7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
Separation and Taking Off in a Fluidized Bed: Comparison between Experimental
Measurements and Threedimensional Simulation Results
Renaud Ansart*! Herv6 Neau *I Philippe Accartti
Alain de Ryckti and Olivier Simonin**
SUniversit6 de Toulouse ; INPT, UPS ; IMFT ; Allee Camille Soula, F31400 Toulouse, France
t CNRS; Institut de Mecanique des Fluides de Toulouse; F31400 Toulouse, France
i:Universit6 de Toulouse ;Mines Albi ; CNRS ; Campus Jarlard, F81013 Albi Cedex 09, France
3 Ecole des Mines Albi, Centre RAPSODEE, Campus Jarlard, F81013 Albi, France >>
ansart~imft.fr, simonin~imft.fr
Keywords: Fluidized bed, twophase flow, CFD, Euler multifluid approach
Abstract
This paper presents experimental measurements of separation and taking off in a fluidized bed from an experimental
equipment designed and built. Simultaneous measurements of pressure profiles and mass of particle takingoff in a
fluidized bed have been realized. These measurements are compared with threedimensional numerical simulation
predictions. Unsteady threedimensional numerical simulations of this column have been carried out with unstructured
parallelized CFD multiphase flow code. The comparison of the taking off and transport of monodispersed glass
beads exhibits a satisfactory agreement between experimental measurements and numerical simulations. However,
the simulation results are dependant on mesh size especially for fine particles and on wall boundary conditions.
Nomenclatu re
Roman symbols
CD drag coefficient ()
d, particle diameter (m)
g gravitational constant (m.s )
K entrainment rate (kg/m s)
P, mean gas pressure (N.m )
q~ mean particle agitation (m .s )
Re, particle Reynolds number ()
Up~ mean velocity of phase k (m.s )
Vf superficial gas velocity (m.s )
Vt terminal settling velocity (m.s )
fluctuating velocity of phase k (m.s
Introduction
Gassolid fluidized beds are used in a wide range of in
dustrial applications such as coal combustion, catalytic
polymerization or uranium fluoridation. Many of flu
idized bed industrial processing involve polydispersed
powder and even multispecies.
In bubbling fluidized bed combustion and catalytic
cracking Kunii and Levenspiel (1991), elutriation is a
major cause of inefficiency, while it may be highly de
sirable in carbon stripper process for Chemical Looping
Combustion. Whether the intention is to quench or pro
mote elutriation, the involved phenomena must be prop
erly known if the process has to be efficiently controlled.
Numerical simulation seems to be a good way to study
the separation and taking off phenomena observed in in
dustrial fluidized bed. In the literature, there is a lack of
experimental data to validate CFD simulations of these
phenomena. Thus, a join experimental and numerical
proj ect between RAPSODEE Centre and IMFT has been
initiated. The object of this paper is first to describe the
Greek
as
Sp
symbols
volume fraction of phase k ()
gas viscosity (kg.m .s )
density of phase k (kg.m )
mean gasparticle relaxation timescale (s)
Subscripts
g
p
gas
particle
Figure 2: Response of mass flow rate controller.
Table 1: Powder properties.
Particle properties Fine Coarse
Density (kg/m"1) 2470 2470
Mean diameter clan (put) 84 213
Span= d05l 0.38 0.414
<143 (put) 85 216
<132 (put) 83 210
Vt (111 s) 0.41 1.51
we can see, the variation between command and mea
surement are very small, even during the ramp, and may
be attributed to PID control.
The particles entrained are collected through a box set
at the outlet of a cyclone. The mass of particles collected
is continuously weighted during the process with a res
olution time of 1 s and an accuracy of 0.01 g.
At the gas outlet of the cyclone a sensor is set to obtain
a measurement of gas temperature and moisture. An
other sensor allows to measure the atmospheric pressure
into the test room. Thus, according to classical thermo
dynamic law, we calculate the gas density and kinetic
viscosity for each experimental trial.
Moreover, pressure variations along the pipe are mon
itored by six sensors located along the column at verti
cal distances from the distributor of 3 <*II, 6 <*II, 9 <*I,
12 <*II, 15 <*II and 18 <*II. We use Honeywell DC se
ries with an accuracy of f0.25 %o of full scale. At
the distances of 3 <*II and 6 <*II, the full scale sensor
is 8 500 Pa and at the other distances the full scale is
2 500 Pa. The resolution time is 0.1 s. The measure
ment of gas pressure on the wall is realized through an
hole about 8 1111 with a filter.
The powder is glas s beads with properties described in
table 1. We use two particle size distributions called fine
(Geldant group A/B, Geldart (1973)) and coarse (Geldant
7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
SMeanuleme
Time (s)
50 60 70
Figure 1: Experimental set up.
experimental set up built and the measurements realized.
Secondly to present a comparison between experimental
data of particle separation and taking off in a fluidized
bed and threedimensional numerical simulation predic
tions.
Experimental set up
The column of laboratory experimental set up is 10 cm
diameter and 59 <*II height (Fig. 1) with a conical out
let. Three kinds of column materials perspexx, per
spex+protection film, stainless steel) are used. As we
describe later, the first measurements with perspex col
umn were disturbed by electric charges.
The bronze distributor has a pressure drop of 6 kPa at
a 0.18 11 sl superficial gas velocity.
Fluidizing air is supplied by a Brooks smart mass
flow meters and controllers 5853S with an accuracy of
f0.7 % of rate and f0.2 % of full scale (2.32 11 s ).
The process is divided into two pants: the first one al
lows the fluidization of particles in order to obtain an ho
mogeneous bubbling mixture; according a linear ramp
up during the second part the fluidization velocity in
creases to take off particles. The Fig. 2 depicts the re
sponse of mass flow controller during the process. As
Table 2: Charge decay time. The corona discharge is
f) kV.
Mateial Relative Mean decay Initial
humidity time voltage
Fine part. 8.7 45 111 1 kv
Coarse part. 8.7 122 111 1.2 kv
Perspex 6 2.15 h 1.8 kv
Fine part. 42 13 111 0.12 kv
Coarse part. 42 10 111 0.11 kv
particles are charged with negative charges, on the other
hand coarse particles are charged with positive charges.
These charge results confirms the experimental observa
tion especially the stuck of fine particles on the surface
of the perspex column. Indeed, the charge of fine par
ticles is negative and plastic surface of the column pro
duces positive charges.
According to a voltage corona discharge (f) kV) the
charge decay time of particles is measured. At cessa
tion of charging the sample is quickly dropped in front
of a fieldmeter and the initial peak surface voltage and
the rate of decay of this voltage are measured. The term
'charge decay' here covers the time taken for this volt
age to fall away to 36.8 % of the initial value. The table
2 shows that under classical relative humidity the glass
beads take a very small voltage about 1/100 of corona
discharge and the charge decay is instantaneous. On
the contrary under dry atmosphere, the powder takes a
higher voltage about 1/8 of corona discharge and the
charge decay is instantaneous. It is noteworthy that the
charge decay of plastic material is very long about 2
hours.
7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
Sva=0.548 m/s T=18.1 0p
Sva=0.548 m/s T=17.8 OC
is gn=0.548 m/s T=17.9 OC
20II 40II 60II 800I lilli 120II 140II
Time (5)
(a) Perspex column. (b) Stainless steal column.
Figure 3: Effect of the wall properties on the mass of particle collected, column in perspex. Initial solid mass 1.5 kg,
cyl 0.18 11 51, rampup=3 s.
group B). The mean diameters of powder are determined
with Mastersizer 2000 with 1.5 bar of dispersion. The
bulk material has been sieved to ensure an almost mono
dispersed distribution. According to the expression of
the drag coefficient, equation (8), the terminal settling
velocity V, of the particle is calculated.
During the bubbling parts, the superficial velocity is
imposed to 0.18 11 51 and in the second part, after the
linear rampup, the superficial velocity chosen is higher
than the terminal settling velocity of particle.
Particle entrainment dependence on wall
properties
The first experiments were realized with a column in
perspex. The Fig. 3(a) presents the mass collected as
a function of time for three successive experiments real
ized for the same operating conditions. As we can see,
there is a high disparity between the results.
This disparity may be attributed to the effects of tri
bocharging of the particles. Actually, the gas injected in
the column through the distributor is very dry about f) %
of relative humidity.
A tribocharging method for testing glass beads has
been used. A solid mass is set into an unearthed rotating
cylinder (!)2 tr 1111) to be charged by rubbing with
the chosen material. The quantity of charge is measured
in a Faraday Pail unit. As the measurements of tribo
electric charging (Fig. 4) detail, glass beads can easily
become electrostatically charged when rubbed against
plastic material such as ertalon. Such triboelectric charg
ing does not appear under classical rate of relative hu
midity (40 It is also interesting to note that fine
Table 3: Effect of the superficial gas velocity on the
timeaveraged maximum of entrainment rate.
7/2 (" ms) 0.538 0.545 0.56
kmax ( kg m s1) 286 309 374
Table 4: Timeaveraged gas pressure drop for fine and
coarse particles. vfl 0.18 m s. Std: Standard
deviation.
Fine Coarse
p Po (Pa) p Po (Pa)
h = 3 cm 835 4.6 826 11.57
h = 6 cm 500 2.5 444 12.75
h = 9 c 114 1.3 77 8.33
h =12 cm 5.7 0.19 0.005 0.007
h =15 cm 1.5 0.15 0.01 0.24
h =18 cm 0 0 0 0
Hbed (cm) 10.02 0.053 9.50 0.11
During the entrainment step when the superficial gas
velocity is lower than the terminal settling velocity, we
can observe a bubbling bed. Accordingly pressure sen
sors set along the stainless steel column, we have deter
mined time averaged gas pressure drop along the wall.
The table 4 shows the time averaged gas pressure as a
function of height corresponding to a superficial veloc
ity of 0.18 m s1 for coarse and fine particles, where
Po is a reference pressure outside the bed at a distance
of 18 cm from the distributor. The value correspond to
a mean data between three experiments. As it is exhib
ited, the standard deviation between the three trials is
lower than the measurement error.
7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
Steel 10% humidty
eEttalon 10% humidity
 Steel 41% humidity
*Ettalon 41% humidity
(a) Fine particles.
Time(s)
(b) Coarse particles.
Figure 4: Triboelectric charging of glass beads.
First of all, we set a sticky film on the inside surface of
the perspex column to isolate the surface from the parti
cles. The first experiments were repeatable and encour
aging but this solution was not retained because of abra
sion of the film by the particles. Thus, for the following
of the study an stainless steel column earthed around its
edge has been used to avoid electric charge.
The Fig. 3(b) presents the evolution of the mass col
lected as a function of time with a stainless steel column
for same operating conditions, even for experiments re
alized on different days. As we can note, the stainless
steel column has highly reduced the disparity between
the results. Hence, all the next experiments will be real
ized with this column.
The Fig. 5 depicts the influence of a variation of su
perficial gas velocity on the entrainment of particles. It
is important to note that all these trials were done for the
same injected gas temperature with the same relative hu
midity and the same atmospheric pressure to ensure the
same thermophysical gas properties. We may note, that
a slightly variation on the superficial gas velocity gener
ates an important variation on the entrainment of particle
and so the entrainment rate (Fig. 5(b)). The entrainment
rate is expressed as following:
k = m(1)
S di '
where m is the mass of particle collected as a function
of time and S the transfer surface of the column. This
entrainment rate is not constant during the trial. At the
beginning of the entrainment, there is a high increase
of this flux to reach a maximum rate. After this maxi
mum rate of separation, the entrainment rate decreases
continuously. The table 3 presents the maximum rate of
entrainment as a function of superficial gas velocity.
Mean
Mean
7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
'=0.538 m/s T=14.8 OC
S=0.545 m/s T=14.9 OC
_ v=0.560 m/s T=15.2 OC
2ill Mill 6iin Sil 10til 12iin 14iin 1600l
Time (5
(a) Mass of particles collected.
(b) Entrainment rate.
Figure 5: Effect of a slightly variation of a superficial gas velocity on the entrainment of fine particles. Initial solid
mass 1.5 kg, vf = 0.18 11 s rampup= 3 s.
Moreover, the time averaged gas pressure on the wall
allows to determine the bed height. Several definitions
of averaged bed height can be found in the literature.
Here, the bed height is defined as the intersection point
of the two linear parts of the gas pressure profile. For a
superficial gas velocity 0.18 11 5 the time averaged
bed height is 10.02 11 for the fine particles and 9.5 11 for
coarse. An increase of the mean particle size diameter
from 84 pin to 213 pri decreases the mean bed height
about 5 o.
Mathematical model
Simulations are carried out using an Eulerian nfluid
modeling approach for turbulent and polydispersed
fluidparticle flows, which is developed and imple
mented by IMFT (Institut de Mecanique des Fluides de
Toulouse) in a specific version of the NEPTUNE_CFD
software, known as NEPTUNE_CFD V1.07@Tlse.
NEPTUNE_CFD is a multiphase flow software devel
oped in the framework of the NEPTUNE project, fi
nancially supported by CEA (Commissariat it l'Energie
Atomique), EDF (Electricit6 de France), IRSN (Institut
de Radioprotection et de Stiret6 Nuceaire) and AREVA
NP.
In the proposed modeling approach, separate mean
transport equations (mass, momentum and fluctuant ki
netic energy) are solved for each phase and coupled
though interphase transfer terms. The transport equa
tions are derived by phase ensemble averaging weighted
by the gas density for the continuous phase and by using
kinetic theory of granular flows supplemented by fluid
and turbulence effects for the dispersed phase.
In the following, when subscript k = g, we refer to
the gas phase and k = p to the particle phase. The mass
transport equation is:
DL i3r
where as is the kth phase volume fraction, pa the den
sity and LT<,; the component of the velocity. In equa
tion (2), the righthandside is equal to zero because no
mass transfer takes place.
The mean momentum transport equation takes the fol
lowing expression:
ax dz + crr pr Y; (3)
+li + )rA a
where .' is the fluctuating part of the instantaneous ve
locity of phase k, P, is the gas pressure, y; the ith com
ponent of the gravity acceleration and IA,; the mean gas
particle interphase momentum transfer without the mean
gas pressure contribution. Finally, 8<,;4 is for k y the
molecular viscous tensor and for k p the collisional
particle stress tensor.
According to large particle to gas density ratio only
the drag force is accounted as acting on the parti
cles. Hence, the mean gasparticle interphase momen
tum transfer is written:
I I arPI(4)
rF
as p 07<
L Of itr ]
7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
where the particle relaxation time scale is given by
Gobin et al. (2003):
C,
Smin(Cd~wy, Od,Erg)
Op > 0.3.
I''
iii; \i
''
hx
where the Ergun's drag coefficient is given by:
a7
CD,Erg = 200 ~3
and Wen & Yu's correlation by
Figure 6: Threedimensional reference mesh containing
428 451 hexahedra.
The numerical simulations have been performed on
parallel computers with 8 cores for the coarse mesh, 64
cores for reference mesh and 128 cores for fine mesh,
because of mesh size and physical time needed, Neau
et al. (2010).
At the bottom (z = 0), the fluidization grid is an inlet
for the gas with imposed superficial velocity correspond
ing to the fluidization velocity vf. For the particles this
section is a wall. At the top of the fluidized, we defined
a free outlet for both the gas and the particles. The wall
type boundary condition is friction for the gas.
A recent study comparing threedimensional numer
ical simulations and experimental data from dense flu
idized bed has shown that the particle wall boundary
condition is crucial for the numerical prediction of the
fluidized bed hydrodynamic Fede et al. (2009). In the
present study two kinds of wall boundary condition for
the particulate phase have been tested. First a pure slip
wall boundary condition,
CDWY = I (1 +0.15Re .687) 1.7
Co~wu i 0.44cr1. 1
Re, < 1000
Re, '> 1000
(8)
The particle Reynolds number is defined by:
ps (vr) d
Re, = as '
The mean relative velocity Vr,i between gas and parti
cle is expressed in terms of the mean gas velocity, mean
particle velocity and drift velocity. The drift velocity
accounts for the correlation between the particle distri
bution and the turbulent velocity Simonin et al. (1993).
In equation (3), the collisional particle stress tensor is
derived in the frame of the kinetic theory of granular
media Boelle et al. (1995). Such a modeling approach
has already been used in industrial fluidized bed such as
for ethylene polymerization reactors Gobin et al. (2003)
and uranium fluoridation Randrianarivelo et al. (2007).
For the turbulence modeling, we use a standard k E
model extended to the multiphase flows accounting for
additional source terms due to the interfacial interac
tions. For the dispersed phase, a coupled transport equa
tion system is solved on particle fluctuating kinetic en
ergy and fluidparticle fluctuating velocity covariance
(9~ 4f p *
Numerical parameters
To study the influence of mesh refinement, we used three
3D meshes based on Ogrid technique. The reference
mesh contains 428 451 hexahedra with approximately
a, Ay a,=az 3.7 mm (Fig. 6). The fine mesh is
uniformally refined by a factor of 1.5 to obtained a mesh
of 1 477 060 cells. On the contrary, the coarse mesh is
uniformally derefined by a factor of 1.5 to get a mesh
of 123 816 cells.
8~n
8n
corresponding to particlewall elastic rebounds on a
flat wall. In equation (10), Up,, is the tangential to the
wall component of mean particle velocity and Up,, the
normal to the wall component of mean particle velocity.
Fede et al. (2009) have shown that noslip wall boun
dary,
Figure 7: Wall distribution of the timeaveraged gas pressure during the bubbling step for coarse particles.
Table 5: Properties of various trials.
Part. vfl (m s )ves (m s )Ramp (s) T (oC) Patm (mbar) Humidity (%b) Solid mass (kg)
Coarse 0.18 1.55 5 15.9 983 7.8 1
Fine 0.18 0.61 5 12.6 993 8.6 1
7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
SReference mesh slip r=0
tReference mesh slip rm/2
tReference mesh noslip r=0
SReference mesh noslip r=7cl2
PPo (Pa)
(a) Comparison between experiment data and numerical simu
lation predictions for the reference mesh with noslip condition.
(b) Effect of the boundary conditions on the numerical results.
reduced from 1.5 kg to 1 kg to reduce the experiment
duration. All the experiment results presented in this
(11) section correspond to an average between three different
experiments realized for the same operating conditions.
8n
gives numerical predictions in good agreement with
experimental measurement obtained by Positron Emis
sion Particle Tracking. It is noteworthy that noslip wall
boundary condition for the mean particle velocity re
mains questionable. Actually, it could mean that a parti
cle bouncing on a wall as an isotropic random direction
after the rebound. This case corresponds to a spherical
particle hitting a rough wall with a very large roughness
or a very irregular particle bouncing a smooth wall, Ko
nan et al. (2009).
In the experiments, the particle phase is slightly poly
dispersed (span 0.4) so the numerical simulations
have been carried out with monodisperse particle dis
tribution having a median diameter equal to the dso
Results and discussion
First of all, a comparison between the numerical pre
dictions for the coarse particles and the experimental
results is realized. Then, the same comparison is car
ried out with the results on fine particles. The operating
conditions for these two studies are described in the ta
ble 5. we can note that the solid initial mass has been
Coarse particles
In this part, the experiment results of gas pressure drop
along the wall during the bubbling step and the mass of
particle collected is compared with the numerical results
realized with the reference mesh and two kinds of wall
boundary conditions.
First of all, we are studying the bubbling step where
the superficial gas velocity vfl is lower than the termi
nal settling velocity. To study the wall gas pressure
drop during this step, the numerical simulation is car
ried out as following: at t 0 the fluidized bed is
fill up of uniform solid mass fraction according to the
experimental solid mass. A transitory step takes place
for t E [0 s, 20 s] corresponding to the destabilization
of the fluidized bed. The statistics are computed for
t e [20 s, 50 s] insuring a statistical convergence.
The Fig. 7(a) presents the experimental results and the
numerical simulation predictions for the wall gas pres
sure drop. The simulation has been done for the ref
erence mesh with noslip condition on the mean parti
cle velocity. As we can see, there is a good agreement
between experiment and simulation. Outside the bed,
7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
3500 Experiment
SExperiment 250
Reference meshnoli
Reference mesh slip 20
Time (s) Time (s)
(a) Mass of particles collected. (b) Entrainment rate.
Comparison between numerical simulation predictions and experimental data on the entrainment of fine
~:i0.4
apmoy0.2
Figure 8:
particles .
0.64
0.48
Slip 0.32 Noslip
0.16
0.00
Figure 9: Effect of the boundary condition on the time
averaged volume fraction and velocity vectors of coarse
particles .
the pressure wall distribution is linear for the numeri
cal simulation and for the experiments, corresponding
to hydrostatic pressure of gas. Inside the bed, the both
distributions are linear. The numerical simulation for the
reference mesh and noslip boundary condition seems to
predict correctly the hydrodynamic of the bed.
The Fig. 7(b) shows the strong influence of the bound
ary conditions on the gas pressure profile predicted.
We observe two different behaviors depending on wall
boundary conditions. When noslip conditions are used,
the pressure profile is linear inside the bed. On the con
trary, for slip wall conditions, the gas pressure distribu
tions is slightly curved inside the bed. As we can see,
this boundary condition generates an asymmetrie of the
fluidized bed. The pressure gas profile on the plan r = 0
Figure 10l: Effect of the boundary condition on the time
averaged radial profile of vertical velocity of coarse par
ticle at z 6 cm.
is not similar to the one of the plan r = ir/2.
We can find again, this observation on the time
averaged volume fraction of particle predicted, Fig. 9.
For the slip condition, we can observe a high vortex in
side the bed which generates a completely asymmetric
bed.
Moreover, the Fig. 10 shows the radial profiles of
mean vertical velocity of particles for the two kinds of
boundary conditions in two different plans. When no
slip condition is used the flow is almost symmetric. On
the other hand, for slip condition as we observed previ
ously the flow is asymmetric. Hence, this boundary con
dition has a strong influence on the bed hydrodynamic
as we can note on velocity vectors of particle.
7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
(a) Comparison between experiment data and numerical simula
tion predictions for the reference mesh with noslip condition.
(b) Effect of mesh refinement,
Figure 11: Wall distribution of the timeaveraged gas pressure during the bubbling step for fine particles.
We may conclude that with the reference mesh and
noslip condition on the particle velocity we are able to
predict the bed height during the bubbling step and the
entrainment of the coarse particles.
Fine particles
For fine particles, the numerical simulations have been
performed on three meshes with noslip condition and
only for reference mesh an effect of the boundary condi
tions has been studied with pure slip wall condition for
the mean particle velocity.
The Fig. 11(a) shows the wall distribution of time
averaged gas pressure for the experiments and for ref
erence mesh and noslip condition on the mean particle
velocity. The numerical predictions give the same shape
of the gas pressure drop profile as the experimental re
sults (linear inside the bed). But, the simulation over
estimates the gas pressure inside the bed.
Moreover, an effect of the mesh size on the simula
tion has been done, Fig. 11(b). The mesh refinement
decreases the gas pressures on the wall especially on
the top part of the fluidized bed. Refining the mesh de
creases the bed height by increasing the volume fraction
until a critical cell size, Parmentier et al. (2008). But,
we clearly see that all the simulations overestimates the
bed height. Even the fine mesh seems not able to capture
the fine structures of the flow. It seems that the cell size
is not enough small. Thus, we have to refine again the
mesh or to use subgrid model to be in a good agreement
with the experimental results and to predict correctly the
bed height for fine particles.
A study of the effect of boundary conditions for the
reference mesh on the numerical predictions has been
pp, m
Figure 12: Influence of the particle wall boundary con
ditions on the gas pressure drop.
Then, we can study the taking off step when the su
perficial gas velocity is higher than the terminal settling
velocity of the coarse particle. The Fig. 8(a) presents
the mass of coarse collected as a function of time. As
we can note, the numerical simulation seems to predict
correctly this mass. On the contrary, the simulation re
alized with slip condition predicts correctly the quantity
of particles entrained at the starting of the trial but after
the mass of particle is strongly underestimated.
The Fig. 8(b) presents the same kind of comparison
on the entrainment rate. The flux of particles entrained is
underevaluated with slip conditions. On the other hand,
the condition of noslip seems to well described this tak
ing off. Nevertheless, the maximum of entrainment rate
is slightly overestimated with noslip condition.
7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
(a) Mass of particles collected. (b) Entrainment rate.
Figure 13: Comparison between numerical simulation predictions
particles .
and experimental data on the entrainment of fine
0.5 Refeencemesh slip
Figure 15: Effect of the boundary condition on the time
averaged radial profile of vertical velocity of fine particle
at z = 6 cm.
condition, Fig. 14.
Then, we can study the taking off step when the su
perficial gas velocity is higher than the terminal settling
velocity of the fine particles. The Fig. 13(a) presents the
mass of fine particles collected as a function of time. A
comparison between experimental results and numerical
predictions with the reference mesh with noslip con
ditions shows that at the starting of the taking off we
have a good agreement. Then, we observe an increasing
gap between the entrained mass of particles measured
and the numerical simulations predictions. At the curve
ending, numerical simulations and experiments do not
present the same shape of the asymptotic convergence
to the initial mass inside the bed. This deviation may be
Noslip
Figure 14: Effect of the boundary condition on the time
averaged volume fraction and velocity vectors of fine
particles .
carried out. When noslip condition is used, the pressure
profile is linear inside the bed. On the contrary, for slip
wall conditions, the gas pressure distribution is slightly
curved inside the bed. As observed on the mean solid
volume fraction of the Fig. 14, the flow is not the same
with these two kinds of boundary conditions. It is note
worthy that for fine particles, the simulation with no slip
generates a fluidized bed almost symmetric, Fig. 15. As
Fede et al. (2010) observed, noslip wall boundary con
dition leads to lower mean solid volume fraction which
leads to increase the bed height. Moreover as Fede et al.
described, we can note a higher solid volume fraction
near wall region with slip than with noslip boundary
alcp~moy
Y 0.60
ii 0.45 I
0.30
0.15
0.00
7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
(a) Effect of mesh. (b) Entrainment of the boundary conditions.
Figure 16: Effect of the mesh size and the boundary conditions on the mass of particles collected.
attributed to the particle size distribution of the particles.
Indeed, we have assume that the distribution was mono
dispersed but during the trial the mean diameter of the
particle remaining in the column increases.
To confirm this assumption, we have stopped an ex
periment when 90 % of the initial solid mass in the col
umn was entrained and realized a particle size analysis
of the particles remained in the column. The mean par
ticle diameter measured is dso 96 pm. This results
have to be compared with the initial PSD of the powder
realized by the same method (Table 1). The mean di
ameter has increased about 14 % from dso = 84 pm to
dso = 96 pm. To make a better comparison, we should
use particle with exactly a monodispersed PSD or tak
ing into account ploydispersed PSD of the particle in
the numerical simulation
The Fig. 13(b) depicts a comparison between experi
mental results and numerical simulation predictions for
reference mesh and noslip conditions on the entrain
ment rate of fine particles. This numerical simulation
seems to pretty well predict the flux of particle entrained
especially at the starting of the process.
Moreover, a study of the mesh refinement with no
slip conditions on the particle collected has been real
ized, Fig. 16(a). The refinement from coarse mesh to
reference mesh decreases the mass of particle collected
as a function of time. On the other hand, the refinement
from reference mesh to fine mesh seems to not have in
fluence on the mass of particle predicted. But, the simu
lation with fine mesh are greatly much expensive about
the CPU time than the one with reference mesh.
A study of the influence of boundary conditions on
the mass of fine particles collected with the reference
mesh has been done, Fig. 16(b). As observed for coarse
particles, the simulation realized with slip condition pre
dicts correctly the quantity of particles entrained at the
starting of the trial but after the mass of particle is under
estimated.
Conclusions
An experimental test equipment has been designed and
built to study particle separation and taking off in a flu
idized bed by measuring the gas pressure along the col
umn and the mass of particles leaving the column. These
experimental results have been compared with three di
mensional unsteady numerical predictions carried out
with the unstructured parallelized CFD multiphase flow
NEPTUNE CFD.
The experimental results have exhibits the strong in
fluence of the tribocharging of the particles on the dis
parity of the measurements. According to a stainless
steel column, the effects of the particle tribocharging
have disappear and we observe no disparity between the
measurements.
In this study, we have compared experimental data
and numerical simulations prediction realized for coarse
particle (Geldart group B). For the bubbling step, the
prediction of the mean gas pressure drop along the wall
with noslip condition on the mean particle velocity
seems to be appropriated to estimate the bed height.
Moreover, we have also a good agreement between ex
periments and numerical simulations on the mass of par
ticle entrained during the taking off step. On the con
trary, the study of the effect of the boundary condition
has shown that the boundary condition of pure slip is
not appropriated to predict the bubbling step and highly
under estimate the mass of particle entrained during the
7th International Conference on Multiphase Flow,
ICMF 2010, Tampa, FL, May 30 June 4, 2010
D. Kunii and O. Levenspiel. Fluidization Engineering.
Butterworth Heinemann, 1991.
H. Neau, J. Lavieville, and O. Simonin. Neptune_cfd
high parallel computing performances for particle laden
reactive flows. h? 7ti' international Conference on Mul
tiphase Flows., 2010.
J.F. Parmentier, O. Simonin, and Delsart O. A numerical
study of fluidization behavior of geldart b, a/b and a par
ticles using an eulerian multifluid modeling approach.
In .I' international Conference on circulating fluidized
beds. Hainbut; Germany:, 2008.
T. Randrianarivelo, H. Neau, O. Simonin, and F. Nico
las. 3d unsteady polydispersed simulation of uranium
hexafluoride production in a fluidized bed pilot. In Proc.
6th Int. Conf on Multiphase Flow, Leipzig (Gerinany'1
paper S6 The A46,, 2007.
O. Simonin, E. Deutsch, and J.P. Minier. Eulerian pre
diction of the fluid/particle correlated motion in turbu
lent twophase flows. Applied Scientific Research, 51,
275283, 1993.
entrainment step.
The same study has been led on the fine particles (Gel
dart group A/B). In this case, the numerical simulations
seem to be not able to predict the bed height even with
a finer mesh. Subgrip model should be applied to cap
ture small structures of flows. But, the numerical simu
lation gives a satisfactory agreement about the mass of
fine particles entrained during the taking off especially
at the starting of the process. Moreover, the simulation
seems to provide with a well agreement the maximum
of entrained rate observed at the starting of the process.
This observation is essential for the following step of the
study. Actually, we are going to study the elutriation of
a mixture of fine and coarse particle.
Acknowledgments
This work was granted access to the HPC resources
of CINES under the allocation 2010026012 made by
GENCI (Grand Equipement National de Calcul Intensif)
and CALMIP (Centre de Calcul MidiPyrendes) under
the allocation P0111.
References
A. Boelle, G Balzer, and O. Simonin. Secondorder pre
diction of the prediction of the particlephase stress ten
sor of inelastic spheres in simple shear dense suspen
sions. h? GasParticle flows, volume 28, ASME FED,
pages 918, 1995.
P. Fede, G. Moula, T. Ingram, and O. Simonin. 3d nu
merical simulation and pept experimental investigation
of pressurized gassolid fluidized bed hydrodynamic. In
in proceedings ofASME 2009 Fluids Engineering Divi
sion Sainier Meeting. ASME., 2009.
P. Fede, H. Neau, O. Simonin, and I. Ghouila. 3d un
steady numerical simulation of the hydrodynamic of gas
phase polymerization pilot reactor. In h? 7"" hIterna
tional Conference on Multiphase Flows., 2010.
D. Geldart. types of gas fluidization. Powder Technol
ogy, 7, pages 285292, 1973.
A. Gobin, H. Neau, O. Simonin, J.R. Llinas, Reiling, and
J.L Selo. Fluid dynamic numerical simulation of a gas
phase polymerization reactor. hIt. J. for Num. Methods
in Fluids, 43, 11991220, pages 11991220, 2003.
N.A. Konan, O. Kannengieser, and O. Simonin.
Stochastic modeling of the multiple rebound effects for
particlerough collisions. international Journal ofMulti
phase Flow, volume 35, Issue 10, pages 933945, 2009.
