Vous êtes sur la page 1sur 22

Chemical Engineering Science 116 (2014) 49–70

Contents lists available at ScienceDirect

Chemical Engineering Science


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

Forces acting on a single introduced particle in a solid–liquid


fluidised bed
Zhengbiao Peng a, Swapnil V. Ghatage b,c, Elham Doroodchi a,n, Jyeshtharaj B. Joshi b,d,
Geoffrey M. Evans a, Behdad Moghtaderi a
a
Discipline of Chemical Engineering, The University of Newcastle, Callaghan, NSW 2308, Australia
b
Department of Chemical Engineering, Institute of Chemical Technology, Matunga, Mumbai 400 019, India
c
Department of Chemical Engineering, Indian Institute of Technology, Gandhinagar, Gujarat 382424, India
d
Homi Bhabha National Institute, Anushaktinagar, Mumbai 400 094, India

H I G H L I G H T S G R A P H I C A L A B S T R A C T

 Particle collision frequency


increased with solid fraction and
particle size ratio.
 The collision force decreased as the
bed voidage increased.
 Magnitude of collision force was
10–50 times greater than that of
weight force.
 Magnitude of collision force was
maximally 9 times greater than that
of drag force.
 A correlation describing collision
force as a function of bed voidage
was developed.

art ic l e i nf o a b s t r a c t

Article history: In a liquid fluidised bed system, the motion of each phase is governed by fluid–particle and particle–
Received 9 February 2014 particle interactions. The particle–particle collisions can significantly affect the motion of individual
Received in revised form particles and hence the solid–liquid two phase flow characteristics. In the current work, computational
17 April 2014
fluid dynamics–discrete element method (CFD–DEM) simulations of a dense foreign particle introduced
Accepted 27 April 2014
Available online 6 May 2014
in a monodispersed solid–liquid fluidised bed (SLFB) have been carried out. The fluidisation hydro-
dynamics of SLFB, settling behaviour of the foreign particle, fluid–particle interactions, and particle–
Keywords: particle collision behaviour have been investigated. Experiments including particle classification velocity
Fluidisation measurements and fluid turbulence characterisation by particle image velocimetry (PIV) were conducted
Collision frequency and collision force
for the validation of prediction results.
Discrete element method
Compared to those predicted by empirical correlations, the particle classification velocity predicted
Fluid–particle interactions
Classification velocity by CFD–DEM provided the best agreement with the experimental data (less than 10% deviation). The
particle collision frequency increased monotonically with the solid fraction. The dimensionless collision
frequency obtained by CFD–DEM excellently fit the data line predicted by the kinetic theory for granular
flow (KTGF). The particle collision frequency increased with the particle size ratio (dP2/dP1) and became
independent of the foreign particle size for high solid fractions when the fluidised particle size was kept
constant. The magnitude of collision force was 10–50 times greater than that of gravitational force and

n
Correspondence to: Priority Research Centre for Advanced Particle Processing & Transport, The University of Newcastle, Australia. Tel.: þ61 2 4033 9066;
fax: þ61 2 4033 9095.
E-mail address: elham.doroodchi@newcastle.edu.au (E. Doroodchi).

http://dx.doi.org/10.1016/j.ces.2014.04.040
0009-2509/& 2014 Elsevier Ltd. All rights reserved.
50 Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70

maximally 9 times greater than that of drag force. A correlation describing the collision force as a
function of bed voidage was developed for Stp 465 and dP2/dP1 r2. A maximum deviation of less than
20% was obtained when the correlation was used for the prediction of particle collision force.
& 2014 Elsevier Ltd. All rights reserved.

1. Introduction the published literature. In the past few decades the focus of the
research in such systems is on the drag applied on the particles
Fluidised beds are widely used in chemical, petrochemical and (Ghatage et al., 2013). The readers may like to read published
process industries. They are preferred over other reactors for literature for greater insight into the turbulence effects on the
carrying out various gas–solid, solid–liquid and gas–solid–liquid particle motion (Brucato et al., 1998; Doroodchi et al., 2008;
processes due to the enhanced contact between the fluid and solid Ghatage et al., 2013; Gidaspow, 1994; Joshi, 1983). In the view
particles. However, the efficient operation of a fluidised bed of present study, the literature review has been presented focus-
requires the accurate choices of design and operating conditions, ing on particle–particle collisions within SLFBs and numerical
and the accuracy of prediction of fluidisation behaviour. With approaches for the simulation of SLFBs.
increasing computational power and development in computa-
tional methods, the use of Computational fluid dynamics–Discrete
element method (CFD–DEM) in the Eulerian–Langrangian frame- 2. Literature review
work for detailed simulations of a fluidised bed is increasing
extensively. It provides important insight into the fluid flow, 2.1. Particle–particle collisions in SLFB
particle flow as well as the effects of fluid drag and particle–
particle collisions on the transport phenomena. Researchers have investigated the settling of a foreign particle
A particle in the solid–liquid fluidised bed (SLFB), mainly in SLFB (see e.g., Grbavcic et al., 2009; Grbavcic and Vukovic, 1991;
experiences two types of forces, i.e. drag applied by the fluid Joshi, 1983; van der Wielen et al., 1996). These authors have
along the fluid streamlines, and collisions with other fluidised studied the effect of fluid drag on the settling of the foreign
particles. The quantification of the relative contribution of drag particle in the presence of fluidised particles. Many have proposed
and collision is very complex but essential. In literature, many correlations to predict the particle classification velocity. For
researchers have studied the effect of turbulence generated by detailed discussion on the effect of fluid drag on the settling of a
fluid on the motion of the particle (see e.g., Brucato et al., 1998; foreign particle in the SLFB and other turbulence devices, readers
Doroodchi et al., 2008; Ghatage et al., 2013; Gidaspow, 1994; Joshi, are referred to Ghatage et al. (2013). Most of these studies
1983). Attempts have also been directed to quantify the effect of enlightened on the effect of fluid drag on the motion of the
fluid drag and propose correlations to predict the motion of foreign particle. However, the importance of particle–particle
individual particles based on theoretical and empirical approaches collisions was often overlooked. Very few attempts have been
(see e.g., Di Felice et al., 1991; Grbavcic et al., 2009; Joshi, 1983; van reported in the literature wherein the estimation of the collisional
der Wielen et al., 1996). However, very few studies have been force was discussed. Some researchers have tried to quantify this
reported discussing the collision effect on the motion of the effect in terms of collisional frequency, number of collisions and
settling particle in the SLFB (Grbavcic et al., 2009; van der collision intensity.
Wielen et al., 1996). It was thought desirable to quantify the Pozo et al. (1993) developed an experimental technique to
particle–particle collision effect and the relative importance of measure the particle collision frequency based on the perturba-
drag and collision forces both experimentally and through tions in the diffusion current of tethered electrodes. A tethered
modelling. particle near its anchorage point was used and gave a good
In this work, a dense foreign particle has been introduced into estimate of the collision frequency of a free floating particle. The
the SLFB. CFD–DEM simulations have been carried out on the collision frequency between an electrode and the bed particles
settling of the foreign particle in the SLFB. The fluid–particle was obtained from the measured number of collisions Nc between
interactions were solved by a fully two-way coupling algorithm the nc tracer particles and the electrode over time t:
(Kafui et al., 2002; Tsuji et al., 1993; Xu and Yu, 1997), i.e., effects of f coll ¼ ðNc =tÞðnt =nc Þ ð1Þ
solid volume fraction on fluid flow and interphase momentum
exchange have been rigorously implemented in the governing where nt is the total number of bed particles which was obtained
equations of the solid–liquid two-phase flow. The particle–particle from the particle mass in the bed. Their experimental results
interactions were solved based on the contact mechanics of rigid showed that the collision frequency monotonically decreased as
bodies (Peng et al., 2013; Tsuji et al., 1993). The motion of the foreign the superficial liquid velocity was increased.
particle was monitored continuously whilst it moved through the Gidaspow (1994) has proposed a correlation for the prediction
bed and the detailed information on particle dynamics was of collision frequency in a binary mixture based on the analogy
recorded. The fluidisation hydrodynamics of SLFB, settling behaviour with the kinetic theory of granular flow (KTGF) as
of the foreign particle, fluid–particle interactions and particle– pffiffiffiffiffiffiffiffi
N12 ¼ 4n1 n2 d12 g 0 πΘ
2
ð2Þ
particle collision behaviour have been investigated. The model
predictions were compared with experimental measurements. where the granular temperature, Θ was defined as
A number of CFD–DEM studies of gas fluidised beds were
1D E
reported in the literature (see e.g., Kafui et al., 2002; Peng et al., Θ ¼ u2p ð3Þ
3
2013; Tsuji et al., 1993; Wu et al., 2006; Xu and Yu, 1997; Ye et al.,
2004), however, similar studies of SLFBs are very scarce. Notably, and g0 was defined as a function of solid fraction as
there have been, to our knowledge, no published numerical     1
g 0 ðαS Þ ¼ 1  αS =αS0 ð4Þ
studies focusing on the settling of a foreign particle in the SLFB.
Therefore, an attempt has been made below to bring out the Gjaltema et al. (1997) assumed homogeneous isotropic turbu-
knowledge gaps and the significant attempts in similar studies in lence and the particle size being larger than the turbulent
Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70 51

microscale in three-phase inverse fluidisation. They proposed a particle collision will not lead to a rebound (ec ¼ 0). Therefore, for
correlation for computing the collision frequency in a binary high solid phase fractions (αS 4 0.3), particle collisions can no
mixture as longer be accurately detected by particle acceleration signals, due
N 12 ¼ π n1 n2 d12 U r
2
ð5Þ to the fact that the total particle collision frequency increases but
collisions with rebound (i.e. acceleration signal) are less frequent.
where Ur is the relative velocity between the fluid and particle. They also claimed that their approach based on the acceleration
For cases of solid–liquid particulate flows where the collisional threshold cannot be applied for high solid fractions (αS 40.3).
effects can be considered dominant, the particle collision beha- van der Wielen et al. (1996) proposed a correlation for
viour was also described by the classic kinetic theory for granular estimating the collision force between a foreign particle and
flow (KTGF) (Ding and Gidaspow, 1990; Koch, 1990; Simonin, fluidised particles in SLFB based on the force balance. This collision
1991). A prediction of a dimensionless collision frequency can be force was defined as
obtained from the KTGF,
rffiffiffiffiffiffi F P  P ¼ F G  F B  F F–P ð9Þ
2
n
f coll ¼ 24 αS g 0 ð6Þ where the fluid–particle interaction force was described as

where g0 is the pair correlation function defined as π 2 ρL U 2S2
F F  P ¼ C D dP2 ð10Þ
  4 2
αS  2:5αSm
g0 ¼ 1  ð7Þ The buoyancy force was defined in literature based on the fluid
αSm
density (Clift et al., 1987; Epstein, 1984; Fan et al., 1987; Joshi et al.,
where αSm is the solid fraction in packed beds of particles. 2001) or mixture density (Foscolo and Gibilaro, 1984; Foscolo et
The results predicted by the above two correlations (Gidaspow, al., 1983; van der Wielen et al., 1996). So, the particle–particle
1994; Gjaltema et al., 1997) and those by KTGF consistently collision (or friction) force was given as
showed that the collision frequency continuously decreased as
the bed voidage increased. Zenit et al. (1997) developed a F ρ
 PP  ¼ 1 M 
0:75C D ρL U 2S2
ð11Þ
measuring technique based on the use of a dynamic pressure V P2 ρP2 g ρP2 dP2 g ρP2
transducer, to measure the collisional particle pressure (character-
Based on the experimentally noted classification velocity, they
ising particle–particle collisions) in a two-phase fluidised bed. The
plotted the ratio of FP–P/FG (from Eq. (11)) against the product of
value of the collisional particle pressure was shown to reach a
liquid hold-up (αL) and particle classification velocity (V2) and
maximum at a solid fraction between 0.37 and 0.43. The increasing
observed a linear dependency. The particle classification velocity
number of collisions and the decreasing collision intensity was
refers to the constant settling velocity of the foreign particle
thought to be responsible for the maximum value of the collisional
through a liquid fluidised bed of different particles. The slope
particle pressure as the solid hold up was increased.
was defined as the interaction coefficient (χ) and its dependency
It is clearly understandable that the number of collisions
on the ratio of dense particle diameter to fluidising particle
increases with the solid fraction, up to a given value where the
diameter was defined as
motion of particles becomes hampered by the high packing
 0:95
density (i.e., in the packed bed). However, Buffière and Moletta d
(2000) carried out experiments using a hydrophone to estimate χ ¼ 10:5 P1 ð12Þ
dP2
the frequency of collisions, which was found to increase with an
increase in solid fraction, attain a peak around αS ¼0.35 and Grbavcic et al. (2009) performed experiments to study the
further decrease as solid fraction approached the packed bed single particle settling velocities in a fluidised bed. Due to colli-
state. Buffière and Moletta (2000) attributed this phenomenon to sions between the foreign particle and fluidised particles, an
the significant effect of particle–particle interactions on the additional collision force was taken into account in their theore-
structure of flow field at high values of solid fraction. However, tical analysis. As a settling particle must displace N fluidised
uncertainties about the validity of the above explanation were particles (i.e. non-elastic collisions), they assumed that the colli-
addressed by the authors, who also stated that more theoretical sion force is proportional to the total drag forces acting on the N
work was needed to understand the underlying mechanisms at a displaced particles moving through the fluidised suspension. The
high solid amount. number of displaced particles N was given as
Aguilar-Corona et al. (2011) conducted experiments to measure  3
dP2
the particle collision frequency by particle tracking in an index- N ¼ ð1  αL Þ ð13Þ
dP1
matched array. Collision detection was based on the acceleration
threshold of the instantaneous speed of coloured tracers. The Therefore, the collision force was expressed as

measured particle collision frequency also showed a peak value at 1 π 2

the solid fraction of 0.3, which apparently deviated from the F P  P ¼ kð1  f ÞN ρf dp U 2S1 ð14Þ
2 4
prediction results by KTGF. Aguilar-Corona et al. (2011) attributed
this behaviour of their experimental results to the detection limits where f is the Ruzicka (2006) transition function. k is the collision
of collisions from the particle acceleration signal process. Specifi- coefficient and was defined as
cally, the particle Stokes number was estimated based on the    1:48  
rDffiffiffiffiffiffiffiffiffiffiffi
E dP2 ρM  2:47
k ¼ 2:2αL 5:07 ð15Þ
particle root-mean-square velocity (i.e., q2p ) to evaluate the dP1 ρeff
effect of collisions that occurred with rebound. The particle Stokes where ρM is the mixture density and ρeff is the effective or relevant
number was defined as density and can be obtained as defined by Ruzicka (2006).
rffiffiffiffiffiffiffiffiffiffiffi
D E Grbavcic et al. (2009) claimed that the collision force can be one
St p ¼ ρp q2p dp =ð9μf Þ ð8Þ order of magnitude higher than the drag force between these
particles and significantly affects the classification velocity of the
For solid fractions higher than 0.3, Stp in their study was smaller foreign particle. They also pointed out that for high bed voidages
than 10, which is considered as the critical value below which a the collision forces between particles of fluidised bed and the
52 Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70

transiting particle are relatively small but their influence increases particles. Such approach can be used to simulate dense, as well as
as the bed voidage decreases. dilute granular systems. However, it was assumed that particle–
particle collisions are impulsive events that do not depend on the
2.2. Numerical approaches for simulation of SLFB local flow field of the continuous phase. This assumption is only
valid in cases where the inertia of the continuous phase is
The motion of both the fluid and the solid particles in SLFBs has negligible compared to that of the dispersed phase (e.g. in gas-
been simulated by the direct numerical simulation (DNS) (see e.g. fluidised beds). Seibert and Burns (1998) used a discrete particle
Hu et al., 2001; Reddy et al., 2010, 2013; Reddy and Joshi, 2008, simulation technique to study structural phenomena in liquid
2009; Sangani and Didwania, 1993; Sangani and Prosperetti, 1993). fluidised beds. Their model, however, was developed in a simpli-
The hydrodynamic forces acting on the particles were directly fied form. The net force acting on a particle that was used to
computed from the surrounding fluid flow, and the motion of the predict particle motion was given by the difference between the
fluid flow and particles were fully coupled. The DNS of the exact equilibrium value of hydrodynamic forces and the calculated value
particle motion may be the only theoretical tool capable of at the current conditions. Moreover, the particle–particle interac-
studying the nonlinear and geometrically complicated phenomena tion forces were not considered in their model. Zhang et al. (1999)
of particle–particle interactions (Hu et al., 2001). However, due to have shown that the generally used collision models need to be
the prohibitive CPU time, the number of particles involved in the modified when the continuous phase is a liquid, in order to
simulation has been largely limited. Brady and Bossis (1988) have account for the drainage of the fluid between colliding particles
developed numerical techniques (Stokesian dynamics) for simu- and the acceleration of the fluid surrounding the particles. The
lating the motion of a relatively large number of particles in Stokes adaptations of discrete particle model (DPM) involved two addi-
flows. Their model was appropriate for colloidal suspensions in the tional forces, i.e. the virtual mass force and the pressure gradient
limit of zero particle Reynolds number. Hu and co-workers (Hu, force. They found that when these two forces were incorporated,
1996; Hu et al., 1992, 1993) developed the arbitrary Lagrangian– the strong influence of the liquid surrounding the particles on the
Eulerian (ALE) particle mover for simulations of fluid–solid sys- particle trajectories before and after a collision can be captured.
tems at finite Reynolds numbers. In this scheme both the fluid and Malone (2007) developed the fully coupled CFD–DEM model
solid equations of motion were incorporated into a single coupled and applied the model in the simulations of hydrodynamics and
variational equation. At each time step a new mesh was generated heat transfer of liquid-fluidised beds. Their simulation results
when the old one became too distorted, and the flow field was indicated that modifications to account for the liquid effect have
projected onto the new mesh. Glowinski et al. (1999) developed a a significant improvement in terms of the micro-scale particle
different approach based on the concept of fictitious domain using mixing behaviour. The CFD–DEM model has since been applied to
a fixed grid, independent of remeshing. Applying this concept to study the liquid–solid interactions in the context of the analysis of
particulate flows, the particle domain was treated as a fluid with gas bubble formation in a gas–liquid–solid system (Chen and Fan,
additional constraints to impose proper rigid body motion. 2004; Li et al., 1999, 2000; van Sint Annaland et al., 2005). Later, Di
Later, Pan et al. (2002) applied the fictitious domain method in a Renzo and Di Maio (2007) applied the CFD–DEM model to
3D simulation with 1204 particles. Peskin (1972) proposed an investigate the characteristics of particulate and aggregative flui-
immersed boundary method (IBM) that allows the use of rectan- disation as well as the transition hydrodynamic stability of gas and
gular grids for arbitrarily complex geometries. Recently, Guo et al. liquid fluidised beds.
(2013) incorporated IBM into their existing CFD–DEM code to Given the above similar studies in the literature and the
simulate slightly compressible gas–solid flows with complex and/ existing knowledge gaps, the objective of present work is to
or moving boundaries. It was demonstrated that the capacity of provide insight into particle–particle and fluid–particle interac-
conventional CFD–DEM has been enhanced with the incorporation tions as well as their relative importance in the settling process of
of IBM. However, as a whole, all of the above methods that are a foreign particle within the SLFB using a CFD–DEM model of high
capable to reveal more details of the fluid flow around a solid fidelity. In the next section, details pertinent to the CFD–DEM
particle are very computationally demanding, which renders them model are presented, followed by the description on the experi-
impractical to simulate some real problems. mental set-up.
The most common and efficient approach for simulating parti-
culate flows is to use the continuum theory (i.e. the Eulerian
approach) that views the solid and the liquid as interpenetrating 3. Computational
mixtures (see e.g. Drew and Passman, 1999; Fan and Zhu, 1998;
Gidaspow, 1994; Lu and Gidaspow, 2003; Reddy and Joshi, 2009). 3.1. Model formulation: Computational fluid dynamics–discrete
However, this approach leads to unknown terms that represent element method
the interactions between the phases. As a result, the nature of
detailed interactions between fluid and particles and those 3.1.1. Governing equations
between particles and particles cannot be understood from the In a dense fluid–solid flow, a single particle is interacting with
application of mixture theories alone. neighbouring particles, surrounding fluid and computational
A second approach for the simulation of SLFBs is Lagrangian domain boundaries. The equations for the translational and rota-
numerical simulation (LNS), which provides a direct description of tional motion of a particle are
the particulate flow by tracking the motion of individual particles, dvi
mi ¼ f c; i þ f f; i þmi g ð16Þ
thus allowing for special focus and treatments on particle–particle dt
and particle–fluid interactions. The fluid flow satisfies the con- and
tinuum equations that are solved on a fixed field in the conven-
tional Eulerian way. d ωi
Ii ¼ Tc;i þTr;i ð17Þ
Andrews and O'Rourke (1996) introduced a multiphase dt
particle-in-cell (MP-PIC) method which used the mapping of A soft-sphere model, namely the linear spring-dashpot model
Langrangian particles to the computational grid. On grid points, (Tsuji et al., 1993; Xu and Yu, 1997), was employed to calculate the
continuum approach was applied for particles considering the collision contact force fc,i and the contact torque Tc,i caused by the
particle phase as the continuum and mapped back to individual tangential component of the contact force. Tr,i is the rolling
Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70 53

resistance torque caused by particle rolling motion (Ai et al., 2011; is calculated based on the Newton's third law by
Zhou et al., 1999) and is calculated by !
Np
N pc
ωij f S–L ¼  ∑ f f ;i' =ΔV c ð28Þ
Tr;i ¼  ∑ μrol jf cn;ij j r ð18Þ i¼1
j¼0 jωij j i
where Np is the total number of particles with a partial or entire
where mrol is the rolling friction coefficient depending on the body locating in the cell and ΔVc is the cell volume. f f;i' is the fluid
particle material. Npc is the total number of the particles that are force acting on the partial particle body (i.e. when the particle is
contacting with particle i at the time instant. fcn, ij is the normal sitting over the cell boundaries) or the entire particle body (i.e.
contact force; ωij is the relative angular velocity between particles when the particle is enclosed by the cell boundaries). In the
i and j as conventional approach, the entire body of a particle is treated to
ωi r i þ ωj r j be in that cell if the particle centroid is inside the cell, thus we
ωij ¼ ð19Þ
have f f ;i' ¼ f f ;i . However, in this study the scenarios where a
ri þ rj
particle is sitting over the cell boundaries were considered and
The fluid–particle interaction force (ff, i) results from the distortion the particle segment volume cut by cell boundaries was calculated
of fluid streamlines passing around the particle and the subse- accurately by a quasi-analytical approach, as to be detailed in
quent variation of local fluid stress tensor at the particle surface. Section 3.1.3. Accordingly, f f ;i'is calculated as
According to Anderson and Jackson (1967), the total fluid force
acting on particle i was expressed as f f;i' ¼ δi f f;i ð29Þ

f f ; i ¼  V i ∇p þV i ð∇ U τf Þ þ αL f d;i ð20Þ where δi is the fractional volume of particle i in the cell (Kuang
et al., 2008).
where τf is the fluid viscous stress tensor and expressed as
 
2
τf ¼ μL ½ð∇uL Þ þ ð∇uL Þ  1  þ λ  μL ð∇ U uL ÞI ð21Þ 3.1.2. Mapping scheme from Eulerian variables to point values
3
The point values of Eulerian variables were calculated based on
It should be noted that the fluid force acting on a particle were their spatial gradient distribution. The point values of these
calculated as point forces at the centroid of a particle to consider variables at the particle centroid position, i.e. ϕp, are thus
the distribution of particles in a cell (van der Wielen et al., 1996; calculated by
Wu et al., 2006; Ye et al., 2004). To this end, the Eulerian variables
of fluid flow field need to be converted to point values at the
φp ¼ φc þ dr U ∇φGR ð30Þ
particle centre (see Section 3.1.2). where dr ¼xp  xc.∇φGR is the gradient of an Eulerian variable that
Compared to the Eulerian–Eulerian approach, no closure is is also stored at the cell centre and computed using the divergence
required for the solid phase stress tensor in the DEM method, as theorem, i.e.,
the motion of individual particles is solved directly. However,
empirical correlations have to be used for the fluid drag force, i ¼ 1 φf;i A f ;i
∑N f

∇φGR ¼ ð31Þ
since the hydrodynamics of fluid phase is resolved on a length ΔV c
scale larger than the particle size. In this study, the Gidaspow drag
where Nf is the number of cell faces; Af,i is the area normal of cell
law (Gidaspow, 1994) was employed for the calculation of fluid
face i. ϕc is the variable value stored at the cell centre and ϕf is the
drag force fd,i,
face value interpolated by adjacent cells following a certain
Vi numerical scheme, e.g. central-differenced, second-order upwind
f d; i ¼ βðuL vi Þ ð22Þ
ð1  αL Þ and third-order Monotone Upstream-centred Scheme for Conser-
where β is the interphase momentum exchanging coefficient and vation Laws (MUSCL).
calculated by the Gidaspow draw law through the following
expressions: 3.1.3. Calculation of cell void fraction
8 Conventionally the cell void fraction was calculated by affiliat-
150ð1  αL Þ2 μL 1:75ð1  αL ÞρL ðuL  vi Þ
>
< þ ðαL r0:8Þ
dp α L
2 dp ing each particle with a cell as the particles were tracked
β¼ ð23Þ
: 0:75C d ρg αL ð1 dαL ÞjuL  vi jαL 2:65
> ðαL 40:8Þ transiently (Tsuji et al., 1993). This affiliation is determined based
p
on the position of the particle centroid, that is, if a particle
where Cd is the drag coefficient and givens as centroid is inside a cell, the entire body of the particle is treated
( to be in that cell and hence the cell void fraction is calculated
0:44 ðRep r1000Þ
based on the total volume of the particle rather than the actual
C d ¼ 24 ð1 þ 0:15Re0:687 Þ ðRe 41000Þ ð24Þ
Rep p p partial volume of the particle within the cell. This treatment may
lead to errors of up to 50% in the calculation of segment volume of
Rep is the particle Reynolds number based on the superficial slip one particle in a computational cell when the particle centroid is
velocity between the particle and the fluid and calculated by near the cell boundaries. Such large errors in particle segment
ρg dp αL juL vi j volume may result in the incorrect prediction of local void fraction
Rep ¼ ð25Þ
μL and consequently lead to strange behaviour in the model outputs
if the particle size is comparable with the computational cell size
The local averaged continuity and momentum equations of the (Peng et al., 2014). In this study a quasi-analytical method based
continuum can be written as (Anderson and Jackson, 1967) on the particle meshing technique was applied to calculate the cell
∂ðαL ρL Þ void fraction.
þ ∇ U ðαL ρL uL Þ ¼ 0 ð26Þ
∂t In the particle meshing method (PMM), the particle is meshed
into small particle grids. Then each particle grid is affiliated with a
∂ðαL ρL uL Þ
þ ∇ U ðαL ρL uL uL Þ ¼ ∇p þ ∇ U τf þ f S–L þ αL ρL g ð27Þ cell based on the position of their centroids. The volume of the
∂t solid phase in a computational cell is calculated by counting the
where fS–L is the local mean particle–fluid interaction force, which number of particle grids and adding up their volumes. The method
54 Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70

to implement PMM in the CFD–DEM simulation is pre- After meshing, the spherical surface of a particle is meshed into
sented below. a number of grids with planar surfaces. The sum of the volume of
Assume a particle is evenly meshed into Npm particle grids with particle grids is thus slightly smaller than the volume of that
each particle grid having a volume of Vpm ¼Vp/Npm. The equivalent particle. For this reason, the volume of each particle grid is
particle grid size is normalised by a scaling factor of γ ¼Vsum/Vp.
rffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffi
3 6V pm 3 6 Vp 3 1
dpm ¼ ¼ ¼ dp ð32Þ 3.1.4. Robust implementation of particle–fluid interactions
π π N pm N pm The dynamic effects of the presence of discrete particles in a
fluid flow are commonly taken into account by the terms involving
Hence, for the same cell size (i.e., Sc), the effective ratio of Sc/dpm is
cell void fraction and particle–fluid interaction forces. These two
much greater than that in the conventional approach (i.e. Sc/dp),
terms are often treated explicitly in the numerical solution (see e.
reducing the error associated with the calculation of particle
g., Kafui et al., 2002; Tsuji et al., 1993; Xu and Yu, 1997). However,
segment volume (Peng et al., 2014).
the explicit treatment of particle–fluid forces may induce conver-
In order to calculate the particle segment volume in computa-
gence difficulty in the solution of a dense particulate flow (Hu and
tional cells, the centroid position and the volume of each particle
Joseph, 1992). In the present study, we calculated the particle–fluid
grid need to be known. To reduce the computational time, a
forces semi-implicitly by treating the particle velocity (vi) impli-
template particle was introduced. The template particle has a
citly, as detailed below.
radius of r0 and the centroid position defined as x0 ¼(0, 0, 0). The
Since the non-linear momentum exchanging coefficient β is a
origin of local Cartesian coordinate is set at the centroid of the
function of uL and vi, we first explicitly linearise β using uL and vi
template particle. The template particle is meshed into a number
at the previous time step according to the drag laws of Gidaspow
of small particle grids. The data of each grid (i.e., centroid xeo and
(1994). Other terms such as pressure gradient and interphase slip
volume Veo) are calculated and stored in arrays of X and V before
velocities are evaluated implicitly. The linearisation of β leads to a
running the CFD–DEM simulation. In each time step, the data
system of linear equations. If only the fluid drag force and pressure
stored in X and V are mapped into the data of real particles that are
gradient force are considered as fluid forces, the particle velocity
moving around in the system. To this end, the coordinate trans-
can thus be calculated by
formation and magnification are conducted to convert the data of !
V i βi
n
template particle grids into the local values of a real particle by
mi ðvni þ 1  vni Þ ¼  V i ∇pn þ 1 þ V i ð∇ U τf Þn þ 1 þ ðunþ1
 v nþ1
Þ Δt s
1  αnL þ 1 L i

xe  xp ¼ ςðxe0  x0 Þ ð33Þ
ð35Þ
Based on the value of vni þ 1 and abiding by Newton's third law, the
V e ¼ ς3 V e0 ð34Þ
fluid–particle interaction forces are obtained. The interphase
where xe and Ve are the local centroid position and volume of momentum exchange terms are subsequently calculated by sum-
particle grids of real particles, respectively. ς is the magnification ming up the fluid–particle interaction forces on each particle in
factor defined as ς ¼ rp/r0 where rp is the real particle size. A the cell, weighted by the particle fractional volume (see Sections
meshed template particle (with 1214 particle grids) and the 2.2 and 2.3). Through the synchronized calculations of particle
acquisition of centroid position and particle grid volume are velocity and liquid flow velocity, the two-way coupling gets
illustrated in Fig. 1. numerically enhanced and hence significantly increase the solu-
tion convergence time (Hu and Joseph, 1992; Wu et al., 2009).

z
3.2. Computational geometry and numerical methodologies
and strategies

y Ve0
xe0 The computational geometry for the fluid flow was kept
consistent with the experimental set-up except the height of the
vertical cylinder, which was reduced to 1.0 m to minimise the
r0 computational time. A non-uniform but symmetric mesh was
x
generated on the bottom inlet face with dense meshes close to
x0 the boundary. The face mesh was then coopered along the axial
direction of the computational domain. Consequently, a total
number of 5600 hexahedral meshes were generated. The compu-
tational geometry and mesh for the calculation of fluid flow are
depicted in Fig. 2.
z The Semi-Implicit Method for Pressure-Linked Equation (SIMPLE)
algorithm was used to solve the pressure–velocity coupling equa-
xe Ve tions of the fluid flow, namely the continuity and momentum
rp
conservation equations (i.e., Eqs. (26) and (27)). The quadratic
xp
y upwind interpolation of convective kinematics (QUICK) scheme
was employed for the spatial discretisation of the convection term.
The diffusion term was discretised by a central-differenced scheme
that always is of second-order accuracy. The Green–Gauss Node
x
Based method was employed to calculate the variable gradients for
O constructing values of a scalar at cell faces and also for computing
Fig. 1. Particle meshing and acquisition of centroid position and volume of particle
secondary diffusion terms and velocity derivatives. In each time step,
grids in particle meshing method: (a) template particle meshed by 1214 grids and the globally scaled residual of 10  5 has been set as the convergence
(b) real particle and its particle grid. criteria for solving fluid phase equations.
Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70 55

0.05 m Table 1
Simulation conditions and parameters.

Fluid phase
Liquid density (kg/m3) 998.2 (water), 1800 (NaI)
Liquid viscosity (Pa s) 0.001 (water), 0.0018 (NaI)
Fluid time step (s) 1  10  4

Solid phase
Foreign particle diameter (mm) 6, 5, 4, 3
Fluidized particle diameter (mm) 3, 5
Foreign particle density (kg/m3) 7800
Fluidized particle density (kg/m3) 2300
Fluidized particle mass (kg) 0.05255
Normal spring-stiffness (N/m) 104
Restitution coefficient (dimensionless) 0.9
Sliding friction coefficient (dimensionless) 0.3
Rolling friction coefficient (dimensionless) 0.0175
Solid time step (s) 5  10  5

simulation. To obtain the steady state of the fluidised bed, 20 s


(with water) or 30 s (with NaI) of simulation was completed before
dropping the dense foreign particle from the top of the computa-
tional domain. A further 15 s of simulation was conducted to allow
the foreign particle to settle through the fluidised bed. However, to
save computational resources, the simulation will automatically
abort if the foreign particle reaches the bottom. For the sake of
accuracy, a value of 104 for the spring stiffness has been chosen for
all computational cases in this study, corresponding to a solid time
step of 5  10  5 s and a maximum particle normal overlap of
0.1074% rp (averaging over 10 s simulation; dP1 ¼ 6 mm, dP2 ¼
1.0 m

5 mm, UL ¼0.06 m/s). The simulation conditions and parameters


used in the simulations are listed in Table 1.
Multi-thread parallel computation (OpenMP) based on the
shared memory was conducted to improve the computational
efficiency. For a typical parallel simulation with 5600 meshes and
16,668 particles using two computational cores (clock speed
2.66 GHz, Smart Cache 12 M and QPI speed 6.4 GT/s), it expends
approximately 66 h to complete the simulation of 23 s solid–
liquid flow.

4. Experimental

4.1. Measurement of classification velocity

The schematic of the experimental set-up of the SLFB is shown


in Fig. 3. It consisted of a glass circular column (1), with an inner
diameter of 50 mm and a height of 1500 mm. The circular test
section of the fluidised bed was encased in a square column
(2) which was filled with the same liquid to ensure proper
photographic images. The distributor was a perforated plate
z (3) containing 128 holes of 2 mm diameter on a triangular pitch
y of 3.1 mm. A calming section (4) packed with either 6 or 8 mm
glass beads of 0.2 m height was provided to homogenise the liquid
x flow before it reached the liquid distributor. The required liquid
flow rate through the column was maintained using a rotameter
(5) and globe valve at the bed inlet. A 2 HP centrifugal pump
(6) was used for pumping the liquid from storage tank (7).
Borosilicate glass beads with mean diameter of 3 70.03 and
Fig. 2. Computational geometry: (a) domain and (b) mesh.
5 70.03 mm have been used as fluidised particles. The glass beads
were purchased from Sigmund Lindner (Germany) under the trade
name Silibeads type P having refractive index of 1.472. Proper
The fluidised particles with the same total mass (i.e. 0.5255 kg) arrangements (8) were made for the insertion of the foreign
in experiments were randomly distributed in the domain. A particle from top of the fluidised bed. A heat exchanger (9) was
simulation on the sedimentation process of the fluidised particles provided in circulation loop to maintain the temperature of fluid.
was conducted separately and the final steady packed bed was Precision-diameter dense (steel) particles with mean diameters
used as the initial condition of the solid phase in each CFD–DEM (72 mm) of 4 mm and 6 mm were used for the classification
56 Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70

Forreign particle was filled in the fluidised bed and the surrounding square column,
insertion 8 with the fluidised bed particles being 3 mm or 5 mm diameter
arrrangementt
Ou
utlet high-precision borosilicate glass beads. Dantec PIV system was
used consisting of Litron LDY 300 laser capable of generating
30 mJ/pulse energy at 1000 Hz. Phantom v640 camera capturing
image pairs with resolution of 1600  1600 pixels at 900 Hz
and BNC 575 synchroniser were employed. The high speed
50
Heat system provided time resolved data, giving a closer look at the
9
50
Exchangeer dynamics liquid velocity within the fluidised bed. The field of view
was 50  50 mm2 covering the complete column diameter. The
Glass column
c 1 refractive index matched sodium iodide solution was seeded with
1500
Outer squ
uare column
n 2
fluorescent seeding particles.
Post-processing of the captured raw PIV images was under-
taken to determine the velocity vectors. The raw PIV images were
processed using the image processing routines programmed in
MATLAB R2011a. Out-of-plane motion of the seeding particles and
Distributoor 3
strong local velocity gradients caused some spurious velocity
vectors. Median filtering, with a threshold value 1.5 times the
200 Calmingg section 4 median of surrounding vectors, was applied to filter the high
spurious vectors. A signal-to-noise ratio of 4 was applied to filter
Vent
the low spurious vectors. Parameters like time difference between
nlet
In
laser pulses, light sheet thickness and seeding density were
optimised so that spurious vectors remained below 2%.
Rotameter
R 5
Wateer
5. Results and discussion
7
PUM
MP 6
5.1. Fluidisation of solid–liquid systems

The hydrodynamics of solid–liquid fluidised bed was first inves-


Fig. 3. Schematic of the experimental set-up of the SLFB.
tigated to ensure the established model is capable to correctly
capture the complex interactions in the multi-particle systems,
namely particle–particle and particle–fluid interactions. Fig. 5 shows
velocity measurements. The settling velocities of single dense the total pressure drop of the liquid fluidised bed (in water) as a
particle were measured over a vertical distance of approximately function of superficial liquid velocity ranging between 0.03 m/s and
100–200 mm from bottom using a high speed video camera 0.24 m/s. As typically exhibited in fluidised beds, the pressure drop
(model: Photron Fastcam Super-10K) and a strong LED lighting keeps increasing until the onset of fluidisation, after which the
behind the bed. The classification velocity was calculated by pressure drop fluctuates around a constant value. As for each liquid
tracking the centre-motion of the dense particle from images superficial velocity, the simulation was run for a sufficient length of
(512  420 pixels) captured at 500 frames per second giving a time time (20 s) to obtain the steady state of fluidisation, the simulation
resolution of 0.002 s. For 5 mm particles, it was observed that the process is equivalent to the de-fluidisation process and the hyster-
dense particle can be easily seen even at high solid concentrations. esis effect was not reflected in Fig. 5.
However, with 3 mm fluidised particles, the refractive index of the As commonly applied in the literature to determine the mini-
particles needed to be matched to that of the liquid in order to mum fluidisation velocity, two fitting lines (y¼51750x  676.42
visualise the foreign particle clearly. Aqueous sodium iodide (NaI) and y¼ 1520) were drawn for the packed bed regime and the
solution was used to match the refractive index with borosilicate fluidisation regime, respectively. The intersection point (i.e. at
spheres. Lab reagent grade sodium iodide powder was purchased x¼ 0.0424) is the minimum fluidisation velocity Umf.
from Sigma Aldrich. To prepare the solution, sodium iodide The theoretical minimum fluidisation velocity can be calculated
powder was weighed in proportion to required amount of water. by equating the expression for the apparent weight of the particles
Powder was gradually added to water and dissolved with rod with the revised Ergun equation (Doroodchi et al., 2012):
!
stirrer. The best RI match was obtained for RI ¼1.472 correspond- μL U mf ð1  αL Þ ρL U 2mf ð1  αL Þ
ing to 58.5% w/w sodium iodide in water. To maintain the clarity of ð1  αL ÞðρP  ρL Þg ¼ 18 2 þ 0:33 ð36Þ
d P
α4:8
L
dP α4:8
L
solution, sodium thiosulphate is added to the solution. Addition of
approximately 0.1 g of sodium thiosulphate is sufficient to clarify where αL is the packed bed voidage, ρP and ρL are the particle
1 l of sodium iodide solution. Due care was taken to ensure that density and the fluid density, respectively; μL is the fluid viscosity.
the particles had reached terminal velocity prior to entering the The predicted Umf by Eq. (36) is 0.0423 m/s, which agrees well
test section and not influenced by the mean flow. The fractional with the value predicted by the CFD–DEM model (i.e. Umf ¼
liquid hold-up was varied from 0.42 to 0.80. Each settling velocity 0.0424 m/s). The total pressure drop calculated based on the
reported is the average of five measurements, where the reprodu- apparent weight of the particles is 1450 Pa, which is 4.6% lower
cibility was observed to be within 75%. than the predicted value, i.e. 1520 Pa. The results verify the
capability of the model to correctly capture the fluidisation
4.2. Turbulence characterisation using particle image hydrodynamics of SLFB.
velocimetry (PIV) Fig. 6 illustrates the transient snapshot of the foreign particle in
the solid–liquid fluidised bed at the end of the calculations for
The PIV measurements were undertaken using the experimen- superficial liquid velocities around the minimum fluidisation
tal set-up shown in Fig. 4. The RI matched sodium iodide solution velocity of the fluidised particles (Umf ¼ 0.042 m/s). The foreign
Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70 57

Fig. 4. Experimental set-up for PIV measurements.

1800
superficial liquid velocity increases due to the increasing upwards
1700 drag force and the consequent smaller inertia of the foreign
y = 1520
particle. At UL ¼ 0.05 m/s (Fig. 6(c)), that is just above the minimum
1600 fluidisation velocity, the foreign particle continues to fall at the
end of the simulation (t ¼15 s). At higher liquid superficial velo-
1500 cities (UL ¼0.06 and 0.07 m/s, i.e. Fig. 6(b) and (a)), the foreign
1400 particle settles to the bottom very quickly, i.e. 4.54 s and 2.79 s for
∆p (Pa)

UL ¼ 0.06 m/s and 0.07 m/s, respectively.


1300 Fig. 7 shows the evolution of vertical position of the foreign
y = 51750x - 676.4 particle for the superficial liquid velocities around the minimum
1200 fluidisation velocity (Umf ¼ 0.042 m/s). It can be seen that at
1100 superficial liquid velocities of UL ¼0.03 and 0.04 m/s, the foreign
x = 0.042443 particle stops falling shortly after reaching the surface of the solid–
1000 liquid bed (H0 ¼ 0.2 m). At UL ¼0.05 m/s the bed expands and the
foreign particle starts falling through the bed of particles. Due to
900 the significant hampered effect of particle–particle interactions, it
800 takes a long period of time (430 s) for the foreign particle to
0.02 0.06 0.1 0.14 0.18 0.22 settle to the bottom. As UL increases further, the bed expands more
UL (m/s) and leads to a higher bed voidage. Subsequently, the foreign
particle settles to the bottom very quickly, although the upwards
Fig. 5. Pressure drop versus liquid superficial velocity (dP1 ¼ 5 mm, liquid: water). drag force increases as UL increases. It implies that in the cases
shown in Fig. 6, the particle collisional effects are dominant over
particle (dP2 ¼6 mm) is coloured in red, and the fluidised particles the fluid viscous effects. The above results also explain the liquid-
(dP1 ¼5 mm) are coloured in white and displayed with 20% like properties of fluidised beds and reveal that the solid hold-up
transparency. The starting point of the timer is when the foreign or bed voidage plays a key role in the settling process of the
particle was released from the top of the column (height: 1 m). It foreign particle.
can be seen the foreign particle rests at the top of the solid–liquid
fluidised bed at UL ¼0.03 and 0.04 m/s (Fig. 6(e) and (d)), i.e. below 5.2. Classification velocity of the foreign particle
the minimum fluidisation velocity. There is a less than 1 mm
penetration depth due to the inertia of the foreign particle when it The data related to the foreign particle were monitored through
reaches the bed surface. The penetration depth decreases as the the entire settling process. Fig. 8 shows the evolution of transient
58 Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70

Fig. 6. Final snapshot of particle settling in the liquid fluidised bed at superficial liquid velocities around the minimum fluidisation velocity (dP1 ¼ 5 mm, dP2 ¼6 mm, liquid:
water, Umf ¼0.042 m/s): (a) UL ¼0.07 m/s; (b) UL ¼ 0.06 m/s; (c) UL ¼0.05 m/s; (d) UL ¼ 0.04 m/s; and (e) UL ¼ 0.03 m/s. (For interpretation of the references to colour in this
figure, the reader is referred to the web version of this article.)

1 liquid fluidised bed. After being released from the top, the particle
0.9 started falling under the effect of gravity. The velocity of the
UL(m/s): particle continuously increased until it reached the terminal value
0.8 when major forces, i.e., buoyancy force, drag force and gravity,
0.03
0.7 0.04 balance. The fluid velocity surrounding the particle was also
observed to be varying slightly due to the motion of the particle.
xp, z(m)

0.6 0.05
After the particle reached the surface of the liquid fluidised bed,
0.06
0.5 there was a significant drop of particle vertical velocity due to
0.07
collisions with fluidised particles. However, the particle re-reached
0.4
the steady state very quickly due to the collisions with fluidised
Packed bed height
0.3 particles (or greater viscosity of the pseudo-fluid) (Gibilaro et al.,
0.2 2007). Thereafter, the foreign particle was settling through the
fluidised bed at the constant settling velocity, i.e., particle classi-
0.1 fication velocity (V2). The local liquid void fraction was fluctuating
0 around 0.55 in the range of 0.46–0.8, which implies the liquid
20 25 30 35 fluidised bed at this superficial liquid velocity (i.e. UL ¼ 0.1 m/s) has
t (s) a heterogeneous structure (as detailed below).
Several approaches can be applied in the simulations to
Fig. 7. Evolution of vertical position (height) of the foreign particle (dP1 ¼ 5 mm, calculate the classification velocity of the foreign particle, e.g.
dP2 ¼ 6 mm, liquid: water).
averaging the particle vertical velocity over a time period or
dividing distance by the total settling time. In experiments, the
data of the foreign particle, including particle vertical position (xp, particle classification velocity was measured through image pro-
z),particle vertical velocity (vp, z), magnitude of particle angular cessing: the time that is taken for the particle to pass through a
velocity (|ωp|), liquid volume fraction (αL, @p) and liquid vertical region with the height of 0.1–0.2 m. To keep consistent with
velocity (vL, z, @p) at the particle position, and particle Reynolds experimental measurements and also ensure that the simulation
number (Rep). The particle experienced three sequent processes: data were sampled in the steady state, the particle classification
acceleration after being released from the top, steady falling velocity was calculated in the same way as that in experiments.
after reaching the terminal velocity, and settling through the Specifically, the particle vertical position (i.e. xp, z) was plotted
Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70 59

Acceleration Steady Settling through SLFB Table 2


7000 Linear fitting to the data of particle vertical position vs. time (dP1 ¼5 mm, dP2 ¼
Rep, - 6 mm, liquid: water).

UL (m/s) Eq. of fitting line R2 UL (m/s) Eq. of fitting line R2

0.06 y¼  0.097xþ 2.269 0.99 0.15 y¼  0.268x þ6.024 0.9779


0 (0.3)
vL, z, @p, m/s 0.07 y¼  0.147x þ 3.324 0.9988 0.16 y¼  0.276xþ 6.335 0.9790
Transient information on the classifying particle (-)

0.08 y¼  0.191xþ 4.290 0.9966 0.17 y¼  0.264x þ6.335 0.954


0.09 y¼  0.229x þ5.103 0.9996 0.18 y¼  0.304x þ6.898 0.9982
0.10 y¼  0.214xþ 4.779 0.9997 0.19 y¼  0.359xþ 8.158 0.9988
0.11 y¼  0.216x þ 4.824 0.9937 0.20 y¼  0.339xþ 7.730 0.9956
1 (-0.1) 0.12 y¼  0.250x þ5.566 0.9907 0.21 y¼  0.372xþ 8.535 0.9877
εL, @p, -
0.13 y¼  0.252x þ5.642 0.9983 0.22 y¼  0.371xþ 8.490 0.9544
0.14 y¼  0.239x þ5.356 0.9803 0.24 y¼  0.422xþ 9.612 0.9936

50 (0.4)
|ωp|, rad/s 0.45

0.40

0 (0) 0.35

0.30
vp, z, m/s

V2 (m/s)
0.25
1 (-1)
xp, z, m 0.20

0.15

Exponential fit of sim. data


0.10
0 Simulation
0 0.5 1 1.5 2
0.05 Experiment
Time (s)
0.00
Fig. 8. Settling of the foreign particle through the liquid fluidised bed column
0.05 0.10 0.15 0.20 0.25
(dP1 ¼ 5 mm, dP2 ¼ 6 mm, UL ¼0.1 m/s, liquid: water).
UL (m/s)

vL (m/s): Fig. 10. Classification velocity as a function of superficial liquid velocity (dP1 ¼
0.24 5 mm, dP2 ¼ 6 mm, liquid: water).
0.06
0.07
0.08
0.22 0.09
0.1
0.11 The particle classification velocities are plotted in Fig. 10 as a
0.2 0.12
0.13 function of superficial liquid velocity. Consistent with the findings
0.14
0.18 0.15 in the literature (see e.g. Grbavcic et al., 2009; van der Wielen et
0.16
0.17 al., 1996), the plot reveals that the particle classification velocity
0.18
increases with the superficial liquid velocity. An exponential
xp, z(m)

0.16 0.19
0.2
0.21 function was applied to fit the scattered simulation data. It shows
0.14 0.22
0.24 that the particle classification velocity increases faster in the early
Linear (0.06)
Linear (0.07) stage (UL o0.09 m/s). At the incipient fluidisation (for superficial
0.12 Linear (0.08)
Linear (0.09) liquid velocities just above the minimum fluidisation velocity) the
0.1
Linear (0.1)
Linear (0.11) packed bed slightly expands. The space between fluidised particles
Linear (0.12)
Linear (0.13) may be even smaller than the foreign particle size. In such cases,
Linear (0.14)
0.08 Linear (0.15) particle–particle collisions are the dominant mechanism control-
Linear (0.16)
Linear (0.17) ling the settling process of the foreign particle. As the superficial
0.06 Linear (0.18)
Linear (0.19) liquid velocity increases, the fluidised bed expands more, and
21.2 21.7 22.2 22.7 Linear (0.2)
Linear (0.21) additional space is available for the particle to pass through.
t(s) Linear (0.22)
Linear (0.24) Therefore, the particle classification velocity increases sharply. At
this moment, the particle is suffering from fluid drag hindrance
Fig. 9. Particle classification velocity based on the evolution of particle vertical
position (dP1 ¼ 5 mm, dP2 ¼ 6 mm, liquid: water). rather than colliding and displacing fluidised particles, i.e. particle
collisional hindrance. As the superficial liquid velocity increases
further, the particle can easily settle through the space between
against time and the slope of the fitting line (i.e. dxp,z/dt) was fluidised particles and thereby the particle classification velocity
considered as the classification velocity of the foreign particle. increases smoothly.
Fig. 9 illustrates the above methodology for the calculation of The experimental data of particle classification velocities for
particle classification velocity for superficial liquid velocities from superficial liquid velocities of 0.06–0.22 m/s are also shown in
0.06 m/s to 0.24 m/s. Table 2 lists the equation and the R-squared Fig. 10. It can be seen that a good agreement between simulation
value of the fitting lines. It can be seen the majority of R-squared and experiment is obtained as the experimental data scatter
values of the fitting lines are around 0.99 with a minimum value of closely around the exponential fitting line. However, it is worth
0.954. It indicates that the above approach can be applied to noting that both predicted results and experimental data of
calculate the particle classification velocity in SLFB. particle classification velocities vary non-monotonically and the
60 Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70

0.8 0.08
Predicted particle classifiication velocit (m/s)

0.7

0.06
0.6

0.5

u'L (m/s)
0.04
0.4 20%
10%
0.3
DEM-CFD Simulation, d_p1=3mm

K&B (1966)
0.02 Experiment (PIV), d_p1=3mm
0.2 K&L (1969) Simulation, d_p1=5mm
Joshi (1983)
Di Felice (1991)
0.1 van der Wielen (1996)
0
Grbavcic (2009)
0.4 0.5 0.6 0.7 0.8 0.9
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 α L (-)
Experimental particle classifiication velocity(m/s) Fig. 12. Averaged value of fluctuating liquid velocity as a function of bed voidage
(liquid: NaI).
Fig. 11. Predicted particle classification velocity versus experimental measured
data (dP1 ¼5 mm, dP2 ¼6 mm, liquid: water).
0
The fluctuating fluid velocity (uL) of fluid flow was calculated by
!
data are more scattered around the fitting line for high superficial 1 Nc 1 t es
uL' ¼ ∑ ∑ ðuL ðc; tÞ  u L ðcÞÞ ð37Þ
liquid velocities, i.e. the maximum fitting error is 8.5% for Nc c ¼ 1 ðt es  t bs Þ=Δt f t ¼ t bs
UL r0.08 m/s and 20.6% for UL Z 0.09 m/s. This is associated with
the transition of solid–liquid fluidisation regime, i.e. from homo- where Nc is the total cell number of the window. u L ðcÞ is the average
geneous to heterogeneous, which is to be discussed in detail in the fluid velocity in computational cell c over a time period and uL(c, t) is
next section. Specifically, at low superficial velocities, a homo- the fluid velocity in cell c at time t. t is the time at each time step,
geneous liquid fluidised bed was obtained. In such cases, the thus is discrete. tbs is the beginning time to process data and tes is
measured classification velocity was time independent, as the the ending time of data processing. In this study, tbs and tes are set as
distribution of fluidised particles in the bed and the fluid flow 10 s and 30 s, respectively. Δtf is the fluid time step.
were almost the same over time. In other words, the particle The fluctuating fluid velocity is shown in Fig. 12 as a function of
settling history through the region of h ¼0.1–0.2 m of the liquid average bed voidage. The first 10 s of simulation time was set for
fluidised bed remained the same for measurements at different the solid–liquid fluidised bed to reach the steady state. It can be
times. However, at high superficial liquid velocities, the local solid seen that for low superficial liquid velocities, i.e. low bed voidages,
concentration and fluid flow structure in the sampled region (i.e. the values of fluctuating fluid velocity are much smaller compared
0.1–0.2 m) varied significantly with time, resulting in the fluctuat- to the values at high bed voidages. It implies the turbulence in the
ing value of particle classification velocity. Therefore the particle solid–liquid fluid bed is nearly homogeneous and stationary in
0
classification velocities appear more scattered for high superficial dense fluidised beds. Remarkably, a sharp increase of uL is
liquid velocities and the deviation between simulation and experi- observed for both dp1 ¼3 mm and 5 mm when the bed voidage
ment becomes larger. increases from 0.52 to 0.62. It would be reasonable to expect that
Fig. 11 plots the particle classification velocities predicted by the solid–liquid flow structure transits from homogeneous to
CFD–DEM simulations and various empirical correlations against heterogeneous after the bed voidage becomes higher than a
the experimental data. It can be seen that the data predicted by critical value, which is sitting between 0.52 and 0.62. The available
CFD–DEM in this work provide the best agreement with the experimental data of the fluctuating fluid velocity measured by
experimental data with all deviation points less than 10%. The PIV is also included in Fig. 12. It can be seen that a good agreement
correlations proposed by Joshi (1983) and Grbavcic et al. (2009) is obtained between simulation and experiment with a maximum
provide a good agreement with experimental data for dilute deviation around 15%. The deviation may be attributed to the ideal
systems. In dense systems (e.g. at the smallest superficial liquid (theoretical) conditions used in the simulations (e.g., uniform fluid
velocity of UL ¼0.06 m/s) a remarkable deviation is observed (far flow profile applied at the inlet of the column, ideal particle size
beyond 20%). The rest correlations developed in literature include and shape, and ideal fluid properties), which are different to some
those by Kennedy and Bretton (1966), Kunii and Levenspiel (1969), extent from practical or actual cases in experiments. Moreover, the
Di Felice et al. (1991) and van der Wielen et al. (1996) provide very limited operation error and inherent flaws of methodologies for
poor predictions due to different operating conditions (e.g. size data acquisition and processing in experiments might have also
and density of particles) used in the experiments for the devel- contributed to the deviation.
opment of these empirical correlations. To explore the relationship between bed voidage and the solid–
liquid fluidisation regime, the superficial velocity used in simula-
tions was increased gradually from 0.06 m/s with a step of 0.01 m/
s, providing a wide range of bed voidage from 0.46 to 0.8. Fig. 13
5.3. Transition of solid–liquid fluidisation regime shows the flooded contour plots of local liquid volume fraction of
the middle xz cross-section (y¼ 0) for superficial liquid velocities
The fluid velocity (uL) of fluid flow in a small window (xϵ of 0.06–0.11 m/s at t¼20 s. At the lowest superficial liquid velocity
[  0.025, 0.025], y¼ 0, zϵ[0.1, 0.2]) of the middle xz cross-section (i.e. UL ¼ 0.06 m/s, Fig. 13(a)), the particles are fluidised and evenly
(y¼ 0) was averaged over a time period of t ¼10–30 s (in NaI). distributed in the column, thus a typical homogeneous fluidisation
Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70 61

0.4 0.4 0.4 0.4 0.4 0.4

α L(-):
0.95
0.9
0.85
0.2 0.2 0.2 0.2 0.2 0.2 0.8
0.75
0.7
0.65
0.6

0 0 0 0 0 0
Fig. 13. Contour plots of liquid volume fraction (dP1 ¼ 5 mm, dP2 ¼ 6 mm, t¼ 20 s, liquid: water): (a) UL ¼0.06 m/s, αL ¼ 0.46; (b) UL ¼ 0.07 m/s, αL ¼0.48; (c) UL ¼0.08 m/s,
αL ¼0.51; (d) UL ¼0.09 m/s, αL ¼0.54; (e) UL ¼0.10 m/s, αL ¼0.56; and (f) UL ¼0.11 m/s, αL ¼ 0.58.

is obtained. As the superficial liquid velocity increases (i.e. foreign particle (dP2 ¼6 mm) is coloured in red and the fluidised
UL ¼0.07–0.08 m/s, Fig. 13(b) and (c)), some tiny local voids appear particles (dP1 ¼3 mm) are coloured in blue. The entire domain is
in the region close to the wall. As the superficial liquid velocity clipped along the y direction, thus a part of the SLFB is shown in
continuously increases, the voids grow bigger and expand to the Fig. 15 in order to have a close look at the interactions between the
central region. For superficial liquid velocities greater than 0.09 m/ foreign particle and surrounding fluidised particles. At the low bed
s (Fig. 13(e) and (f)), the voids spread over the entire fluidised bed, voidage of αL ¼ 0.45 (Fig. 15(a)) the liquid fluidised bed is in the
leading to a heterogeneous solid distribution in the column and homogeneous fluidisation regime with fluidised particles nearly
hence the heterogeneous fluidisation regime. evenly distributed in the bed. The foreign particle collides with
The contour line of vorticity (component at y direction) of the particles below and pushes the particles aside. A void space is
middle xz cross-section (y¼ 0) is calculated and plotted in Fig. 14. A subsequently generated following the motion of the foreign
small window (zϵ[0.1, 0.2]) of the xz cross-section is zoomed in particle, as labelled with ellipses in Fig. 15(a). The trajectory of
with the flooded contours of fluid flow velocity magnitude and the foreign particle is almost linear, which implies that the
flow vector for superficial velocities of 0.06 m/s and 0.11 m/s, as collision intensity is too small to overcome the vertical inertia of
shown in Fig. 14(A) and (B). For superficial liquid velocities lower the foreign particle and change its moving direction. As the bed
than 0.09 m/s (Fig. 14(a)–(c)), the vorticity magnitude is small and voidage increases further to 0.65 (Fig. 15(b)), the bed becomes
the liquid flows upwards in order (Fig. 14(A)). No large gradients of diluter and the solid–liquid flow turns into the heterogeneous
vorticity are observed. The liquid flow is thus homogeneous. As regime. The particle number density around the foreign particle is
the superficial liquid velocity increases beyond 0.09 m/s, the smaller compared to those in Fig. 15(a). The fluidised particles in
vortices distribute asymmetrically in the domain and more small the vicinity of the foreign particle are observed flowing down-
vortices appear in the flow field (Figs. 14(d)–(f)). The magnitude of wards along with the foreign particle, and the particles in other
vorticity varies over a wide range, which indicates the existence of regions mostly are flowing upwards. As the bed voidage increases
heterogeneity. The typical heterogeneous fluid flow field with to 0.79 (Fig. 15(c)), the fluidised bed is very dilute, and the foreign
large vortices can be observed for UL ¼0.11 m/s (Fig. 14(f) and particle moves downwards much faster and more freely (i.e. less
(B)). The results indicate that the structure of the liquid fluidised collisions). In turn, the motion of fluidised particles is not influ-
bed transits from homogeneous to heterogeneous at UL ¼0.09 m/s, enced by the motion of the foreign particle. However, the trajec-
which corresponds to an average bed voidage of 0.54. tory of the foreign particle alters a lot at αL ¼0.79. For instance, the
transverse motion of the foreign particle becomes more intensely,
5.4. Particle–particle collision e.g. more fluidised particles are observed in the region of yZ0.1.
This is due to the increasing relative velocity between the
5.4.1. Particle–particle interactions and collision events particles, which results in the increase in the collision intensity
Fig. 15 shows snapshots of the motion of the foreign particle in and leads to the change in the transient moving direction of the
the LSFB for average bed voidages of 0.45, 0.65 and 0.79. The dense foreign particle.
62 Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70

0.4 0.4 0.4 0.4 0.4 0.4


-1
Vorticity (s ):

90
80
70
60
50
40
30
20
10
0.2 0.2 0.2 0.2 0.2 0.2 0
-10
-20
A B -30
-40
-50
-60
-70
-80

0 0 0 0 0 0

A B
0.2 0.2

0.18 0.18

|uL| (m/s):

0.45
0.16 0.16 0.4
0.35
0.3
0.25
0.2
0.15
0.14 0.14
0.1
0.05

0.12 0.12

0.1 0.1

Fig. 14. Contour plots of y-component vorticity, fluid velocity magnitude and fluid flow vectors (dP1 ¼ 5 mm, dP2 ¼6 mm, t ¼20 s, liquid: water): (a) UL ¼ 0.06 m/s, αL ¼ 0.46;
(b) UL ¼ 0.07 m/s, αL ¼0.48; (c) UL ¼0.08 m/s, αL ¼0.51; (d) UL ¼ 0.09 m/s, αL ¼ 0.54; (e) UL ¼ 0.10 m/s, αL ¼ 0.56; and (f) UL ¼ 0.11 m/s, αL ¼ 0.58.
Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70 63

22.05 s 22.09 s 22.13s

21.67 s 21.71 s 21.75s

21.86 s 21.88 s 21.90s

Fig. 15. Snapshot of the foreign particle motion through the SLFB (dP1 ¼ 3 mm, dP2 ¼ 6 mm, fluid: water): (a) αL ¼0.45, yZ 0; (b) αL ¼0.65, yZ0; and (c) αL ¼0.79, yZ 0.1.
(For interpretation of the references to colour in this figure, the reader is referred to the web version of this article.)

It can be seen that a particle often collides with multiple particles 10 9


at the same time, especially in dense systems. This poses a huge
8
challenge to experimental measurements of collision events. How- N_a, con
ever, in the simulations, the number of contacting particles and any N_max, con
7
new contact of the foreign particle can be detected and recorded to
1
describe the collision behaviour. Fig. 16 shows the average and the 6

Nmax, con(-)
Na, con(-)

maximum number of particles that are contacting with the foreign


particle in the entire settling process. It is shown that the maximum 5
number of contacting particles reaches 8 at αL ¼0.45 and keeps
4
constant at 2 for αL Z0.65. The average number of contacting 0.1
particles decreases from 3.54 to 0.06 as the bed voidage increases 3
from 0.45 to 0.79. The results imply that the particle–particle
collision happens very rarely in dilute systems. 2
Fig. 17 shows the evolution of the cumulative number of
contacts of the foreign particle and the corresponding instant 0.01 1
0.4 0.5 0.6 0.7 0.8
particle velocity. An increase in the cumulative number of contacts
αL (-)
denotes the occurrence of a collision event between the foreign
particle and fluidised particles. It can be seen that the particle Fig. 16. Average and maximum number of fluidised particles that are contacting
collision immediately incurs the pulsation of particle velocity. The with the foreign particle during the entire settling process (dP1 ¼ 3 mm, dP2 ¼ 6 mm,
z-component of the particle velocity (i.e. vpz) responds instantly to liquid: water).
every collision event. However, the x- and y-components of
particle velocity (i.e. vpx and vpy) may not sense some particle describe the particle collision frequency. Moreover, it is worth
collision events, e.g. vpy in the time periods I (t¼1.004–1.008 s) noting that the velocity of the foreign particle varies or decreases
and IV (t¼ 1.068–1.076 s); vpx in the time period III (t¼1.048– linearly before and after the collision sections. It implies that the
1.056 s). It means that the energy dissipated due to particle collisional effects play the dominant role in the particle settling
collisions is mainly reflected by the dampened particle vertical process in the context of this study (with Stp 465). A good
velocity. In other words, the pulsation of particle vertical velocity knowledge of particle collision events is the key to well under-
can be used as a good indication of particle collision events, as stand the particle settling behaviour.
already employed by Aguilar-Corona et al. (2011) in their experi-
ments. However, it should be noted that the collisions often occur
in a group, i.e. more than a single collision. This causes incon- 5.4.2. Collision frequency
sistency between the pulsation of particle velocity and the number In the simulation, the transient velocity of particles was
of collision events if this approach is used in experiments to monitored during the motion of the foreign particle in the SLFB.
64 Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70

Particle fluctuating kinetic energy, (m2/s2)


0.2
v px

0.01
Particle velocity (m/s)

0.2 (-0.5)
v py

0.2 (-0.5)
v pz

0.001
Collisioon number(-)

510 (-0.5) 0.15 0.25 0.35 0.45 0.55


IV
III αS, (-)
I II
Nc V
Fig. 18. Fluctuating kinetic energy of the particle phase as a function of solid
fraction αS (dP1 ¼ 5 mm, dP2 ¼ 6 mm, liquid: water).

430
1 1.02 1.04 1.06 1.08 1.1

t (s) 300

Fig. 17. Particle collisions and consequent variation of particle velocity (dP1 ¼ 5 mm,
dP2 ¼ 5 mm, UL ¼ 0.20 m/s, liquid: NaI). 250

With the data of particle instantaneous velocities and particle size, 200
the dimensionless collision frequency was calculated as (Aguilar-
Corona et al., 2011)
f*col (-)

KTGF (Simonin, 1991)


150
n f dP DEM-CFD simulation
f coll ¼ rcoll
ffiffiffiffiffiffiffiffiffiffiffi
D E ð38Þ
q2p 100
D E
where dP is the particle diameter and q2p is the “small scale”
fluctuating kinetic energy of the solid phase in the fluidised bed. 50
The fluctuating kinetic energy can be calculated as
D E 3D E 0
q2p ¼ up'2 ð39Þ 0 0.1 0.2 0.3 0.4 0.5 0.6
2
D E αS (-)
where up'2 is the particle velocity variance.
Fig. 18 shows the variation of particle fluctuating kinetic energy Fig. 19. Dimensionless collision frequency as a function of the solid fraction (dP1 ¼
(estimated using Eq. (39)) over a wide range of solid fraction (i.e., 5 mm, dP2 ¼6 mm, liquid: water).
0.2–0.6). It can be seen that the particle fluctuating kinetic energy
decreases with solid fraction, which implies the effects of turbu-
lence on the particle motion become significant and the effects of
d P2, mm dP1, mm d P2 /dP1, - Liquid
particle collision events are diminishing due to the decreasing
6 3 2.0 water
solid fraction. 6in4W
6 4 1.5 water
6in5W
6 5 1.2 water
The dimensionless particle collision frequency predicted by
Collision frequency, fcoll (s -1)

5in5W
5 5 1.0 water
CFD–DEM simulations for solid fractions ranging between 0.2 and 3000 4in5W
4 5 0.8 water
3
3in5W 5 0.6 water
0.8 is shown in Fig. 19. The figure shows that the collision 6
6in3N 3 2.0 NaI
frequency increases monotonically with the solid fraction. The 6
6in5N 5 1.2 NaI
dimensionless collision frequency was also calculated by the
classic KTGF (Simonin, 1991), as shown in Fig. 19. The maximum
solid fraction, i.e. αSm used in Eq. (7) for the calculation, is 0.575, 300
which was measured in experiments. It can be seen that the
dimensionless collision frequency obtained by CFD–DEM excel-
lently fit the data line predicted by KTGF.
The variation of collision frequency is shown in Fig. 20 as the
foreign particle size (dP2) and fluidised particle size (dP1) are
altered and the particle size ratio (dP2/dP1) varies between 30
0.1 0.2 0.3 0.4 0.5 0.6
0.6 and 2.0. The data on the effect of fluid medium, namely water
and NaI, has also been included. It can be seen that as the particle αS (-)
size ratio increases, the collision frequency increases for a fixed Fig. 20. Collision frequency as a function of the solid fraction, particle sizes and
solid fraction. This can be easily understood as the foreign particle fluid medium.
Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70 65

50 50

40 40

|Fc |/mg (-)


|Fc |/mg (-)
30 30

20 20

10 10

0 0
21 21.05 21.1 21.15 21.2 21 21.05 21.1 21.15 21.2
t (s) t (s)

50 50

40 40
|Fc |/mg (-)

|Fc |/mg (-)


30 30

20 20

10 10

0 0
21 21.05 21.1 21.15 21.2 21 21.05 21.1 21.15 21.2
t (s) t (s)

50 50

40 40
|Fc |/mg (-)
|Fc |/mg (-)

30 30

20 20

10 10

0 0
21 21.05 21.1 21.15 21.2 21 21.05 21.1 21.15 21.2
t (s) t (s)
Fig. 21. Evolution of collision force acting on the foreign particle (dP1 ¼5 mm, dP2 ¼6 mm, liquid: water): (a) αL ¼ 0.46; (b) αL ¼0.51; (c) αL ¼ 0.60; (d) αL ¼ 0.67; (e) αL ¼ 0.71;
and (f) αL ¼ 0.79.

needs to displace N fluidised particles when it settles through the fluidisation regime is obtained (Section 5.3). The fluidised particles
liquid fluidised bed (Gidaspow, 1994). Since non-elastic collisions are evenly distributed in the domain. Therefore, the chance for the
occur between the particles, N becomes greater as dP2/dP1 increases, foreign particle to come across fluidised particles in a space of
indicating the occurrence of more collisions between the foreign certain volume should be equal, giving rise to a constant collision
particle and fluidised particles. The collision frequency is observed frequency. It can also be seen that the effect of fluid medium on
decreasing steeply at high solid phase fractions (αS 40.4), e.g. the particle collision frequency is negligible with the trend lines
fcoll ¼1649/s at αS ¼0.54 and fcoll ¼101/s at αS ¼0.42. At low solid (fcoll versus αS) overlapping for a fixed particle size ratio (e.g., dP2/
fractions the collision frequency decreases gently as solid fraction dP1 ¼2.0 or dP2/dP1 ¼1.2). This result may be explained by the high
decreases. It can be attributed to the two counteracting effects at low particle Stokes number (Stp 4 65) investigated in this study. Under
solid phase fractions. On the one hand, as the solid fraction the high Stokes number the viscous effects of interstitial fluid on
decreases, the average distance between neighbouring particles the settling of the foreign particle are negligible and the collisional
increases, which leads to a drop in the particle collision frequency. effects can be considered dominant. In such an inertia flow regime,
On the other hand, as the liquid velocity increases, the bed hydro- the motion of the foreign particle is not influenced by the fluid
dynamics becomes more turbulent and the rate at which the medium when the bed voidage of the liquid fluidised bed keeps
particles circulated in the bed increases, which in turn increases constant. Therefore, the change in the fluid medium does not lead
the collision frequency (Pozo et al., 1993). However, the former effect to any significant difference in the particle collisional behaviour.
is more dominant than the latter effect and as a result, the particle
collision frequency decreases as the solid phase fraction decreases,
but not as steeply as those at high solid phase fractions. 5.4.3. Collision force acting on the foreign particle
When the size of fluidised particles is the same, the particle 5.4.3.1. Significance of collision force. The collision force acting on
collision frequency is independent of the foreign particle size, the foreign particle normalised by the weight force of foreign
remaining almost constant at high solid phase fractions particle (mg) was monitored in the simulation at a frequency of
(αS 4 0.54). As discussed above, for αS 40.54 the homogeneous 1000 Hz. Fig. 21 shows the evolution of the normalised collision
66 Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70

0.5
10
0.45 Drag force
9
Pressure gradient force
0.4 Exponential fitting
8
Prediction data
0.35 7
|F|/mg (-)

0.3 6

|Fc|/|Fd| (-)
0.25 5

4
0.2
3
0.15
2
0.1
1
0.05
0
0.4 0.5 0.6 0.7 0.8
0
0.4 0.5 0.6 0.7 0.8 α L(-)
α L(-) Fig. 23. Normalised collision force by fluid drag force as a function of bed voidage
(dP1 ¼5 mm, dP2 ¼6 mm, liquid: water).
Fig. 22. Magnitude of fluid drag force and pressure gradient force as a function of
bed voidage (dP1 ¼ 5 mm, dP2 ¼ 6 mm, liquid: water).

of bed voidage. It can be seen the ratio of collision force to fluid


force for various bed voidages ranging between 0.46 and 0.79. drag force decreases exponentially with the bed voidage. The
Unlike other forces (e.g. drag and gravitational forces), the collision force is the dominant force with the maximum magni-
evolutional line of collision force is not continuous with zero tude of 9 times greater than that of the drag force for the bed
values lasting over a period of time. This is because the foreign voidages (0.45 o αL o 0.8) investigated in this study. These esti-
particle does not collide with any particles in some time periods, mates clearly provide insights about the significance of collision
especially in dilute systems. The non-zero valued data dots denote force which has been overlooked to date in the literature. It is also
the contacting status of the foreign particle with fluidised noted that the force ratio decreases very gently at high bed
particles. It can be seen the non-zero valued data dots are voidages (αL 40.6) due to the two counteracting effect as men-
denser at lower bed voidages, corresponding to more frequent tioned above, i.e. increasing average distance between neighbour-
particle collisional events. However, the magnitude of the ing particles and increasing particle circulation rate. As the
normalised collision force increases with the bed voidage. It is increased superficial liquid velocity also leads to the increase of
worth noting that the magnitude of the normalised collision force fluid drag force, the evolution line of the force ratio (|Fc|/|Fd|)
is much greater than the gravitational force with the ratio |Fc|/mg appears even more flatly than that of the particle collision
ranging between 10 and 50. The results suggest that the collision frequency, as shown in Fig. 20.
force is significant in SLFBs and must be accounted for while
conducting analysis based on the force balance. Also, a simple
correlation should be available to estimate the collision force as a
function of bed voidage. 5.4.3.2. Relationship between collision force and bed voidage. The
Fig. 22 shows the magnitude of fluid drag force and pressure collision force acting on the foreign particle is determined by two
gradient force as a function of bed voidage (0.46 r αL r0.8). The parts: how frequent the particle collides with fluidised particles
magnitude of the forces was calculated as the average value of the (i.e. collision frequency) and how strong the collisions are
total forces acting on the foreign particle since it falls into the SLFB. (i.e. collision intensity). As discussed earlier, the particle collision
The drag force increases at a decreasing rate with the bed voidage frequency is related to local solid concentration or bed voidage.
due to the increased upwards fluid velocity. The magnitude of drag The collision intensity is mainly determined by the relative
force increases from 0.1 mg to 0.5 mg as the bed voidage increases velocity between the foreign particle and fluid (Pozo et al., 1993;
from 0.46 to 0.8. The magnitude of pressure gradient force does van der Wielen et al., 1996) and particle material properties. The
not change significantly, decreasing from 0.11 mg to 0.07 mg as the solid–liquid relative velocity is the difference between particle
bed voidage increases from 0.46 to 0.8. In the homogeneous classification velocity and the interstitial fluid velocity; both of
fluidisation regime (i.e. αL o0.54), the pressure gradient force is them are a function of bed voidage. Considering the particle
merely due to the static pressure gradient (equal to the buoyancy properties are known and identical and the collisional effects are
force of the particle, i.e. ρf/ρp ¼998.2/7800 ¼0.128). The result dominant for Stp 4 65 in this study, the collision force is thus
implies that as the bed voidage increases, the solid distribution described as a sole function of the average bed voidage. Fig. 24
and the fluid flow in the bed are no longer homogeneous. The shows the variation of the normalised collision force (by the
turbulence caused by the increased superficial liquid velocity weight force of foreign particle) as a function of bed voidage for
becomes more significant and leads to the non-uniformly distrib- a 6 mm dense foreign particle settling in the fluidised bed of 5 mm
uted local pressure gradients, which renders the decreasing glass beads in water. It can be seen that the normalised collision
pressure gradient force. force continuously decreases with an increase in bed voidage. As
As the collision force acting on the foreign particle is not conti- discussed above, this is because as the bed expands, the collision
nuous, the magnitude of particle collision force was calculated by frequency decreases, which is more dominant over the increased
averaging the total collision force accumulated during the entire collision intensity as the bed voidage increases. Consequently,
process of particle settling through the SLFB. The comparison of a drop in the net collision force is obtained. The quadratic
collision force and fluid drag force is shown in Fig. 23 as a function fitting method was applied to fit the scattered data to relate the
Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70 67

1 1.0
Present study
Eq. (40) 0.9 van der Wielen (1996)
0.9 Prediction data Grbavcic (2009)
0.8

0.7
0.8
0.6

|Fc |/mg (-)


|Fc |/mg (-)

0.7 0.5

0.4
0.6
0.3

0.2
0.5
0.1

0.4 0.0
0.4 0.5 0.6 0.7 0.8 0.9 0.40 0.50 0.60 0.70 0.80 0.90
αL (-) αL(-)
Fig. 24. Relationship between the normalised collision force and bed voidage Fig. 25. Comparison of the normalised collision force with different correlations
(dP1 ¼ 5 mm, dP2 ¼ 6 mm, liquid: water). from literature (dP1 ¼ 5 mm, dP2 ¼ 6 mm, liquid: water).

normalised collision force to the bed voidage as


provide unreasonable outputs. Moreover, the collision coefficient
jFc j=mp g ¼ 1:08 þ 0:26αL 1:34α2L ðSt p 4 65; dp2 =dp1 r 2Þ ð40Þ
(i.e. k in Eqs. (14) and (15)) depends on the particle size ratio (dP2/
For two extreme cases: dP1) and the properties of the fluidised bed, which may also
contribute to the deviation and inconsistency. The correlation by
(i). Empty bed, where αL ¼1.0, |Fc|/mpg¼ 0, |Fc| ¼0; van der Wielen et al. (1996) was developed based on the force
(ii). Packed bed, where, αL ¼0.425, |Fc|/mpg ¼0.948, |Fc| ¼0.948mpg. balance on the foreign particle, which seems more plausible.
In this case, |Fc| balances the gravity together with fluid forces. However, in a real liquid fluidised bed, especially in the hetero-
As the particles are nearly stationary, the supporting forces geneous regime, the forces acting on the foreign particle may be
from the fluidised particles around the foreign particle are more than those considered in the correlation. Moreover, the
considered as the total collision force acting on the foreign heterogeneous structure of fluid flow and the non-uniform dis-
particle. tribution of solid particles may result in the inapplicability of the
formulations used for the calculation of the forces. For example,
It was thought desirable to compare the estimated normalised the pressure gradient force is equal to the buoyancy force in the
collision force in the present study with those predicted using homogeneous fluidisation regime, but is not the case in the
various correlations from literature. Fig. 25 depicts the comparison heterogeneous fluidisation regime, as shown in Fig. 22.
which clearly shows that the magnitude of the normalised colli- Fig. 26 shows the comparison between the prediction results of
sion force predicted in present study (by Eq. (40)) matches that |Fc|/mg by the correlation (Eq. (40)) and those obtained in the CFD–
predicted by the correlation of van der Wielen et al. (1996). DEM simulations. It can be seen that the majority of the deviation
However, the predictions of Grbavcic et al. (2009) are much less fall into the range of 10% and all the data points are within 20%,
as compared to others. Moreover, the present study clearly shows which verifies the validity of the correlation in the prediction of
the normalised collision force decreases monotonically as the bed collision force as a sole function of the average bed voidage for
voidage increases. Considering the strong decline in the number of these specific cases. The randomness of the sampled data and the
collisions as the bed voidage increases (as shown in Fig. 19), the hydrodynamical stability state of the fluidised bed are considered
effect of the increase in collision intensity would be drown out. the major factors that are responsible for the deviations.
The overall result would be expected to be a decrease in the Eq. (40) correlates the magnitude of the collision force only
normalised collision force. However, the predictions using the with the average bed voidage and should be used with due care in
correlations by van der Wielen et al. (1996) and Grbavcic et al. other situations. As discussed above, in all cases investigated in
(2009) appear non-monotonic and somehow reveal a peak value this study, the particle collisional effects on the settling of the
of the normalised collision force. The inconsistency between the foreign particle are dominant with the high particle Stokes
predictions by correlations and that in the present study by CFD– numbers (Stp 465). The interstitial fluid viscous effects are
DEM simulations may be attributed to the assumptions and deemed negligible. However, in cases where the fluid viscous
operating conditions used in the development of these correla- effects are dominant with a small Stokes number (Stp o5) (Joseph
tions. Grbavcic et al. (2009) developed their correlation based on et al., 2001; Rao et al., 2011), the fluid drag force is playing a
the assumption that the collision force is proportional to the total significant role in determining the particle collision behaviour. As
drag force acting on the N fluidised particles that have been such, the applicability of this correlation needs to be verified in
displaced by the foreign particle during the settling process. This these cases. Moreover, the ratio of the foreign particle size and the
is valid only in the homogeneous fluidisation regime where the fluidised particle size, i.e. dP2/dP1 is between 0.6 and 2. For dP2/
foreign particle must squeeze away a certain number of fluidised dP1 42.0, the correlation may not be valid. In the work by Grbavcic
particles to fall down. In the heterogeneous regime, however, the et al. (2009), the particle collision force tended to be zero for
liquid fluidised bed is dilute and the particle collision number particle size ratios (dP2/dP1) greater than 10. In such cases the
significantly decreases (see Fig. 21). The foreign particle settles correlation developed in this study would be invalid. Furthermore,
mainly through the space between the fluidised particles. In such Eq. (40) was developed based on the data of bed voidages up to
cases, the assumption does no longer hold and the correlation may 0.8. However, as indicated earlier, the correlation by itself is valid
68 Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70

1 (αS 40.4); at low solid fractions the particle collision frequency


decreased gently as the solid fraction decreased. When the size of
0.9 10%
fluidised particles was the same, the particle collision frequency
0.8 was independent of the foreign particle size at high solid phase
fractions (αS 40.54). The effect of fluid medium on the particle
|Fc|/mg in Simulation (-)

0.7 collision frequency was negligible for high Stokes numbers


0.6
(Stp 4 65).
The magnitude of collision force acting on the foreign particle
0.5
20% was much greater than the gravitational force with the ratio of |Fc|/
d , mm d , mm Liquid mg ranging between 10 and 50. The collision force decreased as
0.4 6
6mm 5 water the bed voidage increased. The collision force was the dominant
5
5mm 5 water
0.3 4
4mm 5 water force when the bed voidage was less than 0.8, with the maximum
3
3mm 5 water magnitude being 9 times greater than that of the drag force. The
0.2 6
Series5 4 water collision force was described as a sole function of the average bed
6
6in3mm 3 water
0.1 6 5
6in5mminNaI NaI voidage when the particle collisional effects were dominant. The
6
6in3_NaI 3 NaI correlation proposed to relate the collision force solely to the bed
0 voidage provided a maximum error of 20% in the prediction of
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 collision force for Stp 465 and dP2/dP1 r2. Limitations of the
|Fc |/mg by Eq. (40) (-) correlation have also been addressed.
Fig. 26. Validation of Eq. (40) in the prediction of collision force based on the bed
voidage.
Nomenclature

for cases with the bed voidages up to the ideally maximum bed CD drag coefficient, dimensionless
voidage, i.e., αL ¼1.0 in empty beds. dP1 diameter of particles comprising fluidised bed, m
dP2 diameter of the foreign particle, m
d12 distance between the centre of the particles, m
6. Conclusions dpm equivalent particle grid size, m
D column diameter, m
The settling behaviour of a dense foreign particle in a mono- ec normal restitution coefficient, dimensionless
dispersed solid–liquid fluid bed (SLFB) has been investigated by a f Ruzicka (2006) transition function, dimensionless
fully coupled CFD–DEM model. Specifically, the particle motion fc collision contact forces, N
was solved and tracked by DEM and the fluid flow was solved by ff total fluid forces, N
CFD; the fluid–particle interactions were solved by a full two-way fd fluid drag force, N
coupling algorithm. The fluidisation hydrodynamics of SLFB, set- fS–L forces acting on fluid by the solid particles, N
tling behaviour of the foreign particle, fluid–particle interactions fcoll collision frequency, s  1
and particle–particle collision behaviour have been investigated. fncoll dimensionless collisional frequency
Experiments including particle classification velocity measure- FB buoyancy force, N
ments and fluid turbulence characterisation by PIV were carried FD drag force, N
out to verify the validity of prediction results. FG gravitational force, N
The CFD–DEM model established in this study was capable to g gravitational acceleration, m/s2
correctly capture the solid–liquid fluidisation hydrodynamics. g0 function of solid hold-up, dimensionless
Compared to those predicted by various empirical correlations, I moment of inertia, kg/m2
I unit tensor, dimensionless
the particle classification velocity predicted by CFD–DEM provided
the best agreement with the experimental data with the max- k collision coefficient, dimensionless
imum deviation of less than 10%. When the bed voidage was below n1 number of particles of phase 1 per unit volume, dimen-
0.54, the solid–liquid flow exhibited typical homogeneous flow sionless
characteristics. The particles were evenly distributed in the bed n2 number of particles of phase 2 per unit volume, dimen-
and the fluid flew upwards in order with small vorticity gradients. sionless
The solid–liquid fluidisation transited to the heterogeneous regime N number of displaced particles, dimensionless
for bed voidages (αL) over 0.54 with local voids and large vorticity N12 collision frequency of binary mixture, Hz
gradients spreading over the entire bed. The heterogeneous N
D pmE number of particle grids, dimensionless
regime of solid–liquid fluidisation regime caused the large fluctua- q2p “small scale” fluctuating kinetic energy of the solid
tion of particle classification velocity. phase, m2/s2
The particle collision immediately incurred the pulsation of rp particle radius, m
particle velocity. The vertical component (z-direction) of particle Re liquid Reynolds number, dimensionless
velocity was very sensitive to every collision event, whist the ReP particle Reynolds number, dimensionless
horizontal components (x- and y-directions) of the particle velo- Tc torque due to the tangential component of contact force,
city did not respond to some particle collision events. The colli- Nm
sions often occurred in a group, i.e. more than a single collision, in Tr torque due to the rolling resistance, N m
dense systems. The collision frequency increased monotonically ur relative particle velocity during a collision, m/s
with the solid fraction. The dimensionless collision frequency u’ bulk turbulence velocity, m/s
D E
obtained by CFD–DEM excellently fit the data line predicted u2p mean square velocity of the particles, m/s
by KTGF. As the particle size ratio (dP2/dP1) increased, the collision D E
frequency increased for a fixed solid fraction. The particle colli- up'2 particle velocity variance, m/s
sion frequency increased steeply at high solid phase fractions Vp volume of single particle, m3
Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70 69

Vpm volume of particle grid, m3 Chen, C.X., Fan, L.S., 2004. Discrete simulation of gas–liquid bubble columns and
V2 particle classification velocity, m/s gas–liquid–solid fluidized beds. AIChE J. 50, 288–301.
Clift, R., Seville, J.P.K., Moore, S.C., Chavarie, C., 1987. Comments on buoyancy in
Vsum total particle grid volume, m3 fluidized beds. Chem. Eng. Sci. 42, 191–194.
uL liquid velocity vector, m/s Di Felice, R., Foscolo, P.U., Gibilaro, L.G., Rapagna, S., 1991. The interaction of
up mean velocity of the particles, m/s particles with a fluid–particle pseudofluid. Chem. Eng. Sci. 46, 1873–1877.
Di Renzo, A., Di Maio, F.P., 2007. Homogeneous and bubbling fluidization regimes in
UL superficial liquid velocity, m/s
DEM–CFD simulations: hydrodynamic stability of gas and liquid fluidized beds.
Umf minimum fluidisation velocity, m/s Chem. Eng. Sci. 62, 116–130.
Ur relative velocity between fluid and particle, m/s Ding, J., Gidaspow, D., 1990. A bubbling fluidization model using kinetic theory of
US2 settling velocity of dense particle, m/s granular flow. AIChE J. 36, 523–538.
Doroodchi, E., Evans, G.M., Schwarz, M.P., Lane, G.L., Shah, N., Nguyen, A., 2008.
US1 terminal velocity of particle, m/s Influence of turbulence intensity on particle drag coefficients. Chem. Eng. J. 135,
v particle velocity vector, m/s 129–134.
ω particle angular velocity, rad/s Doroodchi, E., Peng, Z.B., Sathe, M.J., Abbasi-Shavazi, E., Evans, G.M., 2012.
Fluidisation and packed bed behaviour in capillary tubes. Powder Technol.
x particle position, m
223, 131–136.
Drew, D.A., Passman, S.L., 1999. Theory of Multicomponent Fluids. Springer-Verlag,
Greek letters New York.
Epstein, N., 1984. Comments on a unified model for particulate expansion of
fluidized beds and flow in porous media. Chem. Eng. Sci. 39, 1533–1534.
α fractional phase hold-up, dimensionless Fan, L.S., Han, L.S., Brodkey, R.S., 1987. Comments on the buoyancy force on a
αS0 solid hold-up in the fixed bed, dimensionless particle in a fluidized bed. Chem. Eng. Sci. 42, 1269–1271.
Fan, L.S., Zhu, C., 1998. Principles of Gas–Solid Flows. Cambridge University Press,
δ particle fractional volume in a cell, dimensionless Cambridge, UK.
β momentum exchanging coefficient, dimensionless Foscolo, P.U., Gibilaro, L.G., 1984. Fully predictive criterion for the transition
γ scaling factor, dimensionless between particulate and aggregate fluidization. Chem. Eng. Sci. 39, 1667–1675.
Δts solid time step, s
Foscolo, P.U., Gibilaro, L.G., Waldram, S.P., 1983. Unified model for particulate
expansion of fluidized beds and flow in fixed porous media. Chem. Eng. Sci. 38,
λ liquid bulk viscosity, kg/(m s) 1251–1259.
m viscosity of fluid, kg/(m s) Ghatage, S.V., Sathe, M.J., Doroodchi, E., Joshi, J.B., Evans, G.M., 2013. Effect of
ρM mixture density, kg/m3 turbulence on particle and bubble slip velocity. Chem. Eng. Sci. 100, 120–136.
Gibilaro, L.G., Gallucci, K., Di Felice, R., Pagliai, P., 2007. On the apparent viscosity of
ρeff effective or relevant density, kg/m3 a fluidized bed. Chem. Eng. Sci. 62, 294–300.
τf viscous stress tensor, dimensionless Gidaspow, D., 1994. Multiphase Flow and Fluidization: Continuum and Kinetic
χ interaction coefficient, dimensionless Theory Descriptions. Academic Press, Boston.
Gjaltema, A., van Loosdrecht, M.C.M., Heijnen, J.J., 1997. Abrasion of suspended
Θ granular temperature, m2/s2 biofilm pellets in airlift reactors: effects of particle size. Biotechnol. Bioeng. 55,
206–215.
Subscripts Glowinski, R., Pan, T.W., Hesla, T.I., Joseph, D.D., 1999. A distributed Lagrange
multiplier/fictitious domain method for particulate flows. Int. J. Multiphase
Flow 25, 755–794.
1 fluidised particles Grbavcic, Z.B., Arsenijevic, Z.L., Garic-Grulovic, R.V., 2009. Prediction of single
2 dense foreign particle particle settling velocities through liquid fluidized beds. Powder Technol. 190,
283–291.
f fluid Grbavcic, Z.B., Vukovic, D.V., 1991. Single-particle settling velocity through liquid
L liquid fluidised beds. Powder Technol. 66, 293–295.
i index Guo, Y., Wu, C.Y., Thornton, C., 2013. Modeling gas–particle two-phase flows with
complex and moving boundaries using DEM–CFD with an immersed boundary
P particle method. AIChE J 59, 1075–1087.
r relative Hu, H.H., 1996. Direct simulation of flows of solid–liquid mixtures. Int. J. Multiphase
S solid Flow 22, 335–352.
Hu, H.H., Fortes, A., Joseph, D.D., 1993. Experiments and direct simulations of fluid
x, y, z direction component
particle motions. Int. Video J. Eng. Res. 2, 17–28.
1 infinite medium Hu, H.H., Joseph, D.D., 1992. Direct simulation of fluid particle motions. Theoret.
Comput. Fluid Dyn. 3, 285–306.
Hu, H.H., Joseph, D.D., Crochet, M.J., 1992. Direct simulation of fluid particle
motions. Theor. Comput. Fluid Dyn. 3, 285–306.
Hu, H.H., Patankar, N.A., Zhu, M.Y., 2001. Direct numerical simulations of fluid–solid
Acknowledgements systems using the arbitrary Lagrangian–Eulerian technique. J. Comp. Phys. 169,
427–462.
Joseph, G.G., Zenit, R., Hunt, M.L., Rosenwinkel, A.M., 2001. Particle–wall collisions
Thanks to Richard Dear for his assistance with the usage of high
in a viscous fluid. J. Fluid Mech. 433, 329–346.
performance cluster (HPC) facilities at the University of Newcastle. Joshi, J.B., 1983. Solid–liquid fluidised beds: some design aspects. Chem. Eng. Res.
The authors wish to acknowledge the financial support of the Des. 61, 143–161.
University of Newcastle and the Australian Research Council for Joshi, J.B., Dinkar, M., Deshpande, N.S., Phanikumar, D.V., 2001. Hydrodynamic
stability of multiphase reactors. Adv. Chem. Eng. 26, 1–130.
the work presented in this paper. Kafui, K.D., Thornton, C., Adams, M.J., 2002. Discrete particle-continuum fluid
modelling of gas–solid fluidised beds. Chem. Eng. Sci. 57, 2395–2410.
Kennedy, S.C., Bretton, R.H., 1966. Axial dispersion of spheres fluidized with liquids.
References AIChE J. 12, 24–30.
Koch, D.L., 1990. Kinetic theory for a monodisperse gas–solid suspension. Phys.
Aguilar-Corona, A., Zenit, R., Masbernat, O., 2011. Collisions in a liquid fluidized bed. Fluids A 2, 1711–1723.
Int. J. Multiphase Flow 37, 695–705. Kuang, S.B., Chu, K.W., Yu, A.B., Zou, Z.S., Feng, Y.Q., 2008. Computational
Ai, J., Chen, J.F., Rotter, J.M., Ooi, J.Y., 2011. Assessment of rolling resistance models investigation of horizontal slug flow in pneumatic conveying. Ind. Eng. Chem.
in discrete element simulations. Powder Technol. 206, 269–282. Res. 47, 470–480.
Anderson, T.B., Jackson, R., 1967. A fluid mechanical description of fluidized beds – Kunii, D., Levenspiel, O., 1969. Fluidization Engineering. Wiley, New York.
equations of motion. Ind. Eng. Chem. Fund. 6, 527–539. Li, Y., Zhang, J.P., Fan, L.S., 1999. Numerical simulation of gas–liquid–solid fluidiza-
Andrews, M.J., O'Rourke, P.J., 1996. The multiphase particle-in-cell (MP-PIC) method tion systems using a combined CFD–VOF–DPM method: bubble wake behavior.
for dense particulate flows. Int. J. Multiphase Flow 22, 379–402. Chem. Eng. Sci. 54, 5101–5107.
Brady, J.F., Bossis, G., 1988. Stokesian dynamics. Annu. Rev. Fluid Mech. 20, 111–157. Li, Y., Zhang, J.P., Fan, L.S., 2000. Discrete-phase simulation of single bubble rise
Brucato, A., Grisafi, F., Montante, G., 1998. Particle drag coefficients in turbulent behavior at elevated pressures in a bubble column. Chem. Eng. Sci. 55,
fluids. Chem. Eng. Sci. 53, 3295–3314. 4597–4609.
Buffière, P., Moletta, R., 2000. Collision frequency and collisional particle pressure in Lu, H.L., Gidaspow, D., 2003. Hydrodynamics of binary fuidization in a riser: CFD
three-phase fluidized beds. Chem. Eng. Sci. 55, 5555–5563. simulation using two granular temperatures. Chem. Eng. Sci. 2003, 3777–3792.
70 Z. Peng et al. / Chemical Engineering Science 116 (2014) 49–70

Malone, K.F., 2007. Simulation of Particle-scale Motion and Heat transfer in Liquid- Seibert, K.D., Burns, M.A., 1998. Simulation of structural phenomena in mixed-
fluidised Beds (Ph.D. thesis). University of Leeds, UK. particle fluidized beds. AIChE J. 44, 528–537.
Pan, T.W., Joseph, D.D., Bai, R., Glowinski, R., Sarin, V., 2002. Fluidization of 1204 Simonin, O., 1991. Prediction of the dispersed phase turbulence in particle-laden
spheres: simulation and experiments. J. Fluid Mech. 451, 169–191. jets. In: Proceedings of the 4th International Symposium on Gas–Solid Flows.
Peng, Z.B., Doroodchi, E., Alghamdi, Y., Moghtaderi, B., 2013. Mixing and segregation ASME FED, pp. 121–197.
of solid mixtures in bubbling fluidized beds under conditions pertinent to the Tsuji, Y., Kawaguchi, T., Tanaka, T., 1993. Discrete particle simulation of two-
fuel reactor of a chemical looping system. Powder Technol. 235, 823–837. dimensional fluidized bed. Powder Technol. 77, 79–87.
Peng, Z.B., Doroodchi, E., Luo, C.M., Moghtaderi, B., 2014. Influence of void fraction van der Wielen, L.A.M., van Dam, M.H.H., Luyben, K.C.A.M., 1996. On the relative
calculation on fidelity of CFD-DEM simulation of gas-solid bubbling fluidized motion of a particle in a swarm of different particles. Chem. Eng. Sci. 51,
beds. AIChE J. 60, 2000–2018. 995–1008.
Peskin, C.S., 1972. Flow patterns around heart valves: a numerical method. van Sint Annaland, M., Deen, N.G., Kuipers, J.A.M., 2005. Numerical simulation of
J. Comput. Phys. 10, 252–271.
gas bubbles behaviour using a three-dimensional volume of fluid method.
Pozo, M.D., Briens, C.L., Wild, G., 1993. Particle–particle collisions in liquid–solid
Chem. Eng. Sci. 60, 2999–3011.
and gas–liquid–solid fluidized beds. Chem. Eng. Sci. 48, 3313–3319.
Wu, C.L., Berrouk, A.S., Nandakumar, K., 2009. Three-dimensional discrete particle
Rao, A., Curtis, J., Hancock, B., Wassgren, C., 2011. Classifying the fluidization and
model for gas–solid fluidized beds on unstructured mesh. Chem. Eng. J. 152,
segregation behavior of binary mixtures using particle size and density ratios.
AIChE J. 57, 1446–1458. 514–529.
Reddy, R.K., Jin, S., Joshi, J.B., Nandakumar, K., Minev, P.D., 2010. Direct numerical Wu, C.L., Zhan, J.M., Li, Y.S., Lam, K.S., 2006. Dense particulate flowmodel on
simulation of free falling sphere in creeping flow. Int. J. Comp. Fluid Dyn. 24, unstructured mesh. Chem. Eng. Sci. 61, 5726–5741.
109–120. Xu, B.H., Yu, A.B., 1997. Numerical simulation of the gas–solid flow in a fluidized
Reddy, R.K., Joshi, J.B., 2008. CFD simulation of pressure drop and drag coefficient in bed by combining discrete particle method with computational fluid dynamics.
fixed and expanded beds. Chem. Eng. Res. Des. 86, 444–456. Chem. Eng. Sci. 52, 2785–2809.
Reddy, R.K., Joshi, J.B., 2009. CFD modelling of solid–liquid fluidized beds of mono Ye, M., van der Hoef, M.A., Kuipers, J.A.M., 2004. A numerical study of fluidization
and binary particle mixtures. Chem. Eng. Sci. 64, 3641–3658. behavior of Geldart A particles using a discrete particle model. Powder Technol.
Reddy, R.K., Sathe, M.J., Joshi, J.B., Nandakumar, K., Evans, G.M., 2013. Recent 139, 129–139.
developements in experimental (PIV) and numerical (DNS) investigation of Zenit, R., Hunt, M.L., Brennen, C.E., 1997. Collisional particle pressure measurements
solid liquid fluidized beds. Chem. Eng. Sci. 92, 1–12. in solid–liquid flows. J. Fluid Mech. 353, 261–283.
Ruzicka, M.C., 2006. On buoyancy in dispersion. Chem. Eng. Sci. 61, 2437–2446. Zhang, J., Fan, L.S., Zhu, C., Pfeffer, R., Qi, D., 1999. Dynamic behavior of collision of
Sangani, A.S., Didwania, A.K., 1993. Dynamic simulations of flows of bubbly liquids elastic spheres in viscous fluids. Powder Technol. 106, 98–109.
at large Reynolds numbers. J. Fluid Mech. 250, 307–337. Zhou, Y.C., Wright, B.D., Yang, R.Y., Xu, B.H., Yu, A.B., 1999. Rolling friction in the
Sangani, A.S., Prosperetti, A., 1993. Numerical simulation of the motion of particles dynamic simulation of sandpile formation. Phys. A 269, 536–553.
at large Reynolds numbers. In: Roco, M.C. (Ed.), Particulate Two-Phase Flow.
Butterworth-Heinemann, Stoneham, MA.

Vous aimerez peut-être aussi