Chemical Engineering Journal 420 (2021) 127654

Contents lists available at ScienceDirect

Chemical Engineering Journal

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

Development and validation of SuperDEM-CFD coupled model for

simulating non-spherical particles hydrodynamics in fluidized beds
Xi Gao a, b, *, Jia Yu a, Liqiang Lu a, b, Cheng Li a, William A. Rogers a
a
National Energy Technology Laboratory, Morgantown, WV 26506, USA
b
Leidos Research Support Team, Morgantown, WV 26506, USA

A R T I C L E I N F O A B S T R A C T

Keywords: A SuperDEM-CFD coupled model was developed and validated in open-source MFiX for simulating non-spherical
Non-spherical particles fluidization hydrodynamics in fluidized beds. The non-spherical particle was modeled using a super­
Fluidization quadric method and shapes can be modeled by only varying five parameters. DPVM-Satellites interpolation
Superquadric
scheme was developed to distribute the non-spherical particle volume into the neighboring fluid cells. A general
CFD-DEM
MFiX
algorithm to calculate the particle projected area perpendicular to the flow was developed. Two hybrid drag
models that consider the effects of both particle orientation and cell voidage were implemented and compared.
The model was validated by comparing the simulation results with the fluidization experiments of particles of
different shapes, including sphere, cylinder, rod, and cuboid. The solver can correctly simulate the pressure drop,
particle height distribution, and particle orientation distribution. Finally, large-scale systems with 100 million
non-spherical particles were parallel simulated on 6800 cores, which demonstrated the ability of the solver for
industrial-relevant flows simulation.

1. Introduction studies were limited to laboratory scale. Also, it is difficult and expen­
sive to extend these methods for the industrial-scale measurement.
Fluidization of granular materials is widely used in industrial oper­ Computational fluid dynamics (CFD) have gained much attention in
ations, such as drying, mixing, absorption, catalytic reaction, fluid cat­ recent years for fluidized bed simulations because more detailed infor­
alytic cracking (FCC) [1], and biomass/coal thermal chemical mation can be obtained from numerical simulation, and it becomes
conversion [2–4]. Despite the widespread employment in industries, the possible to simulate industrial-scale fluidized bed reactors on super­
underlying fundamental physics is still not fully understood. Most computer owing to the rapid development of computing capability.
granular materials are non-spherical, such as catalysts, sands, pharma­ Several methods have been developed to simulate fluidization flows,
ceutical tablets, biomass pellets, etc. The motion, heat transport, mass such as the Two-Fluid Model (TFM) [11–13], CFD-Discrete Element
transport, and chemical reaction of non-spherical particles are quite Method (DEM) [14–17], Multi-Phase Particle In Cell (MPPIC) [18],
different from spherical particles [2]. Understanding the effect of par­ Particle-Resolved Direct Numerical Simulation (PR-DNS) [19–21], etc.
ticle shape on the fluidization behavior is important for the design, The CFD-DEM coupled method has the ability to tracking individual
optimization, and scale-up of industrial applications. However, relevant particles with a moderate computational cost when compared with
studies on the fluidization of non-spherical particles are extremely other methods mentioned above. This method has been widely used for
lacking in the literature when compared with the fluidization of spher­ spherical particle fluidization simulation. Hundreds of millions of
ical particles [2,5]. spherical particles DEM-CFD simulations have been reported in recent
Several experimental studies were found to investigate the fluidiza­ years [22]. For pure DEM, billions of spherical particle simulation were
tion behavior of non-spherical particles with particle tracking velocim­ reported [23], which shows a great potential of the DEM-CFD for
etry (PTV) [2,6,7], Magnetic particle tracking (MPT) [8], Laser Doppler industrial-scale simulation.
velocimetry (LDV) [9], positron emission particle tracking (PEPT) [10], Several approaches were developed for non-spherical particle DEM
etc. However, limited information was obtained and most of these simulation, including the muti-sphere model [24,25], real geometry

* Corresponding author at: National Energy Technology Laboratory, Morgantown, WV 26506, USA.
E-mail address: [email protected] (X. Gao).

https://doi.org/10.1016/j.cej.2020.127654
Received 22 June 2020; Received in revised form 27 August 2020; Accepted 29 September 2020
Available online 6 November 2020
1385-8947/© 2020 Elsevier B.V. All rights reserved.
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

the capability of the solver of industrial relevant flows simulation.

2. SuperDEM-CFD coupling

2.1. Superquadric particle

In this study, the superquadric method was employed to represent

non-spherical particles instead of the multi-sphere, real geometry, or
polyhedron methods used in other studies. The implicit inside-outside
function of superquadric surface is written as:
[( ) 2 (y)ε2 ]εε21 (z )ε21
x ε
s(x, y, z) = 2
+ 2
+ − 1=0 (1)
a b c

where, a, b, c are the three semi-major axes that control the size of a
Fig. 1. By varying the roundness parameters ε1 and ε2 in Eq.(1), three common
particle. ε1 , ε2 are the roundness parameters that control the smoothness
seen non-spherical particles can be easily generated (semi-axis a = b = 1 and
c = 2). of the particle surface. One of the advantages of the superquadric
method is that almost 80% of all shapes can be molded [31,32] by
varying the five parameters. Fig. 1 shows three commonly seen non-
spherical particles (cuboid, cylinder, and prolate) as examples. The
model [26–28], polyhedron model [29], superquadric model [30], etc.
semi-axes are set to 1, 1, 2, respectively. When ε1 << 1.0 the ε2 << 1.0,
The comparison of different approaches can be found in several studies
a cuboidal particle can be generated. When ε1 << 1.0, ε2 = 1.0, a cy­
[31]. In this study, the superquadric method was employed for non-
lindrical particle can be generated. When ε1 = ε2 = 1.0, a prolate par­
spherical particle DEM simulation because it is a general method to
ticle can be generated.
model nearly all shapes by only varying five parameters [31,32]. In our
previous work [33], a SuperDEM solver based on the superquadric
method was developed for pure granular flow. The solver was validated 2.2. Governing equations for non-spherical particles
by extensive cases, including static packing, rotating drum, hopper
discharge, and so on. The solver was paralleled with MPI and 100 The motion of each particle can be expressed by Newton’s laws in the
million non-spherical particles were parallel simulated on 6800 CPU Lagrangian framework. The governing equations for the position, ve­
cores. locities (translational and rotational) can be expressed as follows:
When coupling the non-spherical DEM with CFD, one needs to
consider the effect of particle shape on the particle–fluid interactions. dXi
= Ui (2)
Different from the drag force for the spherical particle, drag force for dt
non-spherical particles should consider the impacts of both particle
dUi
orientation and cell voidage. Only a few drag models are qualified in the mi = mi g + Fc,i + Fd,i + Fb,i (3)
dt
literature, which was either developed from experimental measurement
[34,35] or DNS simulation [36,37], and the validation of these drag d ωi
models are lacking. Also, the particle volume fraction model, like the Ii + ωi × Ii ωi = τ i (4)
dt
Divided Particle Volume Method (DPVM) for a spherical particle is no
longer accurate for non-spherical particles. Recently, several studies on where, mi and Ii are the particle mass and moment of inertia tensor,
the CFD-DEM simulation of non-spherical particle fluidization were re­ respectively. Ui is the translational velocity. g is the gravity acceleration.
ported. Zhou et al. [38] simulated the fluidization of ellipsoidal parti­ Fc,i is the contact force acting on the particle i due to contacting with
cles. They found that particle shape affects bed permeability and the neighbor particles and wall. ωi and τ i are the angular velocity and total
minimum fluidization velocity significantly. Ma et al. [39] simulated torque acting on the particle i in the body-fixed coordinate system,
rod-like particles in a fluidized bed. They reported that the particle respectively.Fd,i and Fb,i are drag force and buoyance force acting on the
aspect ratio has a significant effect on fluidization behavior. Mahajan particle i due to interacting with the fluid phase.
et al. [40] simulated spherocylindrical particles in a pseudo-2d fluidized The soft-spherical approach was extended for non-spherical parti­
bed. Their results showed that hydrodynamic torque and multi-particle cles. The normal and tangential components of the contact force are
drag is important in CFD-DEM simulation of non-spherical particles. expressed as:
Based on the above discussions, existing studies found that the shapes of
Fn,ij = − kn,ij δn,ij − ηn Un,ij (5)
the particles have a significant effect on the fluidization behavior; drag
force and particle volume division method affect simulation results. { ⃒ ⃒ ⃒ ⃒
⃒ t,ij ⃒⩽μ ⃒Fn,ij ⃒
− kt,ij δ⃒t,ij − ⃒ηt Ut,ij
⃒, F
However, the effect of particle shape on fluidization behavior is not fully Ft,ij = ⃒ ⃒s ⃒ (6)
− μs ⃒Fn,ij ⃒tij , ⃒Ft,ij ⃒ > μs ⃒Fn,ij ⃒
understood. Also, the CFD-DEM solver available for simulating non-
spherical particle fluidization is lacking, especially in an open-source
where δn,ij and δt,ij are the normal and tangential displacement, respec­
framework. Furthermore, large-scale simulation of non-spherical parti­
tively. Un,ij and Ut,ij are the normal and tangential relative velocity,
cle fluidization is not reported yet.
respectively. μs is the friction coefficient. kn,ij and kt,ij are the normal and
This study aims to develop and validate an open-source DEM-CFD
tangential spring stiffness, respectively. ηn and ηt are the normal and
framework for non-spherical particle–fluid fluidization. A new particle
tangential damping coefficient, respectively. The spring stiffness and
volume method (DPVM-satellites) was implemented for non-spherical
damping coefficient were calculated using the Hertzian model, which
particles. Both Di Felice-Ganser hybrid drag and Di Felice-Holzer/
was described in our previous work. Note that the equivalent radius of
Sommerfeld hybrid drag were implemented and compared. The
local curvature was used instead of a volume equivalent radius to better
SuperDEM-CFD solver was validated by comparing it with experimental
calculate the contact force between two non-spherical particles. The
data of different shape particles. Finally, the parallel simulation of 10
equivalent radius of the curvature at the contact point is calculated as
and 100 million non-spherical particles was conducted to demonstrate
follows:

2
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

R = 1/K (7)
∂(αf ρf )
+ ∇⋅(αf ρf Uf ) = 0 (18)
2 2 2 ∂t
1 ∂G ∂G ∂G
K= [∇GT ⋅∇2 G⋅∇G − |∇G|2 ( 2 + 2 + 2 )] (8)
2|∇G|3 ∂X ∂Y ∂Z D(αf ρf Uf )
= − αf ∇P + αf ∇⋅τ f + αf ρf g − If,s (19)
Dt
where ∇G is the gradient at the contact point.
The total torque produced by tangential and normal forces can be where, ρf is the fluid density, uf is the fluid velocity, P is the gas pressure.
calculated as: If,s is the fluid–solid interaction force, mainly the drag force applied in
each fluid cell. The fluid phase stress tensor is written as:
T i = Li × (Fn,ij + Ft,ij ) (9)
τ f = μf ∇Uf + μf (∇Uf )T + λf (∇⋅Uf )I (20)
where Li is the distance vector of the contact point from the center of the
particle.
For a non-spherical particle, the particle orientation must be
2.4. Non-spherical drag forces
modeled. The singularity-free quaternion theory was employed to model
the particle orientation. The transformation between the global coor­
The fluid–solid drag force acting on a single particle can be calcu­
dinate and the local body-fixed coordinate can be achieved through a
lated as:
rotation matrix:
⎡ ⎤ 1 ⃒ ⃒
( ) Fd,i = ρf ⃒Uf,p − Up ⃒CD AV (Uf,p − Up ) = βi (Uf,p − Up )Vi /(αf (1 − αf ))
1 − 2 q22 + q23 2(q1 q2 − q0 q3 ) 2(q0 q2 + q1 q3 ) 2
⎢ ( 2 ) ⎥
Q=⎢ 2 ⎥ (10) (21)
⎣ 2(q0 q3 + q1 q2 ) 1 − 2 q1 + q3 2( − q0 q(1 + q2 q3)) ⎦
2 2
2( − q0 q2 + q1 q3 ) 2(q0 q1 + q2 q3 ) 1 − 2 q1 + q2
where Uf,p is the fluid mean velocity interpolated to the particle location
and will be discussed in a later section. Vi is the particle volume. AV is
where q0 , q1 , q2 and q3 are real numbers and must satisfy q20 + q21 + q22 + the cross-section area of the volume equivalent sphere. CD = CD0 f(αf , Re
q23 = 1.0. A vector can be transformed between different coordinates: , etc) is the drag coefficient. CD0 is the drag coefficient of a single particle.
f(αf , Re, etc) is a function of cell voidage, Reynolds number, etc. to cor­
V G = Q⋅vb (11)
rect the single-particle drag coefficient due to the effect of neighboring
particles.
vb = QT ⋅V G (12)
Several drag coefficient models are available in the literature for
The quaternions need to be updated at each DEM time step as: non-spherical particles, which can be divided into two categories: gen­
eral drag models for a wide range of particle shapes and specific drag
dqi 1
= qi ∘ωi (13) models for a specific particle shape. Ganser drag coefficient [34] and
dt 2
Holzer-Sommerfeld drag coefficient [35] were widely adopted for non-
Rewriting Eq. (13), we can get the following four equations: spherical particle simulations, which were derived from experimental
dqi0 1 data of many different particle shapes and can serve as general drag
= (− qi1 ωix − qi2 ωiy − qi3 ωiz ) (14) coefficients.
dt 2
The Ganser model drag coefficient can be expressed as:
dqi1 1
= (qi0 ωix + qi2 ωiz − qi3 ωiy ) (15) 24 ( ( )0.6567 ) 0.4305K2
dt 2 CGanser
D0 = 1 + 0.1118 Rep K1 K2 + /( ) (22)
Rep K1 1 + 3305 Rep K1 K2
dqi2 1
= (qi0 ωiy + qi3 ωix − qi1 ωiz ) (16) 1 d⊥ 2 − dv
dt 2 K1 = ( + φ 0.5 − 1
) − 2.25 (23)
3 dv 3 D
dqi3 1
= (qi0 ωiz + qi1 ωiy − qi2 ωix ) (17)
dt 2 K2 = 101.8148(− logϕ)0.5743
(24)
A two-phase contact detection algorithm was employed. In the broad
where φ = (dV /dS )2 is the particle sphericity, which is the ratio between
phase contact detection, both bounding sphere and oriented bounding
the surface area of a volume equivalent sphere and the surface area of
box (OBB) was used to reduce the computational cost. In the narrow
the non-spherical particle. D is the diameter of the fluidized bed. d⊥ is
phase contact detection, the deepest point method solved by a
the diameter of a sphere with the equivalent projected area. dV is the
Lagrangian multiplier algorithm was developed to find the maximum
diameter of a sphere with an equivalent volume of the non-spherical
penetration/minimum distance between an interactive pair of particles.
particle.
The contact algorithm was validated in different particulate systems,
The Holzer-Sommerfeld drag coefficient can be written as:
including static packing of oblate chocolate candies in a cylindrical
container, static packing of cylinder in a cylindrical container, dynamics 8 1 16 1 3 1 1
HS
CD0 = √̅̅̅̅̅̅ + √̅̅̅ + √̅̅̅̅̅̅̅ 3/4 + 0.42 × 100.4(− log(φ))0.2
(25)
angle of repose of cylinder in a rotating drum, and chocolate candies Rep φ⊥ Rep φ Rep φ φ⊥
discharge from a hopper. Similarly, a two-phase contact algorithm was
developed and validated for particle and arbitrary wall contact (repre­ where the φ⊥ = d2V /d2⊥ is the crosswise sphericity, which is the ratio
sented by STL facets) detection: bounding sphere was employed in the between the cross-sectional area of a volume equivalent sphere and the
broad phase contact detection and the deepest point method was used in projected cross-section area of the non-spherical particle perpendicular
the narrow phase contact detection. to the flow.
As mentioned before, the single-particle drag coefficient needs to be
2.3. Governing equations for fluid phase corrected to consider the effect of surrounding particles (or cell voi­
dage). The De Felice drag model [41] was employed in several simula­
The fluid phase is described by the Navier-Stokes equation and tions of non-spherical particle fluidization and was adopted here to
solved in the Eulerian framework with a finite volume method. The mass consider the influence of surrounding particles on the single-particle
and momentum governing equations are: drag force. The De Felice drag model is written as follows:

3
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

Fig. 3. Transform the fluid vector into the body-fixed coordinated to calculate
the projected area of the particle perpendicular to the flow.

of a superquadric particle was uniformly divided into N (N = Nx ⋅Ny ⋅Nz )

sub-boxes. Second, the centroid of each sub-box was checked to see
whether it is located inside the superquadric particle or not, and the
number of centroid points Nin inside the superquadric particle can be
obtained. Third, the number of centroid points in the fluid cell j can be
Fig. 2. DPVM-satellites: Divided particle volume method for non-spherical obtained and the spatial weight can be calculated as:
particles based on a satellite point approach (only show 2-D for simplification).
Ni,j
wi,j = (29)
Nin
⃒ ⃒
3 αf (1 − αf )ρf ⃒Uf,p − Up,i ⃒
βi = CD0 α1−f χ
(26)
4 dV 2.6. Projected area of non-spherical particles

(1.5 − logRe)2 The projected area of a non-spherical particle perpendicular to the

χ = 3.7 − 0.65exp( − ) (27)
2 flow depends on the particle shape, particle orientation, and flow di­
rection. To simplify the problem, the projected area can be calculated in
In this study, the Di Felice drag model was coupled with the Ganser
the body-fixed coordinate instead in the global space-fixed coordinate.
drag coefficient and Holzer-Sommerfeld drag coefficient to construct the
The normalized flow vector at the particle’s location was then trans­
Di Felice-Ganser hybrid drag model and Di Felice-Holzer/Sommerfeld
formed into the body-fixed coordinate and was set to u as shown in
hybrid drag model, respectively. Both drag models were employed
Fig. 3. The calculation of the projected area of some common shapes,
and compared in the fluidization simulation of particles with different
like a cylinder, cuboid, and ellipsoid can be found in the literature.
shapes. In both drag models, the surface area and the projected cross-
For a cylinder, the projected area can be calculated as [43]:
section area of a particle perpendicular to the flow needs to be calcu­
lated. For some specific shapes, like a cylinder, a superellipsoid, and a π
A⊥ = LD|sin(ϕ)| + D2 |cos(ϕ)| (30)
cuboid, the analytical expressions can be found in the literature. How­ 4
ever, for a general superquadric particle, the analytical expression is not where, L = 2c and D = 2a = 2b is the length and diameter of the
available, which needs to be calculated numerically. The surface area of cylinder, respectively. ϕ is the angle between the flow vector with the
a general superquadric particle can be calculated by triangulating the symmetry axis.
surface of the particle and the projected area was discussed in a later For a cuboid particle [43], the projected area can be calculated as:
section.
A⊥ = 4bcl + 4acm + 4abn (31)

2.5. Cell voidage model where l = cos(θ)sin(ϕ), m = sin(θ)sin(ϕ), n = cos(ϕ).

For a superellipsoid particle, the projected area can be calculated as
In the CFD-DEM coupled model, the particle volume is required to [44]:
interpolate to the Eulerian fluid grid to calculate the solid volume
fraction (or gas voidage) in a fluid cell. There are two main methods, A⊥ = π(l2 b2 c2 + m2 c2 a2 + n2 a2 b2 )1/2 (32)
namely the centroid method and the divided particle volume method For a general shape particle, an analytical expression of the projected
(DPVM). The centroid method is less accurate than the DPVM method. area is not available and can be calculated numerically [45]. The
In this work, the DPVM method was employed and modified for non- calculation of the projected area needs to determine the rim (boundary
spherical particles as shown in Fig. 2. The voidage in each fluid cell points of the enclosed area) of the particle. The set of points (x, y, z) can
can be calculated as follows: be defined implicitly by:
∑ (33)
N
wi,j VP,i n(x, y, z)⋅u =0
αf ,j = 1 − (28)
Vcell,j
i
The explicit equation of the superquadric particle is defined as:
⎧
⎨ x(ϕ, θ) = asinε1 (ϕ)cosε2 (θ)
Where VP,i is the volume of particle i. Vcell,j is the volume of fluid cell j.
y(ϕ, θ) = bsinε1 (ϕ)sinε2 (θ)
wi,j is the spatial weights (27 elements for 3D, namely divided the par­ ⎩
z(ϕ, θ) = ccosε1 (ϕ)
ticle volume into neighboring 27 fluid cells). Here, a satellite point
method was proposed to calculate the spatial weights, which is similar to π π
− ⩽ϕ⩽ , − π ⩽θ⩽π (34)
the method reported in some other studies [40,42]. To calculate the 2 2
spatial weights, three steps are needed. First, the oriented bounding box The normal of the points can be calculated as follows:

4
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

Fig. 4. Schematic of the staggered grid scheme (only show 2-D for simplification). The fluid fields of neighboring 9 (2D) or 27 (3D) cells are interpolated to the
particle center location to calculate particle drag force.

⎧ 2− ε
⎨ sin 1 (ϕ)cos2− ε2 (ϕ)/a
Table 1
n(ϕ, θ)= sin2− ε1 (ϕ)sin2− ε2 (ϕ)/b (35)
⎩ Particle properties and simulation parameters in the simulation.
cos2− ε1 (ϕ)/c
Shape Sphere Cylinder Rod Cuboid
The equation of the ϕ(θ) can be obtained by combing Eq.(33) and Eq.
Size (mm)D = D= D = 6.1, L D = 3.9, L D = 6.3, L =
(35): diameterL = lengthW 7.2 = 6.2 = 14.0 6.3, W = 5.2
[( ] = width
c (ux 2− ε2 uy )− 1 )1/(2− ε1 )
ϕ(θ)=actan − cos (θ) + sin2− ε2 (θ) (36) Sphericity 1.00 0.87 0.76 0.80
uz a b Density (kg/m3) 772.5 708.5 764.4 746.9
Total mass (kg) 0.535
The rim can be discretized into N points, θi = 2πi/N. Each point Gas viscosity(kg/m.s) 1.83 × 10 -5

(x(ϕ(θi )), y(ϕ(θi )), z(ϕ(θi ))) with a neighboring point and the particle Young’s Modulus (Pa) 1.0 × 107
Poisson’s ratio 0.3
center can form a triangle. The projected area can be calculated by Coefficient of restitution 0.5 (p-p and p-w)
summing the area of all triangles. Coefficient of friction 0.4 (p-p), 0.3 (p-w)

2.7. Interpolation schemes

Vollmari et al. investigated the fluidization behavior of spherical and
non-spherical particles in a rectangular fluidized bed with 0.11 × 0.11
To calculate the particle–fluid drag force, the mean gas velocity at
m2 cross-section area and 0.41 m height. The superficial gas velocity
the particle location is required, which is not available directly because
ranged from 0 m/s to 2.4 m/s, which covers both fixed bed and fluidized
the MFiX fluid solver uses a staggered grid finite volume scheme as
bed regimes. The influences of the gas velocity on the bed pressure drop,
shown in Fig. 4. The scalar fields(like pressure, voidage) represented as
particle height distribution, and particle orientation distribution were
the solid dots are stored at the cell center, while the fluid velocities
experimentally investigated for different shape particles, which is ideal
represented by arrows are stored at the face centers. First, all compo­
for model validation. Four different shapes were simulated in this work,
nents of the fluid velocity were interpolated to the cell center and a cell-
including sphere, cylinder, rod, and cuboid. The roundness parameters
centered fluid velocity can be obtained. Second, the cell-centered gas
were set to ε1 = 0.3, and ε2 = 1.0 in superquadric model for cylindrical
velocity was interpolated to the particle location and the mean gas ve­
particles and ε1 = ε2 = 0.3 for cuboidal particles. The particle properties
locity at the particle location can be calculated as:
and simulation parameters used in the simulation are all consistent with
∑
N
the experiment. The simulation parameters and particle properties are
Uf ,p = wi,j Uf ,j (37) summarized in Table 1. The coefficient of restitution and the coefficient
j
of friction were experimentally measured and reported by Vollmari et al.
where Ug,j is the fluid velocity of the adjacent cells (9 for 2D and 27 for
3D), where the eights wi,j are calculated according to Eq.(29). The cell 3.2. Numerical method
voidage at the particle location can also be calculated similarly.
The fluid solver based on the finite volume method in the open-
3. Simulation condition and numerical method source CFD suite MFiX developed at NETL was used to couple with
SuperDEM for non-spherical particle–fluid fluidization. The Semi-
3.1. Simulation condition Implicit Method for Pressure Linked Equations (SIMPLE) was adopted
for the pressure–velocity coupling. The Crank-Nicholson scheme is used
To validate the Super-DEM solver developed in this work, the for temporal discretization and the second-order upwind scheme was
experiment study by Vollmari et al. [7] was chosen as the test case. adopted for spatial discretization. The new SuperDEM solver developed

5
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

Fig. 5. Comparing predicted pressure drop with experimental data using different drag models, (a) sphere, (b) cylinder, (c) rod, (d) cuboid.

in a previous work was employed to solve particle governing equations. National Energy Technology Laboratory (NETL). The non-spherical
The development, implementation, and validation of the non-spherical particle data generated by SuperDEM simulation was output into VTP
particle–particle, particle–wall contact detection can be found in our files and was loaded into open-source code ParaView for visualization.
previous work. The fluid solver and particle solver are coupled by
interphase forces and voidage model, which was discussed in sections 4. Validation
2.4, 2.5, and 2.7. The CFD time step is set to be 1.0 × 10-4 s, and the DEM
time step is set to be 1/20 of the particle–particle minimum collision 4.1. Pressure drop
( )− 1/2
kn,ij η2n,ij
time (tcol,ij = π meff − 4meff ). The bounding sphere radius is used in The pressure drops obtained for different shape particles (sphere,
cylinder, rod, and cuboid) at different gas velocities (from 0 m/s to 2.4
the minimum collision time calculation instead of the equivalent radius
m/s) in the experiment were employed to validate the SuperDEM-CFD
of curvature. A velocity inlet boundary was used as the inlet and a
coupled solver. Choosing an appropriate gas–solid drag force model is
pressure outlet boundary was used at the outlet. No-slip wall boundary
ultimate for correctly predicted the pressure drop. As discussed in a
was used for the fluid. For unresolved CFD-DEM simulation, Peng has
previous section, two drag models that consider both particle orienta­
identified a lower limit of cell size (1.63 particle diameters) to satisfy the
tion and cell voidage effects were implemented in this study, namely the
assumptions (i.e., local mean fluid variables) of CFD-DEM governing
Di Felice-Ganser hybrid drag and De Felice- Holzer/Sommerfeld hybrid
equations [46]. The cell size used in the simulation is 1.571 × 1.571 ×
drag. The two drag models were tested and compared for different shape
1.571 cm3, which makes the cell size about two times the volume
particles and the results are shown in Fig. 5. For spherical particles
equivalent sphere diameter of a non-spherical particle.
(Fig. 5a), good agreement between simulation and experiment can be
SuperDEM-CFD was parallelized using the messaging passing inter­
observed for the entire regime with the Di Felice- Holzer/Sommerfeld
face (MPI). The computational domain is divided into subdomains (can
drag model. The predicted Umf (1.39 m/s) is close to the experimental
be 1D, 2D, or 3D; uniformly and non uniformly) with specified grids,
value (1.2 m/s). While the Di Felice-Ganser drag model underpredicted
each processor only has the information of the particles/fluid variables
the pressure drop in the fixed bed regime and well predicted the pressure
in the subdomain and the information of ghost particles/fluid variables
drop in the fluidization regime when Ug > 1.5 m/s. The predicted Umf
in ghost cells that replicate information from the adjoining processors.
(1.59 m/s) is significantly larger than the experimental value (1.2 m/s).
The parallelization implement of SuperDEM can be found in our previ­
For non-spherical particles (Fig. 5b-5d), excellent agreement between
ous work, where 100 million non-spherical particles were simulation on
experiment and simulation can be seen over the entire velocity range
6800 CPU cores. In the validation study, each simulation was conducted
using both drag models. In Fig. 5c and 5d, when the superficial gas
using four cores. Fluidization simulation took about three seconds of
velocity is larger than the minimum fluidization velocity, the experi­
physical time to reach statistical steady-state, and the following three
mentally measured pressure drop is lower than the weight of the ma­
seconds simulations were used for postprocessing. The large scale
terial divided by the bottom area (433 Pa). This is because, for the rods
simulation was run on the Joule 2.0 supercomputer located at the
and the cuboids, channels tend to form inside the bed; the gas bypass

6
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

models were developed specifically for non-spherical particles and work

better for particles with lower sphericity. Some other drag models, for
example, the widely used Gidaspow drag model work well for spherical
particles. For the fluidization of each shape, the error of the Di Felice -
Holzer/Sommerfeld drag is lower than that of the Di Felice-Ganser drag.
Thus, the Di Felice- Holzer/Sommerfeld drag was used for later
simulation.

4.2. Fluidization behavior

To comprehensively validate the solver, the fluidization of non-

spherical particles, including sphere, cylinder, rod, and cuboid were
simulated and compared with the experimental data reported in the
work of Vollmari et al [7]. The volume equivalent sphere diameter of the
non-spherical particles are all close to that of the spherical particles
(about 7 mm), and the density of all shape particles are around 750 kg/
m3, which makes the experiment ideal for studying the effect of particle
Fig. 6. The relative error of pressure drop predicted by different drag models shapes on fluidization behavior. In the experiment, the superficial gas
for the sphere, cylinder, rod, and cuboid particles. velocities ranged from 0 m/s to 2.4 m/s, which can be used to validate
the solver for flows in both the fixed bed and fluidization regimes. The
through the channels and not the entire bed is fluidized. While the particles were dropped into the bed with random orientations to
simulation can not reproduce this behavior because the fluid cell size is generate the initial packed bed. After all the particles remain static, the
larger than the particle size, the flow around the particle was not packed bed is used for the initial condition for other simulations with
resolved. Particle-resolved DEM-CFD solver might be able to reproduce non-zero gas velocity. Figs. 7–10 show the snapshots of the packing and
the channel flow in non-spherical particle fluidization, while it is fluidization behavior of different shape particles at 0 m/s, 1.6 m/s, and
extremely computationally expensive. To quantitatively assess the effect 2.4 m/s. The packed bed of particles was formed by dropping particles
of drag models, the absolute relative errors were calculated: with random orientation into the rectangular bed. The surface of the
⃒
⃒
⃒ packed bed is smooth for the spherical particles, while they are bumpy
i ⃒
∑N ⃒pi
sim − pexp ⃒ for non-spherical particles. The fluidization behavior of the non-
Eabs = (38)
i=1
piexp N spherical particles is also quite different from the spherical particles.
For example, the minimum fluidization velocity of the non-spherical
Fig. 6 shows the relative error of the pressure drop prediction using particles is smaller than that of the spherical particles. The expansion
the two drag models for different shape particles. The relative error of bed height of the sphere is much lower than the non-spherical particles.
the Di Felice- Holzer/Sommerfeld is about 8% − 17%. The relative error In Fig. 10 (Ug = 2.4 m/s), it is interesting to see that some cuboidal
of the Di Felice-Ganser drag is about 11%-27%. The relative error for the particles are not fluidized near the corner of the bed and form a structure
sphere is larger than other shapes, which might because the two drag with several particles neatly arranged in the vertical direction.

Fig. 7. Snapshots of the packing and fluidization behavior of spherical particles in the fluidized bed at the superficial gas velocity of 0 m/s, 1.6 m/s and 2.4 m/s,
respectively.

7
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

Fig. 8. Snapshots of the packing and fluidization behavior of cylindrical particles in the fluidized bed at the superficial gas velocity of 0 m/s, 1.6 m/s and 2.4 m/s
respectively.

Fig. 9. Snapshots of the packing and fluidization behavior of rods in the fluidized bed at the superficial gas velocity of 0 m/s, 1.6 m/s and 2.4 m/s, respectively.

4.3. Particle height distribution the model. Fig. 11 shows the comparison of the simulated particle height
distribution with the experimental data for cylinder, rod, and cuboid. To
Analyzing the particle height distribution was conducted to further quantitatively assess the model, the absolute relative errors of predicted
validate the solver for fluidization simulation of different shape parti­ particle hight were calculated (see Eq. (38)). Note that data points
cles. The particle height distribution along the height is a measure of the (<0.01) were not considered, as a very small value in the denominator in
bed expansion height, which was usually used to assess the accuracy of Eq. (38) results in a very large unphysical value. The relative error of

8
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

Fig. 10. Snapshots of the packing and fluidization behavior of cuboid particles in the fluidized bed at the superficial gas velocity of 0 m/s, 1.6 m/s and 2.4 m/s
respectively.

Fig. 11. Comparing predicted particle height distribution with experimental data, (a) cylinder, (b) rod, (c) cuboid.

9
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

Fig. 12. The relative error of predicted particle height for the cylinder, rod, and
Fig. 14. The relative error of predicted particle orientation for the cylinder,
cuboid particles at different gas velocity.
rod, and cuboid particles at different gas velocity.

can be seen. The expansion height was overpredicted in the simulation.

predicted particle height for the cylinder, rod, and cuboid particles at
The derivation shows that the non-spherical drag becomes less accurate
different gas velocity was shown in Fig. 12. It can be seen that the
at high gas velocity. More non-spherical drag models need to be tested in
averaged error is 10.6%,13.2%,14.1% for gas velocity 0 m/s, 1.6 m/s
the future.
and 2.4 m/s, respectively. With the increase of gas velocity, the error
slightly increases. In the packed bed (Ug = 0 m/s), the curves of particle
height distribution match their experimental counterpart very well. The 4.4. Particle orientation distribution
packing pattern of the non-spherical particles directly depends on the
DEM model, which demonstrates the accuracy of the pure DEM part of The particle orientation in the bed directly depends on the parti­
the coupled solver. For the velocity of 1.6 m/s, good agreement between cle–particle contact and the particle–fluid interaction. Comparing the
experimental and numerical results can be observed. For the velocity of particle orientation distribution with experimental data can further
2.4 m/s, a fair agreement between experimental and numerical results assess the accuracy of the solver. Also, analyzing the effect of gas

Fig. 13. Comparing predicted particle orientation distribution (left) with experimental data (right), (a) cylinder, (b) rod, (c) cuboid.

10
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

Table 2 particles increases, while the fraction of horizontally-orientated parti­

Particle properties and simulation parameters in SuperDEM-CFD simulation of cles decreases. Similar trends were also reported in other studies
oblate particles (M&M chocolate candies). [6,47,48]. The SuperDEM-CFD model correctly predicted the trend. In
Parameter Value Parameter Value Fig. 13a and 13c, the ideal cylinder and cuboid do not show a decrease of
Axis, 2a = 2b (mm) 13.56 Coefficient of 0.5
the fraction of laying down particles with the increasing of gas velocity.
friction The same findings can also be found in other studies [49]. To quanti­
Axis, 2c (mm) 7.19 Particle number 10 million, tatively assess the model, the absolute relative errors of predicted par­
100million ticle orientation were calculated (see Eq. (38)). Note that data points
Particle density (kg/ 1338 Gas density (kg/m3) 1.225
(<0.01) were not considered. The relative error of predicted particle
m3)
Young’s modulus (Pa) 1.0 × Gas viscosity(kg/m. 1.83 × 10-5 orientation for the cylinder, rod, and cuboid particles at different gas
107 s) velocity was shown in Fig. 14. It can be seen that the averaged error is
Poisson’s ratio 0.29 DEM time step (s) 2.38 × 10-6 36.5%,31.4%,32.9% for gas velocity 0 m/s, 1.6 m/s and 2.4 m/s,
Coefficient of 0.5 Fluid time step (s) 1.0 × 10-5-1.0 × 10-3 respectively. With the increase of gas velocity, no clear trend was
restitution
observed. Fig. 14 shows that the agreement between simulation and
experiment is fair. Particle-resolved DEM-CFD simulation can be con­
velocity and particle shape on the orientation distribution can gain a ducted in the future to analyze the source of errors.
better understanding of the fluidization behavior of non-spherical par­
ticles. The comparison of the predicted particle orientation distribution 5. Large scale simulation
with the experimental data is shown in Fig. 13. Value of 00 indicates
horizontal/lying particles and value of 900 indicates vertical/standing In the industrial fluid-particle system, the particle number is in the
particles. The range of the orientation angle was divided into bins with order of 100 million to 10 billion or even more. Massively parallelization
10 degrees of each (18 bins for rod and 9 bins for others), which is the of the solver is required for large-scale fluidization simulation. For DEM-
same as the experiment. The number of particle count in each bin was CFD coupled simulation, hundreds of millions of spherical particles have
calculated and then frequency was obtained. In Fig. 13b for rods, with an been reported previously. Xu et al. [50] simulated 127 million spherical
increase of superficial gas velocity, the fraction of vertically-orientated particles in a circulating fluidized bed using an in-house DEM code

Fig. 15. Snapshot of chocolate candies fluidization in a rectangular box at 2 s, (a) 10 million particles, (b) 100 million particles.

11
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

Fig. 16. Pressure drop evolution with time in the large-scale simulation of non-spherical particle fluidization (a) 10 million particles, (b) 100 million particles.

clear to see that the solver could properly simulate the fluidization
behavior of a large number of particles. Fig. 16 shows the pressure drop
evolution with time. The pressure drop first increased and then decrease,
and then gradually approach the expected value (material weight
divided by the bottom area). The simulation results demonstrate the
capability of the SuperDEM-CFD coupled solver for large-scale fluid­
ization systems.
The computational cost of large-scale simulation is reported in
Fig. 17. For 10 million particles, the simulation of one physical time took
about 274.9 h on 200 cores and 109.5 h on 750 cores. The relative
speedup is 2.5 with an efficiency of 67%. For 100 million particles, the
simulation of one physical time took about 267.0 h on 2000 cores and
128.7 h on 6800 cores. The relative speedup is 2.1 with an efficiency of
61%. The large scale simulation in this study indicates that the
SuperDEM-CFD coupled solver can handle at least 100 million non-
spherical particle fluidization simulations on thousands of cores,
which has great potential for industrial-scale non-spherical-particle
systems simulation.

Fig. 17. Wall clock time per one second physical time and speedup in 6. Conclusion
SuperDEM-CFD simulations of 10 million and 100 million oblate chocolate
candies fluidization in rectangular boxes with different numbers of cores.
In this study, a SuperDEM-CFD coupled solver was developed and
validated for non-spherical particulate-fluid fluidization systems in the
open-source CFD suite MFiX. A superquadric particle method was used
coupling with OpenForm. Capecelatro and Desjardins [22] simulated
to model the shape of the non-spherical particles. To correctly calculate
382 million spherical particles in a channel flow using NGA code on
the voidage in the fluid cell, a new interpolation method – Divided
4096 cores with MPI parallelization. However, for non-spherical parti­
Particle Volume Method (DPVM)-Satellites was developed for non-
cle–fluid flow, large-scale simulation with millions of particles are not
spherical particles. Two hybrid drag models, namely the Di Felice-
reported.
Ganosor hybrid drag and Di Felice-Holzer/Sommerfeld hybrid drag
To test the ability of the SuperDEM-CFD coupled solver for large
were implemented. To calculate the projected area of non-spherical
scale particle–fluid simulation, fluidization simulations with 10 million
particles perpendicular to the flow, a general numerical algorithm was
and 100 million oblate particles (M&M candies) were conducted. For
developed. The solver was validated by comparing the simulation results
this simple rectangular domain, a 2D parallelization strategy (in X and Z
with the fluidization experiment of different particle shapes, including
direction) is used to distribute a similar number of particles in each core
sphere, cylinder, rod, cuboid, etc. It was found that the Di Felice –
to balance the load. The particle properties are the same as those re­
Holzer/Sommerfeld drag model performed better than the Di Felice –
ported in our previous work and are summarized in Table 2. For 10
Holzer/Sommerfeld drag model for all simulated shapes. The solver also
million particles fluidization simulation, the simulated domain has a
correctly reproduce the experimentally measured particle height dis­
length of 3.75 m, a height of 5.18 m, and a depth of 3.49 m. The DEM
tribution and orientation distribution. Finally, large-scale fluidization
grid and CFD grid were both set to 86 × 64 × 84. For 100 million par­
systems with 100 million prolate non-spherical particles were parallel
ticles fluidization simulation, the simulated domain has a length of
simulated on 6800 CPU cores, which demonstrates the ability of the
11.85 m, a height of 5.18 m, and a depth of 11.04 m. The DEM grid and
solver for industrial-scale fluidization simulation.
CFD grid were both set to 290 × 128 × 270. The bottom of the domain is
set as the gas inlet with a superficial gas velocity of 3.0 m/s. The top of
the domain is set as the pressure outlet boundary. Two seconds of Declaration of Competing Interest
physical time was simulated in all cases.
The snapshots of particle velocity distribution in the 10 million and The authors declare that they have no known competing financial
100 million particles fluidization simulation at 2 s are shown in Fig. 15. interests or personal relationships that could have appeared to influence
The region with high velocity indicates the formation of bubbles. It is the work reported in this paper.

12
X. Gao et al. Chemical Engineering Journal 420 (2021) 127654

Acknowledgement [18] R. Garg, J. Dietiker, Documentation of open-source MFIX–PIC software for gas-
solids flows, From URS https://mfix. netl. doe. gov/documentation/mfix_pic_doc.
pdf (2013).
This work was performed with support from the CFD for Advanced [19] G. Karthik, V.V. Buwa, Effect of particle shape on catalyst deactivation using
Reactor Design (CARD) project. The Research was executed through the particle-resolved CFD simulations, Chem. Eng. J. 377 (2019), 120164.
NETL Research and Innovation Center’s CARD FWP. Research per­ [20] Q. Zhou, L.-S. Fan, Direct numerical simulation of low-Reynolds-number flow past
arrays of rotating spheres, J. Fluid Mech. 765 (2015) 396–423.
formed by Leidos Research Support Team staff was conducted under the [21] X. Liu, W. Ge, L. Wang, Scale and structure dependent drag in gas–solid flows,
RSS contract 89243318CFE000003. AIChE J. 66 (2020), e16883.
[22] J. Capecelatro, O. Desjardins, An Euler-Lagrange strategy for simulating particle-
laden flows, J. Comput. Phys. 238 (2013) 1–31.
Disclaimer [23] M. Furuichi, D. Nishiura, M. Asai, T. Hori, The first real-scale DEM simulation of a
sand-box experiment using 2.4 billion particles, The International Conference for
This report was prepared as an account of work sponsored by an High Performance Computing, Networking, Storage and Analysis, 2017.
[24] B. Soltanbeigi, A. Podlozhnyuk, J.Y. Ooi, C. Kloss, S.-A. Papanicolopulos,
agency of the United States Government. Neither the United States Comparison of multi-sphere and superquadric particle representation for modelling
Government nor any agency thereof, nor any of their employees, makes shearing and flow characteristics of granular assemblies, EPJ Web of Conferences,
any warranty, express or implied, or assumes any legal liability or re­ EDP Sciences (2017) 06015.
[25] A.M. Lattanzi, J.J. Stickel, Hopper Flows of Mixtures of Spherical and Rod-Like
sponsibility for the accuracy, completeness, or usefulness of any infor­ Particles via the Multi-Sphere Method, AIChE J. (2019).
mation, apparatus, product, or process disclosed, or represents that its [26] Y. Guo, C. Wassgren, B. Hancock, W. Ketterhagen, J. Curtis, Computational study
use would not infringe privately owned rights. Reference herein to any of granular shear flows of dry flexible fibres using the discrete element method,
J. Fluid Mech. 775 (2015) 24–52.
specific commercial product, process, or service by trade name, trade­
[27] K. Kildashti, K. Dong, B. Samali, Q. Zheng, A. Yu, Evaluation of contact force
mark, manufacturer, or otherwise does not necessarily constitute or models for discrete modelling of ellipsoidal particles, Chem. Eng. Sci. 177 (2018)
imply its endorsement, recommendation, or favoring by the United 1–17.
States Government or any agency thereof. The views and opinions of [28] M. Kodam, R. Bharadwaj, J. Curtis, B. Hancock, C. Wassgren, Cylindrical object
contact detection for use in discrete element method simulations, Part I-Contact
authors expressed herein do not necessarily state or reflect those of the detection algorithms, Chemical Engineering Science 65 (2010) 5852–5862.
United States Government or any agency thereof. [29] N. Govender, D.N. Wilke, S. Kok, R. Els, Development of a convex polyhedral
discrete element simulation framework for NVIDIA Kepler based GPUs, J. Comput.
Appl. Math. 270 (2014) 386–400.
References [30] Y. Zhao, H. Xiao, P.B. Umbanhowar, R.M. Lueptow, Simulation and modeling of
segregating rods in quasi-2D bounded heap flow, AIChE J. 64 (2018)
[1] X. Gao, T. Li, W.A. Rogers, K. Smith, K. Gaston, G. Wiggins, J.E. Parks II, Validation 1550–1563.
and application of a multiphase CFD model for hydrodynamics, temperature field [31] B. Soltanbeigi, A. Podlozhnyuk, S.-A. Papanicolopulos, C. Kloss, S. Pirker, J.Y. Ooi,
and RTD simulation in a pilot-scale biomass pyrolysis vapor phase upgrading DEM study of mechanical characteristics of multi-spherical and superquadric
reactor, Chem. Eng. J. 124279 (2020). particles at micro and macro scales, Powder Technol. 329 (2018) 288–303.
[2] X. Gao, J. Yu, C. Li, R. Panday, Y. Xu, T. Li, H. Ashfaq, B. Hughes, W.A. Rogers, [32] W. Zhong, A. Yu, X. Liu, Z. Tong, H. Zhang, DEM/CFD-DEM modelling of non-
Comprehensive experimental investigation on biomass-glass beads binary spherical particulate systems: theoretical developments and applications, Powder
fluidization: A data set for CFD model validation, AIChE J. 66 (2020), e16843. Technol. 302 (2016) 108–152.
[3] L. Lu, X. Gao, M. Shahnam, W.A. Rogers, Coarse Grained CFD-DEM Simulation of [33] X. Gao, J. Yu, R.J.F. Portal, J.-F. Dietiker, M. Shahnam, W.A. Rogers, SuperDEM for
Sands and Biomass Fluidization with a Hybrid Drag, AIChE J. (2019). non-spherical particulate systems using a superquadric particle method, Under
[4] J. Yu, J.D. Smith, H. Golpour, A. Alembath, H. Al-Rubaye, X. Gao, Validation and Review (2020).
Application of a Kinetic Model for Downdraft Biomass Gasification Simulation, [34] G.H. Ganser, A rational approach to drag prediction of spherical and nonspherical
Chemical Engineering & Technology. particles, Powder Technol. 77 (1993) 143–152.
[5] S. Wang, K. Luo, J. Fan, CFD-DEM coupled with thermochemical sub-models for [35] A. Hölzer, M. Sommerfeld, New simple correlation formula for the drag coefficient
biomass gasification: Validation and sensitivity analysis, Chem. Eng. Sci. 217 of non-spherical particles, Powder Technol. 184 (2008) 361–365.
(2020), 115550. [36] S.K. Sanjeevi, J.T. Padding, Hydrodynamic forces on monodisperse assemblies of
[6] Z. Zou, Y.-L. Zhao, H. Zhao, L.-B. Zhang, Z.-H. Xie, H.-Z. Li, Q.-S. Zhu, axisymmetric elongated particles: orientation and voidage effects, AIChE J.
Hydrodynamic and solids residence time distribution in a binary bubbling fluidized (2020).
bed: 3D computational study coupled with the structure-based drag model, Chem. [37] Z. Cao, D.K. Tafti, Investigation of drag, lift and torque for fluid flow past a low
Eng. J. 321 (2017) 184–194. aspect ratio (1: 4) cylinder, Comput. Fluids 177 (2018) 123–135.
[7] K. Vollmari, R. Jasevičius, H. Kruggel-Emden, Experimental and numerical study of [38] Z. Zhou, D. Pinson, R. Zou, A. Yu, Discrete particle simulation of gas fluidization of
fluidization and pressure drop of spherical and non-spherical particles in a model ellipsoidal particles, Chem. Eng. Sci. 66 (2011) 6128–6145.
scale fluidized bed, Powder Technol. 291 (2016) 506–521. [39] H. Ma, L. Xu, Y. Zhao, CFD-DEM simulation of fluidization of rod-like particles in a
[8] K.A. Buist, P. Jayaprakash, J. Kuipers, N.G. Deen, J.T. Padding, Magnetic particle fluidized bed, Powder Technol. 314 (2017) 355–366.
tracking for nonspherical particles in a cylindrical fluidized bed, AIChE J. 63 [40] V.V. Mahajan, T.M. Nijssen, J. Kuipers, J.T. Padding, Non-spherical particles in a
(2017) 5335–5342. pseudo-2D fluidised bed: Modelling study, Chem. Eng. Sci. 192 (2018) 1105–1123.
[9] H. Morikita, K. Hishida, M. Maeda, Measurement of size and velocity of arbitrarily [41] R. Di Felice, The voidage function for fluid-particle interaction systems, Int. J.
shaped particles by LDA based shadow image technique, Developments in Laser Multiph. Flow 20 (1994) 153–159.
Techniques and Applications to Fluid Mechanics, Springer (1996) 354–375. [42] X. Liu, J. Gan, W. Zhong, A. Yu, Particle shape effects on dynamic behaviors in a
[10] Z. Yang, X. Fan, S. Bakalis, D. Parker, P. Fryer, A method for characterising solids spouted bed: CFD-DEM study, Powder Technol. 361 (2020) 349–362.
translational and rotational motions using Multiple-Positron Emission Particle [43] R. Saucier, Resolving the Orientation of Cylinders and Cuboids from Projected Area
Tracking (Multiple-PEPT), Int. J. Multiph. Flow 34 (2008) 1152–1160. Measurements, ARMY RESEARCH LAB ABERDEEN PROVING GROUND MD
[11] X. Gao, T. Li, A. Sarkar, L. Lu, W.A. Rogers, Development and validation of an ABERDEEN PROVING GROUND United States, 2016.
enhanced filtered drag model for simulating gas-solid fluidization of Geldart A [44] G. Vickers, The projected areas of ellipsoids and cylinders, Powder Technol. 86
particles in all flow regimes, Chem. Eng. Sci. 184 (2018) 33–51. (1996) 195–200.
[12] X. Gao, T. Li, W.A. Rogers, Assessment of mesoscale solid stress in coarse grid TFM [45] J. Hilton, L. Mason, P. Cleary, Dynamics of gas–solid fluidised beds with non-
simulation of Geldart A particles in all fluidization regimes, AIChE J. (2018). spherical particle geometry, Chem. Eng. Sci. 65 (2010) 1584–1596.
[13] K.M. Kellogg, P. Liu, C.Q. LaMarche, C.M. Hrenya, Continuum theory for rapid [46] Z. Peng, E. Doroodchi, C. Luo, B. Moghtaderi, Influence of void fraction calculation
cohesive-particle flows: general balance equations and discrete-element-method- on fidelity of CFD-DEM simulation of gas-solid bubbling fluidized beds, AIChE J. 60
based closure of cohesion-specific quantities, J. Fluid Mech. 832 (2017) 345–382. (2014) 2000–2018.
[14] Y. Xu, T. Li, L. Lu, X. Gao, S. Tebianian, J.R. Grace, J. Chaouki, T.W. Leadbeater, [47] H. Ma, Y. Zhao, Investigating the fluidization of disk-like particles in a fluidized
R. Jafari, D.J. Parker, Development and confirmation of a simple procedure to bed using CFD-DEM simulation, Adv. Powder Technol. 29 (2018) 2380–2393.
measure solids distribution in fluidized beds using tracer particles, Chem. Eng. Sci. [48] V.V. Mahajan, J.T. Padding, T.M. Nijssen, K.A. Buist, J. Kuipers, Nonspherical
115501 (2020). particles in a pseudo-2D fluidized bed: Experimental study, AIChE J. 64 (2018)
[15] Y. Xu, X. Gao, T. Li, Numerical study of the bi-disperse particles segregation inside a 1573–1590.
spherical tumbler with Discrete Element Method (DEM), Comput. Math. Appl. (2019). [49] X. Chen, J. Wang, J. Li, Coarse grid simulation of heterogeneous gas–solid flow in a
[16] A. Morris, S. Pannala, Z. Ma, C. Hrenya, Development of soft-sphere contact models CFB riser with polydisperse particles, Chem. Eng. J. 234 (2013) 173–183.
for thermal heat conduction in granular flows, AIChE J. 62 (2016) 4526–4535. [50] J. Xu, X. Liu, S. Hu, W. Ge, Virtual process engineering on a three-dimensional
[17] S. Wang, K. Luo, C. Hu, J. Lin, J. Fan, CFD-DEM simulation of heat transfer in circulating fluidized bed with multiscale parallel computation, J. Adv. Manuf.
fluidized beds: Model verification, validation, and application, Chem. Eng. Sci. 197 Process. 1 (2019), e10014.
(2019) 280–295.

13