Vous êtes sur la page 1sur 9

Powder Technology 254 (2014) 94–102

Contents lists available at ScienceDirect

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

A comparison of two-fluid model, dense discrete particle model and


CFD-DEM method for modeling impinging gas–solid flows
Xizhong Chen a,b, Junwu Wang a,⁎
a
State Key Laboratory of Multiphase Complex Systems, Institute of Process Engineering, Chinese Academy of Sciences, Beijing 100190, PR China
b
University of Chinese Academy of Sciences, Beijing, 100490, PR China

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

Article history: Gas–solid flows have been numerically investigated by various multiphase models, none of which is suitable for
Received 10 October 2013 all the problems encountered in industries. Different multiphase models have been chosen by different
Received in revised form 1 December 2013 researchers to meet their specific requirements; therefore, it is highly desirable to have a comprehensive under-
Accepted 16 December 2013
standing of the merits and drawbacks of these models. In this study, three existing multiphase models, including
Available online 9 January 2014
a two-fluid model (TFM), a dense discrete particle model (DDPM) and a combined computational fluid dynamics
Keywords:
and discrete element model (CFD-DEM) method, are compared by simulating the flow patterns of impinging par-
Multiphase flow ticle jet in a channel. Depending on the solid concentration used, the particle jets can either merge into a single jet
Gas–solid flow or cross through each other (particle trajectory crossing effect) when they are impinging. The TFM and the DDPM
Granular materials methods have the advantage of less computational demanding compared to the CFD-DEM method, with the cost
Fluidization of more uncertainties. Using the simulation results obtained from the CFD-DEM method as the benchmark data, it
Computational fluid dynamics was shown that (i) the TFM fails to predict the well-known particle trajectory crossing effect in any cases as in
Impinging flow previous studies (Desjardins et al., Journal of Computational Physics 2008, 227, 2514–2539) but can reproduce
the merging cases reasonably well; (ii) the DDPM fails to predict the cases where the two particle jets are emerg-
ing due to the over-simplified treatment of particle–particle interactions, highlighting the requirement of a prop-
er way to represent the realistic particle–particle interactions and the importance of volume exclusion effect (the
particles cannot overlap) in dense gas–solid flows; and (iii) quantitative comparisons show there are major dif-
ferences between the results predicted by the three models, highlighting the requirement of further improve-
ment of DDPM and TFM.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction methods. The Eulerian–Eulerian two-fluid model (TFM) treats both


gas and solid phase as interpenetrating continua, the conservation
The dynamical behavior of gas–solid flows continues to receive great equations of mass, momentum and energy are obtained through an ap-
attentions because of its relevance to a wide range of applications in in- propriate averaging process and the constitutive relations for solid
dustries, such as fluid catalytic cracking, airslide flows [1] and circulat- phase are usually closed using kinetic theory of granular flow (KTGF)
ing fluidized bed combustion [2]. The design and scale-up of these [4]. The TFM method normally requires much less computational re-
industrial devices motivate a better understanding of the dynamical be- sources compared to Eulerian–Lagrangian approaches. Therefore, it
havior of gas–solid flows inside the reactors. Thanks to the rapid ad- can be used to model and study pilot scale and industrial scale reactors
vancement in computer hardware, numerical algorithms and physical [5–7]. Despite the advantages, the discrete character of the solid phase is
understanding, computational fluid dynamics (CFD) has become a pow- lost in the TFM method owing to the continuum description of the dis-
erful tool to provide both qualitative and quantitative insight into the persed phase. This limitation can be overcome by discrete approaches
complex gas–solid flows [3]. Gas–solid flows have been numerically in- such as discrete element method (DEM) [8,9], in which solid particles
vestigated by various multiphase models, none of which is suitable for are tracked individually according to Newton's laws of motion with
all the problems we faced, due to the existence of multiple spatiotempo- detailed particle–particle and particle–wall collisions. One of the main
ral scales in gas–solid flow and the fact that our understanding of the drawbacks of DEM method is the high computational demands [10],
underlying physics of gas–solid flow is far from complete. which restricts its applications to small scale, fundamental investiga-
Two approaches that are frequently used to model the gas–solid tions. To avoid this restriction, the dense discrete phase model
flows problems are the Eulerian–Eulerian and Eulerian–Lagrangian (DDPM) [11] and other similar methods such as multiphase particle-
in-cell (MP-PIC) method [12–14] have been developed in which the de-
⁎ Corresponding author. Tel.: +86 10 82544842; fax: +86 10 62558065. tails of particle–particle and particle–wall collisions are not explicitly
E-mail address: jwwang@home.ipe.ac.cn (J. Wang). tracked anymore; instead, a force is used to represent the details of

0032-5910/$ – see front matter © 2014 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.powtec.2013.12.056
X. Chen, J. Wang / Powder Technology 254 (2014) 94–102 95

particle–particle and particle–wall collisions. Furthermore, the concept multiscale structures such as bubbles and clusters. In response to the ex-
of parcel is used to reduce the numbers of particles involved in compu- istence of multiple length and time scales, CFD methods suitable for
tations, resulting in a significant acceleration of the speed of simula- simulating gas–solid flows at different spatiotemporal scales have
tions. Although the DDPM and other similar methods look promising, been developed. As already mentioned in preceding section, three dif-
since it has the benefits of Lagrangian methods and is applicable to ferent multiphase models are selected to simulate impinging gas–solid
large systems, it demands further tests and validations. flow, where the CFD-DEM method is used to generate the benchmark
In addition to comparing the simulation results with experimental data because the method has been extensively proven to be effective
data for validating multiphase models, it is possible to shed light on [22,23], although its computational demands are high. TFM is selected
the merits and disadvantages of multiphase models by comparing the because it is the most-widely used method targeting on industrial appli-
results obtained using different multiphase models. For example, cations [6,24]; however, it is not easy to incorporate the effect of realistic
Goldschmidt et al. [15] compared a two-fluid model and a hard- particle size distribution in Eulerian–Eulerian method, although it plays
sphere discrete particle model with experimental data obtained from an very important role in determining the hydrodynamic characteristics
a pseudo-two-dimensional gas-fluidized bed. They showed that a sig- of gas–solid flow [25,26]. DDPM method [11] is proposed to overcome
nificant improvement can be achieved by incorporating an additional some of the main drawbacks of both Eulerian–Eulerian and CFD-DEM
source of dissipation in the KTGF considering the effects of roughness. methods, it takes the advantages of easy implementation of realistic
Tsuji et al. [16] used a direct simulation Monte Carlo method to predict particle size distribution and tracking the discrete nature of particles,
the cluster patterns in the riser of a circulating fluidized bed and com- its computational cost is also less than that of Eulerian–Eulerian method
pared to the simulation results of two-fluid model [17]. They concluded [27], due to the fact that coarse computational grid and larger time step
that the results of the two models were qualitatively similar but quanti- can be used to achieve grid-size-independent simulations and the use of
tatively different. Ibsen et al. [18]compared the multi-fluid model and the concept of parcel.
discrete particle model to the experimental findings, where it was The commercial software Ansys Fluent and the open source code
shown that the discrete particle model gave a better agreement with ex- MFIX (www.mfix.org) are used to model the problem. The TFM model
perimental observation. Benyahia and Galvin [19] compared the results and CFD-DEM model are carried out on the platform of MFIX and
obtained from the MP-PIC method and the CFD-DEM method; it was DDPM in Ansys Fluent is used. The details of theory and numerical
shown that both methods gave qualitatively results, but a fairly quanti- methods can be obtained from the MFIX documentation [28,29] and
tative difference exists. A detailed comparison of the results of the axial the documentation of Fluent. A summary of the models used is given
and radial solid concentration profiles, solids circulation patterns, pres- as follows.
sure drop and granular temperature in a dense fluidized beds was given
by Wang et al. [20] using both two-fluid and discrete particle model. 2.1. Two-fluid model (TFM)
Ryan et al. [21] compared with the TFM, DDPM and MP-PIC methods
in predicting the hydrodynamics in a solid sorbent carbon capture reac- Both gas phase and solid phase are treated as a continuous phase in
tor. They showed that the DDPM were unstable for the given reactor de- the two-fluid model and the mass conservation equations for gas and
sign, while the TFM method and the MP-PIC method provided a stable solid phases are
solution. These works give us a good understanding of the relevant mul-
∂   
tiphase models, but still much work needs to be done to fully under-
stand the merits and drawbacks of various multiphase models. ε ρ þ ∇  ε g ρg !
ug ¼ 0 ð1Þ
∂t g g
Present study attempts to extend the understanding of the advan-
tages and disadvantages of various multiphase models. There is no in-
tention to improve the models. Therefore, TFM, DDPM and CFD-DEM, ∂
ðε ρ Þ þ ∇  ðεs ρs !
u sÞ ¼ 0 ð2Þ
all of which have been available in existing software (FLUENT and ∂t s s
MFIX), are used to simulate impinging particle jets in a channel. The im-
pinging particle jets are chosen not only because of its simplicity and where εi, ρi and !
u i (i = g or s) represent the local volume fraction, den-
similarity to the crossing cluster flow that happens frequently in circu- sity and velocity vector, respectively. The momentum conservation
lating fluidized bed risers but also because the merits and drawbacks equations for gas and solid phases are
of the three models can be well highlighted in this type of flow. Noting
that a comparison of the TFM, DDPM and CFD-DEM methods is in
some sense more suitable for identifying the uncertainty of the models ∂   
!
ε g ρg ! ! !
u g þ ∇  ε g ρg u g u g ¼ −ε g ∇p þ ∇  τ g þ ε g ρg g
than with experiments, since in the first case the boundary conditions ∂t  
such as the inlet solid velocity and volume fraction and the wall bound- þβ ! !
u s− u g ð3Þ
ary condition for solid phase, are hard to be controlled precisely in the
real experiments. Moreover, it ensures that all systems have exactly
∂ !
the same drag model, gas and particle characteristics (perfectly non- ðε ρ ! ! !
u Þ þ ∇  ðε s ρs u s u s Þ ¼ −ε s ∇p þ ∇  τ s þ ε s ρs g
∂t s s s  
frictional mono-disperse sphere in present study). Also in numerical
þβ ! !
u g− u s ð4Þ
simulations, the possible effects of humidity, inter-particle cohesive
forces, electrostatics, polydispersity, non-sphericity and so on can be ex-
cluded. When these models are applied to simulate the hydrodynamics !
where p, ps and g are the gas pressure shared by both phases, solid pres-
of real fluidized beds, one will face more uncertainties and often get a
sure and gravitational acceleration, respectively. Furthermore, τ g, τ s and
satisfactory simulation result compared to the experimental data by
β are the shear tensor of gas phase and solid phase and the drag coeffi-
adjusting some of the model input, such as the drag force correlation
cient, respectively. The stress–strain tensor for gas and solid phases are
and/or wall boundary condition for solid phase.
  2  
! T
2. Mathematical models τg ¼ εg μ g ∇!
ug þ ∇ug − εg μ g ∇  !
ug I ð5Þ
3
It is well known that due to the nonlinear gas–solid interaction and    2

dissipative particle–particle and particle–wall interaction, the hydrody- ! T
τs ¼ −ps I þ μ s ∇!
us þ ∇us þ λs − μ s ð∇  !
u s ÞI ð6Þ
namic of gas–solid flow is quite complex due to the formation of 3
96 X. Chen, J. Wang / Powder Technology 254 (2014) 94–102

The interphase drag coefficient is calculated according to Wen and The terms in the right-hand side of Eq. (15) are the force contributions
Yu [30] due to buoyant gravity, inter-phase drag force and particle collisions,
  respectively. The !
F KTGF is computed from particle pressure predicted by
! !  the kinetic theory of granular flows as given in Section 2.1:
3 ρg εg εs  u g − u s  −2:65
β ¼ CD εg ð7Þ
4 dp
!
8   F KTGF ¼ −∇  τ s ð16Þ
   ! 
< 0:687 εg ρg dp !u g − u s
ð24=ReÞ 1 þ 0:15Re ; Reb1000
CD ¼ ; Re ¼
: 0:44; Re≥1000 μg
The conservation equation for the granular temperature and rele-
ð8Þ vant parameters are all the same as the TFM.
One point that deserves to mark is that the drag coefficient used in
Conservation equation of granular temperature is given as follows, Eq. (15) (β/εg) is based on the so-called type-B drag model. It is well
  known that different types of momentum equations (type-A vs type-
3 ∂ðεs ρs Θs Þ ! B) have been used in numerical simulations of gas–solid flow [35]. Pre-
þ ∇  ð ε s ρs ! !
u s Θs Þ ¼ −∇  q þ τ s : ∇ u s − J s −3βΘs ð9Þ
2 ∂t vious studies [36,37] have shown that the momentum equation and the
inter-phase drag correlation used must match each other to make sure
Where the first term on the right-hand side represents the conduc- that the inter-phase interaction force (drag force plus pressure gradient
tion of the fluctuating energy, the second term is the generation due force) is equal to the effective weight of gas–solid suspension. In present
to solids stresses, the third term is the sink due to inelastic particle col- study, type-A momentum equation is used as shown in Eq. (3), whereas
lisions and the last term is the sink due to the presence of gas phase. The type-B model is used to calculate the particle motion as shown in
algebraic formulation of the equation is solved by neglecting the con- Eq. (15); therefore, type-A drag correlation (β) is modified into type-B
vection and conduction terms. The solids pressure and the solid shear drag correlation (β/εg) according to the study of Gidaspow [35].
and bulk viscosities are calculated according to Lun et al. [31]

2
ps ¼ εs ρs Θs þ 2ρs ð1 þ eÞεs g 0 Θs ð10Þ 2.3. CFD-DEM method
rffiffiffiffiffi pffiffiffiffiffiffiffiffi  
4 2 Θs εs ρs dp πΘs 2 In the DEM method, the mass and momentum conservation equa-
μs ¼ ρs dp εs g0 ð1 þ eÞ þ 1 þ ð1 þ eÞð3e−1Þε s g0 ð11Þ
5 π 6ð3−eÞ 5 tions for gas phase are also given by Eqs. (1) and (3). However, the dif-
rffiffiffiffiffi ferent treatments of particulate phase in the TFM and CFD-DEM result in
4 Θs one of the major differences between them (i.e., the way how the inter-
λs ¼ εs ρs dp g 0 ð1 þ eÞ ð12Þ
3 π phase drag force term is calculated). In the TFM, the drag force is only a
function of averaged parameters (local solid volume fraction, gas and
The collisional dissipation of granular energy is given as follows: solid velocities) as well as gas and particle properties. In contrast, in
  the CFD-DEM method, the drag force exerted on each particle in a
12 1−e2 g 0 2 3=2
fluid cell is calculated and summed over all particles within that specific
Js ¼ pffiffiffi ρs εs Θs ð13Þ cell to obtain the drag force term in the momentum equation of gas
dp π
phase. Fortunately, those two different methods of calculation of inter-
phase drag force have a negligible effect on the simulation results [38].
where g0 is the radial distribution function. There is no unique formula-
tion in the literatures. For example, a new formulation based on the mo-
lecular dynamics simulations is recommended for dense solid fraction
[32]. The formulation of the radial distribution used in present study is
given as follows,
" !1=3 #−1
εs
g 0 ¼ 1− ð14Þ
εs; max

It should be noted that the constitutive correlations are under active


research [32,33] and the formulations are slightly different from the
default formulations in MFIX [34]; therefore, they are modified in
order to be consistent with the formulations in the commercial software
FLUENT.

2.2. Dense discrete particle model (DDPM)

The mass and momentum conservation equations for gas phase are
almost the same as the TFM, except that the interphase exchange term
in momentum equation is obtained from a summation of the drag forces
acting on all the discrete particles in a fluid computational cell. Howev-
er, in DDPM, the particle phase is described by the Newtonian equations
of motion for each particle (or parcel) in the system. The trajectory of
the particles is calculated by a force balance

d!   β 
us ! ! ! !
ρs ¼ g ρs −ρg þ u g − u s þ F KTGF ð15Þ
dt εg Fig. 1. The channel geometry used for the calculations of two impinging particle jets.
X. Chen, J. Wang / Powder Technology 254 (2014) 94–102 97

Fig. 2. Particle flow pattern with a jet superficial gas velocity of 10 m/s and a solid volume fraction of 0.3: (a) TFM, (b) DDPM and (c) CFD-DEM.

The equation of motion solved for particle a in MFIX is Eq. (22) makes sure that the restitution coefficient (e) is same in both
TFM and CFD-DEM method. Finally, we note that the normal spring
stiffness (kn) is selected arbitrarily, provided that the maximum overlap
d!ua ! βV a ! !  !
ma ¼ −V a ∇p þ ma g þ u g − u a þ F c;a ð17Þ between contacting particles is less than 1% of particle diameter at any
dt 1−ε g
time step.

The terms in the right-hand side of Eq. (17) are the force contribu-
3. Simulation set up and layout
tions due to pressure gradient, gravity, interphase drag and particle col-
lisions, respectively. In the TFM, the particle–particle and particle–wall
Impinging particle jets are chosen. The channel geometry used for
friction is not considered; therefore, in the CFD-DEM simulation, only
the calculations of two impinging particle jets is shown in Fig. 1. The
the normal particle–particle and the normal particle–wall collisions
gas flows into the simulation domain from the bottom in plug flow
were included to maintain consistency with KTGF. A linear spring-
with specified velocity and leaves from the top side of channel, where
dashpot model is used. In addition to its simplicity, the linear spring-
atmospheric pressure is prescribed. Two particle flows are injected in
dashpot model leads to a constant restitution coefficient as used in the
the two opposite sides of the channel with 45° angels, and the particle
TFM method, which makes it equitable. In the KTGF, quasi-
velocity is specified as the same as the gas flow in the bottom side.
instantaneous particle collisions are assumed; however, in the so-
The gas density is 1.2 kg/m3, the gas viscosity is 1.8 × 10−5 kg/(m s −1),
called soft sphere model used here, particle collisions take a finite
the particle density is 1000 kg/m3 and the particle size is 6.5 × 10−4 m.
time. Fortunately, a previous study [39] has shown that an excellent
All the simulations are run without gravity and with Wen-Yu drag [30].
agreement can be found between the results obtained from soft-
The jets' solid volume fractions and velocity magnitudes are varied in
sphere model and hard-sphere model. The particle contact force is cal-
different cases. Unless specified otherwise, the restitution coefficients
culated by
for particle–particle and particle–wall interactions are 0.8 and 1.0, respec-
tively. The particle stiffness coefficient used in the CFD-DEM method is
!
F ab ¼ −kn δn ! !
n ab −ηn v ab;n ð18Þ 1000 N/m and the grid number used in all three methods is 30 × 60.

where the overlap is given by

δn ¼ ðRa þ Rb Þ−j! !
r b− r aj ð19Þ

where Ra and Rb denote the radii of the interacting particles, ! !


r a and r b
denote the position vector of the particles. The normal unit vector is de-
fined as
! !
! r b− r a
n ab ¼ ð20Þ
j! !
r b− r aj

The relative velocity of particle a and b is


!
v ab;n ¼ ðð! ! ! !
v a − v b Þ  n ab Þ n ab ð21Þ

After that, we can calculate the normal damping coefficient as follow


[40]
pffiffiffiffiffiffiffiffiffiffiffiffiffi
−2 mab kn ln e Fig. 3. The voidage distribution along the nearby central axis of the channel
ηn ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ð22Þ
(x = 0.0145 m) with a jet superficial gas velocity of 10 m/s and a solid volume fraction
π2 þ ð ln eÞ2
of 0.3.
98 X. Chen, J. Wang / Powder Technology 254 (2014) 94–102

models as shown in Fig. 5. All flow patterns predicted by the three


models are not exactly the same, especially the DDPM method predicts
the jets are crossing in the impingement region, bouncing off the per-
fectly elastic wall and then further crossing and moving outward.
These phenomena were also reported by Cloete et al. [41]. However,
both TFM and DEM methods do not have such apparent crossing result.
The possible reason is that in DDPM method, the particle–particle inter-
action force calculated from kinetic theory of granular flow is too small
to represent the real situation, which results in the negligible effect of
particle–particle collisions. Therefore, the particles can flow in the sim-
ulation domain as if the gas–solid flow is in the dilute limit (i.e., the par-
ticle–particle interaction can be neglected and particles can overlap
each other during numerical simulations), although the solid concentra-
tion is actually as high as 0.1. Furthermore, in the studied gas–solid sys-
tem, since the injected particles are inertia-dominant, they do not
strictly follow gas flow as inertia-free particles will, and if the collisions
between particles have been neglected, the jets can obviously cross each
other as if they do not impinge. This result highlights that volume exclu-
Fig. 4. The velocity distribution along the nearby central axis of the channel sion effect is very important in numerical simulations of dense gas–solid
(x = 0.0145 m) with a jet superficial gas velocity of 10 m/s and a solid volume fraction flow and cannot be neglected as concluded in many previous studies
of 0.3.
[42]. Furthermore, the result of DDPM is very similar with the reported
simulation results obtained using MP-PIC method without the consider-
4. Result and discussion ation of collisions[43,44]. Recent studies by O'Rourke and Snider [44]
have shown that the model for inter-particle collision has a paramount
Fig. 2 shows the particle flow pattern using different models, where impact on the simulation results, by using the most advanced collisional
the jet superficial gas velocity is 10 m/s and solid volume fraction is 0.3. model, it is possible to correctly predict the merging of two impinging
All the three models predict a similar flow pattern: the two jets merge jets at a very similar situation. However, even with the most advanced
into one in the impinging region and then flow towards the outlet of collisional model, when two impinging jets[44] with a solid concentra-
channel. However, there are some differences predicted by the three tion of 0.01 (see Fig. 9 of their article) is carried out, the two jets are
models. Both TFM and CFD-DEM methods predict some particles have going to merge; however, the CFD-DEM method shows that under
opposite velocity to the main jet flow in the impinging region as this situation, the two jets should cross each other as shown in Fig. 8.
highlighted in the figure, while this phenomenon was not captured by Based on limited results present here, we may conclude that (1) the
DDPM method. This difference is quantitatively shown in Fig. 3, advanced MP-PIC method[44] is better than the DDPM method avail-
where the voidage along the nearby central axis of the channel able in FLUENT and (2) the model for particle–particle collision needs
(x = 0.145 cm) is shown. It can be seen that although all the models further improvement. Finally, although the basic flow pattern in Figs.
predict a similar trend, the voidage distributions are quantitatively dif- 5a and 5c is the same, there is also an obvious difference. In the result
ferent, especially in the region of bottom. Both TFM and DDPM predict a of DEM method, collisions cause the jets to spread and scatter, which
nearly constant voidage, which is close to the minimum voidage set in is not observed in the result of the TFM method. Similarly, O'Rourke
the simulation, while DPM predicts that the voidage increases towards and Snider[43] also reported that MP-PIC simulation do not observe
the end of simulation domain due to the expansion of merged jet. Fig. 4 the phenomenon that jets spread and scatter. Fig. 6 shows the voidage
shows the gas velocity distribution along the nearby central axis of the distribution along the nearby central axis of the channel (x = 0.145 m),
channel (x = 0.145 cm). Figs. 2–4 clearly indicate that much effort is and the DDPM predicts a peak of solid volume fraction at about
needed to improve the TFM and DDPM. height = 0.02 m and height = 0.05 m, respectively, in accordance
As we keep the inject velocity and decrease the solid volume fraction with Fig. 5. Although both TFM and CFD-DEM predict a merging jet,
of the jets flow to 0.1, different flow patterns are predicted by three the trend of the voidage distribution after merging is totally different.

Fig. 5. Particles flow pattern with a jet superficial gas velocity of 10 m/s and a solid volume fraction of 0.1: (a) TFM, (b) DDPM, (c) CFD-DEM (Kn = 1000 N/m) and (d) CFD-DEM
(Kn = 10 N/m).
X. Chen, J. Wang / Powder Technology 254 (2014) 94–102 99

Because of the main differences between the DDPM and the CFD-
DEM methods, (a) a concept of parcel has been used in DDPM and
(b) the detailed particle–particle interactions in the CFD-DEM method
have been replaced by a force gradient predicted by kinetic theory of
granular flow. We try to identify where the source of the apparent dif-
ference is. Fig. 7 shows the effect of the number of particles in a parcel
in DDPM method. The results are almost the same with decreasing the
number of particles in a parcel from 50 to 1. Parcels are found to overlap
significantly in the impinging region although there is only one particle
in a parcel, which is clearly an unphysical phenomenon due to the
unrealistic treatment of particle–particle interactions in the DDPM
method. Such unphysical overlap between particles is not possible in
the TFM and the CFD-DEM methods (there is a slightly overlap between
particles in CFD-DEM method, typically, less than one percent of particle
diameter), that is to say the volume exclusion effect has been consid-
ered in the TFM and CFD-DEM methods. Therefore, the use of parcel is
not the main source of the qualitative difference of DDPM, and the sim-
plified particle–particle interactions is the right answer. In general,
Fig. 6. The voidage distribution along the nearby central axis of the channel compared to DDPM, TFM gives a better agreement with the CFD-DEM
(x = 0.0145 m) with a jet superficial gas velocity of 10 m/s and a solid volume fraction
of 0.1.
results for the specific case. With respect to this, methods like similar
particle assembly model [45], large-scale discrete element method
[46] and discrete cluster method[47] may appear to be a better way to
reduce the computational cost of the CFD-DEM method while keeping
TFM predicts that particles trend to accumulate and then shorten the a reasonable accuracy, where the concept of parcel is used, but the in-
width of the merging jet, while CFD-DEM predicts the particles spread teraction between particles is still modeled using a DEM-type tech-
after that the jets are merged. nique. Of course, the computational cost is relatively high compared
Compared to TFM and CFD-DEM methods, DDPM has another major to DDPM method or MP-PIC method.
drawback when it is applied to dilute gas–solid flow or the situation The advantage of the DEM approach over the other two models lies
where part of the simulated domain is dilute (for example, gas–solid in its explicit treatment of the particle–particle collisions. The interac-
flow in risers where the bottom part of the bed is dense, but the upper tions between two particles are represented as spring and dashpot in
part of the bed is usually dilute). Due to the fact that KTGF is used to pre- DEM method, where the spring causes the rebound off the particles
dict !
F KTGF in DDPM method, the values of the state variables (averaged and the dashpot mimics the dissipation of the kinetic energy due to in-
solid concentration, granular temperature and solid velocity) of each elastic effect. If the spring stiffness coefficient is very small, it is possible
computational grid have to be evaluated from the corresponding dis- that a large overlap between particles can exist. Fig. 5d shows the case
crete values of particles at that grid; however, in dilute flow, it is possi- where the stiffness coefficient is decreased from 1000 N/m to a small
ble that there is no particle (or parcel) in a grid, which results in that the value (10 N/m). It can be seen that the jets indeed cross each other.
state variables are zero, leading to the impossibility to evaluation of the Such a parametric study further confirms the conclusion made in pre-
values of particle velocity gradient and granular temperature gradient. ceding paragraph and highlights the importance of volume exclusion
Therefore, special treatment is needed in this case as in the study of effect.
Cloete et al. [41]. Note that in a similar method (MP-PIC method), this As we still keep the inject velocity and decrease substantially the
drawback does not exist, because empirical correlation is used to solid volume fraction of the jets flow to 0.01, the flow patterns predicted
calculate ! F KTGF . Clearly, simplified CFD-DEM methods, such as DDPM by three models are shown in Fig. 8. In this case, the particle loading is
and MP-PIC, should find a realistic way to properly represent the true low and the particle inertia is high, one would expect that all three
particle–particle interactions. models predict the particle trajectory crossing effect [48]. However,

Fig. 7. Particles flow pattern with different particles number in a parcel using DDPM method: (a) 50, (b) 10 and (c) 1.
100 X. Chen, J. Wang / Powder Technology 254 (2014) 94–102

Fig. 8. Particle flow pattern with a jet superficial gas velocity of 10 m/s and a solid volume fraction of 0.01: (a) TFM, (b) DDPM and (c) CFD-DEM.

TFM method does not predict such a phenomena (we have run a lot of 5. Conclusion
cases with different inject velocity and solid volume fraction using
TFM method although they are not reported in the article, none of Three multiphase models are used to investigate the flow patterns of
them shows the particle trajectory crossing effect). Since it is well two impinging gas-particle jets in a channel. The TFM method and
known that the scale resolution has a significant effect on the simulation DDPM method have more potential for industrial applications since
results [49,50], we further test the effect of the size of computational they are less computational demanding. However, the limitations of
cell. Fig. 9 shows that further refinement of computational cell used in these methods should be understood. By studying the gas–solid flow
TFM model does not help to capture the particle trajectory crossing ef- in the extreme situation, some of the main drawbacks of both TFM
fect; even the grid size has already smaller than the particle size in method and DDPM method have been highlighted, it was shown that
case of Fig. 9c. The conclusion that the TFM cannot predict the particle TFM method always predicted a converging flow and failed to predict
trajectory crossing effect is not new since it has been concluded in the particle trajectory crossing effect in dilute system, while the DDPM
previous studies[41,51]. And the underlying reason for the failure method failed to predict the cases where the two particle jets are
has also been clarified, that is, in TFM, only the mean particle velocity emerging. At the same time, both TFM and DDPM successfully
is tracked (i.e., only one value is used to represent the real particle reproduced the main features of impinging flow at some cases. As the
velocity distribution), while it has been shown that using multiple conclusion that TFM failed to predict the particle trajectory crossing ef-
values (at least two values) to represent the real particle velocity fect in dilute system was not new, the main contribution of present
distribution is mandatory in order to capture the particle trajectory study appeared to be the fact that DDPM had the major drawback due
crossing effect [52,53]. to the over-simplified treatment of particle–particle interactions,
As a final note, we have carried out extensive parametric studies on highlighting the requirement of a proper way to represent the realistic
the effects of jet superficial gas velocity (from 0.1 m/s to 10 m/s) and particle–particle interactions and the importance of volume exclusion
solid volume fraction (from 0.01 to 0.3). The results are more or less effect in dense gas–solid flows. Efforts should be focused on the estab-
the same as these reported in present section. We therefore prefer not lishment of a proper model for representing particle–particle interac-
to report them for the purpose of simplicity. tion as in the study of O'Rourke and Snider.

Fig. 9. Particle flow pattern with different grid resolutions using TFM method: (a) 15 × 30, (b) 30 × 60 and (c) 60 × 120.
X. Chen, J. Wang / Powder Technology 254 (2014) 94–102 101

Notation [10] M. Xu, F. Chen, X. Liu, W. Ge, J. Li, Discrete particle simulation of gas–solid two-phase
flows with multi-scale CPU-GPU hybrid computation, Chem. Eng. J. 207 (2012)
746–757.
[11] B. Popoff, M. Braun, A Lagrange Approach to Dense Particulate Flows, In International
Conference on Multiphase Flow, 2007, (Leipzig, Germany).
[12] M.J. Andrews, P.J. Orourke, The multiphase particle-in-cell (MP-PIC) method for
TFM two-fluid model dense particulate flows, Int. J. Multiphase Flow 22 (1996) 379–402.
DDPM dense discrete particle model [13] D. Snider, P. ORourke, M. Andrews, An incompressible two-dimensional multiphase
DEM discrete element model particle-in-cell model for dense particle flows, Los Alamos National Lab, NM (United
States), 1997.
dp particle diameter, m [14] F. Li, F. Song, S. Benyahia, W. Wang, J. Li, MP-PIC simulation of CFB riser with
e coefficient of restitution between particle–particle interaction EMMS-based drag model, Chem. Eng. Sci. 82 (2012) 104–113.
ew coefficient of restitution between particle and wall interaction [15] M. Goldschmidt, R. Beetstra, J. Kuipers, Hydrodynamic modelling of dense
! gas-fluidised beds: comparison and validation of 3D discrete particle and continu-
g gravitational acceleration, m/s2
um models, Powder Technol. 142 (2004) 23–47.
g0 radial distribution function [16] Y. Tsuji, T. Tanaka, S. Yonemura, Cluster patterns in circulating fluidized beds pre-
p gas pressure, Pa dicted by numerical simulation (discrete particle model versus two-fluid model),
ps particle pressure, Pa Powder Technol. 95 (1998) 254–264.
[17] Y.P. Tsuo, D. Gidaspow, Computation of flow patterns in circulating fluidized beds,
Ug superficial gas velocity, m/s AICHE J. 36 (1990) 885–896.
Uslip slip velocity, m/s [18] C.H. Ibsen, E. Helland, B.H. Hjertager, T. Solberg, L. Tadrist, R. Occelli, Comparison of
! !
u g; u s gas and solid velocity vectors, m/s multifluid and discrete particle modelling in numerical predictions of gas particle
flow in circulating fluidised beds, Powder Technol. 149 (2004) 29–41.
I unit tensor [19] S. Benyahia, J.E. Galvin, Estimation of numerical errors related to some basic
kn normal spring stiffness, N/m assumptions in discrete particle methods, Ind. Eng. Chem. Res. 49 (2010)
! 10588–10605.
n ab normal unit vector
! [20] J. Wang, M.A. van der Hoef, J.A.M. Kuipers, Comparison of two-fluid and discrete
v ab;n relative velocity of particle particle modeling of dense gas-particle flows in gas-fluidized beds, Chem. Ing.
Tech. 85 (2013) 290–298.
[21] E.M. Ryan, D. DeCroix, R. Breault, W. Xu, E.D. Huckaby, K. Saha, S. Dartevelle, X. Sun,
Multi-phase CFD modeling of solid sorbent carbon capture system, Powder Technol.
Greek symbols 242 (2013) 117–134.
Β drag coefficient for a control volume, kg/m3 s [22] N.G. Deen, M.V.S. Annaland, M.A. Van der Hoef, J.A.M. Kuipers, Review of discrete
εg voidage particle modeling of fluidized beds, Chem. Eng. Sci. 62 (2007) 28–44.
[23] H.P. Zhu, Z.Y. Zhou, R.Y. Yang, A.B. Yu, Discrete particle simulation of particulate sys-
εs solid volume fraction tems: a review of major applications and findings, Chem. Eng. Sci. 63 (2008)
εs,max solid volume fraction at packed condition 5728–5770.
μg, μs fluid and solid viscosity, Pa · s [24] A. Nikolopoulos, D. Papafotiou, N. Nikolopoulos, P. Grammelis, E. Kakaras, An ad-
vanced EMMS scheme for the prediction of drag coefficient under a 1.2 MWth
λs solid bulk viscosity CFBC isothermal flow—part I: numerical formulation, Chem. Eng. Sci. 65 (2010)
Θs granular temperature, m2/s2 4080–4088.
ρg, ρs fluid density and solid density, kg/m3 [25] J.W. Chew, C.M. Hrenya, Link between bubbling and segregation patterns in
gas-fluidized beds with continuous size distributions, AIChE J. 57 (2011) 3003–3011.
δn overlap between particles [26] J.W. Chew, D.M. Parker, C.M. Hrenya, Elutriation and species segregation character-
ηn normal damping coefficient istics of polydisperse mixtures of group B particles in a dilute CFB riser, AIChE J. 59
(2013) 84–95.
[27] S. Cloete, S.T. Johansen, M. Braun, B. Popoff, S. Amini, Evaluation of a Lagrangian Dis-
crete Phase Modeling Approach for Resolving Cluster Formation in CFB Risers Fluid-
ized and Circulating Fluidized Beds, 7th International Conference on Multiphase
Acknowledgement Flow - ICMF 2010 Proceedings, 2010.
[28] M. Syamlal, W. Rogers, T.J. O'Brien, MFIX documentation: theory guide, Technical
Note, DOE/METC-94/1004, NTIS/DE94000087, National Technical Information Ser-
This study is financially supported by the National Nature Science vice, Springfield, VA, 1993.
Foundation of China (grant nos. 21206170 and 91334106) and the [29] M. Syamlal, MFIX documentation: numerical technique, Rep. DOE/MC31346–5824,
“Strategic Priority Research Program” of the Chinese Academy of Sci- DE98002029, 1998.
[30] C.Y. Wen, Y.H. Yu, Mechanics of fluidization, Chem. Eng. Prog. Symp. Ser. 62 (1966)
ences (grant no. XDA07080200).
100–111.
[31] C.K.K. Lun, S.B. Savage, D.J. Jeffrey, N. Chepurniy, Kinetic theories for granular flow:
inelastic particles in Couette flow and slightly inelastic particles in a general flow
References field, J. Fluid Mech. 140 (1984) 223–256.
[32] L. Oger, S.B. Savage, Airslide flows. Part 2—flow modeling and comparison with ex-
[1] S.B. Savage, L. Oger, Airslide flows, part 1: experiments, review and extension, Chem. periments, Chem. Eng. Sci. 91 (2013) 22–34.
Eng. Sci. 91 (2013) 35–43. [33] A. Nikolopoulos, N. Nikolopoulos, N. Varveris, S. Karellas, P. Grammelis, E. Kakaras,
[2] A. Nikolopoulos, K. Atsonios, N. Nikolopoulos, P. Grammelis, E. Kakaras, An advanced Investigation of proper modeling of very dense granular flows in the recirculation
EMMS scheme for the prediction of drag coefficient under a 1.2MWth CFBC isother- system of CFBs, Particuology 10 (2012) 699–709.
mal flow—part II: numerical implementation, Chem. Eng. Sci. 65 (2010) 4089–4099. [34] S. Benyahia, M. Syamlal, T. O'Brien, Summary of MFIX Equations 2012–1, 2012.
[3] W. Ge, W. Wang, N. Yang, J. Li, M. Kwauk, F. Chen, J. Chen, X. Fang, L. Guo, X. He, X. [35] D. Gidaspow, Multiphase Flow and Fluidization: Continuum and Kinetic Theory De-
Liu, Y. Liu, B. Lu, J. Wang, J. Wang, L. Wang, X. Wang, Q. Xiong, M. Xu, L. Deng, Y. Han, scriptions with Applications, ACADEMIC PressINC, 1994.
C. Hou, L. Hua, W. Huang, B. Li, C. Li, F. Li, Y. Ren, J. Xu, N. Zhang, Y. Zhang, G. Zhou, G. [36] Y.Q. Feng, A.B. Yu, Assessment of model formulations in the discrete particle simu-
Zhou, Meso-scale oriented simulation towards virtual process engineering (VPE)—the lation of gas–solid flow, Ind. Eng. Chem. Res. 43 (2004) 8378–8390.
EMMS paradigm, Chem. Eng. Sci. 66 (2011) 4426–4458. [37] Z.Y. Zhou, S.B. Kuang, K.W. Chu, A.B. Yu, Discrete particle simulation of particle–fluid
[4] J. Ding, D. Gidaspow, A bubbling fluidization model using kinetic theory of granular flow: model formulations and their applicability, J. Fluid Mech. 661 (2010) 482–510.
flow, AIChE J. 36 (1990) 523–538. [38] Y. Liu, EMMS-based Simulation of Gas–Solid Flow, (Doctoral thesis) Institute of Pro-
[5] X.-Z. Chen, D.-P. Shi, X. Gao, Z.-H. Luo, A fundamental CFD study of the gas–solid cess Engineering, Chinese Academy of Sciences2011.
flow field in fluidized bed polymerization reactors, Powder Technol. 205 (2011) [39] M. Ye, M.A. van der Hoef, J.A.M. Kuipers, From discrete particle model to a continu-
276–288. ous model of Geldart A particles, Chem. Eng. Res. Des. 83 (2005) 833–843.
[6] N. Zhang, B. Lu, W. Wang, J. Li, 3D CFD simulation of hydrodynamics of a 150MWe [40] N.G. Deen, Annaland M. Van Sint, M.A. Van der Hoef, J.A.M. Kuipers, Review of dis-
circulating fluidized bed boiler, Chem. Eng. J. 162 (2010) 821–828. crete particle modeling of fluidized beds, Chem. Eng. Sci. 62 (2007) 28–44.
[7] A. Nikolopoulos, N. Nikolopoulos, A. Charitos, P. Grammelis, E. Kakaras, A.R. Bidwe, [41] S. Cloete, S. Amini, S.T. Johansen, Performance evaluation of a complete Lagrangian
G. Varela, High-resolution 3-D full-loop simulation of a CFB carbonator cold KTGF approach for dilute granular flow modelling, Powder Technol. 226 (2012)
model, Chem. Eng. Sci. 90 (2013) 137–150. 43–52.
[8] Y. Tsuji, T. Kawaguchi, T. Tanaka, Discrete particle simulation of two-dimensional [42] S. Elghobashi, On predicting particle-laden turbulent flows, Appl. Sci. Res. 52 (1994)
fluidized bed, Powder Technol. 77 (1993) 79–87. 309–329.
[9] B.H. Xu, A.B. Yu, Numerical simulation of the gas–solid flow in a fluidized bed by [43] P.J. O'Rourke, D.M. Snider, An improved collision damping time for MP-PIC calcula-
combining discrete particle method with computational fluid dynamics, Chem. tions of dense particle flows with applications to polydisperse sedimenting beds and
Eng. Sci. 52 (1997) 2785–2809. colliding particle jets, Chem. Eng. Sci. 65 (2010) 6014–6028.
102 X. Chen, J. Wang / Powder Technology 254 (2014) 94–102

[44] P.J. O'Rourke, D.M. Snider, Inclusion of collisional return-to-isotropy in the MP-PIC [49] K. Agrawal, P.N. Loezos, M. Syamlal, S. Sundaresan, The role of meso-scale structures
method, Chem. Eng. Sci. 80 (2012) 39–54. in rapid gas–solid flows, J. Fluid Mech. 445 (2001) 151–185.
[45] M.A. Mokhtar, K. Kuwagi, T. Takami, H. Hirano, M. Horio, Validation of the similar [50] J. Wang, M.A. van der Hoef, J.A.M. Kuipers, Why the two-fluid model fails to predict
particle assembly (SPA) model for the fluidization of Geldart's group A and D parti- the bed expansion characteristics of Geldart A particles in gas-fluidized beds: a ten-
cles, AIChE J. 58 (2012) 87–98. tative answer, Chem. Eng. Sci. 64 (2009) 622–625.
[46] M. Sakai, Y. Yamada, Y. Shigeto, K. Shibata, V.M. Kawasaki, S. Koshizuka, Large-scale [51] R.O. Fox, A quadrature-based third-order moment method for dilute gas-particle
discrete element modeling in a fluidized bed, Int. J. Numer. Methods Fluids 64 flows, J. Comput. Phys. 227 (2008) 6313–6350.
(2010) 1319–1335. [52] O. Desjardins, R. Fox, P. Villedieu, A quadrature-based moment method for dilute
[47] X. Liu, X. Xu, Modelling of dense gas-particle flow in a circulating fluidized bed by fluid-particle flows, J. Comput. Phys. 227 (2008) 2514–2539.
distinct cluster method (DCM), Powder Technol. 195 (2009) 235–244. [53] A. Vié, E. Masi, O. Simonin, M. Massot, On the direct numerical simulation of
[48] S. Hari, Y.A. Hassan, Computational fluid dynamics simulation of virtual impac- moderate-Stokes-number turbulent particulate flows using algebraic-closure-based
tor performance: Comparison to experiment, Powder Technol. 188 (2008) and kinetic-based moment methods, Presented at Proceedings of the Summer Pro-
13–22. gram, 2012, p. 355.

Vous aimerez peut-être aussi