Computers and Chemical Engineering 31 (2007) 334–345

CFD phenomenological model of solid–liquid mixing in stirred vessels

∗
Louis Fradette a , Philippe A. Tanguy a, , Franc¸ois Bertrand a , Francis Thibault a ,
Jean-Benoˆıt Ritz b, Eric Giraud b
a
URPEI, Department of Chemical Engineering, Ecole Polytechnique, P.O. Box 6079, Montreal H3C 3A7,
Canada b SME, BP 57, 33166 St-Me´dard-en-Jalles Cedex, France
Received 17 April 2006; received in revised form 18 July 2006; accepted 21 July 2006
Available online 7 September 2006

Abstract
Particle migration in a concentrated viscous suspension subjected to a non-homogeneous shear field was computed using a 3D extension of
the diffusion model of Phillips et al. [Phillips, R. J., Armstrong, R. C., Brown, R. A., Graham, A. L., & Abott, J. R. (1992). A constitutive
equation for concentrated suspensions that accounts for shear-induced particle migration. Physics of Fluids A, 4, 30–40]. The numerical results
were compared to experimental data from the literature for simple flows and to our own data in the case of two helical ribbon based mixing
systems. It is shown that this type of diffusion model, which was developed to predict the behavior of concentrated suspensions dominated by
particle–particle interactions, can predict migration trends but definitely requires additional improvements.
© 2006 Elsevier Ltd. All rights reserved.

Keywords: Mixing; Particle migration; Concentrated suspensions; CFD; Simulations

1. Introduction comparable to that obtained in many polymerization

The makedown of viscous suspensions at high solids con-
tent is a unit operation involved in the production of coatings,
specialty chemicals, drugs and space fuels. This operation is
generally carried out in batch or semi-batch mode in stirred
vessels. The very viscous nature of the media and the homo-
geneity requirements for the resulting mixture impede the use
of classical impeller mixers. Hence, adapted mixing technolo-
gies such as planetary mixers with kneading blades (Zhou,
Tanguy, & Dubois, 2000) and multishaft mixers that combine in
a single apparatus a high-speed dispersing turbine, a pumping
impeller and a wall scraping blade (Tanguy, Bertrand, Thibault,
& Galy-Jammou, 2001) are common options. From an industrial
perspective, however, these mixers are costly to acquire,
operate and maintain.
In the polymer industry, it is well known that viscous
mixing is best achieved with close-clearance impellers like
helical rib- bon impellers. For the makedown of high solids
content viscous suspensions, as the level of viscosity induced
by the high solids concentration and sometimes by the
suspending matrix itself is
∗
Corresponding author. Tel.: +1 514 340 4017; fax: +1 514 340 4105.
E-mail address: philippe.tanguy@polymtl.ca (P.A. Tanguy).
processes, it would be tempting to use helical ribbon
impellers. In practice, however, this approach is not used by
fear of particle segre- gation in poorly mixed regions and
particle migration under non-uniform shear, which might lead
to suspension demixing and therefore mixture quality
problems. As an example of indus- trial impact, shear-induced
migration of conductive particles is sometimes responsible for
the loss of electrical conductivity in polymer compounds
(Jana, 2003).
Since the first observations of particle diffusion under shear
deformation by Gadala-Maria and Acrivos (1980), several con-
tributions have led to a better understanding of the hydrody-
namic phenomena that induce demixing within concentrated
suspensions (Altobelli, Givler, & Fukushima, 1991; Leighton
& Acrivos, 1986, 1987a,b). It has been shown in particular
that the diffusion is related to the particle collision frequency
as well as the variations in the suspension viscosity. In 1992,
Phillips, Armstrong, Brown, Graham, and Abott (1992) pro-
posed a conservation equation for the solid phase that includes
convective transport, collision-induced diffusion, and
viscosity effects. These authors used the Krieger and
Dougherty model (Krieger & Dougherty, 1959) to express the
variation of the sus- pension viscosity as a function of the
solid volume fraction. They combined this equation with the
momentum conservation equation for the continuous phase to
investigate the diffusion of

0098-1354/$ – see front matter © 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2006.07.013
 L. Fradette et al. / Computers and Chemical Engineering 31 (2007) 334–
345
Nomenclature
aparticle diameter (m)
cimpeller distance to the bottom of the tank
Dimpeller diameter
DTtank diameter
Hliquid height in tank
Kconsistency index (Pa sn)
Kcempirical constant
Kμempirical constant
nshear-thinning or power-law index
Nrotational speed (rev/s)
Nmigmigration number
ppressure (Pa)
ttime (s)
Ttorque (N m)
Vvelocity (m/s)
Greek letters
γ˙rate-of-strain
η ρ φ φm Ω tensor (s−1) viscosity (Pa s)
density (kg/m3)
particle concentration (v/v)
maximum particle concentration (v/v) computational domain
concentrated monodisperse suspensions of PMMA
(polymethyl methacrylate) particles in silicon oil (45 < φ < 55
vol.%). The particles were non-colloidal (a = 675 µm) and
neutrally buoy- ant. Couette and Poiseuille flows were studied.
It was shown that for these two ‘academic’ flows, a
concentration gradient was generated in the radial direction.
Model results were shown to compare well with experimental
particle concentration pro- files obtained by nuclear magnetic
resonance (NMR).
The phenomenon of particle migration was also modeled in
the case of particle resuspension in horizontal ducts, in the
lami- nar and transition regimes (Zhang & Acrivos, 1994). The
authors complemented the model of Phillips et al. (1992) with a
flux term to account for particle settling and thus make it
usable for non- neutrally buoyant particles. Here again, the
prediction of the solids concentration was compared with
NMR measurements from Altobelli et al. (1991). This study
highlighted the effect of the Reynolds number and the
suspension concentration on the resuspending mechanism.
From a qualitative standpoint, the solid–liquid interface was
correctly captured by the numerical model.
Particle migration along solid walls was also investigated
(Jana, Kapoor, & Acrivos, 1995). An apparent wall slip veloc-
ity coefficient was determined experimentally, and was shown
to be insensitive to the shear rate while being rather strongly
influenced by the solids content of the suspension.
Finally, a few studies have used the model of Phillips et
al. (1992) to simulate particle migration within concentrated
suspensions flowing between parallel plates and in the gap of
off- centered cylinders. In particular, Fang and Phan-Thien
(1995)
33
showed that the model predictions are in very good agreement
with the analytical solution of the flow between parallel
plates. However, the results on the solid volume fraction in the
core of the eccentric cylinder problem has not been validated.
The model of Phillips et al. (1992) is not the only model that
can be used to simulate the flow of concentrated suspensions.
In a recent paper by Tanguy, Thibault, Ascanio, and Brito-
De La Fuente (2006), the performance of the network-of-zone
approach (Mann & Hackett, 1988) used in combination with the
virtual finite element method (Bertrand, Tanguy, & Thibault,
1997) to analyze the complex flow and suspension
mechanisms in a coaxial mixer was investigated. Experiments
carried out at lab scale confirmed the validity of the
predictions. However, this model could not capture the shear-
induced particle migration, and as a result, the agreement
between the numerical and the experiments in the intensely
sheared zones was unsatisfactory.
In all the above approaches, the solid phase is considered
as a continuum. However, it is possible to use direct numerical
sim- ulation to describe the motion of each individual particle
in the liquid phase. The strategy consists of solving the
Navier-Stokes equations in the liquid phase while taking into
account the pres- ence of the particles through its physical
properties and their motion through Newton’s second law of
motion for each par- ticle (Glowinski, Pan, & Perieux, 1994;
Hu, 1996; Hu, Joseph, & Crochet, 1992; Maury & Glowinski,
1997). This approach allows for particle–particle and particle–
fluid interactions within the suspension to be modeled in a
physically correct manner. In practice, due to computer
limitations, the number of particles is limited to a few
thousand, whereas in a typical mixing applica- tion this
number is several orders of magnitude larger.
The objective of the present study is to assess the accuracy
of the model proposed by Phillips et al. (1992) and augmented
as per Zhang and Acrivos (1994), to capture the particle
suspend- ing phenomenon and particle migration in solid–
liquid mixing. We call this model the shear-induced migration
model (SIMM). Three mixing systems will be considered: a
propeller and two helical ribbon geometries. Since these
mixing systems lead to complex three-dimensional flows,
there are no analytical solu- tions to the corresponding flow
fields so that experimental results will be used for validation.
The present work can be consid- ered as a real, asymmetrical
three-dimensional application of the model of Phillips et al.
(1992) to a mixing situation. To our knowledge, there does not
exist a similar application in the lit- erature, the closest being
the one presented by Subia, Ingber, Mondy, Altobelli, and
Graham (1998).
2. Experimental systems
The first experimental set-up (Fig. 1) consists of a small
reser- voir (glass beaker) with diameter DT = 7.2 cm and
height H = 6.5, and a down-pumping marine propeller with
diameter D = 1.8 cm that provides agitation. The propeller is
located at a distance c = 2.6 cm from the bottom.
The investigation of the particle motion inside this mixer
involved a Newtonian solution of corn syrup with a viscosity
of 1.05 Pa s and a density of 1360 kg/m3. The particles were
red Ballotini glass beads. The spherical beads had a uniform
33 L. Fradette et al. / Computers and Chemical Engineering 31 (2007) 334–
diame- 345
 Fig. 1. Dimensions of the marine propeller mixing system (in cm).

ter of 1 mm and a density of 2500 kg/m3. The maximum

random packing concentration (φm) for this particle type is
0.68.
In order to investigate the suspending phenomenon, two
parameters were varied: the propeller rotational speed and the
particle volume concentration. The conditions were as
presented in Table 1.
In each experiment, the particles were initially at rest at the
vessel bottom. Once the agitator was started up, the particles
motion was recorded with time until equilibrium was obtained
on the basis of torque and visual observation of the particles
dispersion.
The other two mixing systems make use of the same
D = 27 cm diameter ribbon. Two tank diameters were used in
the study, providing a wide clearance mixing system (impeller-
to- tank diameter ratio, D/DT = 0.4) and a close clearance
mixing system (D/DT = 0.87). Impeller pitch was equal to 1 in
both cases. A schematic of the close clearance system is given
in Fig. 2. The wide clearance system differs only by the tank Fig. 2. Small clearance mixing system with an impeller-to-tank diameter ratio
diameter which was 65 cm compared to the 31 cm of the close of 0.87.
clearance system. The counterclockwise rotational speed was,
respectively, 50 and 16 rpm for the wide and close clearance
3. Numerical model
impellers. The close clearance mixer was instrumented with a
torque meter and a speed encoder directly mounted on the
For the prediction of shear-induced particle migration, the
actu- ated shaft to allow the real-time measurement of torque
SIMM was used, which states that, in a viscous concentrated
during the mixing process. The fluid was a suspension of
sus- pension, small but non-Brownian particles migrate from
monodisperse spherical glass beads in polybutadiene, a
regions of high shear rate to regions of low shear rate, and
slightly shear-thinning fluid obeying a power-law model:
from regions of high concentrations to regions of low
−1 concentrations in addi- tion to which settling by gravity is
η(γ˙ ) = K|γ˙ |n , (1)
added. In the case of a mixing process, owing to the action of
where γ˙ is the shear rate, K the consistency index, and n is the shear and inertia, the particles may segregate and demix,
power-law index. For the entire experimental program, several thereby generating concentration gra- dients in the vessel. This
glass bead diameters ranging between 0.1 and 3 mm were con- shear-induced migration phenomenon can be simulated either
sidered. The properties of these glass beads and those of the at the microscopic scale, in which case the particles are
suspending solution of polybutadiene are given in Table 2. considered as interacting entities making up the dispersed
phase (Brady & Bossis, 1985) or at the macroscopic
Table 1
Experimental conditions for the glass beaker agitation system
Table 2
Physical properties of the solid–liquid system used in the helical ribbon mixers
Experiment Rotational speed (rpm) Bead concentration (vol.%)
Specific gravity Rheological parameters
1 173 2.8
2 230 7.1 Glass beads 2.48 –
3 350 11.9 Polybutadiene (PBHT) 0.9 K = 500 Pa sn, n = 0.8
scale, where the suspension is modeled as one continuous
phase through a viscosity law (Phillips et al., 1992). If
microscopic scale simulations should yield more accurate
results, they are still today very costly and out of reach for
predicting the behav- ior of industrial process suspensions. For
this reason, the motion of particle suspensions is accounted for
in this work using a macroscopic model.
Because it was only slightly shear-thinning as shown in
Table 2, the fluid was assumed to be a Newtonian suspension
of monodisperse spherical particles in the numerical model.
Note that, as evidenced by the work of Kim (2001), the
rheological properties of a dispersing medium can sometimes
have a signif- icant effect on the flow behavior. The governing
equations were as follows:
∂v
ρm . + v · grad vΣ + grad p − div(ηm grad v) = 0, (2)
∂ Mondy, Baer, Al t obell i , a nd S t e ph en s ( 20 02 ) , the
t hindered settling velocity does not go down to zero when the
div v = 0, (3)
where ρm = φρparticle + (1−φ)ρliquid denotes the density of the
suspension and where ηm, its viscosity, is modeled through the
Krieger–Dougherty correlation (Krieger & Dougherty, 1959):
where α equals 1.82 and where
Dφ ∂φ
= + v · grad φ (8)
Dt ∂t
is the material derivative.
In Eq. (7), the solid and liquid phases are treated as a
con- tinuum and the viscosity helps keep the particle
concentration φ below the selected maximum concentration
φm. As noted by Fang and Phan-Thien (1995), numerical
problems are some- times encountered in the low shear
zones with particle con- centrations close to − φM. Indeed, the
term (φM φ) then tends
towards zero, which can generate numerical instabilities in
these zones. These problems can be further magnified if, as
noted by Rao, Mondy, Sun, and Altobelli (2002), and Rao,
volume fraction reaches φM. The model in Eq. (7) is empirical
since constants Kc and Kμ have been determined from
experimental measurements of the concentration profiles in
Couette and Poiseuille flows (Hampton, Mammoli, Tetlow, &
Altobelli, 1997; Phillips et a l . , 1992). It must be
remembered that they have been obtained for concen- trations
ranging between 45 and 55 vol.%. Since their respective

ηm = ηliquid . Σ
φ −1.82
values are almost identical for both flow types, they are
1− m . (4) generally considered to be constant. These same values were
φ used in all the simulations in this work, that is Kc = 0.41 and
Kμ = 0.62. Never-
In this equation, ηliquid represents the viscosity of the theless, Tetlow et al. (1998) showed that Kc could be related to
suspending fluid, φm the maximum particle concentration in the local concentration and physical characteristics of the solid
the suspension and φ the particle concentration that can be particles. These authors also observed a dependence of the
computed by solv- ing: con- centration as the cube of the particle diameter. Because
dφ they were experimental data developed in one dimension in
2 2 the case of a wide-gap Couette apparatus, these results were
+ v · grad φ = a Kc div(φ grad|γ˙ | + φ|γ˙ | grad φ)
dt . 2 Σ not included in the present work.
2 φ |γ˙ | ∂ηm Boundary conditions are required for mathematical well-
+ a Kμ div grad φ , (5) posedness. As no particle can penetrate or exit the domain, the
ηm ∂φ
sum of fluxes through the wall and the free surface is zero, that
where |γ˙ | denotes the magnitude of the rate-of-strain tensor, a is: Σ
the particle diameter, and Kc and Kμ are model constants that Σ. αφ2
2
account for the particle/particle and fluid/particle interactions, n· a2 Kμ | + a Kc φ| grad φ
respectively. γ˙ | φM − γ˙ |
φ
Σ
Following along the lines of Zhang and Acrivos (1994), an
n 2 2
extra term is added to express the sedimentation flux by means + vsφ(1 − φ) RZ + a Kcφ grad| . (9)
of the settling velocity of the non-buoyant particles owing to γ˙ |
gravity: As for the initial conditions in the tank volume (Ω), two
options were considered: all the particles are at rest and
vφ = vs(1 − φ) , nRZ
(6)
located at the tank bottom (Ωbottom) or are evenly dispersed
where vφ is the classical hindered settling velocity as defined according to the average concentration. In the first case, the
by Richardson and Zaki (1954). Unlike in Zhang and Acrivos solid volume fraction corresponds to the maximum volume
(1994), the exponent nRZ was kept constant in the computations. fraction (φm):
Its value depends on the size of the settling particles and
ranges from 4.5 for a = 100 µm, to 3 for a = 700 µm. φt=0 = φm in Ωbottom, (10)
Combining (4) and (5) leads to the following mass and
conservation equation for
Ω
the solid phase: φt=0 = 0 in Ωbottom , (11)
 Σ Σ
Dφ 2 αφ2 2 In the second case:
φt=0 = φavg in Ω, (12)
+ div[a Kc φ grad|γ˙ |] + div[v
2 2
s φ(1 − φ)
nRZ
], (7)
so that the solids mass conservation condition over the whole
domain Ω reads as:
nelt
Σ Vj
φ¯ = j
(13)
V
φt=0 t

j=
1
where Vt represents the total volume, Vj the volume of the
finite element Ωj, φt=0 the initial concentration in this element,
j
and ‘nelt’ the number of elements.
Problems (2), (3) and (7) were discretized using the finite ele-
ment method. Velocity and pressure were approximated with
the stable quadratic discontinuous-pressure P2+–P1 tetrahedral
ele- ment (Bertrand, Gadbois, & Tanguy, 1992). Such a second-
order finite element was required∇so that γ˙ in (7) could be
computed accurately. Concentration φ was approximated by
means of lin- ear P1 tetrahedral elements.
The solution of coupled Eqs. (2), (3) and (7) was obtained in
a decoupled fashion using a Picard-like successive substitution
algorithm. Given a concentration field, mixed problem (2)
and
(3) was solved for velocity and pressure by means of a
conjugate- gradient based Uzawa algorithm (Bertrand &
Tanguy, 2002). Given the velocity field, the advection-
diffusion Eq. (7) was solved for concentration using the
GMRES algorithm (Saad, 1985). This strategy was iterated
upon until a stable solution was reached. Transient simulations
Fig. 3. Contours of the particle concentration in an axially symmetric 4:1
were performed in all cases and, whenever a steady-state
expan- sion. Concentration maps of the exit flow for (a) 500 µm and (b) 100
solution existed, it was obtained as the asymptotic time
µm particles at a 50 vol.% average concentration. (Experimental data from
solution. Altobelli et al., 1997, reproduced with permission from Journal of Rheology.)
The solution of (7) requires some discussion since in some
convection-dominated situations, reaching a numerically satis-
In the present work, the validation is extended to the case
fying solution may be difficult. The Pe´clet number (Pe)
of a 4:1 sudden expansion, as shown in Fig. 3. For this simple
is a dimensionless group that expresses the importance of
convec- tive mass transfer with respect to conductive mass geometry, no analytical solution exists to describe the velocity,
transfer. For the problem at hand, it is the inverse of the so- pressure, and concentration profiles. In the expansion, the flow
called migration number (Ritz, Bertrand, Thibault, & Tanguy, is directed from right to left. The NMR experimental concen-
2000):
Nmig a trations are those of Altobelli, Fukushima, and Mondy (1997)
= . Σ2, (14)
and correspond to an average particle concentration of 50 vol.
D %.
which shows that shear-induced particle migration (diffusion) are successfully compared to the analytical solutions of Couette
is proportional to the square of the particle diameter compared and Poiseuille flows (Ritz et al., 2000).
to a characteristic length. It must be noted that Nmig appears
explicitly in the right hand side of Eq. (7) when expressed in
dimensionless form.
When diffusion dominates (Nmig 1),≥ the problem is ellip- tic
and a standard Galerkin method is adequate for solving the
conservation equation. However, when transport by convec-
tion becomes predominant, Eq. (7) becomes hyperbolic and
the Galerkin method is then known to generate solutions with
spu- rious oscillations (Johnson, 1992). In this work, a stable
and consistent SUPG method (Grygiel & Tanguy, 1991) was
used to compute the solution of (7) for all values of Nmig.

4. Validation

A quantitative validation of the above described method

has been published elsewhere, where the numerical solutions
This imaging technique outputs bitmap images that cannot
be made more precise than what is presented in their paper.
The par- ticle diameter is 500 µm in Fig. 3a, and 100 µm
in Fig. 3b. One can readily note that the model can capture
the salient features of such flow. In the case of the 500 µm
particles, the depletion in concentration in white and light
grey along the solid wall where the shear is important and
the transport of this low concentration downstream from
the expansion are clearly predicted by the sim- ulation.
This is revealed by the drop-shaped pale zone extending
from the corner after the expansion. Moreover, the increase
in concentration along the perpendicular wall behind this
depleted
zone is also captured. This increase in concentration results
from the circulation between the wall and the drop-shaped
zone. A similar particle compaction phenomenon was also
noted both experimentally and numerically in a journal
bearing apparatus (Fang & Phan-Thien, 1995; Tetlow et al.,
1998).
For the 100 µm particles, the drop-shaped zone is clearly
more slender than in the previous case. The very thin
depleted layer at the wall in the reduced-radius section,
again indicated by the pale coloration, is transported
through the expansion and
follows the same path as in the experiment. The different
shapes representing the depleted zones generated with the
small and larger particles are a good indication that the model
correctly scales according to the size of the particles. Note that
the changes in concentration are fairly steep, which renders the
solution pro- cedure more difficult but feasible with the
algorithms described in the previous section.
5. Particle migration in mixing systems
5.1. Beaker
For this experimental mixing system, the following qualita-
tive observations about the resuspending mechanism were
made and are illustrated in Fig. 4. At start-up, the top layer of
solid particles starts sliding. Under the action of the impeller,
the par- ticles pile up and are eventually dragged upward by
the flow before being propelled to the wall of the beaker.
Some of these particles get suspended in the upper part of the
fluid while others settle to the bottom. At low rotational speed,
a large proportion
Fig. 4. Resuspension mechanism at 230 rpm and 7.1 vol.% of glass beads: (a) t =0 s, (b) t = 25 s, (c) t = 50 s, (d) t = 90 s, and (e) equilibrium.
of particles remain dormant at the bottom where the volume
fraction is at its maximum (φm). This proportion of quiescent
particles decreases as the rotational speed is increased.
The mesh used for the simulations is shown in Fig. 5,
where only the elements on the impeller and the tank surfaces
are dis- played. It contains 19,879 P2+–P1 finite elements for
velocity and pressure (249,626 velocity equations) and 19,879
P1 finite elements for the concentration. A Lagrangian frame
of reference was used for the simulations, which simplified
the imposition of boundary conditions and allowed for the
solution of a steady- state problem. Convergence of the
computations was obtained after six coupling iterations of the
momentum, continuity and concentration equations. The
simulated conditions are identical to the experimental
conditions presented in Table 1. The phys- ical properties of
the dispersed phase are those of Table 2. The 1 mm particle
size combined to the characteristic dimension of the mixing
system in the order of 1 cm leads to a migration num-
ber in the order of Nmig = 10−2. For this value of the migration
number, it is expected that shear induced migration will play a
significant role on the equilibrium concentration field.

Fig. 5. Finite element mesh for the axial marine propeller system.
Fig. 6 displays the equilibrium values of the volume frac-
tion and the norm of the rate-of-strain tensor on a mid-plane,
both obtained from simulations. Also appearing in this figure
are snapshots of the particle distribution in the beaker
obtained
Fig. 6. Volume fraction of solids at equilibrium for three sets of operating conditions in the beaker: (a) 173 rpm—2.8 vol.% solids; (b) 230 rpm—7.1 vol.% solids;
(c) 350 rpm—11.9 vol.% solids.
experimentally. It is interesting to analyze these results based
on the fact that the SIMM predicts that the migration of the
parti- cles occurs from high shear regions towards low shear
regions. The first comment is that the model does not predict
with good accuracy the packing at the bottom of the beaker for
the lower rotational speeds (173 and 230 rpm). As a matter of
fact, the max- imum concentration calculated ranges between
15 and 25 vol.%, far from the theoretical value of 68 vol.%.
However, the model is able to predict the movement of the
particles towards the low shear zone under the impeller. The
calculated concentration field at 350 rpm appears to be in
closer agreement with the experi- mental results. In all cases,
the particles move under the impeller in the low shear zone.
The model also predicts a close to zero concentration at the
free surface, which is also in accordance with our
observations. Finally, the model gives the correct trend in
predicting a more effective circulation when rpm is increased.
5.2. Close clearance helical ribbon
The mesh of the helical ribbon mixing apparatus is
presented in Fig. 7. The mesh comprises 12,532 P2+–P1 finite
elements for

Fig. 7. Finite element mesh for the close clearance helical ribbon mixing
appa- ratus.
velocity and pressure (154,923 velocity equations), and
12,532 P1 finite elements for the concentration. The simulated
case cor- responds to an initial non-uniform particle volume
distribution with 5 vol.% located at the bottom, 21 vol.%
elsewhere in the volume for an average concentration of 26
vol.%. The rotational speed was constant at 16 rpm. The
simulations were performed with neutrally buoyant particles
but gravity was nevertheless considered in the simulations.
The simulated beads had a diam- eter of 100 µm to mimic the
real particles.
The motion of the solid particles is presented in Fig. 8. It
can be seen that the particles are transported by the fluid and
that there are no zones where the particles are trapped. Also
of interest is the resuspending mechanism of the solid particles
that are initially located at the bottom of the tank. These parti-
cles are dragged by the fluid up along the central shaft and
back to the bottom of the tank in the gap region between the
ribbon tip and the tank walls. Migration, if any, in this system
is unim- portant owing to the absence of dead zones and the
efficient top-to-bottom pumping features associated with this
impeller. This situation is totally different from that of the
beaker, and Fig. 8 shows the efficiency of this close-clearance
impeller sys- tem for mixing a viscous suspension. As a matter
of fact, the migration number in the case of the helical ribbon
is in the order

Fig. 8. Contours of the particle concentration in the tank equipped with a close clearance ribbon impeller (D/DT = 0.87; N = 16 rpm; particle size, 100 µm; average
concentration, 25 vol.%).

Fig. 9. Experimental and numerical torques vs. time for a close clearance heli-
cal ribbon (D/DT = 0.87; 16 rpm; particles size, 100 µm; average
concentration, 26 vol.%).
of 10−8, a value a million times lower than the one calculated
for the beaker. This very low value indicates that shear-induced
migration is extremely low and, consequently, only the global
circulation of the fluid generates noticeable particle movement.
F i g. 9 shows the evolution of experimental and numerical
torques with respect to time at start-up. In order to avoid the
complicated and often inaccurate task of summing the forces on
the surface of the impeller, the numerical torque was computed
by a macroscopic energy balance:
∫ observed here.
(ηm |γ˙ |2 ) dΩ
T= Ω . (15)
2πN
It can be noticed that, notwithstanding the oscillations in the
experimental signal, the model is capable of predicting with
good accuracy the torque with time and its final value after
300 s. This is a good indication that, for this transport-
dominated flow, the rheological behavior of the suspension is
well described by the Krieger–Dougherty model (3), all the
more so within the homogeneous layer located initially at the
bottom of the tank.
The notable difference in the torque behavior after the start-
up is an indication of some limits in the way the viscosity is
dealt with in the model. While the numerical curve flattens,
the experimental torque shows a decreasing oscillatory pattern
Fig. 11. Contour of the particle concentration generated with a large clearance helical ribbon impeller rotating at 50 rpm after 1, 20, and 100 s (D/DT = 0.4; average
concentration, 50 vol.% of neutrally buoyant particles; a, 3 mm).
Fig. 10. Finite element mesh for the large clearance helical ribbon mixing
appa- ratus.
before stabilizing, the numerical value then being close to the
experimental one. Two aspects of the model should indeed be
considered here. First, the solid phase is considered as a
contin- uum and nothing allows for the creation of clusters of
particles that may arise from strong particle–particle
interactions. It is likely that the real system is in fact not
uniform in concentration immediately after start-up and that
some particle agglomer- ates are initially present and partly
responsible for the torque variations. Moreover, the way
torque measurements are made, without short time averaging
for instance, may also have an influence on the torque versus
time curve shape and generate oscillations like the ones
5.3. Large clearance helical ribbon
In this experiment, pumping is expected to be less efficient
than with the close clearance ribbon, leaving poorly mixed
regions along the tank walls because of the reduced size of the
impeller. Consequently, particles should migrate from the
high- est shear rate region swept by the helical ribbon towards
the periphery of the tank where the lowest shear rates are
experi- enced.
The mesh of the large clearance geometry is presented in
Fig. 10. It contains 18,602 P2+–P1 finite elements for velocity
and pressure (233,779 velocity equations), and 18,602 P1
finite elements for the concentration (3714 equations). The
simulated
Fig. 12. Evolution of the envelope circumscribing the region within which the concentration reaches 45 vol.% in solids at times 1, 20, and 100 s and 50 rpm in the
large clearance ribbon impeller apparatus (D/DT = 0.4; average concentration, 50 vol.%; a, 3 mm).

Fig. 13. Concentration profiles in the large clearance helical ribbon apparatus (D/DT = 0.4; 50 rpm; average concentration, 26 vol.%). Top, 300 µm; bottom, 3 mm.
case corresponds to an initial uniform particle volume distribu-
tion of 50 vol.%. The rotational speed was constant at 50 rpm.
The simulations were performed with neutrally buoyant parti-
cles but gravity was however included in the calculations. The
simulated beads had a diameter a of 3000 µm.
Shear induced particle migration is evidenced in Fig. 11,
which depicts concentration contours at three different times.
The resulting concentrations vary from 25 to 59 vol.% in the
most concentrated regions. This is close to the maximum
pack- ing limit of 68 vol.%. At equilibrium, the particles are
located in low shear regions, that is near the walls and around
the agitation shaft leaving a cavern-like region with a lower
average concen- tration where mixing is intense. In fact, it can
be observed that the lowest concentration occurs in the
vicinity of the impeller, as expected.
Fig. 12 shows the variation with time of the envelope
within which the concentration is equal to 45 vol.% in solids.
One can readily see that the particles escaping the high shear
rate are near the impeller and migrate towards the lower shear
rate zones so that the envelope becomes larger and larger until
it reaches its maximum size at equilibrium.
Fig. 13 shows the effect of the particle size (3 and 0.3 mm,
respectively) on the migration phenomena for an average con-
centration of 26 vol.% and the same impeller speed of 50 rpm.
The 3 mm particle size was chosen so that the migration num-
ber would be a hundred times larger than for the 0.3 mm case.
The results point out that shear-induced migration also occurs
with smaller particles although with much less significance.
The upper part of the figure (a = 300 µm) reveals a
relatively
uniform concentration ranging from 24 to 28 vol.%, the low-
est and highest concentrations being located at the same place
as in the case of large particles (a = 3 mm) presented on the
lower part of the figure. In this latter case, the concentra-
tions vary from 13 up to 33 vol.%, which gives a range that
is five times wider than that with smaller particles. The com-
parable torque values obtained from the two cases seem to
indicate that torque is not significantly affected by the migra-
tion phenomenon. This question will need to be investigated
thoroughly.
6. Additional remarks and conclusion
This paper aimed at simulating the flow of suspensions in
stirred tanks using an implementation of the model of Phillips
et al. (1992). For this purpose, a numerical method based on a
macroscopic model and the finite element method was
validated and used.
The model was developed to take into account the
migration of particles due to particle collisions and viscosity
variations. From this investigation, it can be concluded that the
model is able to fairly predict the average behavior of a
suspension in a three dimensional situation.
The calculations performed during this research required
CPU times in the order of days and typically ran on machines
like or comparable to IBM RS/6000 pSeries 630 with AIX 5.1
as OS. The problems size was always in the order of 150MB
of RAM for solving 150,000 equations.
The results obtained in this work revealed that the model is
adequate for predicting the transport-dominated flow in a
small- gap (close-clearance) system. On the other hand,
simulations performed in the case of a large gap between a
helical ribbon and the tank walls have shown that the SIMM
need to be improved. Rao, Mondy, Sun, et al. (2002), and Rao,
Mondy, Baer, et al. (2002) expressed some concerns about the
capability of this model to provide acceptable solutions based
on the observation that, in wide-gap Couette geometries, the
shear rate and the vis- cosity terms are both large in magnitude
and of opposite signs. The predicted results become very
sensitive to errors, which may lead to large discrepancies with
respect to the experimental flow and concentration fields. To
overcome this weakness in the model, Fang and Phan-Thien
(1995) used a ‘flow aligned ten- sor’ to render the particle
migration more anisotropic. However, this addition did not
improve the behavior of the model close to the walls where
migration is known to be the most intense. It is also in these
regions that the simple particle–particle collision model of the
SIMM hits its validity limit due to the intense multi- body
collisions taking place. This limitation was also noted in
Hampton et al. (1997). The second aspect of the model that needs
to be discussed is that related to its potentially erratic behavior
in close-to-zero shear rate and high concentration zones.
Based on the results shown here in which the maximum
packing is never reached while the experimental results show
the opposite, it is evident that the correct description of the
maximum pack- ing remains a problem. Future investigations
should point at the viscosity model, the only way maximum
packing can interfere with the concentration field calculations.
Finally, it was noticed in this work that the numerical
torque values could be made closer to the experimental ones
by arbitrar- ily changing the values of the empirical constants
Kc and Kμ. However, doing so resulted in concentration
profiles that did not comply with the experimental
observations. Moreover, and despite the very interesting work
of Tetlow et al. about the vari-
ation of the parameters with respect to concentration (Tetlow
et al., 1998), changing of these parameters in order to better fit
the experimental values is not a satisfying solution from a
scientific standpoint. Rao, Mondy, Sun, et al. (2002), and Rao,
Mondy, Baer, et al. (2002) observed recently that the shear
parameter (Kμ in Eq. (5)) is the one that must be modified to
simulate more accurately the flow of a suspension in a large
Couette device.
In addition, it must be said that the presence of zones of
very low velocity combined to concentrations close to
maximum packing fraction may generate important numerical
convergence problems leading to non-physical results. The
maximum con- centration can indeed be exceeded in those
regions because the model does not prevent the concentration
from going above the selected maximum packing fraction.
The only protection against concentration overshoots comes
from the rheological model that however becomes ineffective
when the velocities are very small. The settling of particles by
gravity in low velocity zones only augments this problem.
Another aspect that will need to be addressed following the
work of Kim (2001) is the influence of the rheological
behavior of the dispersing medium on the migration
phenomena. Two additional issues are worth investigating,
namely the values of
the constants Kc and Kμ in the SIMM – Tetlow reported better
modeling results with a linear variation of their ratio (Tetlow
et al., 1998) with concentration – and the formation of particle
clusters on the migration phenomenon.
Acknowledgements
The financial assistance of NSERC is gratefully acknowl-
edged. The authors also want to thank Dr. Steve Altobelli for
kindly providing the original pictures from his work.
References
Altobelli, S. A., Givler, R. C., & Fukushima, E. (1991). Velocity and
concentra- tion measurements of suspensions by nuclear magnetic
resonance imaging. Journal of Rheology, 35(5), 721–735.
Altobelli, S. A., Fukushima, E., & Mondy, L. A. (1997). Nuclear magnetic
res- onance imaging of particle migration in suspensions undergoing
extrusion. Journal of Rheology, 41, 1105–1115.
Bertrand, F., & Tanguy, P. A. (2002). Krylov-based Uzawa algorithms for the
solution of the stokes equations using discontinuous pressure tetrahedral
finite elements. Journal of Computational Physics, 181, 617–638.
Bertrand, F., Gadbois, M., & Tanguy, P. A. (1992). Tetrahedral elements for
fluid flow. International Journal for Numerical Methods in Engineering,
33, 1251–1257.
Bertrand, F., Tanguy, P. A., & Thibault, F. (1997). A three-dimensional
fictitious domain method for incompressible flow problems. International
Journal for Numerical Methods in Fluids, 25, 719–736.
Brady, J. F., & Bossis, G. (1985). The rheology of concentrated suspensions
of spheres in simple shear flow by numerical simulation. Journal of Fluid
Mechanics, 155, 105–129.
Fang, Z., & Phan-Thien, N. (1995). Numerical simulation of particle
migration in concentrated suspensions by a finite volume method. Journal
of Non- Newtonian Fluid Mechanics, 58, 67–81.
Gadala-Maria, F., & Acrivos, A. (1980). Shear-induced structure in a
concentrated suspension of solid spheres. Journal of Rheology, 24, 799–
814.
Glowinski, R., Pan, T.-W., & Perieux, J. (1994). A fictious domain method for
Dirichlet problem and applications. Computer Methods in Applied
Mechan- ics and Engineering, 111, 283–303.
Grygiel, J. M., & Tanguy, P. A. (1991). Finite element solution for advection-
dominated thermal flows. Computer Methods in Applied Mechanics and
Engineering, 93, 277–289.
Hampton, R. E., Mammoli, A. A., Tetlow, N., & Altobelli, S. A. (1997).
Migra- tion of particles undergoing pressure-driven flow in a circular
conduit. Journal of Rheology, 41, 621–640.
Hu, H. H. (1996). Direct simulation of flows of solid–liquid mixtures.
Interna- tional Journal of Multiphase Flow, 22(2), 335–352.
Hu, H. H., Joseph, D. D., & Crochet, M. J. (1992). Direct simulation of fluid
particle motions. Theoretical and Computational Fluid Dynamics, 3, 285–
306.
Jana, S. C. (2003). Loss of surface and volume electrical conductivities in
poly- mer compounds due to shear-induced migration of conductive
particles. Polymer Engineering and Science, 43(3), 570–579.
Jana, S. C., Kapoor, B., & Acrivos, A. (1995). Apparent wall slip velocity
coef- ficients in concentrated suspensions of non-colloidal particles.
Journal of Rheology, 39(6), 1123.
Johnson, C. (1992). Numerical solution of partial differential equations by the
finite element method. Cambridge University Press.
Kim, C. (2001). Migration in concentrated suspension of spherical particles
dis- persed in polymer solution. Korea-Australia Rheology Journal, 13, 19–
27. Krieger, I. M., & Dougherty, T. J. (1959). A mechanism for non-
Newtonian flow in suspension of rigid spheres. Transactions of the Society of
Rheology, 3,
137–152.
Leighton, D., & Acrivos, A. (1986). Viscous resuspension. Chemical
Engineer- ing Science, 41(6), 1377–1384.
Leighton, D., & Acrivos, A. (1987a). The shear-induced migration of particles
in concentrated suspensions. Journal of Fluid Mechanics, 181, 415–439.
Leighton, D., & Acrivos, A. (1987b). Measurement of shear-induced self-
diffusion in concentrated suspensions of spheres. Journal of Fluid
Mechan- ics, 177, 109–131.
Mann, R., & Hackett, L. A. (1988). Fundamentals of gas–liquid mixing in a
stirred vessel: An analysis using network of backmixed zones. In
Proceed- ings of the sixth European conference on mixing (pp. 321–328).
Maury, B., & Glowinski, R. (1997). Fluid-particle flow: A symmetric
formula- tion. Me´thode Nume´riques de la Me´canique, 1–6.
Phillips, R. J., Armstrong, R. C., Brown, R. A., Graham, A. L., & Abott, J. R.
(1992). A constitutive equation for concentrated suspensions that accounts
for shear-induced particle migration. Physics of Fluids A, 4, 30–40.
Rao, R., Mondy, L., Sun, A., & Altobelli, S. (2002). A numerical and experi-
mental study of batch sedimentation and viscous resuspension.
International Journal for Numerical Methods in Fluids, 39, 465–483.
Rao, R. R., Mondy, L. A., Baer, T. A., Altobelli, S. A., & Stephens, T. S.
(2002). NMR measurements and simulations of particle migration in non-
Newtonian fluids. Chemical Engineering Communications, 189(1), 1–22.
Richardson, J. F., & Zaki, W. N. (1954). Sedimentation and fluidisation: Part 1.
Transactions of the Institution of Chemical Engineers, 32, 35–53.
Ritz, J.-B., Bertrand, F., Thibault, F., & Tanguy, P. A. (2000). Shear-induced
particle migration in a short-dwell coater. Chemical Engineering Science,
55, 4857–4867.
Saad, Y. (1985). GMRES: A generalized minimal residual algorithm for
solving non-symmetric linear systems. SIAM Journal of Scientific and
Statistical Computing, 7, 850–869.
Subia, S. R., Ingber, M. S., Mondy, L. A., Altobelli, S. A., & Graham, A. L.
(1998). Modeling of concentrated suspensions using a continuum constitu-
tive equation. Journal of Fluid Mechanics, 373, 193–219.
Tanguy, P. A., Bertrand, F., Thibault, F., & Galy-Jammou, P. (2001).
Simulation of kneading mechanisms in a three blade planetary mixer,
Enyclopedia of materials science and technology. Elsevier, pp. 7425–
7428.
Tanguy, P. A., Thibault, F., Ascanio, G., & Brito-De La Fuente, E. (2006).
Solid- liquid mixing: Numerical simulation and physical experiments. In
Sunggyu Lee (Ed.), Enc. Chem. Proc. (pp. 2753–2768). M. Dekker.
Tetlow, N., Graham, A. L., Ingber, M. S., Subia, S. R., Mondy, L. A., &
Altobelli,
S. A. (1998). Particle migration in a Couette apparatus: Experiment and
modeling. Journal of Rheology, 42, 307–327.
Zhang, K., & Acrivos, A. (1994). Viscous resuspension in fully developed lam-
inar pipe flows. International Journal of Multiphase Flow, 20(3), 579–591.
Zhou, G., Tanguy, P. A., & Dubois, C. (2000). Power consumption in a double
planetary mixer with non-Newtonian and viscoelastic materials. Chemical
Engineering Research and Design, 78, 445–453.