J. Fluid Mech. (2010), vol. 661, pp. 482–510.

c Cambridge University Press 2010
doi:10.1017/S002211201000306X

Discrete particle simulation of particle–fluid

flow: model formulations and their applicability
Z. Y. Z H O U, S. B. K U A N G, K. W. C H U
A N D A. B. Y U†
Laboratory for Simulation and Modelling of Particulate Systems, School of Materials Science and
Engineering, The University of New South Wales, Sydney NSW 2052, Australia

(Received 3 September 2009; revised 28 May 2010; accepted 28 May 2010;

first published online 25 August 2010)

The approach of combining computational fluid dynamics (CFD) for continuum fluid
and the discrete element method (DEM) for discrete particles has been increasingly
used to study the fundamentals of coupled particle–fluid flows. Different CFD–DEM
models have been used. However, the origin and the applicability of these models
are not clearly understood. In this paper, the origin of different model formulations
is discussed first. It shows that, in connection with the continuum approach, three
sets of formulations exist in the CFD–DEM approach: an original format set I,
and subsequent derivations of set II and set III, respectively, corresponding to the
so-called model A and model B in the literature. A comparison and the applicability
of the three models are assessed theoretically and then verified from the study of
three representative particle–fluid flow systems: fluidization, pneumatic conveying and
hydrocyclones. It is demonstrated that sets I and II are essentially the same, with
small differences resulting from different mathematical or numerical treatments of a
few terms in the original equation. Set III is however a simplified version of set I.
The testing cases show that all the three models are applicable to gas fluidization
and, to a large extent, pneumatic conveying. However, the application of set III is
conditional, as demonstrated in the case of hydrocyclones. Strictly speaking, set III is
only valid when fluid flow is steady and uniform. Set II and, in particular, set I, which
is somehow forgotten in the literature, are recommended for the future CFD–DEM
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

modelling of complex particle–fluid flow.

Key words: fluidized beds, granular media, particle–fluid flows

1. Introduction
Coupled particle–fluid flow can be observed in almost all types of particulate
processes which are widely used in industry. Understanding the fundamentals
governing the flow and formulating suitable governing equations and constitutive
relationships are of paramount importance to the formulation of strategies for process
development and control. This necessitates a multi-scale approach to understanding
the phenomena at different time and length scales (see e.g. Villermaux 1996; Xu &
Yu 1997; Li 2000; Tsuji 2007; Zhu et al. 2007). The existing approaches to modelling
particle flow can generally be classified into two categories: the continuum approach
at a macroscopic level and the discrete approach at a microscopic level. In the

† Email address for correspondence: [email protected]

 Discrete particle simulation of particle–fluid flow 483
continuum approach, the macroscopic behaviour is described by balance equations,
e.g. mass and momentum, closed with constitutive relations together with initial
and boundary conditions (see e.g. Anderson & Jackson 1967; Ishii 1975; Gidaspow
1994; Enwald, Peirano & Almstedt 1996). The discrete approach is based on the
analysis of the motion of individual particles, i.e. typically by means of the discrete
element method (DEM) (Cundall & Strack 1979). The method considers a finite
number of discrete particles interacting by means of contact and non-contact forces,
and every particle in a considered system is described by Newton’s equations of
motion. On the other hand, fluid flow can be modelled at different time and length
scales from discrete (e.g. molecular dynamic simulation, lattice Boltzman, pseudo-
particle method) to continuum (direct numerical simulation, large-eddy simulation
and other conventional computational fluid dynamics (CFD) techniques). Different
combinations of models for the particle phase and fluid phase can be made, and their
relative merits in describing particle–fluid flow have been discussed (see e.g. Yu 2005;
Zhu et al. 2007).
Two popular combinations are widely used to describe particle–fluid flow: the
two-fluid model (TFM) and CFD–DEM. In TFM, both fluid and solid phases are
treated as interpenetrating continuum media in a computational cell which is much
larger than individual particles but still small compared with the size of process
equipment (Anderson & Jackson 1967). However, its effective use heavily depends
on the constitutive or closure relations for the solid phase and the momentum
exchange between phases, which are often difficult to obtain within its framework;
this is particularly true when dealing with different types of particles that should
be treated as different phases. In CFD–DEM, the motion of discrete particles is
obtained by solving Newton’s second law of motion as used in DEM, and the flow of
continuum fluid by solving the Navier–Stokes equations based on the concept of local
average as used in CFD, with the coupling of CFD and DEM through particle–fluid
interaction forces (Xu & Yu 1997). The main advantage of CFD–DEM is that it
can generate detailed particle-scale information, such as the trajectories of and forces
acting on individual particles, which is key to elucidating the mechanisms governing
the complicated flow behaviour. With the rapid development of computer technology,
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

the CFD–DEM approach has been increasingly used by various investigators to study
various particle–fluid flow systems as, for example, reviewed by Zhu et al. (2007, 2008).
The implementation of a CFD–DEM model, as pointed out by Feng & Yu (2004a),
lies in three aspects: the formulation of governing equations, the coupling scheme for
numerical computation and the calculation of particle–fluid interaction forces. The
latter two have been well discussed, particularly for monosized particles (Feng & Yu
2004a; Zhu et al. 2007). However, the first aspect is not well established. In fact,
two models, called model A and model B, have been used by different investigators,
and there are conflicting views regarding their applications (Hoomans et al. 1998;
Xu & Yu 1998; Kafui, Thornton & Adams 2002, 2004; Feng & Yu 2004a,b; Di
Renzo & Di Maio 2007). Xu & Yu (1998) pointed out that the interpretation of the
fluid–particle interaction force is different for both models, but both the methods can
meet the requirement that the bed pressure drop balances the bed weight at minimum
fluidization and hence are valid for fluidization. Kafui et al. (2002) discussed model
A and model B, and summarized the models used by different research groups.
Nevertheless, their claim that model A best captures the essential features of a
fixed/fluidized bed was questionable, as noted by Feng & Yu (2004b) and Kafui
et al. (2004). It is now clear that the two models have little significant difference
when applied to gas fluidization of monosized particles. It is not clear which is
 484 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu
better when applied to the modelling of fluidization of particle mixtures, but this
problem is related to the calculation of particle–fluid interaction forces. In fact, how
to calculate the fluid drag force acting on a particle in a mixture is still an open
problem (Feng & Yu 2004a; Beetstra, van der Hoef & Kuipers 2007). More recently,
the two models were further discussed by Di Renzo & Di Maio (2007). These authors
claimed that model B is only valid at the minimum fluidization condition, contrary
to those by other investigators (Kafui et al. 2004; Feng & Yu 2004b). Therefore,
although the CFD–DEM approach is now widely used, its theoretical background
is not fully established. In fact, to date, there are still some basic questions which
need to be answered. For example, what are the origins of different models such as
model A and model B? What are the exact differences between those models? What
is the applicability or limitation of a particular model, and which model is the most
appropriate for modelling different particle–fluid flow systems?
This paper provides our answers to those questions. Firstly, the origins of different
formulations in the continuum approach are explored. It is argued that three models,
rather than two, exist in the continuum approach. Then, it is demonstrated that
corresponding to the continuum approach, there are three models in the CFD–
DEM approach. The relationships between the three models are discussed, and
their applicability is analysed theoretically and verified in a comparative study of
three representative particle–fluid flow systems: fluidization, pneumatic conveying
and hydrocyclones.

2. Theoretical treatments
2.1. Origin of different model formulations in the continuum approach
The model formulation in the continuum approach to describing particle–fluid flow,
focused on gas–solid flow in fluidization, has been proposed since the 1960s, including
those, for example, by Anderson & Jackson (1967), Ishii (1975), Gidaspow (1994),
Enwald et al. (1996) and Jackson (1997). In practice, different sets of governing
equations from different resources have been used (e.g. Arastoopour & Gidaspow
1979; Lee & Lyczkowski 2000; van Wachem et al. 2001). However, the area has
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

been well established, as recently summarized by Prosperetti & Tryggvason (2007).

Unfortunately, this is not the case in the CFD–DEM approach, as described in
§ 1. On the other hand, the derivation of CFD–DEM models is closely related to
the continuum approach due to the fact that fluid flow is still modelled at the
macroscopic local average level. Thus, it is helpful to start our discussion with the
TFM approach.
In the continuum approach, both fluid and solid phases are treated as continuous
media. Anderson & Jackson (1967) used the local average method to directly derive
the fluid governing equation on the basis of the point equation of motion of the fluid,
and the solid phase governing equation on the basis of the equation of motion for the
centre of mass of a single particle. They obtained the first set of governing equations
(set I):
ρf εf [∂(u)/∂t + ∇ · (uu)] = ∇ · ξ − n f i + ρf εf g (fluid phase), (2.1)
ρs εs [∂(v)/∂t + ∇ · (vv)] = nΦ − ∇ · S + n f i + ρs εs g (solid phase), (2.2)
where εf and εs (= 1 − εf ) are, respectively, volume fractions of fluid and particles.
ξ is fluid stress tensor, Φ is the local mean value of particle–particle interaction
force, S is the tensor representing ‘Reynolds stresses’ for the particle phase, f i is
 Discrete particle simulation of particle–fluid flow 485
the local mean value of the force on particle i by its surrounding fluid and n is the
number of particles per unit volume. This cannot be used unless the undetermined
terms or dependency of ξ , Φ, S and f i on the voidage, the local mean velocities and
the pressure are known. In order to solve the problem, Anderson & Jackson (1967)
derived some constitutive equations, including: (i) combination of nΦ and −∇ · S into
−∇ · ξ s which represents the solid stress tensor; (ii) ξ and ξ s are analogous to that
for the stress tensor in a Newtonian fluid, and written into ξ = −pδk + f (λ, µ, u),
where p is the local mean fluid pressure and λ, µ are, respectively, the effective bulk
and shear viscosities; and (iii) decomposition of n f i into two components, namely, a
component due to ‘macroscopic’ variations in the fluid stress tensor on a large scale
compared with the particle spacing, together with the other component representing
the effect of detailed variations of the point stress tensor as the fluid flows around a
particle. That is,

n f i = n(Vp ∇ · ξ )/V + n f i = εs ∇ · ξ + n f i , (2.3)

where Vp is the volume of the particle and n f i represents the part of the total fluid–
particle interaction force per unit bed volume arising from the detailed variations
in the stress tensor induced by fluctuations in velocity as the fluid passes around
individual particles and through the interstices between particles. It mainly includes
the drag force in the direction of the relative velocity (ui − v i ), and virtual mass force
proportional to the mass of fluid displaced by a particle. Other forces such as the
lift force can also be included. Thus, the interaction force n f i in (2.1) and (2.2) is
replaced by (2.3) together with the consideration of nΦ−∇ · S = −∇ · ξ s , giving the
second set of equations (set II):

ρf εf [∂(u)/∂t + ∇ · (uu)] = εf ∇ · ξ − n f i + ρf εf g (fluid phase), (2.4)

ρs εs [∂(v)/∂t + ∇ · (vv)] = εs ∇ · ξ + n f i + ρs εs g + ∇ · ξ s (solid phase). (2.5)

On comparing the governing equations in sets I and II, it can be seen that the
difference is caused by the introduction of those constitutive equations. Anderson &
Jackson (1967) commented that set I is derived directly from the basic equations of
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

fluid mechanics for the system, the subsequent set II reflects their own opinion of the
most appropriate form for the undermined terms in set I.
Based on set II, particle–fluid interaction force can be further written in another
format (Jackson 1963; Anderson & Jackson 1967). In particular, to eliminate the fluid
stress tensor term, an equation is obtained by multiplying (2.5) by (1 − εf )/εf and
subtracting from (2.4), giving

ρs εs [∂(v)/∂t + ∇ · (vv)] = n f i /εf − ρf εs g + ρf εs [∂(u)/∂t + ∇ · (uu)] + ρs εs g + ∇ · ξ s .

(2.6)

This is an equation of motion for particles which does not contain the fluid stress
tensor ξ . When compared with (2.5), it can be seen that the elimination of fluid stress
tensor ξ has introduced a buoyancy term (−ρf εs g) and a term ρf εs [∂(u)/∂t + ∇ · (uu)]
which represents the fluid acceleration into the particle equation of motion. The
magnitude of the fluid acceleration term depends on flow conditions. If this term
approaches zero or much smaller than (n f i /εf − ρf εs g), according to (2.6), the total
particle–fluid interaction force acting on particles can be written as

n f i = n f i /εf − ρf εs g. (2.7)
 486 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu
Incorporating (2.7) into (2.1) and (2.2) and considering nΦ − ∇ · S = −∇ · ξ s give
the third set of governing equations (set III):
ρf εf [∂(u)/∂t + ∇ · (uu)] = ∇ · ξ − [n f i /εf − ρf εs g] + ρf εf g (fluid phase), (2.8)
ρs εs [∂(v)/∂t + ∇ · (vv)] = ∇ · ξ s + [n f i /εf − ρf εs g] + ρs εs g (solid phase). (2.9)
However, it should be pointed out that the derivation of this set of equations is
conditional. Strictly speaking, the following conditions for the fluid phase should be
satisfied:
ρf εs [∂(u)/∂t + ∇ · (uu)] = 0. (2.10)
This indicates that the fluid flow through the particle phase should be steady and
uniform (Anderson & Jackson 1967; Gidaspow 1994).
On the other hand, hydrodynamics models with concepts of the so-called models
A and B have been widely used for the particle–fluid flow, as discussed by Bouillard,
Lyczkowski & Gidaspow (1989), and later by Gidaspow (1994) and Enwald et al.
(1996). The difference between models A and B depends on the treatment of the
pressure source term in the governing equations. Generally speaking, if the pressure
is attributed to the fluid phase alone, it is referred to as model B. If the pressure
is shared by both the fluid and solid phases, it is referred to as model A. Bouillard
et al. (1989) attributed the origin of model A to Nakamura & Capes (1973) and
Lee & Lyczkowski (1981), and that of model B to Rudinger & Chang (1964) and
Lyczkowski (1978). The drag coefficients βA and βB are, respectively, defined in models
A and B to calculate the particle–fluid interaction force. The applications of those two
sets of governing equations and their comparison have been assessed for pneumatic
conveying (Arastoopour & Gidaspow 1979) and fluidization process (Bouillard et al.
1989), showing insignificant difference between the two models. Bouillard et al. (1989)
commented that the main problem with model A is its stability, and being conditional
on the absence of all viscous stresses and without the solid elastic modulus g(εf ),
while model B, which possesses all real characteristics, makes the set of equations
well-posed. But Enwald et al. (1996) later clarified that nobody has yet proved
well-posedness for a multi-dimensional initial-boundary value problem. To date, most
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

researchers prefer model A, as reflected by the fact that commercial software packages
FLUENT and CFX both use model A.
On comparing the three formulations (sets I, II and III) with those hydrodynamics
models A and B, it can be seen that in principle, model A is consistent with set II,
and model B with set III. However, sets II and III are more general than models A
and B but less detailed, which is another reason why the concepts of models A and
B are more popular in TFM modelling. Mathematically, sets I and II (model A) are
identical, as seen from the derivation of set II. Set III (or model B) is a simplified
form of set I with the assumption of (2.10). It should be noted that, according to the
definition of model B, set I is also in a form of model B. Thus, model B has two types:
an original model B (set I) and a simplified model B. The deficiency of simplified
model B has been realized by some investigators (e.g. Anderson & Jackson 1967;
Gidaspow 1994). However, the difference between original model B and simplified
model B has not been fully recognized, and the original model B (set I) is somehow
forgotten. The two models are mixed up, and only simplified model B is commonly
used. This has created some conceptual problems. For example, although model B
or its treatment is argued to be well-posed, model A, which may be ill-posed, is
more widely used due to its convenience in numerical implementation. When applied
to CFD–DEM modelling, some investigators feel that the model B treatment is
 Discrete particle simulation of particle–fluid flow 487
better than model A. However, because a simplified model B was used, the treatment
experiences problems, as discussed in § 2.3.
Nevertheless, the governing equations in the continuum approach have been well
established, particularly if the expressions for different source terms are ignored. The
challenge remaining is to develop closure laws to determine solid flow parameters
including dynamic/bulk viscosities and particle pressure, and interfacial momentum
transfer in multi-sized system (Bouillard et al. 1989; Enwald et al. 1996; Arastoopour
2001; van Wachem et al. 2001). Two approaches are commonly used to achieve
this goal. One is to formulate empirical models mainly based on particle properties
and (local) voidage. However, those models vary significantly, depending on the
conditions, as reviewed by Enwald et al. (1996). The other way is to use the so-called
kinetic theory (e.g. Gidaspow 1994; Iddir, Arastoopour & Hrenya 2005). However,
its general application is still questioned (see e.g. Campbell 2006; Goldhirsch 2008).
Particles exhibit three flow regimes: quasi-static, fast and in-between; to date, the
success of TFM is largely limited to the fast flow regime. This difficulty does not exist
in the CFD–DEM approach, as discussed below.

2.2. Model formulations in the computational fluid dynamics–discrete element

method approach
Corresponding to the three set models in the continuum approach, CFD–DEM also
has three models. However, the CFD–DEM approach is quite different from the
traditional TFM. In CFD–DEM, one has to consider the coupling between DEM
at the particle scale and CFD at the computational cell scale. The main difference
between the CFD–DEM and TFM approaches lies in the treatment of the particle
phase. In CFD–DEM, for the particle phase, based on the soft sphere model originally
proposed by Cundall & Strack (1979), a particle in a particle–fluid flow system can
have two types of motion: translational and rotational. The governing equations for
the translational and rotational motion of particle i with radius Ri , mass mi and
moment of inertia Ii can be written as

dv i
c k
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

mi = f pf,i + ( f c,ij + f d,ij ) + mi g, (2.11)

dt j =1

dωi
k
c

Ii = (M t,ij + M r,ij ), (2.12)

dt j =1

where v i and ωi are, respectively, the translational and angular velocities of the
particle, and kc is the number of particles in interaction with the particle. The forces
involved are: the particle–fluid interaction force f pf,i , the gravitational force mi g,
and inter-particle forces between particles which include the elastic force f c,ij and
viscous damping force f d,ij . The torque acting on particle i by particle j includes two
components: M t,ij , generated by the tangential force, and M r,ij , commonly known
as the rolling friction torque. The equations used to calculate the particle–particle
interaction forces and torques have been well established in the literature (Zhu et al.
2007). Many of these have been used in our previous studies of particle–fluid flow
(Xu & Yu 1997; Zhou et al. 1999; Xu et al. 2000; Feng & Yu 2004a; Feng et al.
2004; Feng & Yu 2007; Chu & Yu 2008; Kuang et al. 2008; Zhou, Yu & Zulli 2009).
The particle–fluid interaction force f pf , similar to f i in the continuum approach, is
the sum of all types of particle–fluid interaction forces acting on individual particles
 488 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu
by fluid, including the so-called drag force f d , pressure gradient force f ∇p , viscous
force f ∇·τ due to the fluid shear stress or deviatoric stress tensor, virtual mass force
f vm , Basset force f B and lift forces such as the Saffman force f Saff and Magnus force
f Mag (Crowe, Sommerfeld & Tsuji 1998). Unless otherwise specified in later discussion,
the buoyancy force is included in the pressure gradient force f ∇p . Therefore, the
total particle–fluid interaction force on an individual particle i can be written
as

f pf,i = f d,i + f ∇p,i + f ∇·τ ,i + f vm,i + f B,i + f Saff ,i + f Mag,i . (2.13)

Many correlations have been proposed to calculate the particle–fluid interaction

forces, particularly the drag force which can be based on the equation of Ergun
(1952) and Wen & Yu (1966) equations, and correlation of Di Felice (1994) or others.
Details of those correlations can be found elsewhere (e.g. Crowe et al. 1998; Zhu
et al. 2007).
For the fluid phase, its flow is essentially governed by the Navier–Stokes equation
to be satisfied at every point of the fluid. As discussed earlier, the present interest is
more focused on the particle behaviour, not fluid phase. The flow of fluid can thus be
determined at a large scale, such as a CFD cell, which may contain many particles.
Consequently, the governing equations for fluid phase are obtained based on the local
averaged method as used in TFM. The fluid governing equations corresponding to
sets I, II and III are summarized in table 1. Note that the equations to calculate the
volumetric particle–fluid interaction force F pf differ for different sets, although they
are all related to the particle–fluid interaction force f pf .
It should be noted that τ = µ[∇u + (∇u)−1 ] − (2/3)µ(∇ · u)δ k for Newtonian fluids.
Corresponding to the volumetric particle–fluid interaction force terms in sets I, II
and III, those force terms in (2.15)–(2.17) are respectively written by F set I
pf (= n f i ),
F pf (= n f i ) and F pf (= n f i /εf − ρf εs g). The definitions of n f i and n f i can be
set II  set III 

found in § 2.1, and their determination in CFD–DEM is described below.

The coupling of CFD and DEM is achieved mainly through the particle–fluid
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

interaction force, which is at the computational cell level for the fluid phase (F pf in
(2.15)–(2.17)) and at the individual particle level for the solid phase ( f pf in (2.13)).
Three coupling schemes have been identified (Feng & Yu 2004a). In scheme 1, the
force on the fluid phase from particles is calculated by a local-average method as
used in the TFM, whereas the force on a particle from the fluid phase is calculated
separately according to individual particle velocity. In scheme 2, the force on the fluid
phase from particles is first calculated at a local-average scale as used in scheme 1,
then this force is distributed among individual particles according to a certain average
rule. In scheme 3, at each time step, the particle–fluid interaction forces on individual
particles in a computational cell are calculated first, and the values are then summed
to produce the particle–fluid interaction force at the cell scale. Theoretically, scheme
1 is problematic because Newton’s third law of motion may not hold in describing
the particle–fluid interaction. This problem is not there for schemes 2 and 3, but the
implementation of scheme 2 needs to introduce an extra assumption or numerical
treatment at a CFD cell level to distribute the particle–fluid forces among the particles
in the cell. Because scheme 3 represents the basic features of CFD–DEM modelling
from particle scale to computational cell scale, it is more reasonable and logical. In
fact, it has been widely used since its introduction by Xu & Yu (1997). Thus, the total
volumetric particle–fluid interaction force n f i in a computational cell of volume V
 Discrete particle simulation of particle–fluid flow 489

Mass conservation: ∂(εf )/∂t + ∇ · (εf u) = 0. (2.14)

Momentum conservation (and corresponding particle–fluid interaction force).
Set I:
∂(ρf εf u)/∂t + ∇ · (ρf εf uu) = −∇p − F set pf + ∇ · τ + ρf εf g,
I
(2.15a)
1
 
n
where F set I
pf = f d,i + f ∇p,i + f ∇·τ ,i + f i ,
V i=1
and f pf,i = f d,i + f ∇p,i + f ∇·τ ,i + f  . (2.15b)
Set II:
∂(ρf εf u)/∂t + ∇ · (ρf εf uu) = −εf ∇p − F set pf + εf ∇ · τ + ρf εf g,
II
(2.16a)
1
n
 
where F setpf
II
= f d,i + f i , and f pf,i = f d,i + f ∇p,i + f ∇·τ ,i + f  . (2.16b)
V i=1
Set III:
∂(ρf εf u)/∂t + ∇ · (ρf εf uu) = −∇p − F set pf
III
+ ∇ · τ + ρf εf g, (2.17a)
1
n
  1

n


where F setpf
III
= f d,i + f i − ρf Vp,i g ,
εf V i=1 V i=1
and f pf,i = ( f d,i + f i )/εf − ρf Vp,i g. (2.17b)
Notes. (1) The governing equations for particle phase are given by (2.11) and (2.12).
(2) f i = f vm,i + f B,i + f Saff ,i + f Mag,i is the sum of particle–fluid interaction forces on particle i,
other than the drag, pressure gradient and viscous forces which are often regarded as the dominant
forces in particle–fluid flow.
Table 1. Formulations of different models in the CFD–DEM approach.

can be determined by
1
1

n n
n fi = ( f pf,i ) = (f + f ∇·τ ,i + f d,i + f vm,i + f B,i + f Saff ,i + f Mag,i ).
V i=1 V i=1 ∇p,i
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

(2.18)
Equation (2.18) represents the total particle–fluid interaction force in a cell
determined at a particle scale. According to (2.3), the total force in a cell in the
continuum approach can be further rewritten as

n f i = −εs ∇p + εs ∇ · τ + n f i . (2.19)
Equations (2.18) and (2.19) should be consistent. Then

1

n
n f i = n f i − (−εs ∇p + εs ∇ · τ ) = ( f + f vm,i + f B,i + f Saff ,i + f Mag,i ).
V i=1 d,i
(2.20)
Thus, the volumetric particle–fluid interaction force in each model can be written
as
1

n
F set I
pf = n f i = (f + f ∇·τ ,i + f d,i + f vm,i + f B,i + f Saff ,i + f Mag,i ),
V i=1 ∇p,i
(2.21)
 490 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu
1

n
F set II
= n f i = ( f + f vm,i + f B,i + f Saff ,i + f Mag,i ), (2.22)
pf
V i=1 d,i
1

n
F set III
=n f i /εf − ρf εs g = ( f + f vm,i + f B,i + f Saff ,i + f Mag,i )
pf
εf V i=1 d,i
1

n
− (ρf Vp,i g). (2.23)
V i=1

When coupling CFD with DEM, or vice versa, the governing equations should be
consistent, as noted by Xu & Yu (1998). Generally speaking, for all the three models,
the governing equations for particles can be the same, shown in (2.11)–(2.13). However,
in order to satisfy the Newton’s third law of motion, for set III, the particle–fluid
interaction force acting on particles, instead of (2.13), can be obtained from (2.23),
and written as
1  
f pf,i = f d,i + f vm,i + f B,i + f Saff ,i + f Mag,i − ρf Vp,i g. (2.24)
εf

2.3. Comments on different computational fluid dynamics–discrete element method

models
Clearly, from the above discussion, there are three sets of governing equations in
the CFD–DEM modelling of particle–fluid flow. They correspond to those in TFM.
By reference to the discussion presented in § 2.1, the relationship and applicability of
these models in the CFD–DEM approach can be obtained as discussed below.
Firstly, set II is derived from set I mainly with the decomposition of particle–fluid
interaction force n f i as shown in (2.3). Physically speaking, in set II, the pressure
gradient force and viscous force on particles are separated from the volumetric
particle–fluid interaction force F pf , while set I does not. However, such a treatment
will not cause any significant difference in the simulated results because they are
mathematically the same, which will be verified in § 3.
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

Secondly, set III is a simplified form of set I under the assumption of (2.10). In the
fluid governing equations, F set I set III
pf and F pf represent the total volumetric particle–fluid
interaction force. The difference between them is that F set I set III
pf is explicit while F pf is
implicit and lumps the drag force and pressure gradient force (excluding those caused
by buoyancy) together, as seen in table 1. Both models are identical only when (2.10)
is satisfied.
Thirdly, when comparing set III with model B in the literature, a slight difference
related to the viscous part exists (Xu et al. 2000; Kafui et al. 2002; Feng & Yu 2004a).
In the literature, the particle–fluid interaction force F pf and f pf in model B includes
a component of the viscous force εs ∇ · τ , but it has been excluded in the present set
III. From its derivation as shown in (2.6)–(2.9), it can be seen that the viscous part
together with the pressure has been hidden in the expression of (n f  /ε-ρεs g) in (2.8)
and (2.9).
Most importantly, it should be noted that set III is not a general model, and
its application is conditional. Strictly speaking, it can be used only when (2.10) is
satisfied. Alternatively, from the viewpoint of forces, the following condition obtained
from (2.4) and (2.10) should be satisfied in a CFD cell:
F resid = εs ∇ · ξ − n f  εs /εf + ρf εs g = 0. (2.25a)
 Discrete particle simulation of particle–fluid flow 491
Note that
⎫
1
1

n n
⎪
εs = (Vp,i ), ∇ · ξ = (−Vp,i ∇p + Vp,i ∇ · τ ), ⎪
⎪
⎬
V i=1 V i=1
(2.25b)
1

n
⎪
⎪
n f i = ( f d,i + f vm,i + f B,i + f Saff ,i + f Mag,i ), ⎪
⎭
V i=1
then (2.25a) can be further written as
1
1

n n
F resid = (f )= [−Vp,i ∇p − ( f d,i + f vm,i + f B,i + f Saff ,i
V i=1 resid V i=1
+ f Mag,i )εs /εf + Vp,i ∇ · τ + ρf Vp,i g] = 0. (2.25c)
According to (2.7), the total particle–fluid interaction force in a cell should be
1
  1
 
n n
F total = f total ,i = f effective,i + f resid ,i , (2.26)
V i=1 V i=1
where f effective,i = ( f d,i + f vm,i + f B,i + f Saff ,i + f Mag,i )/εf − ρf Vp,i g according to
(2.24).
Strictly speaking, the fluid flow in a real particle–fluid system is non-uniform and
unsteady. That is, F resid calculated according to (2.25c) is not equal to zero. Thus, the
applicability of set III can be assessed by the value of F resid relative to F total in (2.26).
That is, a parameter η to assess the applicability of set III can be defined as
η = | f resid |/| f total | × 100 %, (2.27)
where f resid and f total are, respectively, the forces acting on individual particles
determined by (2.25c) and (2.26). η for each particle in a considered system can be
traced in a CFD–DEM simulation.

3. Applicability of different CFD–DEM model formulations

https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

Three typical particle–fluid flow systems are used to test the applicability of the three
models in the present simulation: fluidization, pneumatic conveying and hydrocyc-
lones. Fluidization and pneumatic conveying have been extensively studied, therefore
our focus here is to examine the reliability of the published results based on the differ-
ent CFD–DEM models. A hydrocyclone gives a more complicated flow, offering a very
representative example to examine the applicability of different model formulations.
To eliminate the effects of CFD–DEM algorithms and other unexpected factors on
numerical results, the simulations for each particle–fluid system are carried out based
on the following conditions: (i) the same CFD–DEM code except for the necessary
minor changes for the implementation of different models; (ii) the same equations to
calculate the particle–particle and particle–fluid interaction forces as listed in table 2,
where the drag force is based on the correlation formulated by Di Felice (1994); (iii) the
same initial and boundary conditions. Ideally, all simulations should be performed
with one computational code. However, this is difficult to achieve at the moment.
On the other hand, our aim is to examine the difference of the three models for a
given system. This aim can be achieved by ensuring that the same code is used for
all the models for each of the flow systems considered. This treatment can better
correspond to the real application because different investigators often use different
codes, although they are developed based on the same principle. The results analysis is
 492 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu

Forces and torques Equations∗

√
Normal elastic force ( f cn,ij ) − 43 E ∗ R ∗ δn3/2 n
 √ 1/2
Normal damping force ( f dn,ij ) −cn 8mij E∗ R ∗ δn v n,ij
Tangential elastic force ( f ct,ij ) −µs  f cn,ij  (1 − (1 − δt /δt,max )3/2 )δ̂ t (δt < δt,max )
1/2
Tangential damping force ( f dt,ij ) −ct 6µs mij | f cn,ij | 1 − |δt |/δt,max /δt,max
×v t,ij (δt < δt,max )
 
Coulumb friction force ( f t,ij ) −µs  f cn,ij  δ̂ t (δt > δt,max )
 
Torque by tangential forces (M t,ij ) Rij × f ct,ij + f dt,ij
  n
Rolling friction torque (M r,ij ) µr,ij  f n,ij  ωij
Drag force ( f d,i ) 0.125Cd 0,i ρf πdi2 εi2 |ui − v i | (ui − v i )εi−χ
Pressure gradient force ( f ∇p,i ) −∇p · Vp,i
Viscous force ( f ∇·τ ,i ) −(∇ · τ )Vp,i
ωnij
where 1/mij = 1/mi + 1/mj , 1/R ∗ = 1/|Ri | + 1/|Rj |, E ∗ = E/2(1 − ν 2 ), ωnij =

,
|ωnij |
δ̂ t = δ t /|δ t |, δt,max = µs ((2 − ν)/2(1 − ν))δn , v ij = v j − v i + ωj × Rj − ωi × Ri ,
v n,ij = (v ij · n) · n, v t,ij = (v ij × n) × n, χ = 3.7 − 0.65 exp[−(1.5 − log10 Rei )2 /2],
−1
Cd0,i = (0.63 + 4.8/Re0.5 i ) , Rei = ρf dpi εi |ui − v i | /µf , τ = µf [(∇u) + (∇u) ],
2

εi = 1 − ni=1 Vp,i /V . Note that tangential forces ( f ct,ij + f dt,ij ) should be replaced by
f t,ij when δt > δt,max .
∗
Parameters in these equations are explained in tables 3–5.
Table 2. Components of forces and torques acting on particle i.

mainly carried out in terms of flow patterns, forces, and parameter η defined by (2.27).
For convenience, the simulation conditions for each case are described, respectively.
The methods for the numerical solution of CFD and DEM have been well
established in the literature (e.g. Xu & Yu 1997). For convenience, it is briefly
described here. An explicit time integration method is used for DEM to solve the
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

translational and rotational motions of discrete particles (Cundall & Strack 1979). The
conventional semi-implicit method for pressure-linked equations (SIMPLE) method is
used for CFD to solve the fluid governing equations (Patankar 1980). The governing
equations are discretized in finite volume form on a uniform staggered grid. The
second-order central difference scheme is used for the pressure gradient and divergence
terms, the first-order up-wind scheme for the convection term, and a second-order
Crank–Nicolson scheme for the time derivative. The CFD–DEM coupling scheme 3
as discussed earlier, is used in all the simulation cases. At each time step, DEM will
give information, such as the positions and velocities of individual particles, for the
evaluation of porosity and volumetric fluid drag force in a computational cell. CFD
will then use these data to determine the gas flow field which then yields the fluid
drag forces acting on individual particles. Incorporation of the resulting forces into
DEM will produce information about the motion of individual particles for the next
time step. All the simulations are carried out on the Intel䊊R Xeon䊊R CPU5130 2.0 GHz,
and the simulation time varies from hours for a case of gas fluidization or pneumatic
conveying to days for a case of hydrocyclones.
3.1. Fluidization
Fluidization is one of the most popular particle–fluid flow systems in the CFD–DEM
modelling. The conflicting views, if any, about the applicability of difference model
 Discrete particle simulation of particle–fluid flow 493

Variables Values
Bed geometry:
Bed width 150 mm
Bed thickness 24 mm
Initial bed height 240 mm
Total CFD cells 15 × 120
Cell size (x × z) 10 mm × 10 mm
Particle properties:
Number of particles (N) 15 000
Particle diameter (dp ) 4 mm
Particle density (ρp ) 2500 kg m−3
Particle–particle/wall sliding friction (µs ) 0.4
Particle–particle/wall rolling friction (µr ) 1 %dp mm
Particle–particle/wall damping (cn = ct ) 0.3
Particle Young’s modulus (E) 1.0 × 107 Pa
Particle Poisson ratio (ν) 0.3
Time step (t) 1.75 × 10−7 s
Fluid properties:
Gas density 1.2 kg m−3
Gas viscosity 1.8 × 10−5 Pa s
Table 3. Parameters used in the present gas fluidization.

formulations mainly stem from the study of this flow system (Hoomans et al. 1998;
Xu & Yu 1998; Kafui et al. 2002, 2004; Feng & Yu 2004a,b; Di Renzo & Di Maio
2007). However, previous studies did not examine this issue theoretically but focused
on the result comparison. The outcomes are not as general and convincing. This
deficiency can be overcome with the support of the theoretical arguments presented
in § 2. Therefore, as the first step to examine the applicability of different models
listed in table 1, fluidization is taken as our first case study. The simulation condition
is similar to that of Feng & Yu (2004b), but the structure used in this work has a
thickness of six-particle diameter with front and rear wall boundary conditions. It
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

should be noted that CFD for the fluid flow is assumed to be two-dimensional due
to the thin bed thickness. However, DEM for the particle flow is three-dimensional.
Such a technique of coupling two-dimensional CFD with three-dimensional DEM
has been successfully used in the literature (e.g. Xu & Yu 1998; Kafui et al. 2002,
2004; Feng & Yu 2004a,b). It is adopted in the present study of fluidization. Table
3 lists the parameters used. It should be noted that the initial bed conditions for all
the simulation cases of fluidization are identical. The bed is produced by randomly
dropping particles into the rectangular box as done elsewhere (Zhou et al. 2009), and
its porosity is 0.418.
Figure 1 shows the initial response of particles to the introduction of gas using the
three models when the gas superficial velocity is 3.0 m s−1 . It can be observed that the
flow patterns obtained are quite comparable, although the bed height generated by set
III is slightly higher than those generated by sets I and II. This is consistent with the
previous findings (Kafui et al. 2004; Feng & Yu 2004b). The relationship between the
pressure drop and gas superficial velocity is an important feature in gas fluidization.
Figure 2 shows such a relationship generated by the three models, indicating again
that different model formulations do not yield any significant differences.
The examination of different particle–fluid interaction forces is useful in identifying
any difference between the three models at a more fundamental level. It should be
noted that the virtual mass, Basset and lift forces are not considered in all fluidization
 494 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu
(a)
0.252 s 0.707 s 0.959 s 1.111 s 1.464 s 1.767 s 2.374 s

(b)
0.252 s 0.707 s 0.959 s 1.111 s 1.464 s 1.767 s 2.374 s
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

(c)
0.252 s 0.707 s 0.959 s 1.111 s 1.464 s 1.767 s 2.374 s

Figure 1. Snapshots showing the flow patterns of particles at the early stage for different
CFD–DEM models when gas superficial velocity is 3.0 m s−1 : (a) set I, (b) set II and (c) set III.
 Discrete particle simulation of particle–fluid flow 495
4000

3500 Bed weight 3424 Pa

3000
Pressure drop (Pa)

2500

2000

1500

1000 Set I
Set II
Set III
500

0 1 2 3 4
Gas superficial velocity (m s–1)
Figure 2. The relationship between bed pressure drop and gas superficial velocity for the
three models in gas fluidization.

(a) (b)
Total particle–fluid interaction force (N)

10 20
U = 2.8 m s–1
9 18 U = 2.8 m s–1
Pressure gradient force (N)

8 16
7 14
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

6 12
5 U = 1.6 m s–1 10
4 U = 1.6 m s–1
8
3 6
Set I Set I
2 Set II Set II
U = 0.8 m s–1 4 Set III
1 U = 0.8 m s–1
2

0 2 4 6 8 10
0 2 4 6 8 10
Time (s)
Time (s)
Figure 3. Variation of (a) the pressure gradient force with time for sets I and II; and (b)
the total particle–fluid interaction force (= N i=1 ( f d,i + f ∇p,i + f ∇·τ ,i ) for sets I & II; and
N
i=1 ( f d,i /εf − ρf Vp,i g) for set III) for the three models.

cases in this work, because they are insignificant compared with the pressure gradient
force and drag force, particularly in gas fluidization (Crowe et al. 1998; Zhu et al.
2007). The pressure gradient force and drag force are two dominant particle–fluid
interaction forces, and their variations with time are given in figure 3. As shown in
 496 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu
figure 3(a), the difference in the pressure gradient force between sets I and II exists,
even in the fixed bed. However, such a difference is still small, which cannot be reflected
by flow patterns (figures 1a and 1b). The difference is mainly caused by the different
treatments of the pressure gradient force in the model formulation and hence the
numerical scheme. In set II, the pressure gradient force is incorporated into the fluid
governing equations, based on the cell properties such as porosity and pressure drop.
However, in set I, the pressure gradient force acting on the individual particles in a
cell are summed together with the drag force. It is treated as a property on the particle
scale. In theory, as discussed by Feng & Yu (2004a), information transfer from particle
scale to cell scale is more reasonable as it can avoid the unnecessary assumption of
distributing a cell property among particles. Therefore, set I should be more reliable.
Figure 3(b) shows the total particle–fluid interaction forces (including the drag force,
pressure gradient force and viscous force) for the three models in the vertical direction,
thus showing some slight differences. It can be seen that the particle–fluid interaction
force generated by set III is larger than the other two, explaining why the bed height
from set III is slightly higher as shown in figure 1. It indicates that, in set III, the
total particle–fluid interaction force represented by (2.23) is slightly overestimated.
This must be caused by the assumption made in (2.10) or (2.25).
The assumption as given by (2.25) can be examined by parameter η. For each
particle in the bed, the total particle–fluid interaction force f total can be determined
by (2.26) while the ignored force f resid in set III is calculated by (2.25c). Thus,
according to (2.27), each particle has a value of η. Figure 4(a) shows the variations
of bed averaged η with time when gas superficial velocity is 3.4 m s−1 . It can be seen
that η is small, fluctuating around 9 %. Figure 4(b) further shows a snapshot of solid
flow patterns with spatial distributions of η, together with the solid and gas flow fields
and porosity spatial distribution. It can be seen that large η mainly locates in areas
where voids or bubbles exist, indicating a larger variation in porosity, pressure and
velocity. Such a spatial variation of η can explain the observed differences of set III
with other two models such as higher bed height or particle–fluid interaction force.
It indicates that (2.25) is not always satisfied, particularly in those regions with gas
voids and bubbles. Nevertheless, the overall differences between the three models in
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

gas fluidization are not significant.

It should be noted that the simulation cases above are all carried out in a narrow
fluidized bed. The solid flow in such a narrow bed is more one-dimensional. However,
in the real fluidization process, it is featured with the formation of gas bubbles, particle
clusters and rigorous two-dimensional or three-dimensional motion of particles. It is
necessary to examine the effect of bed geometry on η. Thus, a case is chosen with the
bed width 3 times larger than the one in table 3, and total 60 000 particles involved.
Figure 5(a) shows the variations of bed averaged η at around 7 %, slightly smaller
than that in the narrow bed (figure 4a). A snapshot of solid flow patterns is also
shown in figure 5(b) with η spatial distribution, together with the corresponding solid
and gas flow fields and porosity distribution. It can be seen that the large η still
mainly locates in the boundary region of gas bubbles and the interface between dense
and dilute particle flow regions, which is consistent with those in figure 4(b).
In gas fluidization, the drag force is almost equally as important as the pressure
gradient force where the buoyancy force is negligible. However, in liquid fluidization,
the buoyancy force becomes more important. The flow features of liquid fluidization
are also different from those in gas fluidization. Thus, as part of the present study,
liquid fluidization is also considered. The three models show results consistent with
the gas fluidization. Thus, for brevity, they are not presented here.
 Discrete particle simulation of particle–fluid flow 497
(a) 14
13
12
11

Parameter η
10
9
8
7
6
5
4
0 2 4 6 8 10
Time (s)

(b) 20 m s–1
0 20 40 60 80 100
5m s–1 0.40 0.56 0.73 0.89
0.65 2.61 4.57 6.53
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

Figure 4. (a) Variations of bed averaged η when gas superficial velocity is 3.4 m s−1 , and
(b) a snapshot at t = 6.341 s showing the solid flow patterns with spatial distribution of η,
and the corresponding particle velocity field, gas flow field and porosity distribution (data are
generated by set I).

Therefore, it can be concluded that the differences in the simulated results based
on sets I, II and III are very small. All the three models can be applied to such a
system, although theoretically set I is most acceptable. They also confirm that the
results reported in the literature, produced by either set II or set III (i.e. models A or
B), are reasonable.
 498 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu
(a) 12

11

10

9
Parameter η

4
0 2 4 6 8
Time (s)
(b)
20 m s–1
10 m s–1
0 20 40 60 80 100 0.44 0.66 0.89
0.8 3.4 5.9 8.5 11.0
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

Figure 5. (a) Variations of bed averaged η when gas superficial velocity is 3.4 m s−1 in a wider
bed, and (b) a snapshot at t = 5.853 s showing the solid flow patterns with spatial distribution
of η, and the corresponding particle velocity field, gas flow field and porosity distribution (data
are generated by set I).

3.2. Pneumatic conveying

Pneumatic conveying is one of the most commonly used methods of transporting
granular materials from one place to another. Depending on solid loading and
conveying speeds, one can distinguish between dilute phase, dispersed and dense
 Discrete particle simulation of particle–fluid flow 499

Variables Values
Pipe geometry:
Diameter 50 mm
Length 0.8 m
Number of body-fitted cells 7 × 7 × 100
Particle properties:
Number of particles (N ) 33 000
Particle diameter (dp ) 3 mm
Particle density (ρp ) 1000 kg m−3
Particle–particle/wall sliding friction (µs ) 0.4
Particle–particle/wall rolling friction (µr ) 1 %dp mm
Particle–particle/wall damping (cn = ct ) 0.1
Particle Young’s modulus (E) 5.0 × 109 Pa
Particle poisson ratio (ν) 0.33
Time step (t) 3.8 × 10−6 s
Gas properties:
Density 1.2 kg m−3
Viscosity 1.8 × 10−5 Pa s
Table 4. Parameters used in the present simulation of pneumatic conveying.

phase slug flow regimes where the latter is most commonly used. To understand the
fundamentals of slug flow, the CFD–DEM approach in a set II or set III format
has been increasingly used in recent years (see e.g. the review of Zhu et al. 2008 and
Kuang et al. 2008). It is therefore important to know the applicability of different
model formulations to this system. Table 4 shows the parameters used in the present
simulations of the horizontal pneumatic conveying. For computational efficiency, a
0.8 m horizontal pipe with internal diameter of 0.04 m is chosen as the computation
domain, facilitated by periodic boundaries for gas and particles in the flow direction
as used by Kuang et al. (2008). In all the simulations, the same initial particle
configuration is used, which consists of a stationary slug and a stationary settled
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

layer. The flow of both gas and solid phases is three-dimensional.

Figure 6 shows the snapshots of particle flow patterns in the slug flow regime for
the three models. It can be seen that throughout the entire physical time considered,
there is no significant difference in the shape of slugs, and the height of settled layers
predicted by the three models. In each simulation, a slug locates at almost the same
axial position in the pipe with the evolution of time. This shows that the three models
predict the same slug velocity. Note that slug velocity as well as slug shape and settled
layer height are the three key process parameters usually used to characterize slug
flow. The variations of pressure drop with time for the three models are also traced,
and shown in figure 7. It can be observed that the differences in the pressure drop
generated by the three models are small. The variations of particle–fluid interaction
forces in the horizontal direction are also examined and shown in figure 8. Again,
it can be seen that the three models produce comparable results with negligible
differences.
Figure 9(a) shows the spatial distribution of η. It can be seen that η is close to
zero within a slug, but may have a relatively large magnitude within settled layers.
This result suggests that set III is not so suitable in the simulation of inhomogeneous
pneumatic conveying, such as stratified flow and dispersed flow with clusters. This
is because gas mainly flows over settled layers, and its flow is non-uniform, hence
 500 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu
(a) t = 0.450 s

t = 1.245 s

t = 2.550 s

t = 3.405 s

0 0.2 0.4 0.6 0.8

(b) t = 0.450 s

t = 1.245 s

t = 2.550 s

t = 3.405 s

0 0.2 0.4 0.6 0.8

(c) t = 0.450 s
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

t = 1.245 s

t = 2.550 s

t = 3.405 s

0 0.2 0.4 0.6 0.8

Figure 6. Side views of particle flow patterns in slug-flow pneumatic conveying when gas
superficial velocity is 2.1 m s−1 , calculated by (a) set I, (b) set II and (c) set III..
 Discrete particle simulation of particle–fluid flow 501
4000

Set I
Set II
3500 Set III

Pressure drop (Pa)

3000

2500

2000

1500

1000
0.4 1.4 2.4 3.4 4.4
Time (s)
Figure 7. Variation of the pressure drop with time for the three models under the conditions
corresponding to figure 6.
Total particle–fluid interaction force (N)

(a) 4.0 (b) 7.0

Set I Set I
Set II
Pressure gradient force (N)

Set II
3.5 Set III
6.0

3.0
5.0
2.5

4.0
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

2.0

1.5 3.0
0.4 1.4 2.4 3.4 4.4 0.4 1.4 2.4 3.4 4.4
Time (s) Time (s)
Figure 8. Variation of: (a) the pressure gradient force with time for sets I and II; and (b),
the total particle–fluid interaction force (= N i=1 ( f d,i + f ∇p,i + f ∇·τ ,i ) for sets I and II; and
N
i=1 ( f d,i /εf − ρf Vp,i g) for set III) for the three models.

the assumption induced by (2.10) or (2.25) is not satisfied. However, it is relatively

uniform within a slug, as shown in figure 9(b). Hence, set III can be applied in this
flow regime. It should be pointed out that in a slug flow, slug behaviour depends
on particle motion and gas flow within a slug, e.g. the pressure drop in the whole
conveying pipeline is the result of the pressure difference when gas flows across slugs,
as demonstrated in figure 9(c). As a result, although η deviates from zero within
settled layers, set III still produces gas and solid flow patterns similar to those given
by sets I and II.
 502 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu

(a) 10 21 33 44 56 67 79 90

(b) 5 m s–1

(c)

2834.39 2834.39 1800.00 400.00 –1.01 –1.01

0 0.2 0.4 0.6 0.8

Figure 9. (a) Snapshots produced by set I showing spatial distributions of η; (b) gas velocity
profile; (c) pressure drop at x = 0.02 m with 4 particle diameter thickness, corresponding to
figure 6(a).
Total particle–fluid interaction force (N)
(b)
(a) t = 2.715 s (Set I)

104 t = 2.715 s (Set II)

101 Set I
Pressure drop (Pa)

t = 2.715 s (Set III) Set II

Set III
103 0.2 0.3 0.4 0.5 0.6 100
Set I
Set II
Set III
102 10_1

101 10_2
0.4 1.4 2.4 3.4 0.4 1.4 2.4 3.4
Time (s) Time (s)
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

Figure 10. Variations of (a) pressure drop and (b) the total particle–fluid interaction force
(= N i=1 ( f d,i + f ∇p,i + f ∇·τ ,i ) for sets I and II; and
N
i=1 ( f d,i /εf − ρf Vp,i g) for set III) for
−1
non-slug flow when gas superficial velocity is 10 m s . The inset in (a) shows a comparison
of solid flow patterns for non-slug flow at one snapshot.

Figure 10 shows the variations of the pressure drop and particle–fluid interaction
forces for non-slug flow when gas superficial velocity is 10 m s−1 . It can be observed
that, at the initial stage of slug collapsing, the difference is minor. However, once
the flow reaches its macroscopically steady state, set III generates a relatively large
pressure drop, which corresponds to a large pressure gradient force, and the motion
of particles is slightly more vigorous. This indicates that set III should be used with
caution when applied to the non-slug flow regimes in pneumatic conveying. It is for
this reason that set I (or set II) was used in our recent studies of gas–solid flow in
pneumatic conveying (Kuang, Yu & Zou 2009).
3.3. Hydrocyclones
A hydrocyclone separates particles in a liquid suspension or slurry (normally a
mixture of water and particles) mainly according to their densities or sizes depending
 Discrete particle simulation of particle–fluid flow 503
Dc
(a) Do (b)

Lv
Di
Lc

Du

Figure 11. Geometry (a) and mesh representation (b) of the simulated hydrocyclone.

on operations. It is widely used in industry, particularly in mineral and chemical

processing because of its simplicity in design, high capacity, low maintenance and
operational cost (Svarosky 1984). The flow behaviour in a hydrocyclone is very
complicated. For example, the strong rotational flow of liquid suspension can
create a low-pressure axial core and a free liquid surface. Such a low-pressure
core may communicate directly with the atmosphere at the outlets with air filled.
The air is inhaled through the apex and forms an air core, which often results
in poorer performance. Moreover, the suspension that is fed to a hydrocyclone is
thickened partially by the effect of centrifugal forces. The suspension is withdrawn
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

at the underflow with a high solid concentration, and the clarified liquid leaves the
hydrocyclone by the overflow through the vortex finder. Such a complex flow is very
useful in examining the applicability of the three models, particularly set III.
In this work, the particle–liquid–air flow in a hydrocyclone is simulated by both sets
I and III. The diameter of the hydrocyclone considered is 75 mm and its geometry
and mesh representation are shown in figure 11 and table 5. Parameters used in
the DEM for the current case are also listed in table 5. Figure 11(b) shows the
computational domain, containing 87 500 cells. The whole computational domain is
divided by hexahedron grids. Trial numerical results demonstrate that the solution is
independent of the characteristics of the mesh size. The pressure at the two outlets
(vortex finder and apex) is set to 1 atm. The inlet water velocity and the particle
velocity are both 2.25 m s−1 . The particle diameter is 2 mm and particle density ranges
uniformly from 800 to 1200 kg m−3 . To better evaluate the proposed models, the
fluid–solid interaction forces considered are two dominant forces, i.e. the drag force
and pressure gradient force. Other fluid–solid forces such as the Basset force and lift
forces are ignored because: (i) the aim of this work is to investigate the difference
between sets I and III; and (ii) their effects are very minor from our trial test, but
their involvement may induce uncertainty in computation (NB equations to calculate
some of these forces are not yet generally established). It should also be noted that,
 504 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu

Variables Values
Hydrocyclone geometry:
Diameter of the body (Dc) 75 mm
Diameter of inlet (Di ) 25 mm (same area quadrate is used)
Diameter of vortex finder (Do) 25 mm
Diameter of apex (Du) 12.5 mm
Length of cylindrical part (Lc) 75 mm
Length of vortex finder (Lv ) 50 mm
Included angle (a) 20◦
Particle properties:
Solid flow rate 20 kg s−1
Particle density distribution (ρ) Uniform in the range of 800–1200 kg m−3
Particle diameter (dp ) 2 mm
Particle–particle/wall rolling friction (µr ) 0.005 mm
Particle–particle/wall sliding friction (µs ) 0.3
Poisson’s ratio (ν) 0.3
Young’s modulus (E ) 1 × 107 Pa
Damping coefficient (ct = cn ) 0.3
Particle velocity at inlet 2.25 m s−1
Fluid properties:
Gas density 1.225 kg m−3
Gas viscosity 1.8 × 10−5 Pa s
Water density 998.2 kg m−3
Water viscosity 0.001 Pa s
Water velocity at inlet 2.25 m s−1
Table 5. Parameters used in the present simulation of hydrocyclone flow.

unlike the other two case studies, where the simulations are performed with in-house
codes, the CFD computation for this system is based on the commercial software
FLUENT, achieved by incorporating our DEM code into FLUENT through its User
Defined Functions. This approach has been used in the study of fluidization and
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

pneumatic conveying (Chu & Yu 2008). Currently, it is difficult to implement set II

under the FLUENT platform. Therefore, our simulations are carried out only by sets
I and III. However, as discussed in § 2, sets I and II are largely equivalent, which has
been confirmed from the results of fluidization and pneumatic conveying. The other
details of the CFD and DEM treatments can be found from our recent work on
hydrocyclones and dense medium cyclones (Wang, Chu & Yu 2007; Chu et al. 2009).
Figure 12 shows the solid flow patterns. It can be seen that the flow pattern obtained
from set III is quite different from that obtained from set I. There is separation of
particles by density according to the results produced by set I as shown in figure
12(a). Heavy particles go to the spigot of the hydrocyclone and light particles go to
the vortex finder of the hydrocyclone. However, such a phenomenon does not happen
when set III is used, as shown in figure 12(b) where all particles exit from the bottom.
Moreover, although the initial and boundary conditions are the same, the number of
particles present in the hydrocyclone differs significantly: 5194 in the simulation by
set I and 9217 by set III. These results indicate that set III cannot capture the flow
features in this complicated system.
Different flow properties can also be observed, corresponding to the different solid
flow patterns in figure 12. The variation of the pressure drop with time for sets I and
III is first examined and shown in figure 13(a). The pressure drop generated by set
 Discrete particle simulation of particle–fluid flow 505
(a) (b)

Particle Particle
density density
(kg m–3) (kg m–3)
1150 1150
1100 1100
1050 1050
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

1000 1000
950 950
z z
900 900
850 850
x y x y

Figure 12. Comparison of simulated particle flow patterns in a hydrocyclone by use of

different models: (a), set I; and (b), set III. Particles are coloured by particle density.

III is much lower than that generated by set I after 0.5 s. In addition, it has a large
fluctuation. The total particle–fluid interaction forces, given as the sum of the pressure
gradient force and the drag force on particles, are further examined and shown in
figure 13(b). This illustrates that the total particle–fluid interaction force generated
by set III is quite different from that generated by set I, and further confirms that set
III is not valid in the simulation of hydrocyclone flow. The reason for this is that the
assumption induced by (2.25) cannot be met for the flow in a hydrocyclone. This can
be illustrated from the spatial distribution of η as shown in figure 14. The values of
parameter η for almost all particles are larger than 90 %.
 506 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu
(a) 60 000 (b) 200

Normalized–fluid force (mg)

50 000 Set I
Set I
150
40 000
Pressure (Pa)

30 000 100

20 000 Set III Set III

50
10 000

0 0.5 1.0 1.5 2.0 0 0.5 1.0 1.5 2.0

Time (s) Time (s)
Figure 13. Variation of (a) the pressure drop with time for different model formulations,
and (b) averaged total particle–fluid interaction force on individual particles (= N i=1 ( f d,i
N
+ f ∇p,i + f ∇·τ ,i ) for set I; and i=1 ( f d,i /εf − ρf V p,i g) for set III) with time for different
model formulations.

A A

B B
A–A
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

Percentage
(%) C C

100
90 B–B
80
70
60
50
40
30
20
C–C
10

Figure 14. Snapshots showing the spatial distribution of η in a hydrocyclone,

generated by set I.
 Discrete particle simulation of particle–fluid flow 507
(a) (b)

Pressure (Pa)
50 000
43 750
37 500
31 250
25 000
18 750
12 500
6250
0
10 m s–1

(c) (d )

100 m s–1 20 m s–1

Figure 15. (a) Static pressure distribution; (b) velocity distribution; (c) vectors of total
particle–fluid interaction force (= N i=1 ( f d,i + f ∇p,i )) acting on individual particles at section
B–B (figure 14) generated by set I; and (d) vectors of total particle–fluid interaction forces =
N
i=1 ( f d,i /εf − ρf Vp,i g)) at the same section generated by set III. The force on a particle is
normalized by dividing its weight.
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

It is worthwhile to examine why set III fails to predict the separation of particles
by density. For this purpose, figure 15 shows the internal flow information. It can
be seen from figure 15(a) that the pressure decreases gradually from the wall to the
centre of the cyclone, indicating that the pressure gradient force will point towards
the centre. On the other hand, figure 15(b) shows that the fluid velocity vector directs
mainly tangentially. Therefore, the direction of the pressure gradient force is quite
different from that of the drag force. The pressure gradient force is a radial force,
and its magnitude is more than 30 times the gravity force in most of the region. This
large radial force can overcome the centrifugal force and move light particles towards
the centre of the cyclone where the fast upward flow can lift particles up towards the
vortex finder. Figures 15(c) and 15(d) show the total particle–fluid interaction forces
calculated by sets I and III, respectively. Clearly, the particle–fluid interaction force
towards the centre is significantly underestimated by set III.
Generally speaking, the difference between sets I and III lies in the different
treatment of the particle–fluid interaction forces. According to (2.21), each type of
particle–fluid interaction force has been explicitly considered in set I, including the
pressure gradient force and the drag force, which are dominant for many particle–fluid
flow systems. However, as implied in (2.23), the pressure gradient force, whereby the
 508 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu
buoyancy force is excluded, and the drag force are lumped together in set III. This
treatment is not so problematic when the two forces act in the same direction. In
fluidization or pneumatic conveying, the particle–fluid flow is largely one-dimensional.
Consequently, sets I and III do not show any significant differences, although the
conditions of (2.10) or (2.25) are not always met locally, e.g. in the areas with voids or
bubbles in fluidization or the non-plug flow regime in pneumatic conveying. However,
the flow in a hydrocyclone is much more complicated and truly three-dimensional
for fluid and solid phases. In particular, the pressure gradient and drag forces act in
different directions. Consequently, the conditions of (2.10) or (2.25) cannot be met
globally or locally, and set III fails to predict the particle–fluid flow and hence the
separation performance in a hydrocyclone.
Therefore, caution must be taken in applying set III in particle–fluid flow modelling.
In spite of the success in the previous studies of fluidization and pneumatic conveying
reported in the literature, set III should not be used in the future CFD–DEM
modelling of particle–fluid flow. This consideration equally applies to the so-called
model B in the TFM approach in the literature. In other words, for general application,
model B should be replaced by its origin, i.e. set I as given by (2.1) and (2.2).

4. Conclusions
There are three, rather than two, models in the continuum description of particle–
fluid flow, including set I, given by (2.1) and (2.2), set II, given by (2.4) and (2.5), and
set III given by (2.8) and (2.9). Sets I and II result from different treatments of the
fluid stress tensor term in the original equation, but they are essentially the same. Set
III is however a simplified version of set I, obtained when the condition of (2.10) is
met. Consequently, the application of set III is conditional.
Corresponding to the continuum treatments, there are also three sets of formulations
in CFD–DEM modelling, as listed in table 1. However, unlike the continuum
approach, there is a difference between sets I and II because CFD and DEM are
modelled at different length scales. In theory, set I is more logical and hence more
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

acceptable. The implementation of set II needs to implement a method to distribute

(some of) the particle–fluid forces among the particles in a CFD cell, which may
induce a difference in the simulated results. However, the difference is small, and
can be treated as a numerical error. Therefore, set II and, in particular, set I, which
is somehow forgotten in the literature, are recommended for future CFD–DEM
modelling. The application of set III is, however, conditional. Strictly speaking, it
is only valid when the condition described by (2.10) or (2.25) is met, which means
that fluid flow is steady and uniform or the residual force acting on particles is
zero.
A new index resulting from the theoretical analysis, given by (2.27), is proposed to
assess the applicability of model formulation. The test cases show that the three models
are all applicable to fluidization and to a large extent, pneumatic conveying, the two
systems studied most extensively thus far in the literature. The results produced by
set III in the previous studies are therefore valid, although there are some minor
differences when comparing with those produced by set I or set II. However, caution
must be taken when applying set III to complicated flow systems like hydrocyclones,
where the condition of (2.10) or (2.25) cannot be met globally or locally. In spite of
its successful application reported in the literature, set III should not be used in the
future work.
 Discrete particle simulation of particle–fluid flow 509
The authors are grateful to the Australian Research Council for the financial
support of this work.

REFERENCES
Anderson, T. B. & Jackson, R. 1967 A fluid mechanical description of fluidized beds. Ind. Engng
Chem. Fundam. 6, 527–539.
Arastoopour, H. 2001 Numerical simulation and experimental analysis of gas/solid flow systems:
1999 Fluor–Daniel Plenary Lecture. Powder Technol. 119, 59–67.
Arastoopour, H. & Gidaspow, D. 1979 Vertical pneumatic conveying using four hydrodynamic
models. Ind. Engng Chem. Fundam. 18, 123–130.
Beetstra, R., van der Hoef, M. A. & Kuipers, J. A. M. 2007 Numerical study of segregation
using a new drag force correlation for polydisperse systems derived from lattice-Boltzmann
simulations. Chem. Engng Sci. 62, 246–255.
Bouillard, J. X., Lyczkowski, R. W. & Gidaspow, D. 1989 Porosity distributions in a fluidized-bed
with an immersed obstacle. AIChE J. 35, 908–922.
Campbell, C. S. 2006 Granular materials flows: an overview. Powder Technol. 162, 208–229.
Chu, K. W. & Yu, A. B. 2008 Numerical simulation of complex particle–fluid flows. Powder Technol.
179, 104–114.
Chu, K. W., Wang, B., Yu, A. B. & Vince, A. 2009 CFD–DEM modelling of multiphase flow in
dense medium cyclones. Powder Technol. 193, 235–247.
Crowe, C. T., Sommerfeld, M. & Tsuji, Y. 1998 Multiphase Flow with Droplets and Particles. CRC.
Cundall, P. A. & Strack, O. D. L. 1979 A discrete numerical model for granular assemblies.
Geotechnique 29, 47–65.
Di Felice, R. 1994 The voidage function for fluid–particle interaction systems. Intl J. Multiph. Flow
20, 153–159.
Di Renzo, A. & Di Maio, F. P. 2007 Homogeneous and bubbling fluidization regimes in DEM–
CFD simulations: hydrodynamic stability of gas and liquid fluidized beds. Chem. Engng Sci.
62, 116–130.
Enwald, H., Peirano, E. & Almstedt, A. E. 1996 Eulerian two-phase flow theory applied to
fluidization. Intl J. Multiph. Flow 22, 21–66.
Ergun, S. 1952 Fluid flow through packed columns. Chem. Engng Process. 48, 89–94.
Feng, Y. Q., Xu, B. H., Zhang, S. J., Yu, A. B. & Zulli, P. 2004 Discrete particle simulation of gas
fluidization of particle mixtures. AIChE J. 50, 1713–1728.
Feng, Y. Q. & Yu, A. B. 2004a Assessment of model formulations in the discrete particle simulation
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

of gas–solid flow. Ind. Engng Chem. Res. 43, 8378–8390.

Feng, Y. Q. & Yu, A. B. 2004b Comments on ‘Discrete particle-continuum fluid modelling of
gas-solid fluidised beds’ by Kafui et al. [Chem. Engng Sci. 57 (2002) 2395–2410]. Chem. Engng
Sci. 59, 719–722.
Feng, Y. Q. & Yu, A. B. 2007 Microdynamic modelling and analysis of the mixing and segregation
of binary mixtures of particles in gas fluidization. Chem. Engng Sci. 62, 256–268.
Gidaspow, D. 1994 Multiphase Flow and Fluidization. Academic.
Goldhirsch, I. 2008 Introduction to granular temperature. Powder Technol. 182, 130–136.
Hoomans, B. P. B., Kuipers, J. A. M., Briels, W. J. & van Swaaij, W. P. M. 1998 Comments
on the paper ‘Numerical simulation of the gas–solid flow in a fluidized bed by combining
discrete particle method with computational fluid dynamics’. Chem. Engng Sci. 53, 2645–
2646.
Iddir, H., Arastoopour, H. & Hrenya, C. M. 2005 Analysis of binary and ternary granular
mixtures behavior using the kinetic theory approach. Powder Technol. 151, 117–125.
Ishii, M. 1975 Thermo-Fluid Dynamics Theory of Two-Phase Flow. Eyrolles.
Jackson, R. 1963 The mechanics of fluidized beds. Part I. The stability of the state of uniform
fluidization. Trans. Inst. Chem. Engng 41, 13–21.
Jackson, R. 1997 Locally averaged equations of motion for a mixture of identical spherical particles
and Newtonian fluid. Chem. Engng Sci. 52, 2457–2469.
Kafui, K. D., Thornton, C. & Adams, M. J. 2002 Discrete particle-continuum fluid modelling of
gas–solid fluidised beds. Chem. Engng Sci. 57, 2395–2410.
 510 Z. Y. Zhou, S. B. Kuang, K. W. Chu and A. B. Yu
Kafui, K. D., Thornton, C. & Adams, M. J. 2004 Reply to comments by Feng and Yu on ‘Discrete
particle-continuum fluid modelling of gas–solid fluidised beds’ by Kafui et al. Chem. Engng
Sci. 59, 723–725.
Kuang, S. B., Chu, K. W., Yu, A. B., Zou, Z. S. & Feng, Y. Q. 2008 Computational investigation
of horizontal slug flow in pneumatic conveying. Ind. Engng Chem. Res. 47, 470–480.
Kuang, S. B., Yu, A. B. & Zou, Z. S. 2009 Computational study of flow regimes in vertical
pneumatic conveying. Ind. Engng Chem. Res. 48, 6846–6858.
Lee, W. H. & Lyczkowski, R. W. 2000 The basic character of five two-phase flow model equation
sets. Intl. J. Numer. Meth. Fluids 33, 1075–1098.
Li, J. H. 2000 Compromise and resolution: exploring the multi-scale nature of gas–solid fluidization.
Powder Technol. 111, 50–59.
Lyczkowski, R. W. 1978 Transient propagation behavior of two-Phase flow equations. In Heat
Transfer: Research and Application (ed. J. C. Chen), AIChE Symposium Series, vol. 75, Issue
174, pp. 175–174. American Institute of Chemical Engineers, New York.
Nakamura, K. & Capes, C. E. 1973 Vertical pneumatic conveying: theoretical study of uniform
and annular particle flow models. Can. J. Chem. Engng 51, 39–46.
Patankar, S. V. 1980 Numerical Heat Transfer and Fluid Flow. Hemisphere.
Prosperetti, A. & Tryggvason, G. 2007 Computational Methods for Multiphase Flow. Cambridge
University Press.
Rudinger, G. & Chang, A. 1964 Analysis of nonsteady 2-phase flow. Phys. Fluids 7, 1747–1754.
Svarovsky, L. 1984 Hydrocyclones. Technomic Publishing Inc.
Tsuji, Y. 2007 Multi-scale modeling of dense phase gas-particle flow. Chem. Engng Sci. 62, 3410–
3418.
Villermaux, J. 1996 New horizons in chemical engineering. In Proceedings of the Fifth World
Congress of Chemical Engineering, San Diego, CA, pp. 16–23.
van Wachem, B. G. M., Schouten, J. C., van den Bleek, C. M., Krishna, R. & Sinclair, J. L.
2001 Comparative analysis of CFD models of dense gas–solid systems. AIChE J. 47, 1035–51.
Wang, B., Chu, K. W. & Yu, A. B. 2007 Numerical study of particle–fluid flow in a hydrocyclone.
Ind. Engng Chem. Res. 46, 4695–4705.
Wen, C. Y. & Yu, Y. H. 1966 Mechanics of fluidization. AIChE Ser. 62, 100.
Xu, B. H. & Yu, A. B. 1997 Numerical simulation of the gas–solid flow in a fluidized bed by
combining discrete particle method with computational fluid dynamics. Chem. Engng Sci. 52,
2785–2809.
Xu, B. H. & Yu, A. B. 1998 Comments on the paper ‘Numerical simulation of the gas–solid flow
in a fluidized bed by combining discrete particle method with computational fluid dynamics’:
reply. Chem. Engng Sci. 53, 2646–2647.
https://doi.org/10.1017/S002211201000306X Published online by Cambridge University Press

Xu, B. H., Yu, A. B., Chew, S. J. & Zulli, P. 2000 Numerical simulation of the gas–solid flow in a
bed with lateral gas blasting. Powder Technol. 109, 13–26.
Yu, A. B. 2005 Powder processing: models and simulations. In Encyclopedia of Condensed Matter
Physics (ed. F. Bassani, G. L. Liedl & P. Wyder), Chapter 5 56, vol. 4, pp. 401–414. Elsevier.
Zhou, Y. C., Wright, B. D., Yang, R. Y., Xu, B. H. & Yu, A. B. 1999 Rolling friction in the
dynamic simulation of sandpile formation. Physica A 269, 536–553.
Zhou, Z. Y., Yu, A. B. & Zulli, P. 2009 Particle scale study of heat transfer in packed and bubbling
fluidized beds. AIChE J. 55, 868–884.
Zhu, H. P., Zhou, Z. Y., Yang, R. Y. & Yu, A. B. 2007 Discrete particle simulation of particulate
systems: theoretical developments. Chem. Engng Sci. 62, 3378–3396.
Zhu, H. P., Zhou, Z. Y., Yang, R. Y. & Yu, A. B. 2008 Discrete particle simulation of particulate
systems: a review of major applications and findings. Chem. Engng Sci. 63, 5728–5770.