 (1993) 7987.
[10] T.B. Anderson, R. Jackson, Fluid mechanical description of uidized beds.
Equations of motion, Ind. Eng. Chem. Fundam. 6 (1967) 527539.
[11] K. Luo, M. Fang, S. Yang, K. Zhang, J. Fan, LESDEM investigation of an
internally circulating uidized bed: effects of gas and solid properties, Chem.
Eng. J. 228 (2013) 583595.
[12] M. Xu, F. Chen, X. Liu, W. Ge, J. Li, Discrete particle simulation of gassolid twophase ows with multi-scale CPUGPU hybrid computation, Chem. Eng. J.
207208 (2012) 746757.
[13] C.L. Wu, A.S. Berrouk, K. Nandakumar, Three-dimensional discrete particle
model for gassolid uidized beds on unstructured mesh, Chem. Eng. J. 152
(2009) 514529.
[14] H. Li, Y. Li, F. Gao, Z. Zhao, L. Xu, CFDDEM simulation of material
motion in air-and-screen cleaning device, Comput. Electron. Agric. 88
(2012) 111119.
[15] Y. Guo, C.-Y. Wu, K.D. Kafui, C. Thornton, 3D DEM/CFD analysis of size-induced
segregation during die lling, Powder Technol. 206 (2011) 177188.
[16] J.J. Monaghan, An introduction to SPH, Comput. Phys. Comm. 48 (1988)
8996.
[17] S. Koshizuka, A. Nobe, Y. Oka, Moving-particle semi-implicit method for
fragmentation of incompressible uid, Nucl. Sci. Eng. 123 (1996) 421434.
[18] M. Sakai, Y. Shigeto, X.S. Sun, T. Aoki, T. Saito, J. Xiong, S. Koshizuka,
LagrangianLagrangian modeling for a solidliquid ow in a cylindrical tank,
Chem. Eng. J. 200202 (2012) 663672.
[19] Y. Yamada, M. Sakai, LagrangianLagrangian simulations of solidliquid ows
in a bead mill, Powder Technol. 239 (2013) 105114.
[20] X.S. Sun, M. Sakai, Y. Yamada, Three-dimensional simulation of a solidliquid
ow by the DEM-SPH method, J. Comput. Phys. 248 (2013) 147176.
[21] Y. Shigeto, M. Sakai, Parallel computing of discrete element method on multicore processors, Particuology 9 (2011) 398405.
[22] X. Li, Y. Zhang, X. Wang, W. Ge, GPU-based numerical simulation of multiphase ow in porous media using multiple-relaxation-time lattice Boltzmann
method, Chem. Eng. Sci. 102 (2013) 209219.
[23] M. Sakai, S. Koshizuka, Large-scale discrete element modeling in pneumatic
conveying, Chem. Eng. Sci. 64 (2009) 533539.
[24] M. Sakai, Y. Yamada, Y. Shigeto, K. Shibata, V.M. Kawasaki, S. Koshizuka, Largescale discrete element modeling in a uidized bed, Int. J. Numer. Meth. Fluids
64 (2010) 13191335.
[25] M. Sakai, H. Takahashi, C.C. Pain, J.-P. Latham, J. Xiang, Study on a large-scale
discrete element model for ne particles in a uidized bed, Adv. Powder
Technol. 23 (2012) 673681.
[26] K. Washino, C.H. Hsu, K. Kawaguchi, Y. Tsuji, Similarity model for DEM
simulation of uidized bed, J. Soc. Powder Technol. Japan 44 (2007) 198205
(in Japanese).
[27] K. Kuwagi, H. Takeda, M. Horio, The similar particle assembly (SPA) model, an
approach to large-scale discrete element (DEM) simulation, Fluidization IX
243250 (2004).

M. Sakai et al. / Chemical Engineering Journal 244 (2014) 3343


[28] L.L. Baxter, P.J. Smith, Turbulent dispersion of particles: the STP model, Energy
Fuels 7 (1993) 852859.
[29] M.J. Andrews, P.J. ORourke, The multiphase particle-in-cell (MP-PIC) method
for dense particle ow, Int. J. Multiphase Flow 22 (1996) 379402.
[30] A. Vir, J. Xiang, F. Milthaler, P.E. Farrell, M.D. Piggott, J.-P. Latham, D. Pavlidis,
C.C. Pain, Modelling of uidsolid interactions using an adaptive-mesh uid
model coupled with a combined nite-discrete element model, Ocean Dyn. 62
(2012) 14871501.

43

[31] S. Ergun, Fluid ow through packed columns, Chem. Eng. Progr. 48 (1952)
8994.
[32] C.Y. Wen, Y.H. Yu, Mechanics of uidization, Chem. Eng. Progr. Sympos. Series
62 (1966) 100111.
[33] X. Lv, H. Li, Q. Zhu, Simulation of gassolid ow in 2D/3D bubbling uidized
beds by combining the two-uid model with structure-based drag model,
Chem. Eng. J. 236 (2014) 149157.

You might also like