Chemical Engineering Journal 223 (2013) 875–890

Contents lists available at SciVerse ScienceDirect

Chemical Engineering Journal

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

CFD simulations of dense solid–liquid suspensions in baffled

stirred tanks: Prediction of solid particle distribution
A. Tamburini a, A. Cipollina a, G. Micale a,⇑, A. Brucato a, M. Ciofalo b
a
Dipartimento di Ingegneria Chimica, Gestionale, Informatica, Meccanica, Università di Palermo, Viale delle Scienze Ed. 6, 90128 Palermo, Italy
b
Dipartimento dell’Energia, Università di Palermo, Viale delle Scienze Ed. 6, 90128 Palermo, Italy

h i g h l i g h t s

 Solid–liquid suspensions in stirred tanks are investigated by CFD.

 Solid particle distribution phenomenon is studied under partial-to-complete suspension conditions.
 The validity of neglecting the radial profiles of solids concentration is extensively evaluated.
 For the first time, axial profiles of solids concentration under partial suspension conditions are predicted via CFD.

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

Article history: Industrial tanks devoted to the mixing of solid particles into liquids are often operated at an impeller
Received 3 August 2012 speed N less than Njs (defined as the lowest speed allowing the suspension of all particles): under such
Received in revised form 9 January 2013 conditions the distribution of solid-particles is very far from being homogeneous and very significant
Accepted 12 March 2013
concentration gradients exist. The present work is devoted to assessing the capability of Computational
Available online 19 March 2013
Fluid Dynamics (CFD) in predicting the particle distribution throughout the tank.
The CFD model proposed by Tamburini et al. [58] and successfully applied to the prediction of the sed-
Keywords:
iment amount and shape was adopted here to simulate the particle distribution under partial-to-com-
Multi Fluid Model (MFM)
Computational Fluid Dynamics (CFD)
plete suspension conditions. Both transient (via the Sliding Grid approach) and steady state (via the
Turbulence closure Multiple Reference Frame approach) RANS simulations were carried out for the case of a flat bottomed
Solid–liquid suspension baffled tank stirred by a Rushton turbine.
Partial suspension Results show that the model can reliably predict the experimental particle distribution at all investi-
Drag force gated impeller speeds. Transient simulations were found to predict slightly better the experimental data
Stirred tank with respect to steady state simulations. Radial gradients of solids concentration, usually neglected in the
Particle distribution literature, where found to be significant in the presence of unsuspended solid particles settled on the ves-
Unsuspended Solid Criterion (USC)
sel bottom (i.e. incomplete suspension conditions).
Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction
Solid–liquid mixing within tanks agitated by stirrers can be eas-
ily encountered in many industrial processes. It is common to find
an industrial tank operating at an impeller speed N lower than the
minimum agitation speed for the suspension of all solid particles,
Njs [44,49,59]. As a matter of fact, the loss in active interfacial area
due to an impeller speed N below Njs as an operating condition is
somewhat counterbalanced by a lower power requirement (the
power is proportional to the cube of the impeller speed) which
may ultimately result in a more economically competitive process.
Nevertheless, most literature works focus on the assessment
[71,41,2,20,7,60,64] and many others or on the prediction of Njs
⇑ Corresponding author. Tel.: +39 09123863780.
E-mail address: giorgiod.maria.micale@unipa.it (G. Micale).
[22,40,61]. Only recently, attention has been paid to the incom-
plete suspension conditions. In particular Tamburini et al. [58] pro-
posed a CFD model able to reliably predict the suspension curves
(for different particle sizes and concentrations), i.e. the mass frac-
tion of suspended particles (xsusp) as a function of N, on the basis of
a suitable computation of the quantity of solids resting motionless
on the vessel bottom. This model was found also to be able to pre-
dict Njs [61] and the impeller speed for sufficient suspension con-
ditions Nss [63]. Most solid–liquid mixing operations require a
knowledge of the impeller speed at which the solid particles be-
come fully suspended and/or the amount of solids suspended at
different N below Njs. In many cases information is also required
on the quality of the solids distribution within the tank, since the
particle distribution may largely affect the process performance.
In such cases, a reliable prediction of the solids distribution is of
crucial importance for accurate design and testing of the pertaining

1385-8947/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.cej.2013.03.048
876 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890
Nomenclature
A impeller blade width (m)
B impeller blade height (m)
C impeller clearance (m)
CD drag coefficient (–)
D impeller diameter (m)
dp particle mean diameter (m)
ft correction factor due to the effect of liquid background
turbulence on CD (–)
H liquid height (m)
k turbulent kinetic energy (m2 s2)
M inter-phase momentum transfer term (N m3)
N rotational impeller speed (rpm)
nr number of computational cells along the radial direction
(–)
Njs just suspension speed (rpm)
Nss sufficient suspension speed (rpm)
r volumetric fraction (–)
rb_av average solid volumetric fraction value (–)
rb_r-av radially averaged solid volumetric fraction value (–)
rb_packed solid volumetric fraction maximum packing value (–)
R radial coordinate (m)
solid–liquid stirred systems [57,65]. Surprisingly, it is not easy to
find such data in the literature for partial suspension conditions,
despite the interest expressed so far at the industrial level for this
particular regime [19].
Experimental data on particle distribution in a liquid within a
stirred tank are usually presented in the form of axial and radial
profiles of solids concentration. Such data were collected in the
past by employing sampling withdrawal techniques [8,4]. Also op-
tic [1], conductivity [45,54] and acoustic probes [3] were used to
measure point-wise particle concentration. In recent years non-
intrusive techniques were employed: in particular tomographic
techniques, among which the Electrical Resistance Tomography
(ERT) [31,13], can provide information on particle concentration
distribution throughout the vessel under either complete or partial
suspension conditions, as recently documented in the literature
[17].
Particle distribution can be also measured by means of non
intrusive techniques as the light attenuation technique [5,28,36].
This technique makes use of a light emitting diode as light source
and a silicon photo-diode as receiver and measures the attenuation
that a light beam undergoes while crossing (horizontally along a
chord) the vessel. The results obtainable by this method concern
the average particle concentration along the line covered by the
beam. Such average particle concentration measurements along
the chord have been considered, though as an approximation [1],
to be representative of the average concentration in the whole hor-
izontal section, on the basis of the assumption that radial profiles
were fairly flat [4]. This last specific approximation can be easily
found in the literature [28–30]. Micheletti et al. [35] proposed a
method to assess the suspension curve. The particle concentration
at a specific radial and azimuthal location and at different axial
positions were measured by means of a conductivity probe. They
integrated all local concentrations measured above the unsuspend-
ed solids bed by assuming that each local concentration was repre-
sentative of the average solids concentration at the relevant vessel
cross section.
The distribution of solid particles in a stirred vessel is a quite
complex function of velocity field, turbulence characteristics and
liquid-particle interactions. Thus, the soundness of the former
T tank diameter (m)
U velocity (m s1)
W baffle width (m)
xsusp mass fraction of suspended solids (–)
Z axial coordinate (m)
Greek letters
e turbulent dissipation (W kg1)
k Kolmogorov length scale (m)
l viscosity (Pa s)
m kinematic viscosity (m2 s1)
q density (kg m3)
r variation coefficient (–)
rr radial variation coefficient (–)
1 suspension quality (–)
Subscripts
a liquid phase
b solid phase
approximation depends on several factors, such as geometrical
configuration and suspension properties: for example radial impel-
lers provide larger concentration gradients than axial impellers,
also, the higher the particle size and concentration, the higher
the concentration gradients [4].
The present work is devoted to the investigation via CFD of the
particle distribution in a dense suspension ranging from partial to
complete suspension conditions. In particular, the CFD model by
Tamburini et al. [58] is the only one purposely developed to deal
with partial suspension conditions. It has been fully validated in
previous works and found to reliably predict integral data in the
form of (i) suspension curves [58], (ii) Njs [61] and (iii) Nss [63].
Such essential data concerning the particle suspension phenome-
non, however, do not provide any information on local details since
they are intrinsically lumped. Investigation of particle distribution
completes the description of the solid–liquid suspension by adding
a further level of detailed information throughout the whole vessel
volume. Here, the model by Tamburini et al. [58] is tested in order
to evaluate its capability to deal with local particle concentration
distribution under incomplete suspension conditions. Notably, to
the authors’ knowledge, no literature work addressed this specific
topic so far: all the CFD models proposed in the literature are gen-
erally validated against experimental axial profiles of solids con-
centration collected at impeller speeds equal or higher than Njs
(as examples see [33,36,69,37,38,43,21,70]).
2. System investigated
The data employed for the validation of the CFD simulations
derive from the literature [35]. In particular, a flat bottomed
vessel with a diameter T = 0.29 m was investigated. The liquid
level height H was equal to T. The tank was stirred by a
six-bladed Rushton turbine with a diameter D equal to T/3. The
offset of the impeller from the vessel bottom C was equal to
T/3. A sketch of the system under investigation is depicted in
Fig. 1. A dense solid–liquid suspension was studied: distilled
water (qa = 1000 kg/m3) and mono-dispersed glass particles
(dp = 600–710 lm; qb = 2470 kg/m3) with a concentration of 25%
wsolid/wliquid were used for all cases.
 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890 877

W=T/10 The experimental data to be predicted are axial profiles of solid

concentrations. More precisely, local steady state normalized sol-
ids concentrations measurements at a radial position R/T = 0.35
(see Fig. 2), midway between baffles and at different heights of
the tank, were collected by Micheletti et al. [35] using a conductiv-

B=D/5
ity probe.
H=T

Specific measurements were carried out to evaluate the liquid

volume fraction of the particle bed lying on the bottom under no
agitation conditions: this value was found to be about 40%. The
minimum impeller speed for complete suspension according to
C=T/3

Zwietering’s criterion was estimated by Micheletti et al. [35] and

found equal to 988 rpm.
Further details on the geometrical and physical features of this
D=T/3 system can be found in Micheletti et al. [35].

T
3. Modelling

Two different approaches can be adopted to simulate solid–

liquid suspensions. Some researchers employ the Eulerian–
Lagrangian approach, based either on RANS models or on
Large Eddy Simulation (LES) for the continuous phase and on a
Lagrangian model for tracking the particles [11]. This method pre-
sents the advantage of using a more accurate approach for solving
A=D/4 the fluid N–S equations, thus providing potentially better results.
However, computational requirements are very large and directly
related to the number of particles to be individually tracked.
Other approaches are based on Eulerian–Eulerian frameworks
T = 0.29 m along with the Multi Fluid Model (MFM). The two phases are trea-
ted as two interpenetrating continua: the continuity and momen-
Fig. 1. Micheletti et al. [35] system.
tum equations are solved for each phase, thus obtaining separate
flow field solutions for the liquid and the solid phase simulta-
neously. The two phases share the same pressure field. MFM mod-
els are usually preferred for their simplicity [42]; [34,38,56], which
allows lower computing requirements.
Accounting for a distribution of particle sizes in the Euleri-
an–Eulerian Multi Fluid Model would require the definition of
a number of different solid phases with different characteristic
diameters thus leading to very large computational require-
ments. In the present work, only mono-dispersed glass particles
were simulated (dp = 600–710 lm) thus limiting the associated
complications.
Reynolds Averaged Navier Stokes simulations were carried out
by means of the commercial CFD code AnsysÒ CFX4.4 [9].
A number of modelling approaches have been tested in the
present work in order to predict the particle distribution under
partial-to-complete suspension conditions: the main features of
such models and results are reported in Tab. 1. Only some of these
(including the best ones) are shortly described in the present pa-
per. Further details can be found in Tamburini [55] and in Tambu-
rini et al. [58].
Also, the continuity and momentum equations are the same re-
ported in the paper by Tamburini et al. [58] and are not presented
here for the sake of brevity.
In the momentum equation a molecular viscosity l equal to
that of the liquid was chosen for the solid phase as suggested by
the literature [34,40].
The momentum inter-phase transfer term M was considered to
be equal to the drag force, while contributions due to other forces
(such as the Basset, lift and virtual mass forces) were neglected as
suggested by the literature [25,14,38,40,46,21,56,15,18,50,52]. In
particular, Kasat et al. [21] stated that the Basset force is much
smaller than the inter-phase drag force. Similarly, according to
Tatterson [66], virtual mass and lift force should be considered
as much smaller than the drag force and can be neglected when
Fig. 2. Monitoring position of particle concentration. qb/qa > 2, as it occurs in the present work.
878 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890

A standard formulation was adopted for the drag force:
 
~ ab 3 C D;slip ft ~b  U
~ a j ðU
~b  U
~a Þ
M ¼ rb qa jU
4 dp
where the subscripts a and b refer to the continuous and dispersed
phases respectively, r is their volumetric fraction, q is density, CD is
particle drag coefficient, U is mean velocity and ft is a correction fac-
tor accounting for the influence of liquid background turbulence on
the particle drag coefficient. The particle drag coefficient CD,slip was
estimated by means of the Clift et al. [10] correlation on the basis of
the slip velocity between the two phases.
A universally validated and accepted correlation for the modifi-
cation of the particle drag coefficient due to the liquid background
turbulence is lacking in the literature on solid–liquid suspensions.
Brucato et al. [6] and Pinelli et al. [48] correlated the drag correc-
tion with the ratio between particle size and Kolmogorov length
scale, and either of their correlations has been adopted in a very
large number of papers on suspensions. The two correlations,
which were both tested in the present work, are:
Brucato et al. [6]:
" low values for all the cases investigated. Only for the case with
 3 #
dp
ft;Brucato ¼ 1 þ 8:76  104
k
Pinelli et al. [48]:
  2
k were considered sufficient to reach steady state conditions in all
ft;Pinelli ¼ 0:6 þ 0:4 tanh 16  1
dp
In which k is the Kolmogorov length scale, calculated by
k ¼ ðm3 =eÞ0:25
The local dissipation of turbulent kinetic energy e is provided by the
turbulence model.
Other papers exist [24,12] where some dependence on either
the Stokes number or the Richardson number is invoked.
All these studies refer to highly dilute suspensions, while no
experimental study on dense suspensions (and therefore on the
possible interaction between background turbulence correction
and hindered settling) has been conducted so far.
However, Khopkar et al. [23] investigated the phenomenon via
CFD: they simulated the flow of water through an array of cylin-
ders, with different packing features representing the particle con-
centration and computed the drag of the cylinders. The
Kolmogorov length-scale was controlled by imposing a turbulence
energy source term. They proposed a correlation having the same
form as that of Brucato et al. [6], but with a coefficient reduced
by a factor of 10; no clear packing effects were found so that the
correlation does not include the particle concentration. Despite
the geometrical difference between the above 2-D geometry and
a real particle suspension [15], this modified correlation was em-
ployed also in other papers [21,51,68]. However, this crude reduc-
tion of the corrective coefficient does not appear to be physically
justified and was not tested in the present simulations.
The asymmetric ke turbulence model [58] was employed: only
the turbulence of the liquid phase was accounted for, while no tur-
bulence was assumed in the solid phase. This choice had previously
been found suitable to model dense solid–liquid suspensions in
stirred tanks under partial-to-complete suspension regimes
[55,58]. The relevant equations can be found in Tamburini et al.
[58].
diffusive terms. The QUICK discretization scheme was employed
for the convective terms. Notably, similar results (not shown in
ð1Þ the following for brevity) were obtained by employing the simpler
hybrid-upwind differencing scheme, which was also tested.
As concerns the treatment of the impeller-baffle relative rota-
tion, both the steady state Multiple Reference Frame (MRF) by
Luo et al. [26] and the time dependent Sliding Grid (SG) algorithm
by Murthy et al. [39] were adopted in the present work.
The position of both the sliding surface (SG) and the surface
separating the two reference frames (MRF) was set in accordance
with the assumptions of Luo et al. [26] at the radial position
0.62T/2.
For all the SG simulations a time step equal to the time the
impeller needs to sweep an azimuthal angle equal to a cell (one cell
time step) was adopted. Such a choice corresponds to a Courant
number equal to 1 along the azimuthal direction. A time step half
as large was also tested, but practically identical results were
found [58].
In MRF simulations typically 12,000 SIMPLEC iterations were
found to be sufficient to allow variable residuals to settle to very
the lowest impeller speed (i.e. 400 rpm) some additional iterations
ð2Þ (4000 was more than sufficient) were necessary to allow the ESVC
algorithm to operate a sufficient number of times and thus avoid-
ing any over-packing issue.
As far as the SG simulations are concerned, 100 full revolutions
ð3Þ
cases, consistently with what can be found in the literature for
similar systems [34,56,58]. The number of SIMPLEC iterations per
time step was set to 30 with the aim of allowing residuals to settle
ð4Þ before moving to the next time step.
As regards the initial condition, particles were considered to be
motionless on vessel bottom and at their maximum packing vol-
ume fraction rb_packed.
For both phases no slip boundary conditions were assumed for
all tank boundaries with the exception of the top surface where
free slip conditions were imposed.
As far as the system’s discretization is concerned a structured
grid totally composed of hexahedrons was employed according
to the simple geometry of the system under investigation (Rushton
impellers geometry is very simple with respect to, for instance,
hydrofoil impellers) and in order to keep low the number of com-
putational cells [52,62]. In accordance with system’s axial symme-

4. Numerical details

The segregated SIMPLEC algorithm was adopted to couple pres-

sure and velocity. Central differences were employed for all Fig. 3. The computational grid employed.
 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890
try, only one half of the tank was included in the computational do-
main and two periodic boundaries were imposed along the azi-
muthal direction. The structured grid chosen for the
discretization of this half-tank encompasses 74,592 cells distrib-
uted as 72  37  28 along the axial, radial and azimuthal direc-
tion respectively (Fig. 3).
The computational grid is finer in the vicinity of the impeller
where the largest gradients of velocities and other flow variables
are expected. In the SG simulations, the above mentioned 100
impeller revolutions correspond to 5600 time steps, resulting in
168,000 SIMPLEC iterations, a remarkable number in comparison
with the 8000 SIMPLEC iterations of MRF simulations.
Grids up to eight times finer were employed to simulate the
system under investigation and no appreciable differences in the
results were found [55]. In particular the mean value of the dis-
crepancy between the finest grid and the grid adopted here was
found to be 1.0% for rb which is the main variable taken into ac-
count in Section 5.
Under highly incomplete suspension conditions, overpacking is-
sues may occur during the simulation thus leading to unphysical
results: when no solid pressure term is included in the solid phase
momentum equation, the solids volume fraction may exceed the
maximum physically allowed rb_packed value (=0.6). In order to
solve this problem, in the present work alternative approaches
were tested, as reported in Table 1 (simple solid pressure model,
granular kinetic theory), leading to the adoption of the Excess Solid
Volume Fraction Correction (ESVC) algorithm (proposed by Tambu-
rini et al. [56] and successively optimized by Tamburini et al. [58]).
Only a negligible amount of solid-excess was found in the simula-
tions even at the very lowest impeller speeds: in particular in all
cases and inside all computational cells rb was always found to
be lower than 0.605.
When partial suspension conditions are dealt with, it is also
necessary to allow the code to distinguish the suspended particles
from the unsuspended ones. The Unsuspended Solids Criterion
(USC) by Tamburini et al. [58] addresses this issue and it was con-
sequently employed in the present work.
Both the ESVC and USC algorithms were found suitable to deal
with solid–liquid suspensions in baffled stirred tank under partial
to complete suspension regimes [58,61]. Other details on these
algorithms can be found in previous works [58,61].
5. Results and discussion
5.1. Choice of the impeller modelling technique
As already mentioned in the introduction section, the model
proposed by Tamburini et al. [58] was found able to predict the
amount and the shape of settled solids, as well as the minimum
impeller speed for complete suspension. In the present work it is
further tested with the aim of evaluating its capability to predict
also local quantities, i.e. the local axial profiles of solids concentra-
tion at different impeller speeds ranging from partial to complete
suspension conditions.
Notably, these CFD simulations represent the first attempt in
the literature to predict the solid distribution in a stirred tank un-
der partial suspension conditions.
Simulation results are presented in Fig. 4 for the case described
in Section 2.
At 400 rpm (solid line in Fig. 4a), i.e. an impeller speed much
lower than Njs, the model along with the SG algorithm manages
to predict with high accuracy the experimental profile: only slight
differences among CFD results and experiments are observable.
The presence of particles near the vessel bottom with a volumetric
fraction corresponding to the maximum allowed one (i.e.
879
rb_packed = 0.6) is very well predicted by CFD simulations, thus con-
firming the effectiveness of the ESVC algorithm. Actually, the CFD
simulation for the 400 rpm cases slightly underestimates the par-
ticle concentration in the upper region of the vessel (Z/H P 0.4)
with respect to the experimental data.
Similar considerations can be inferred from the relevant results
shown in Fig. 5a where the solids resting on the bottom: most solid
particles are not suspended laying on the vessel bottom (sediment)
are shown in grey: the sediment appears to be only slightly drawn
by the flow field; also, very few particles are suspended and they
do not reach a height higher than the impeller plane.
The steady state simulation performed via the MRF approach for
the same test case (dashed line in Fig. 4a) provides results quite
similar to the SG ones, especially near the free surface (i.e. Z/
H > 0.7) and bottom of the vessel (i.e. Z/H < 0.1), conversely, some
differences can be observed in the vicinity of the impeller plane:
in particular the experimental data are underestimated by the
MRF simulation at Z/H = 0.2, also, the MRF profiles shows a spike
in concentration at about Z/H = 0.35T not exhibited by the experi-
mental profile. Notably, the results relevant to the SG approach
corresponds to an instantaneous system position when the impel-
ler blade is midway between two subsequent baffles; the frozen
positions investigated by the MRF is equal to the SG instantaneous
one for comparison purposes. However, when local axial profiles
are obtained via SG by averaging data over an entire impeller rev-
olution no differences were found (Fig. 6).
At 500 rpm (Fig. 4b) the SG-model predicts the experimental
data with a fair accuracy: the figure shows that the experimental
axial trend of solids concentration is correctly predicted at all ves-
sel heights but at Z/H = 0.1, where a lower value of rb was found.
Conversely, the MRF fails to predict the experimental profile of
rb: the MRF results largely over-predict the experimental rb values
above the impeller plane and largely under-predict them below the
impeller plane.
As concerns the unsuspended particle distribution throughout
the whole tank, clearly the amount of sediment is reduced as the
impeller rotational speed increases from 400 rpm to 500 rpm. As
it can be seen in Fig. 5b, at 500 rpm more particles are suspended
and some of them reach the upper part of the vessel: this is in
agreement with the local data provided in Fig. 4b. Also, at this
speed the flow discharged by the impeller starts to significantly
profile the sediment shape: near the lateral wall the sediment is
present only on the trailing side of the baffles which shield it from
the water flow.
At 600 rpm (Fig. 4c), below the impeller plane both the SG and
the MRF approaches yield a large underestimation of the experi-
mental data, which may be related either to an inadequate model-
ling of the inter-phase forces or to an inaccurate prediction of the
details of the sediment shape. In the upper part of the vessel only
very slight differences between experimental data and computa-
tional SG results are observable. In this region, MRF profiles are dif-
ferent from SG ones and in worse agreement with the
experimental data.
At 700 rpm (Fig. 4d) a good agreement between the experimen-
tal profile and the SG-model predictions is visible above the impel-
ler plane: only the experimental point at the highest elevation is
not perfectly predicted by the CFD simulation. Below the impeller
plane, the figure shows an underestimation compared with the
experimental data whose amplitude appears to be lower than in
the 600 rpm case. The MRF model results exhibit a slightly larger
underestimation of the experimental data below the impeller
plane and a wrong shape of the concentration profile above it.
The Njs calculated by Micheletti et al. [35] by means of Zwieter-
ing’s correlation was found to be 988 rpm so that all the experi-
mental profiles discussed so far are relevant to partial suspension
conditions. On the other hand, the impeller speed of 1100 rpm
 880

Table 1
Modelling approaches tested for the predictions of suspension curves.

Case Turbulence dispersion Turbulence model Hindered settling modelling Free stream turbulence Overpacking issues treatment Particle distribution prediction results
number correction on the base
drag coefficient
None Addtional terms Asymmetric Homogeneous None Gidaspow’s Piecewise None Brucato Pinelli ESVCe Solid Granular
in continuity k–e k–e (Eq. dense correlationd et al. et al. pressure kinetic
equationb (6)) particle (1998a) [48] modelf theoryg
effectc
1      Inconsistent results for high dp/ka
2      Good for all the impeller speeds investigateda
3      Fair underestimation of the particle distribution
degreea
4      Signifact overpacking issues however present at
the lowest impeller speeds
5      Some numerical convergence issues or
unphysical results at the lowest impeller speeds
6      Overpredictions of the particle distribution
degree at suspension starting (i.e. intermediate
impeller speeds, 500 rpm)
7      Overpredictions of the particle distribution
degree at suspension starting (i.e. intermediate
impeller speeds, 500 rpm)
8     Good for all cases. Results very similar to case
#2. Overpredictions of Njs
9     Large overpredictions at the lowest impeller
speeds. Good results for all other cases

Notes: all cases use the Clift et al. [10] correlation to compute the base value for the drag coefficient (CD,slip).
a
Modelling approaches described and discussed in the present paper.
b
Last term of LHS of Eq. (1) in Montante and Magelli [38]. A turbulent Prandtl number equal to 0.8 was employed for these terms.
A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890

c
Eq. (13) in Tamburini et al. [56] for any rb value.
d
Eqs. (13)–(15) in Tamburini et al. [56] depending on rb.
e
Excess Solid Volume Correction algorithm by Tamburini et al. [58] briefly described in the present work.
f
Named also as the Simple Solid Pressure Model: see Eq. (15) in Fletcher and Brown [15]. A coefficient equal to either 20 [15] or 600 [16] was employed.
g
Van Wachem et al. [67]. A restitution coefficient equal to either 0.6 or 0.9 was tested.
 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890 881

1.0 1.0
0.9 Experimental_400rpm 0.9
Experimental_500rpm
0.8 0.8
SG_Brucato et al. correlation
0.7 0.7 SG_Brucato et al. correlation
0.6 MRF_Brucato et al. correlation 0.6
Z/H [-]

MRF_Brucato et al. correlation

Z/H [-]
0.5 0.5
0.4 0.4
0.3
(a) 0.3
(b)
0.2 0.2
0.1 0.1
0.0 0.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6
rβ [-] rβ [-]

1.0 1.0
0.9 0.9 Experimental_700rpm
Experimental_600rpm
0.8 0.8
0.7 SG_Brucato et al. correlation 0.7 SG_Brucato et al. correlation
0.6 0.6
Z/H [-]

MRF_Brucato et al. correlation

Z/H [-]
MRF_Brucato et al. correlation
0.5 0.5
0.4 0.4
0.3 0.3
0.2 (c) 0.2 (d)
0.1 0.1
0.0 0.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.0 0.1 0.2 0.3 0.4 0.5 0.6
rβ [-] rβ [-]

1.0
0.9 Experimental_1100rpm
0.8
0.7 SG_Brucato et al. correlation
0.6
Z/H [-]

0.5 MRF_Brucato et al. correlation

0.4
0.3
0.2 (e)
0.1
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
rβ [-]

Fig. 4. SG/MRF simulations versus experimental local axial profiles of rb (midway between subsequent baffles and at R = 0.35T) at different impeller speeds: (a) 400 rpm, (b)
500 rpm, (c) 600 rpm, (d) 700 rpm, (e) 1100 rpm. Experiments were collected by Micheletti et al. [35] (Njs = 988 rpm).

Fig. 5. 3-D sediment volume plot upon contour plots of solid volumetric fractions on a vertical diametrical plane at three different impeller speeds for the case of the SG
simulations: (a) 400 rpm, (b) 500 rpm, (c) 1100 rpm. The arrow indicates the direction of the impeller rotation.
882 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890

1.0 1.0
0.9 Experimental_400rpm 0.9 Experimental_1100rpm
0.8 0.8
SG_Brucato et al. correlation SG_Brucato et al. correlation
0.7 0.7
0.6 SG_Brucato et al. correlation_impeller 0.6
Z/H [-]

SG_Brucato et al. correlation_impeller

Z/H [-]
revolution average revolution average
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.0 0.1 0.2 0.3 0.4 0.5 0.6
rβ [-] rβ [-]

Fig. 6. SG simulations versus experimental [35] local axial profiles of rb (midway between subsequent baffles and at R = 0.35T) at some different impeller speeds. Comparison
between instantaneous time-averaged SG data.

(Fig. 4e) is higher than Njs but lower than the speed necessary for Table 2
the achievement of homogeneous suspension conditions: in other Influence of liquid free stream turbulence on drag coefficient.

words, the case of 1100 rpm is representative of the commonly N (rpm) Brucato et al. correlation Pinelli et al. correlation
investigated regime of complete suspension. As a matter of fact, dp/k ft dp/k ft
for this speed Fig. 5c does not show any fillets on the vessel bottom
400 14.45 3.61 14.49 2.43
thus confirming that the suspension is complete. It also shows that 500 17.52 5.66 17.34 3.09
particles distribute following the liquid flow discharged by the 600 20.42 8.38 20.47 3.78
impeller and assume the typical double loop configuration. As 700 22.67 11.10 22.63 4.23
concerns the local concentration profile, a very good agreement 800 25.13 14.76 25.00 4.69
900 27.51 19.05 27.32 5.09
between the SG-model prediction and the experimental profile is
1000 29.82 23.98 29.57 5.45
observable at this impeller speed in Fig. 4e. Notably, the under- 1100 32.07 29.60 31.76 5.77
predictions formerly seen below the impeller plane for the
600 rpm and 700 rpm (Fig. 4c and d) cases completely disappear
in the present case. Only the experimental point corresponding
corrections provided by the two correlations are summarized in
to the impeller plane height is slightly under-predicted. The
Table 2 while the results obtained are shown in Fig. 7.
MRF-model predictions are very similar to the SG-model ones,
At 400 rpm (Fig. 7a) the two correlations provide practically
even if larger discrepancies (both below and above the impeller
identical axial profiles, as expected in relation to the quite similar
plane) can be observed, especially near the free surface.
enhancements of the drag coefficient observable in Table 2.
Tamburini et al. [58] found that very similar results are pro-
As a difference from the Brucato et al. correlation results, at
vided by SG and MRF in terms of mass fraction of solids resting
500 rpm, Fig. 7b, the drag coefficient increase predicted by the
on the bottom (i.e. xsusp). Conversely differences in the local axial
Pinelli et al. correlation is not sufficient to follow well the experi-
rb profiles were found at all speeds in the present work, as shown
mental profile above the impeller plane. More precisely, the Pinelli
in Fig. 4. Summarizing, it can be stated that for the case of partial
et al. correlation results show a concentration equal to zero above
suspension conditions, integral data can be predicted with very
the impeller plane, while the Brucato et al. correlation predict non
similar accuracy by SG and MRF simulations, while local informa-
negligible values in accordance with the experimental data. Con-
tion is better predicted by employing the SG approach. This is not
versely, at Z/H = 0.1 the experimental data is very well predicted
surprising, since the CFD prediction of local data concerning solid
by the Pinelli et al. correlation, while a large underestimation is ob-
concentration values at different vessel heights requires a more
tained by employing the Brucato et al. one.
accurate calculation. As a matter of fact, in accordance with the rel-
At 600 rpm, Fig. 7c, and 700 rpm, Fig. 7d, both models underes-
evant literature [32,42,46], a transient CFD simulation approach
timate the experimental data below the impeller plane and exhibit
based on the fully predictive SG algorithm accounts for the tempo-
only moderate differences from each other.
ral variations in the mixing tank thus providing better predictions
At 1100 rpm (complete suspension conditions, Fig. 7e) the two
of the liquid flow field and solid suspension than the MRF steady
correlations provide similar results above the impeller plane, the
state framework.
main difference concerns the highest experimental point which
is better predicted by the Pinelli et al. correlation. Conversely, be-
low the impeller plane the results obtained by means of the corre-
5.2. Drag-turbulence correlation comparison lation by Pinelli et al. underestimates the corresponding
experimental data. Fig. 7e reports also results obtained by neglect-
Tamburini et al. [58,61] investigated the influence of different ing the effect of the liquid background turbulence on the particle
correlations accounting for the effect of the liquid background tur- drag (i.e. ft = 1). The corresponding uncorrected particle drag coef-
bulence on the particle drag coefficient for the case of a very sim- ficient is the quantity CD,slip given by the Clift et al. [10] correlation.
ilar system aiming at the prediction of integral data (xsusp, Njs). In This choice yields a severe under-prediction of the particle distri-
the present work the assessment of such choice has been per- bution degree, with a large underestimation of the experimental
formed in order to evaluate the local particle distribution. The data both above and below the impeller plane and an improbable
CFD results presented in the former sub-section were obtained very high rb value on the vessel bottom.
by adopting the Brucato et al. [6] correlation, Eq. (2). In this sub- Notably, at impeller speeds higher than 500 rpm, the Pinelli
section, these results are compared with corresponding ones ob- et al. correlation predicts solid volume fractions on the vessel bot-
tained by the Pinelli et al. [48] correlation, Eq. (3). The different tom which are higher than those predicted by the Brucato et al.
 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890 883

1.0 1.0
0.9 Experimental_400rpm 0.9
Experimental_500rpm
0.8 0.8
SG_Brucato et al. correlation SG_Brucato et al. correlation
0.7 0.7
SG_Pinelli et al. correlation

Z/H [-]
0.6 0.6 SG_Pinelli et al. correlation
Z/H [-]

0.5 0.5
0.4
0.4
(a) 0.3
(b)
0.3
0.2 0.2
0.1 0.1
0.0 0.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6

rβ [-] rβ [-]

1.0 1.0
0.9 0.9
Experimental_600rpm Experimental_700rpm
0.8 0.8
0.7 SG_Brucato et al. correlation 0.7 SG_Brucato et al. correlation
0.6

Z/H [-]
0.6
Z/H [-]

SG_Pinelli et al. correlation SG_Pinelli et al. correlation

0.5 0.5
0.4 0.4
0.3 0.3
0.2
(c) 0.2
(d)
0.1 0.1
0.0 0.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0 0 0 0 1 1

rβ [-] rβ [-]

1.0
0.9 Experimental_1100rpm
0.8
SG_Brucato et al. correlation
0.7
0.6 SG_Pinelli et al. correlation
Z/H [-]

0.5 SG_No drag coefficient correction

0.4
0.3
0.2 (e)
0.1
0.0
0 0.1 0.2 0.3 0.4 0.5 0.6

rβ [-]

Fig. 7. SG simulations versus experimental [35] local axial profiles of rb (midway between subsequent baffles and at R = 0.35T) at some different impeller speeds: (a) 400 rpm,
(b) 500 rpm, (c) 600 rpm, (d) 700 rpm, (e) 1100 rpm. Comparison of different correlations for the drag coefficient correction due to the background liquid turbulence.

correlation. This would lead to smaller vessel bottom zones where by the Pinelli et al. correction are lower than those obtained via
rb < rb_packed thus yielding lower values of xsusp. Such hypothesis is Brucato et al. correction. This result is in accordance with the find-
confirmed by the suspension curve (i.e. xsusp versus N) shown in ings of Tamburini et al. [58] who adopts the Pinelli et al. correlation
Fig. 6: at intermediate impeller speeds the xsusp values provided for the case of the largest particle size they investigated (500–
600 lm) finding xsusp values which were lower and closer to the
experimental data with respect to the Brucato et al. correction
predictions.
1.0
Fig. 8 shows also that neglecting the effect of liquid background
0.9
turbulence on drag coefficient leads to a fraction of suspended par-
0.8
0.7 SG_Brucato et al. correlation ticles different from one at impeller speeds clearly corresponding
to complete suspension conditions, thus resulting in a non physical
xsusp [-]

0.6 SG_Pinelli et al. correlation

0.5 outcome. Therefore, both solids suspension and distribution phe-
SG_No drag coefficient correction
0.4 nomena are significantly affected by the drag coefficient correction
0.3 Njs_Zwietering
and they can be predicted only if this effect is properly accounted
0.2 for.
0.1
Summarizing the comparison between the two correlations
0.0
300 400 500 600 700 800 900 1000 1100 1200 tested, it can be stated that the Brucato et al. correlation is capable
to predict with good accuracy the local axial profiles of rb, but xsusp
N [rpm]
may be overestimated for large particles. The Pinelli et al. correla-
Fig. 8. SG suspension curve data: comparison of different correlations for the drag tion provides a lower enhancement of the particle drag coefficient
coefficient correction due to the background liquid turbulence. at high dp/k so that it can predict well the amount of still particles
884 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890
Fig. 9. SG contour plots of solids volumetric fraction on horizontal planes at different vessel heights (i.e. 0.1H, 0.3H, 0.5H, 0.7H, 0.9H) at different impeller speeds: (a) 400 rpm,
(b) 500 rpm, (c) 600 rpm, (d) 700 rpm, (e) 1100 rpm.
lying on the bottom and thus the suspension curve, but is not com-
parably successful in predicting the particle distribution through-
out the vessel. Similar differences have been already reported in
the literature [23,21].
We are aware that the approaches tested in the present work
have not a universal validity, but neglecting the effect of back-
ground turbulence on the drag coefficient altogether leads to unre-
liable results (as it was largely demonstrated in other works), and
at the present time there are no more reliable alternatives available
in the literature. Further research on this topic would certainly be
desirable, but clearly this is not the focus of the present work.
Moreover, up to now very few papers have included compari-
sons between different correlations: most of the studies presented
so far adopt only one of them and compare the results with exper-
imental data, so that it is not clear which correlation should be re-
garded as being the most reliable.
5.3. Radial profiles of particle concentration
All the axial profiles presented so far refer to a specific radial
and azimuthal location: in the literature it is easy to find similar lo-
cal data, especially when solids concentration is measured by
intrusive techniques making use of a probe. Such local information
is sometimes considered as being representative of the whole ra-
dial direction [27,4,53,30,29,35] or even of the whole horizontal
plane (e.g. this is the case of measurements taken by the light
attenuation technique, [28]. This choice is based on the assumption
that either the radial gradients of solids concentration or both ra-
dial and azimuthal gradients are negligible.
In order to qualitatively evaluate the reliability of this approxi-
mation, computed contour plots of the solids volumetric fraction
on horizontal planes at different vessel heights (i.e. 0.1H, 0.3H,
0.5H, 0.7H, 0.9H) are reported in Fig. 9 for different rotational
speeds.
As it can be seen in Fig. 5a, at the lowest impeller speed (i.e.
400 rpm), the tank is practically divided in two parts separated
by an almost flat interface: the upper part is full of liquid only,
the lower one is characterized by the presence of still solids exhib-
iting a rb = rb_packed. In these conditions, as shown in Fig. 9a, radial
and azimuthal gradients are important only in the proximity of the
interface between these two zones, while they can be reasonably
neglected in the rest of the tank.
At 500 rpm, three different zones separated by two widespread
interfaces can be recognized: the sediment zone, the suspension
zone and the almost-clear liquid layer zone (Fig. 9b). As concerns
the sediment-suspension interface, at this higher agitation speed
the sediment is profiled by the liquid flow and exhibits a more
complex shape. This three-dimensional shape of the sediment-sus-
pension interface makes the assumption of negligible radial and
azimuthal rb gradients unacceptable. The contour plot of rb at
Z = 0.1H, reported in Fig. 9b, exhibits the typical shape of a four-
arm star, thus proving the prominence of the rb gradients. Similar
 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890 885

considerations can be done for the case of the interface separating
the suspension and the almost-clear liquid layer although radial
and azimuthal gradients are less marked.
As the impeller speed increases and more particles are sus-
pended (i.e. 600 rpm and 700 rpm, Fig. 9c and d), the extension
of the suspension zone increases: significant gradients are however
observable in the proximities of the two interfaces, especially near
the sediment-suspension interface.
When complete suspension conditions are achieved (i.e.
1100 rpm) the whole vessel volume is occupied only by the so-
lid–liquid suspension region and far less significant concentration
gradients are present (Fig. 9e).
The presence of non-negligible radial concentration gradients
near the suspension-clear liquid interface is in accordance with
the findings of [30] who measured the local solids concentration
by means of a conductivity probe for the case of a conical-bot-
tomed baffled tank stirred by a down-pumping pitched blade tur-
bine. They dealt with partial-to-complete suspension conditions
and collected radial profiles of rb at vessel heights higher than
the one relevant to the impeller plane. They found nearly negligible
radial gradients of rb everywhere except for the proximity of the
above mentioned interface.
Moreover, it is worth noting that the azimuthal gradients, if any,
are significant only in the proximities of the baffles, while they are
less important in the inner parts of the vessel.
In Fig. 10 the local axial profiles already shown in Fig. 4 are
compared with corresponding radially averaged profiles. In accor-
dance with the comments made in Fig. 9, at 400 rpm (Fig. 10a)
some discrepancies between the two axial profiles are observable
only in the proximity of the sediment interface. Also the compari-
son at 500 rpm (Fig. 10b) confirms the considerations made on the
corresponding contour plots of Fig. 5b: the local and the radially
averaged profiles mainly differ in the vicinity of the two wide-
spread interfaces (sediment-suspension and suspension-almost
clear liquid layer). At higher impeller speeds (i.e. 600 rpm and
700 rpm, lower than Njs), Fig. 10c and d, the difference between
the local and the radially averaged profiles decreases as the impel-
ler speed increases both in the upper and in the lower part of the
vessel. As regards the upper part, as the impeller speed increases,
more particles reach the vessel top causing the liquid layer to be

1.0 1.0
0.9 0.9
SG_Brucato et al. correlation_local SG_Brucato et al. correlation_local
0.8 0.8
0.7 0.7
SG_Brucato et al. correlation_radially SG_Brucato et al. correlation_radially
0.6
Z/H [-]

0.6
Z/H [-]

averaged averaged
0.5 0.5
0.4 0.4
0.3
(a) 400rpm 0.3
(b) 500rpm
0.2 0.2
0.1 0.1
0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6

rβ [-] rβ [-]

1.0 1.0
0.9 0.9 SG_Brucato et al. correlation_local
SG_Brucato et al. correlation_local
0.8 0.8
0.7 0.7
SG_Brucato et al. correlation_radially SG_Brucato et al. correlation_radially
0.6
Z/H [-]

0.6
Z/H [-]

averaged averaged
0.5 0.5
0.4 0.4
0.3 0.3
0.2 (c) 600rpm 0.2 (d) 700rpm
0.1 0.1
0.0 0.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.0 0.1 0.2 0.3 0.4 0.5 0.6

rβ [-] rβ [-]

1.0
0.9
0.8 SG_Brucato et al. correlation_local
0.7
0.6
Z/H [-]

SG_Brucato et al. correlation_radially

0.5 averaged
0.4
0.3
0.2 (e) 1100rpm
0.1
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6

rβ [-]

Fig. 10. Comparison of local and radially averaged SG axial profiles of rb (midway between subsequent baffles and at R = 0.35T) at some different impeller speeds: (a) 400 rpm,
(b) 500 rpm, (c) 600 rpm, (d) 700 rpm, (e) 1100 rpm.
886 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890
completely replaced by the suspension and the suspension-liquid
layer interface to disappear. As concerns the region below the
impeller, the difference between the two profiles with the impeller
speed is reduced, although radial concentration gradients are far
from being negligible.
At impeller speeds higher than Njs (i.e. 1100 rpm, Fig. 10e)
although all particles are suspended and no fillet is present, slight
differences are however evident in the lower part of the tank, while
the radial gradients appear not to play any longer a significant role
in the upper part.
Notably, the integration of radially averaged data also along the
axial and azimuthal directions provided the correct value of rb_av
(=0.092) confirming that total solids mass is conserved during
the CFD simulation.
The local variation of rb along some radial monitoring lines
placed at different vessel heights (see Fig. 11) was also investigated
and the data are presented in Fig. 12.
As it can be seen in Fig. 12a, the largest gradients can be ob-
served near the vessel bottom at Z = 0.1H: neglecting the radial
variations of rb at this vessel height may lead to substantial
mistakes practically at all impeller speeds: in particular at
some N the variation of rb covers the entire allowable range (i.e.
0rb_packed). Notably, at 500 rpm (Fig. 12a) and low R/T, the local
rb is lower than rb_packed as a difference with the cases at 400,
600 and 700 rpm where rb = rb_packed allegedly because (i) at
500 rpm the sediment height is lower than at 400 rpm (as also
observable in Fig. 5a and b) while (ii) at N = 600–700 rpm the
sediment concentrates in the central part of the tank due to the
radial flow generated by the higher impeller speed thus reaching
a height higher than that relevant to 500 rpm.
Fig. 11. Monitoring lines for radial profiles of solids concentration (i.e. Z = 0.1H,
0.2H, 0.3H, 0.4H, 0.5H, 0.6H, 0.7H, 0.8H, 0.9H).
Lower yet still significant radial gradients are observable below
the impeller plane (i.e. Z = 0.2 H, 0.3 T; Fig. 12b and c) especially at
intermediate impeller speeds.
Conversely, the radial gradients are much less pronounced
above the impeller plane (Z P 0.4H) practically at all impeller
speeds: at low N the solids are mainly located in the lower part
of the vessel resulting in very low and almost constant rb values
in the upper part of the tank as the radial profiles at 400 rpm
and for Z P 0.4H show. At higher N, the profiles show only slight,
yet non negligible variations especially near the shaft and in the
proximities of the vessel lateral wall.
Moreover, the higher N, the closer the rb value to the average so-
lid concentration rb_av. Clearly, at very low radial positions, rb is al-
ways equal to zero over the impeller plane because of the presence
of the shaft.
The extent of particle dispersion along the radial direction is
here quantitatively assessed by means of the radial variation coef-
ficient rr defined as in the following:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2ffi
1 Xnr r bi
rr ¼ 1 ð5Þ
nr i¼1 r
b rav
where rb_rav is the radially averaged solid volumetric fraction va-
lue, i denotes the generic finite volume in CFD work (or a generic
measurement radial-point in experimental work), and nr is the total
number of computational cells along the radial direction. rr values
relevant to the data of Fig. 12 are shown in Fig. 13. Above the impel-
ler plane (Z = H/3), the higher the impeller speed, the higher the
particles dispersion degree (i.e. lower rr). This is as expected due
to the higher liquid turbulence levels. Conversely, a non monotonic
behaviour is observable in the lower part of the tank, below the
impeller plane. In particular at the lowest impeller speed and vessel
height (i.e. 400 rpm and 0.1H) the solids constitute an almost-
homogeneously distributed sediment (see also Fig. 5a) correspond-
ing to a rb = rb_packed in most of the radial positions of the line. This
leads to a very low rr. At 500 rpm the sediment assumes the star-
like shape thus resulting in a higher rr. At higher impeller speeds
(i.e. 600 rpm, 700 rpm yet lower than Njs) more particles are sus-
pended causing rb to cover a wider range of values and rr to conse-
quently increase. As a difference, at N > Njs all the mechanical power
is employed to distribute more homogeneously the particles
throughout the vessel thereby reducing the concentration gradients
everywhere and in particular also near the vessel bottom: the con-
centration becomes closer to the average value and rr consequently
decreases.
5.4. Homogeneous suspension conditions
The attainment of homogeneous suspension conditions requires
high impeller speeds and a corresponding very large mechanical
power (the power is proportional to the cube of impeller speed),
nevertheless it can be required in some industrial applications
(e.g. crystallization and solid catalyzed reaction). The ‘‘homogene-
ity’’ or ‘‘suspension quality’’ f is a parameter which is very often
employed in the literature to quantify the distribution of solid par-
ticles within the stirred tank. According to Hosseini et al. [17], it
can be defined as:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2ffi
1 Xn rb i
f¼1r¼1 1 ð6Þ
n i¼1 r
b av
Although the relation between suspension conditions and specific
features of the particle distribution depends on particle type (e.g.
mono- or poli-dispersed), size and concentration and on system
configuration (impeller type, clearance, diameter, etc.), according
to the Bohnet and Niesmak [5] classification, homogeneous
 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890 887

Z=0.1H Z=0.2H
0.7 0.7 (b)
400rpm 500rpm 600rpm 700rpm 1100rpm
0.6 0.6
0.5 (a) 0.5 400rpm
500rpm
r β [-]

r β [-]
0.4 0.4 600rpm
0.3 0.3 700rpm
1100rpm
0.2 0.2
0.1 0.1
0.0 0.0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
R/T [-] R/T [-]

0.7
Z=0.3H (c) 0.7
Z=0.4H (d)
0.6 0.6
0.5 400rpm 0.5 400rpm
500rpm 500rpm
rβ [-]

rβ [-]
0.4 0.4
600rpm 600rpm
0.3 700rpm 0.3 700rpm
1100rpm 1100rpm
0.2 0.2
0.1 0.1
0.0 0.0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50

R/T [-] R/T [-]

0.7 Z=0.5H (e) 0.7 Z=0.6H (f)

0.6 0.6
0.5 400rpm 0.5 400rpm
500rpm 500rpm
r β [-]

rβ [-]

0.4 0.4
600rpm 600rpm
0.3 700rpm 0.3 700rpm
1100rpm 1100rpm
0.2 0.2
0.1 0.1
0.0 0.0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50

R/T [-] R/T [-]

0.7 Z=0.7H (g) 0.7 Z=0.8H (h)

0.6 0.6
0.5 400rpm 0.5 400rpm
500rpm 500rpm
rβ [-]

0.4
r β [-]

0.4
600rpm 600rpm
0.3 700rpm 0.3 700rpm
1100rpm 1100rpm
0.2 0.2
0.1 0.1
0.0 0.0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50

R/T [-] R/T [-]

0.7 Z=0.9H

0.6
(i)
0.5 400rpm
500rpm
0.4
r β [-]

600rpm
0.3 700rpm
1100rpm
0.2
0.1
0.0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50

R/T [-]

Fig. 12. SG radial profiles of solids concentration at different vessel heights and impeller speeds.
888 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890

1.0 degree of homogeneity decreases when the terminal velocity of the

0.9 400RPM
solid particles increases. For the case of the present work where a
0.8 500RPM
radial Rushton turbine is employed, f appears to increase as the
600RPM
impeller speed increases in a monotonic way. Conversely the adop-
0.7
tion of high efficiency impellers (as the Lightnin A310) leads to the
0.6 700RPM
Z/H [-]

presence of a maximum in the function f versus N [17]: once the

1100RPM
0.5 suspension quality f has attained this peak value, any further in-
0.4 crease in impeller speed is not beneficial but might be dramatically
0.3 detrimental.
0.2 The above considerations relevant to Fig. 14 are further con-
0.1 firmed by observing Fig. 15 where local (the same radial and azi-
muthal position of the data by Micheletti et al. [35]) rb values at
0.0
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80 2.00 different vessel heights are reported. As it can be seen, at N  Njs
local values of rb are quite different along the axial direction, and
σ r [-] the higher N, the smaller the above difference: the particle concen-
Fig. 13. Radial variation coefficient at different vessel heights and impeller speeds. trations at all vessel heights approach the (‘‘homogeneous’’) value
of rb_av (i.e. 0.092) as N increases, although at 1350 rpm this condi-
tion appears still far from being attained.

6. Conclusions

Reynolds Averaged Navier Stokes (RANS) simulations of dense

solid–liquid suspensions within a flat bottomed vessel stirred by
ζ

a standard Rushton turbine were performed with a finite volume

Fig. 14. SG suspension quality as a function of impeller speed.
suspension conditions can be considered to be achieved when
r < 0.2 (i.e. f > 0.8).
The results shown in the previous section have shown that at
N > Njs, the higher N, the higher the particle distribution degree
throughout the entire vessel. Despite this finding, by observing
Fig. 14 it can be stated that homogeneous suspension conditions
are far from being achieved even at very high and economically
expensive impeller speeds, especially when the Pinelli et al. corre-
lation is employed. In this regard, it is well known that similar con-
ditions are difficult to attain when dense suspensions [30]) and/or
large particle sizes [17,18] are employed (as in the present work).
As a matter of fact, in accordance with Peker and Helvaci [47] the
0.40
0.35
0.30
0.25
rβ [-]
code by adopting the fully predictive Eulerian–Eulerian Multi Fluid
Model in conjunction with the k–e turbulence model for the con-
tinuous (liquid) phase. The specific modelling and numerical de-
tails employed were those adopted in Tamburini et al. [58,61,63],
where they reliably predicted global suspension quantities such
as the sediment amount and shape [58] and the impeller speed
for complete [66] and sufficient [68] suspension conditions. Here,
this model was further tested in order to evaluate its capability
of predicting also the three-dimensional particle distribution phe-
nomenon. Both the time-dependent Sliding Grid (SG) method and
the steady state Multiple Reference Frame technique (MRF) were
used. Moreover, in order to account for background turbulence
influence on particle drag coefficient, two different correlations
were also tested.
Results showed that experimental local axial profiles of solid
concentrations (a typical information characterizing particle distri-
bution) at different impeller speeds ranging from partial to com-
plete suspension conditions can be predicted with high accuracy
by the Tamburini et al. [58] CFD model. In particular transient SG
simulations were found to be in slightly better agreement with
experiment with respect to steady state MRF simulations.
A drag coefficient correction accounting for liquid background
turbulence was found to be necessary to obtain reliable predic-
tions. As concerns the comparison between the two corrections
tested, the results obtained in the present work along with the
Z/H=0.10 findings of Tamburini et al. [58] show that for small particles (dp/
Z/H=0.20
Z/H=0.30 k < 10) the two approaches yield practically identical results
Z/H=0.40 and both satisfactorily reproduce experimental suspension curves
Z/H=0.50 (i.e. fraction of suspended particles as a function of impeller speed).
Z/H=0.60

Z/H=0.70 For the case of large dp/k, the Pinelli et al. [48] correction is more
0.20
Z/H=0.79 suitable for the prediction of the suspension curves, while the Bru-
0.15 Z/H=0.88 cato et al. [6] correction guarantees a better prediction of the spa-
0.10 tial particle distribution. Clearly, swapping from a correlation to
0.05 another in dependence of the phenomenon simulated cannot be
the right ending: rather, the present findings should be considered
0.00
300 400 500 600 700 800 900 1000 1100 1200 1300 1400 as a stimulus to further and broader investigation on this topic.
Although solid concentration radial profiles are often neglected
N [rpm]
in the literature, the present results showed that this approxima-
Fig. 15. SG local rb (midway between subsequent baffles and at R = 0.35T) values at tion is far from being realistic in some zones of the tank, especially
different vessel heights as a function of impeller speed. under partial suspension conditions.
 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890 889

Acknowledgement interrogate mixing processes at plant scale, Chem. Eng. Sci. 52 (1997) 2087–
2097.
[32] G. Micale, A. Brucato, F. Grisafi, M. Ciofalo, Prediction of flow fields in a dual-
This work was carried out under financial support by Ministero impeller stirred vessel, AIChE J. 45 (1999) 445–464.
dell’Università e della Ricerca, PRIN-2006 Contract No. [33] G. Micale, G. Montante, F. Grisafi, A. Brucato, J. Godfrey, CFD simulation of
particle distribution in stirred vessels, Chem. Eng. Res. Des. 78 (2000) 435–
2006091953_004.
444.
[34] G. Micale, F. Grisafi, L. Rizzuti, A. Brucato, CFD Simulation of particle
suspension height in stirred vessels, Chem. Eng. Res. Des. 82 (2004) 1204–
References 1213.
[35] M. Micheletti, L. Nikiforaki, K.C. Lee, M. Yianneskis, Particle concentration and
mixing characteristics of moderate-to-dense solid–liquid suspensions, Ind.
[1] R. Angst, M. Kraume, Experimental investigations of stirred solid/liquid
Eng. Chem. Res. 42 (2003) 6236–6249.
systems in three different scales: particle distribution and power
[36] G. Montante, G. Micale, F. Magelli, A. Brucato, Experiments and CFD
consumption, Chem. Eng. Sci. 61 (2006) 2864–2870.
predictions of solid particle distribution in a vessel agitated with four
[2] P.M. Armenante, E.U. Nagamine, J. Susanto, Determination of correlations to
pitched blade turbines, Chem. Eng. Res. Des. 79 (2001) 1005–1010.
predict the minimum agitation speed for complete solid suspension in agitated
[37] G. Montante, F. Magelli, Modelling of solids distribution in stirred tanks:
vessels, Can. J. Chem. Eng. 76 (1998) 413–419.
analysis of simulation strategies and comparison with experimental data, Int.
[3] J.A. Bamberger, M.S. Greenwood, Using ultrasonic attenuation to monitor
J. Comput. Fluid Dyn. 19 (2005) 253–262.
slurry mixing in real time, Ultrasonics 42 (2004) 145–148.
[38] G. Montante, F. Magelli, Mixed solids distribution in stirred vessels:
[4] A. Barresi, G. Baldi, Solid dispersion in an agitated vessel, Chem. Eng. Sci. 42
experiments and computational fluid dynamics simulations, Ind. Eng. Chem.
(12) (1987) 2949–2956.
Res. 46 (2007) 2885–2891.
[5] M. Bohnet, G. Niesmak, Distribution of solids in stirred suspensions, Ger. Chem.
[39] J.Y. Murthy, S.R. Mathur, D. Choudhury, CFD simulation of flows in stirred tank
Eng. 3 (1980) 57–65.
reactors using a sliding mesh technique, IChemE Symp. Ser. 136 (1994) 341–
[6] A. Brucato, F. Grisafi, G. Montante, Particle drag coefficients in turbulent fluids,
348.
Chem. Eng. Sci. 53 (1998) 3295–3314.
[40] B.N. Murthy, R.S. Ghadge, J.B. Joshi, CFD simulations of gas–liquid–solid stirred
[7] A. Brucato, A. Cipollina, G. Micale, F. Scargiali, A. Tamburini, Particle suspension
reactor: prediction of critical impeller speed for solid suspension, Chem. Eng.
in top-covered unbaffled tanks, Chem. Eng. Sci. 65 (2010) 3001–3008.
Sci. 62 (2007) 7184–7195.
[8] C. Buurman, G. Resoort, A. Plaschkes, Scaling-up rules for solids suspension in
[41] A.W. Nienow, Suspension of solid particles in turbine-agitated baffled vessels,
stirred vessels, Chem. Eng. Sci. 41 (1986) 2865–2871.
Chem. Eng. Sci. 23 (1968) 1453–1459.
[9] CFX-4 Documentation. AnsysÒ, 2009.
[42] A. Ochieng, A.E. Lewis, CFD simulation of solids off-bottom suspension and
[10] R. Clift, J.R. Grace, M.E. Weber, Bubbles, Drops and Particles, Academic Press,
cloud height, Hydrometallurgy 82 (2006) 1–12.
New York, San Francisco, London, 1978.
[43] A. Ochieng, M.S. Onyango, Drag models, solids concentration and velocity
[11] J.J. Derksen, Numerical Simulation of Solid Suspension in a Stirred Tank, AICHE
distribution in a stirred tank, Powder Technol. 181 (2008) 1–8.
J. 49 (2003) 2700–2714.
[44] J.Y. Oldshue, Fluid Mixing Technology, McGraw-Hill, New York, NY, 1983.
[12] E. Doroodchi, G.M. Evans, M.P. Schwarz, G.L. Lane, N. Shah, A. Nguyen,
Chapter 5.
Influence of turbulent intensity on particle drag coefficients, Chem. Eng. J. 135
[45] A. Paglianti, S. Pintus, The impedance probe for the measurements of liquid
(2008) 129–134.
hold-up and mixing time in two/three-phase stirred tank reactors, Exp. Fluids
[13] T. Dyakowski, L.F.C. Jeanmeure, A.J. Jaworski, Applications of electrical
31 (2001) 417–427.
tomography for gas–solids and liquid–solids flows – a review, Powder
[46] R. Panneerselvam, S. Savithri, G.D. Surender, CFD modeling of gas–liquid–solid
Technol. 112 (2000) 174–192.
mechanically agitated contactor, Chem. Eng. Res. Des. 86 (2008) 1331–1344.
[14] L. Fan, Z. Mao, Y. Wang, Numerical simulation of turbulent solid–liquid two-
[47] S.M. Peker, S.S. Helvaci, Solid–Liquid Two Phase Flow, Elsevier, Amsterdam,
phase flow and orientation of slender particles in a stirred tank, Chem. Eng. Sci.
2008.
60 (2005) 7045–7056.
[48] D. Pinelli, M. Nocentini, F. Magelli, Solids distribution in stirred slurry reactors:
[15] D.F. Fletcher, G.J. Brown, Numerical simulation of solid suspension via
influence of some mixer configurations and limits to the applicability of a
mechanical agitation: effect of the modeling approach, turbulence model
simple model for predictions, Chem. Eng. Sci. 188 (2001) 91–107.
and hindered settling drag law, Int. J. Comput. Fluid Dyn. 23 (2009) 173–187.
[49] F. Rieger, P. Ditl, O. Havelkova, Suspension of solid particles –
[16] J.W. Holbeach, M.R. Davidson, An Eulerian–Eulerian model for the dispersion
concentration profiles and particle layer on the vessel bottom, in:
of a suspension of microscopic particles injected into a quiescent liquid, Eng.
Proceedings of the 6th European Conference on Mixing, Pavia, Italy, 24–
Appl. Comput. Fl. Mech. 3 (2009) 84–97.
26 May 1988. pp. 251–258.
[17] S. Hosseini, D. Patel, F. Ein-Mozaffari, M. Mehrvar, Study of solid–liquid mixing
[50] M.V. Sardeshpande, V.A. Juvekar, V.V. Ranade, Solid suspension in stirred
in agitated tanks through electrical resistance tomography, Chem. Eng. Sci. 65
tanks: UVP measurements and CFD simulations, Can. J. Chem. Eng. 89 (2011)
(2010) 1374–1384.
1112–1121.
[18] S. Hosseini, D. Patel, F. Ein-Mozaffari, M. Mehrvar, Study of solid–liquid mixing
[51] M.V. Sardeshpande, V.V. Ranade, Simulation of settling of solid particles due to
in agitated tanks through computational fluid dynamics modelling, Ind. Eng.
sudden impeller stoppage, Ind. Eng. Chem. Res. 51 (10) (2012) 4112–4118.
Chem. Res. 49 (2010) 4426–4435.
[52] J. Scully, P. Frawley, Computational fluid dynamics analysis of the suspension
[19] R. Jafari, J. Chaouhi, P.A. Tanguy, A comprehensive review of just suspended
of nonspherical particles in a stirred tank, Ind. Eng. Chem. Res. 50 (2011)
speed in liquid–solid and gas–liquid–solid stirred tank reactors, Int. J. Chem.
2331–2342.
React. Eng. 10 (R1) (2012).
[53] P.A. Shamlou, E. Koutsakos, Solids suspension and distribution in liquids under
[20] T. Jirout, J. Moravec, F. Rieger, V. Sinevic, M. Spidla, V. Sobolic, J. Tihon,
turbulent agitation, Chem. Eng. Sci. 44 (1989) 529–542.
Electrochemical measurement of impeller speed for off-bottom suspension,
[54] M. Spidla, V. Sinevic, M. Jahoda, V. Machon, Solid particle distribution of
Inzynieria chemiczna i procesowa 26 (3) (2005) 485–497.
moderately concentrated suspensions in a pilot plant stirred vessel, Chem.
[21] G.R. Kasat, A.R. Khopkar, V.V. Ranade, A.B. Pandit, CFD simulation of liquid-
Eng. J. 113 (2005) 73–82.
phase mixing in solid–liquid stirred reactor, Chem. Eng. Sci. 63 (2008) 3877–
[55] A. Tamburini, Suspension phenomena in solid–liquid agitated systems,
3885.
Doctoral thesis, University of Palermo, Fotograf, Palermo, Italy, 2011. ISBN:
[22] R.C.S. Kee, R.B.H. Tan, CFD simulation of solids suspension in mixing vessels,
9788895272979.
Can. J. Chem. Eng. 80 (2002) 721–726.
[56] A. Tamburini, A. Cipollina, G. Micale, M. Ciofalo, A. Brucato, Dense solid–liquid
[23] A.R. Khopkar, G.R. Kasat, A.B. Pandit, V.V. Ranade, Computational fluid
off-bottom suspension dynamics: simulation and experiment, Chem. Eng. Res.
dynamics simulation of the solid suspension in a stirred slurry reactor, Ind.
Des. 87 (2009) 587–597.
Eng. Chem. Res. 45 (2006) 4416–4428.
[57] A. Tamburini, L. Gentile, A. Cipollina, G. Micale, A. Brucato, Experimental
[24] G.L. Lane, M.P. Schwarz, G.M. Evans, Numerical modelling of gas–liquid flow in
investigation of dilute solid–liquid suspension in an unbaffled stirred vessels
stirred tanks, Chem. Eng. Sci. 60 (2005) 2203–2214.
by a novel pulsed laser based image analysis technique, Chem. Eng. Trans. 17
[25] M. Ljungqvist, A. Rasmuson, Numerical simulation of the two-phase flow in an
(2009) 531–536.
axially stirred vessel, Chem. Eng. Res. Des. 79 (2001) 533–546.
[58] A. Tamburini, A. Cipollina, G. Micale, A. Brucato, M. Ciofalo, CFD Simulations of
[26] J.Y. Luo, R.I. Issa, A.D. Gosman, Prediction of impeller-induced flows in mixing
dense solid–liquid suspensions in baffled stirred tanks: predictions of
vessels using multiple frames of reference, IChemE Symp. Ser. 136 (1994) 549–
suspension curves, Chem. Eng. J. 178 (2011) 324–341.
556.
[59] A. Tamburini, A. Cipollina, Micale, CFD simulation of solid liquid suspensions
[27] V. Machon, I. Fort, J. Skrivanek, Local solids distribution in the space of a stirred
in baffled stirred vessels below complete suspension speed, Chem. Eng. Trans.
vessel, in: 4th European Conference on Mixing, Noordwijkerhout, BHRA,
24 (2011) 1435–1440.
Cranfield, UK, Apr 27–29, 1982, pp. 289.
[60] A. Tamburini, A. Cipollina, G. Micale, A. Brucato, Dense solid–liquid
[28] F. Magelli, D. Fajner, M. Nocentini, G. Pasquali, Solid distribution in vessels
suspensions in top-covered unbaffled stirred vessels, Chem. Eng. Trans. 24
stirred with multiple impellers, Chem. Eng. Sci. 45 (1990) 615–625.
(2011) 1441–1446.
[29] A.T.C. Mak, PhD thesis, University College London, London, UK, 1991.
[61] A. Tamburini, A. Cipollina, G. Micale, A. Brucato, M. Ciofalo, CFD simulations of
[30] A.T.C. Mak, S.W. Ruszkowski, Scaling-up of solids distribution in stirred
dense solid–liquid suspensions in baffled stirred tanks: prediction of the
vessels, IV Fluid Mix. AIChE Symp. Ser. 121 (1990) 379–395.
minimum impeller speed for complete suspension, Chem. Eng. J. 193–194
[31] R. Mann, F.J. Dickin, M. Wang, T. Dyakowski, R.A. Williams, R.B. Edwards, A.E.
(2012) 234–255.
Forrest, P.J. Holden, Application of electrical resistance tomography to
890 A. Tamburini et al. / Chemical Engineering Journal 223 (2013) 875–890
[62] A. Tamburini, G. La Barbera, A. Cipollina, M. Ciofalo, G. Micale, CFD simulation
of channels for direct and reverse electro dialysis, Des. Water Treat. 48 (2012)
370–389.
[63] A. Tamburini, A. Brucato, A. Cipollina, G. Micale, M. Ciofalo, CFD predictions of
sufficient suspension conditions in solid–liquid agitated tanks, Int. J. Nonlinear
Sci. Num. Sim. 13 (6) (2012) 427–443.
[64] A. Tamburini, A. Cipollina, G. Micale, A. Brucato, Measurements of Njs and
power requirements in unbaffled bioslurry reactors, Chem. Eng. Trans. 27
(2012) 343–348.
[65] A. Tamburini, A. Cipollina, G. Micale, A. Brucato, Particle distribution in dilute
solid–liquid unbaffled tanks via a novel laser sheet and image analysis
technique, Chem. Eng. Sci. 87 (2013) 341–358.
[66] G.B. Tatterson, Fluid Mixing and Gas Dispersion in Agitated Tanks, McGraw-
Hill, New York, 1991.
[67] B.G.M. Van Wachem, J.C. Schouten, R. Krishna, C.M. Van Den Bleek, Eulerian
simulations of bubbling behaviour in gas–solid fluidised beds, Comput. Chem.
Eng. 22 (Suppl.1) (1998) S299–S306.
[68] D. Wadnerkar, R.P. Utikar, M.O. Tade, V.K. Pareek, CFD simulation of solid–
liquid stirred tanks, Adv. Powder Technol. 23 (2012) 445–453.
[69] F. Wang, W. Wang, Z. Mao, Numerical study of solid–liquid two-phase flow in
stirred tanks with Rushton impeller. (I) Formulation and simulation of flow
field, Chin. J. Chem. Eng. 12 (2004) 599–609.
[70] L. Wang, Y. Zhang, X. Li, Y. Zhang, Experimental investigation and CFD
simulation of liquid–solid–solid dispersion in a stirred reactor, Chem. Eng. Sci.
65 (2010) 5559–5572.
[71] T.N. Zwietering, Suspending of solid particles in liquids by agitators, Chem.
Eng. Sci. 8 (1958) 244–253.