CHERD-1399; No.

of Pages 19
ARTICLE IN PRESS
chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx

Contents lists available at ScienceDirect

Chemical Engineering Research and Design

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

Influence of drag and turbulence modelling on

CFD predictions of solid liquid suspensions in
stirred vessels

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 Energia, Ingegneria dell’Informazione e Modelli Matematici, Università di Palermo, Viale delle

Scienze Ed. 6, 90128 Palermo, Italy

a b s t r a c t

Suspensions of solid particles into liquids within industrial stirred tanks are frequently carried out at an impeller
speed lower than the minimum required for complete suspension conditions. This choice allows power savings
which usually overcome the drawback of a smaller particle-liquid interfacial area. Despite this attractive economical
perspective, only limited attention has been paid so far to the modelling of the partial suspension regime.
In the present work two different baffled tanks stirred by Rushton turbines were simulated by employing the
Eulerian-Eulerian Multi Fluid Model (MFM) along with either the Sliding Grid algorithm (transient simulations) or
the Multiple Reference Frame technique (steady state simulations). In particular, a comparison of alternative mod-
elling approaches for inter-phase drag force and turbulence closure is presented. The results are evaluated against
a number of experimental data concerning sediment features (amount and shape) and local axial profiles of solids
concentration, with emphasis on the partial suspension regime.
Results show that some of the approaches commonly adopted to account for dense particle effects or turbulent
fluctuations of the volumetric fractions may actually lead to substantial discrepancies from the experimental data.
Conversely simpler models which do not include such additional effects give the best overall predictions in the whole
range of partial to complete suspension conditions.
© 2013 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.

Keywords: Computational Fluid Dynamics (CFD); Stirred tank; Solid liquid suspension; Drag force; Turbulence model;
Multiphase flow

1. Introduction and literature review complete suspension conditions, Njs is the most investigated
topic since it is known to represent a good compromise
1.1. Partial and complete suspension conditions between the reduction of agitation costs and the enhance-
ment of mass transfer processes. Many studies focus on
Many research efforts have been devoted to the inves- the proposal of methods to measure Njs (Zwietering, 1958;
tigation of solid–liquid stirred tanks in the last decades Musil and Vlk, 1978; Bohnet and Niesmak, 1980; Rewatkar
(Zwietering, 1958; Nienow, 1968; Bourne and Sharma, 1974; et al., 1991; Micale et al., 2002; Zhu and Wu, 2002; Jirout
Yamazaki et al., 1986; Barresi and Baldi, 1987; Oldshue and et al., 2005; Ren et al., 2008; Brucato et al., 2010; Tamburini
Sharma, 1992; Armenante et al., 1998; Micale et al., 2000; et al., 2011c, 2012c; Selima et al., 2008) or CFD models
Rieger, 2000; Angst and Kraume, 2006; Sardeshpande et al., for its prediction (Kee and Tan, 2002; Wang et al., 2004;
2009; Tamburini et al., 2009b, 2013a; Jirout and Rieger, 2011; Murthy et al., 2007; Hosseini et al., 2010; Tamburini et al.,
Montante et al., 2012). The minimum impeller speed for 2012a).

∗
Corresponding author. Tel.: +39 09123863780; fax: +39 09123860841.
E-mail address: [email protected] (G. Micale).
Received 30 July 2013; Received in revised form 3 October 2013; Accepted 16 October 2013
0263-8762/$ – see front matter © 2013 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.cherd.2013.10.020

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
2 chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx

associated losses may become drastic only when the system is

Nomenclature operated at an N < 0.6 Njs (under gassed condition). Jafari et al.
(2012) reported data concerning the operation of an industrial
gd3p ˛ (ˇ −˛ )
Ar Archimedes number: Ar = slurry reactor for the gold cyanidation process at N = 0.8 Njs ; a
2˛
B impeller blade height (m) choice which (i) did not appreciably affect the reaction selec-
C impeller clearance (m) tivity and process yield and (ii) allowed a substantial reduction
C˛ˇ inter-phase drag term (kg m−3 s−1 ) of energy consumption with respect to complete suspen-
CD drag coefficient sion conditions. Also, they reported that a decrease of ∼0.55
C␮ , C1 C2 k–ε model parameter M$/year in product value due to operating at N = 0.5 Njs cor-
D impeller diameter (m) responded to an energy saving of ∼1 M$/year, with a ∼0.45
dp particle mean diameter (m) M$/year increase in net profit.
g gravitational constant (m s−2 ) In this regard, the availability of data on the partial suspen-
H liquid height (m) sion conditions may be very useful for the industry. Such infor-
hsed height sediment assumes on the lateral wall mation could allow economical analyses to be performed lead-
midway two subsequent baffles (m) ing to the optimal choice of N. Notwithstanding the interest
M inter-phase momentum transfer term (N m−3 ) expressed so far at the industrial level for partial suspension
n exponent depending on the value of conditions, only a few data can be found in the literature for
Archimedes number this particular regime and many of them were collected with
N rotational impeller speed (rpm) different aims. For instance, except for a few cases (Tamburini
Njs just suspension speed (rpm) et al., 2011a,b, 2012b), all literature works presenting CFD data
Np power number on systems under partial suspension conditions are devoted to
P pressure (N m−2 ) the evaluation of Njs (Kee and Tan, 2002; Oshinowo and Bakker,
r volumetric fraction 2002; Wang et al., 2004; Ochieng and Lewis, 2006; Murthy et al.,
rˇ-av average solid volumetric fraction value 2007; Kasat et al., 2008; Panneerselvam et al., 2008; Hosseini
rˇ min low cut-off value for the Piecewise correlation et al., 2010): most of the criteria adopted in the literature
approach to estimate Njs require data at N < Njs and N > Njs thus lead-
rˇ max high cut-off value for the Piecewise correlation ing to the necessity of performing simulations under partial
approach suspension conditions. However, none of these works fully
rˇ-packed solid volumetric fraction maximum packing investigated the partial suspension regime.
value A universal CFD model capable to manage all the types of
R radial coordinate (m) solid–liquid suspensions within stirred tanks does not exist
Rep particle Reynolds number yet. For the case of complete suspension, it is possible to find
T tank diameter (m) in the literature different formulations for the inter-phase drag
t time (s) force treatment as well as for the turbulence closure.
U velocity (m s−1 )
W baffle width (m) 1.2. Inter-phase drag force
w/w mass fraction: wsolid /wliquid (%)
xsusp mass fraction of suspended solids The inter-phase drag force is one of the most crucial factors
z axial coordinate (m) affecting both solids suspension and distribution. Gidaspow’s
dense particle effect (Gidaspow, 1994) is the most accepted
Greek letters formulation of the inter-phase drag force which takes into
ε turbulent dissipation (W kg−1 ) account the particle–particle interaction effect (Gidaspow,
 Kolmogorov length (m) 1994; Ochieng and Onyango, 2008; Scully and Frawley, 2011):
 viscosity (Pa s) according to this formulation, the higher the local solid vol-
 kinematic viscosity (m2 s−1 ) ume fraction, the higher the inter-phase drag force as a
 density (kg m−3 ) consequence of more intense particle–particle interactions.
 turbulent Prandtl number When the solid loading is very high, the adoption of Ergun’s
equation (Ergun, 1952) for the inter-phase drag force is com-
Subscripts monly suggested (CFX-4 Documentation, 1994; Ochieng and
␣ liquid phase Onyango, 2008) although it was originally formulated for the
AS avoidance of settling case of fixed beds of particles. In some cases, the two previous
ˇ solid phase approaches are simultaneously employed for the simula-
BL bottom lifting tion of the same system, the former being employed in the
t turbulent domain regions with low solid volume fractions and the lat-
ter in the domain regions with higher solid volume fractions
(Gidaspow, 1994; CFX-4 Documentation, 1994; Ochieng and
Despite this traditional interest for the assessment of Onyango, 2008) thus leading to a discontinuity in the rela-
Njs , in many industrial solid–liquid stirred reactors the best tion between inter-phase drag force and solid volume fraction.
compromise between cost reduction and process efficiency Recently, Tamburini et al. (2009a) proposed a piecewise cor-
is achieved by operating at an impeller speed lower than relation employing the two above formulations along with a
Njs (Oldshue, 1983; Rieger et al., 1988; Van der Westhuizen linear interpolation between them for an intermediate range
and Deglon, 2007; Jafari et al., 2012; Wang et al., 2012). Van of solid volume fractions thereby avoiding any discontinuity
der Westhuizen and Deglon (2007) investigated a mechani- and providing a monotonic behaviour in the inter-phase drag
cal flotation cell and stated that particles sedimentation and force vs solid volume fraction relation.

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx 3

1.3. Turbulence closure W=T/10

As far as the turbulence closure is concerned, the k–ε tur-

bulence model (along with the Eulerian-Eulerian treatment
of the two phase system) is the most widely employed

B=D/5
for solid–liquid systems. In particular, four main two-phase
extensions of the standard k–ε turbulence model (homogeneous,

H=T
per phase, dispersed and asymmetric) can be encountered in the
literature.

C=T/3
• In the homogeneous approach, only one k and one ε equations
are solved, where the physical properties of the mixture
are adopted: the two phases share the same k and ε value
and the transport equations for k and ε have no inter-phase
turbulence transfer terms. D=T/2
• In the per phase, or phase-specific, formulation, the turbu- T
lence model equations are solved for each phase. Additional
terms referring to the modelling of inter-phase transport of
k and ε have to be included in the equations relevant to each
phase.
• Alternatively, in the dispersed approach, suitable for dilute
suspensions, fluctuating quantities of the dispersed phase
are computed as functions of the mean characteristics of
the continuous phase and the ratio of the particle relaxation
time and eddy-particle interaction time (Gosman et al.,
1992). Continuous phase turbulence is modelled using the
standard k–ε model including extra terms which account for A=D/4
the influence of the dispersed phase on the continuous one
(Feng et al., 2012). Predictions of turbulence quantities for
the dispersed phases are obtained using the Tchen theory of
dispersion of discrete particles by homogeneous turbulence
T = 0.19 m
(Hinze, 1975).
Fig. 1 – System A.
• Recently Tamburini et al. (2011a) proposed the adoption
of the asymmetric k–ε turbulence model and applied it to
both consisting of solid–liquid suspensions in a cylindrical
dense solid–liquid suspensions at N ≤ Njs : according to this
flat-bottomed baffled tank stirred by a standard six-bladed
approach, since many particles may be unsuspended under
Rushton turbine.
partial suspension conditions, only the turbulence of the
liquid phase was accounted for and no turbulent viscosity
was calculated for the solid phase. (A) The first tank, sketched in Fig. 1, has internal diameter
T equal to 0.19 m, impeller diameter D equal to T/2 and
impeller clearance C equal to T/3, and is filled up to a
Montante and Magelli (2005) compared the first three for-
height H = T.
mulations. They observed that using more computationally
The experimental data (Tamburini et al., 2011a) regarding
demanding approaches like the per phase formulation does not
this system are relevant to three different suspensions of
lead to any significant improvement over the homogeneous
glass ballottini (ˇ = 2500 kg/m3 ) in deionized water with
formulation. Furthermore, the homogeneous k–ε turbulence
diameter range and mass fractions of:
model provides a satisfactory representation of the solids
- 212–250 ␮m, 33.8% wsolid /wliquid (referred as w/w in the
distribution throughout the vessel for a number of cases
following);
involving dense suspensions in stirred tanks provided that
- 212–250 ␮m, 16.9% w/w;
N ≥ Njs (Montante et al., 2001; Micale et al., 2004; Montante and
- 500–600 ␮m, 33.8% w/w.
Magelli, 2005; Khopkar et al., 2006; Kasat et al., 2008; Tamburini
These data provide the suspension curve (i.e. the mass
et al., 2009a).
fraction of suspended particles xsusp against the impeller
On the basis of the above considerations, the present work speed N) and the height of the sediment hsed visible on the
aims at comparing different CFD models including alternative lateral wall midway between two subsequent baffles.
formulations of either the inter-phase drag force or the turbu- (B) The second tank, investigated by Micheletti et al. (2003), is
lence closure in order to identify the best modelling procedure a flat bottomed tank with T = H = 0.29 m, impeller diameter
to deal with the incomplete suspension regime. D equal to T/3 and impeller clearance C equal to T/3.
The experimental data regard a suspension in deionized
2. Experimental data water of 600–710 ␮m glass particles (ˇ = 2470 kg/m3 ) with
25% w/w.
The data employed for the validation of the CFD simulations
derive from experiments made by the authors (Tamburini These data provide axial profiles of solid concentration.
et al., 2011a) and from the literature (Micheletti et al., More precisely, local steady state solid concentrations mea-
2003). Such data are relevant to two very similar systems surements were performed by a conductivity probe at a radial

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
4 chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx

position R/T = 0.35, midway between subsequent baffles and

Asymmetric k–ε
at different heights of the tank.
The above experimental data are the only ones available
in the literature concerning particle distribution under par-
tial suspension conditions. This is the main reason why all
the presently reported simulations were limited to a Rushton

X
X
X

X
turbine.

Turbulence Model
The void fraction of the particle bed lying on the bottom

Homogeneous
under no agitation conditions was estimated to be about 40%
for both systems on the basis of the literature (Tamburini et al.,
2011a) and specific measurements.

k–ε

X
X
3. Modelling

RANS simulations of the systems described in the previous

section were performed by adopting the Eulerian-Eulerian
Multi Fluid Model (MFM), available as a standard option in
the commercial finite volume CFD code Ansys(R) CFX4.4. With

Interphase drag force in momentum equations

correlation
Gidaspow-
respect to more recent releases, this code version offers a

Piecewise
greater flexibility in designing user-defined routines for non

Ergun
standard problems, thus, it has been extensively applied by

X
the present authors to a variety of non standard CFD problems
(Micale et al., 1999; Di Piazza and Ciofalo, 2002). In the present
work, the choice of this code allowed the implementation of

Gidaspow’s
the Excess Solid Volume Correction solids redistribution algo-

particle
rithm described in section 4.

dense

effect
According to the MFM fundamentals, the two phases are

X
treated as two interpenetrating continua: the continuity and
momentum equations are solved for each phase, thus obtain-
ing separate flow field solutions for the liquid and the solid
phase simultaneously. The two phases share the same pres-
Standard

sure field.
(Eq. (6))

Different modelling approaches were implemented and

tested in the present work: all the combinations simulated
X

X
X
X
are summarized in Table 1, while the main features of each
combination (i.e. tested modelling approach) are classified and
described in the following paragraphs.
Both phases

3.1. Reference Model

Additional terms arising from turbulence

A very simple model was employed to evaluate its capabil-

X

ity of predicting the whole set of the experimental data. As it

will be better described in the following, this model accounts
Table 1 – Summary of the modelling approaches features.

only for the turbulence of the liquid phase and adopts only
closure in continuity equation

phase only

a very standard formulation of the drag force. Moreover, also

the turbulence fluctuations of the volumetric fractions are not
Liquid

taken into account. As a second step, more refined modelling

X

approaches will be also tested. This simple model represents

the reference modelling approach with respect to which, all
the other approaches can be regarded as modifications. It was
already employed by the present authors (Tamburini et al.,
No terms

2011a) and will be only briefly described in the following.

X
X
X
X

3.1.1. Continuity equations

Assuming both phases to be incompressible, for each phase,
Modelling approach

the continuity equation is written as a function of the relevant

Homogeneous no terms
Homogeneous 2 terms

volume fraction:
Asymmetric 1 term
Reference Model

∂  ˛) = 0
 · (r˛ U
˛ (r˛ ) + ˛ ∇ (1)
∂t
PwC
DPE

∂  ˇ) = 0
 · (rˇ U
ˇ (rˇ ) + ˇ ∇ (2)
∂t

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx 5

Table 2 – Influence of liquid free stream turbulence on drag coefficient.

System A System B

231 ␮m 550 ␮m 655 ␮m

N (RPM) Brucato et al. Correlation N (RPM) Brucato et al. correlation N (RPM) Brucato et al. correlation

dp / CD,turb /CD,slip dp / CD,turb /CD,slip dp / CD,turb /CD,slip

121 2.78 1.02 122 6.89 1.28 400 14.54 3.66

157 3.33 1.03 161 8.24 1.49 500 17.58 5.71
209 4.18 1.06 213 9.91 1.84 600 20.55 8.53
258 5.03 1.11 263 11.79 2.42 700 23.10 11.69
299 5.64 1.16 316 13.83 3.29 800 25.45 15.30
362 6.44 1.23 368 15.71 4.36 900 27.87 19.77
493 8.21 1.48 413 17.08 5.32 1000 29.68 23.68
460 18.62 6.60 1100 32.57 30.95
510 20.21 8.15
624 23.64 12.46
678 25.20 14.87

where the subscripts ˛ and ˇ refer to the continuous and CD,slip = 0.4 (for Rep > 1000) (7b)
dispersed phases, respectively, r is volumetric fraction,  is
density and U is mean velocity. where Rep is the slip velocity Reynolds number:
Clearly,
 
˛ dp U  ˛
ˇ − U
r˛ + rˇ = 1 (3) Rep = (8)
˛

3.1.2. Momentum equations The influence of the free stream turbulence on particle drag
Assuming both phases to behave as Newtonian fluids, one has was accounted for by employing the correlation by Brucato
et al. (1998),

 ˛)
∂(r˛ U   3 
˛ + ˛ ∇ ˛ ⊗ U
 · (r˛ U  ˛) = ∇
 · (r˛ (˛ + t˛ )(∇
U  ˛ )T ))
 ˛ + (∇ U dp
∂t −4
CD,turb = CD,slip 1 + 8.76 × 10 (9)

− r˛ ∇P  ˛ˇ
 + r˛ ˛ g + M (4)

where  is the Kolmogorov length scale, calculated by

 ˛)  0.25
∂(rˇ U ˇ ⊗ U
 ˇ) = ∇  ˇ + (∇  ˇ )T )) 3
ˇ  · (rˇ U
+ ˇ ∇  · (rˇ ˛ (∇
U U = (10)
∂t ε
− rˇ ∇P  ˛ˇ
 + rˇ ˇ g − M (5)
where the dissipation of turbulent kinetic energy ε is provided
where g is gravity acceleration,  is viscosity, t is turbulent by the turbulence model.
viscosity, P is pressure and M is the inter-phase momen- Table 2 shows the correction of CD,slip provided by Eq. (9), for
tum transfer term. Notably, M was considered to be equal to the case of the Reference model. The values provided in the table
the drag force, while contributions due to other forces were are relevant to the average value of ε in the computational
neglected as suggested by the literature for similar systems domain. Clearly, the correction is crucial for large particles and
(Ljungqvist and Rasmuson, 2001; Fan et al., 2005; Fletcher and high impeller speeds.
Brown, 2009; Sardeshpande et al., 2011). In particular, accord-
ing to Fletcher and Brown (2009) the virtual mass and the 3.1.4. k–ε transport equations
lift forces have a negligible effect on the system dynamics;
∂

 
similarly, the Basset force in most cases is found to be much
˛ (r˛ k˛ ) + ∇  ˛ k˛ − r˛ ˛ + t˛
 r˛ ˛ U  ˛
∇k
smaller than the drag force (Khopkar et al., 2006; Kasat et al., ∂t k
2008). Also, Tatterson (1991) estimated virtual mass and lift U ˛ (∇
U ˛ + (∇
U T
 ˛ ) ) − ˛ ε˛ ]
= r˛ [t˛ ∇ (11)
forces to be much smaller than the drag force and, thus, to be
negligible when ˇ /˛ > 2 (as in the present work).

3.1.3. Interphase drag

 
 force  ∂  ˛ ε˛ − r˛ ˛ + t˛
 r˛ ˛ U  ˛
3 CD,turb   ˛ (r˛ ε˛ ) + ∇ ∇ε
rˇ ˛ U  ˛  (U
 ˛ˇ = C˛ˇ (U
ˇ − U
 ˛) = ˇ − U ˇ − U
 ˛) ∂t ε
M (6)
4 dp  
ε˛ 2
= r˛ C1 U
t˛ ∇  ˛ (∇
U ˛ + (∇
U ˛ )T ) − C2 ˛ ε (12)
k˛ k
The particle drag coefficient based on two phases slip veloc-
ity was estimated by means of the Clift et al. correlation (1978):

24 k2˛
CD,slip = (1 + 0.2Re0.63
p ) (for Rep ≤ 1000) (7a) t˛ = ˛ C (13)
Rep ε˛

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
6 chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx

The present version of the k–ε model was named Asymmet- C␣ˇ vs rˇ (Tamburini et al., 2009a).
ric k–ε turbulence model by Tamburini et al. (2011a).
Alternatives to specific aspects of the reference modelling  ˛ˇ = C˛ˇ (U
M ˇ − U
 ˛)
approach will be presented later on. In particular alternative  
C˛ˇ (rˇ max ) − C˛ˇ (rˇ min )
inter-phase drag force formulations and different turbulence = C˛ˇ (rˇ min ) + (rˇ − rˇ min )
rˇ max − rˇ min
closures will be tested.
ˇ − U
× (U  ˛) (15c)
3.2. Modifications to the Inter-phase Drag Force

3.2.1. Dense Particle Effect (DPE) 3.3. Modification to the turbulence closure
The Dense Particle Effect (DPE) modification, following
Gidaspow (1994), introduces an additional factor E in the inter- This sub-section is devoted to the description of the mod-
phase drag force equation to account for particle–particle elling approaches which present some modification in the
interactions in dense suspensions (i.e. high solids loading) so turbulence closure with respect to the Reference Model. The
that Eq. (6) is rewritten as: turbulence closure concerns both the continuity equations
(because of the products between the fluctuating volumetric
 
3 CD,turb   fractions and the fluctuating velocity components) and the
M ˇ − U
 ˛ˇ = C˛ˇ (U  ˛) = rˇ ˛ U  ˛  E (U
ˇ − U ˇ − U
 ˛)
momentum equation (because of the products between the
4 dp
(14a) fluctuating velocity components). In the Reference Model no tur-
bulence closure is adopted for the continuity equation and, as
far as the momentum equation is concerned, only the turbu-
where
lence closure of the liquid phase is addressed (asymmetric k–ε
−1.65 turbulence model).
E = (1 − rˇ ) (14b)
The phase specific treatment, in which k and ε fields are sep-
arately computed for each phase, was not taken into consider-
As shown by Eq. (14b), larger solid volume fractions lead to
ation in the present study because, as anticipated in the intro-
an enhancement of the drag force as a consequence of more
duction, it involves a greater complexity in the modelling of
frequent particle–particle collisions.
inter-phase momentum transfer terms and does not offer sig-
nificant advantages with respect to the homogeneous and the
3.2.2. Piecewise correlation (PwC) asymmetric models tested here (Montante and Magelli, 2005).
This inter-phase drag force modelling was proposed by
Tamburini et al. (2009a) and successfully applied for the case 3.3.1. Homogeneous k–ε turbulence model
of the start-up dynamics of a dense solid–liquid suspension. 3.3.1.1. Momentum equations. As concerns the modification
Three different equations are used for the computation of the of the reference momentum equation, both phases are con-
inter-phase drag force, each one for a specific solids volumetric sidered here to be turbulent and a turbulent viscosity appears
fraction range. in the momentum equations of both phases. As already men-
tioned in the introduction, the homogeneous k–ε turbulence
• low volumetric fractions (0 < rˇ < rˇ min ): model was found to provide a fair representation of the solid
equations 14 are applied, yielding: distribution throughout the vessel for a number of cases deal-
ing with dense suspensions in stirred tank (Montante et al.,
 ˛ˇ = C˛ˇ (U
M ˇ − U
 ˛) 2001; Micale et al., 2004; Montante and Magelli, 2005; Khopkar
  et al., 2006; Kasat et al., 2008). Therefore, a homogeneous k–ε tur-
3 CD,turb  
= rˇ ˛ U  ˛  (1 − rˇ )−1.65 (U
ˇ − U ˇ − U
 ˛) (15a) bulence model was employed here for comparison purposes
4 dp
with the Reference Model. k and ε transport equations remain
practically the same with respect to the single-phase case, but
• high volumetric fractions (rˇ max < rˇ < rˇ packed ): no volume fractions are present and all physical properties
the model adopts the inter-phase force initially proposed by appearing in these equations are the “mixture” averaged prop-
Ergun (1952) to deal with closely packed fixed-bed systems: erties. The interphase drag force formulation is equal to that
of the Reference Model.

 ˛)
∂(r˛ U
 ˛ˇ = C˛ˇ (U
M ˇ − U
 ˛) ˛ ˛ ⊗ U
 · (r˛ U
+ ˛ ∇  ˛) = ∇
 · (r˛ (˛ + t˛ )(∇
U  ˛ )T ))
 ˛ + (∇ U
 ∂t
 
rˇ2 ˛ rˇ ˛ U  ˛
ˇ − U  ˛ˇ
 + r˛ ˛ g + M
− r˛ ∇P (16)
= 150 + 1.75 ˇ − U
(U  ˛) (15b)
(1 − rˇ )d2p dp

 ˛)
∂(rˇ U
ˇ + ˇ ∇ ˇ ⊗ U
 · (rˇ U  ˇ) = ∇
 · (rˇ (˛ + tˇ )(∇
U ˇ + (∇
U ˇ )T ))
• intermediate volume fractions (rˇ min < rˇ < rˇ max ): ∂t
for this range a linear interpolation of the two previous − rˇ ∇P  ˛ˇ
 + rˇ ˇ g − M (17)
equations is used in order to avoid any discontinuity in
the C˛ˇ − rˇ relation: this unphysical behaviour would occur
k2
if only equations (15a) and (15b) were employed. Adopt- t˛ = ˛ C (18a)
ε
ing this interpolation with rˇ min and rˇ max set to 0.35 and
0.45, along with a CD computed via slip velocity (equa- k2
tˇ = ˇ C (18b)
tion 7), provides in most cases a monotonic dependence of ε

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx 7

3.3.1.2. k–ε transport equations. All these variants are relevant to the Reynolds averaging
of the governing (continuity and momentum) equation. The
∂

 
 k+∇  −  + t
 Uk 
∇k  ∇
 U(
= t ∇  + (∇
U  T ) − ε
 U) (19)
Favre averaging is a well-known alternative: in this case, no
∂t k additional terms arise in the continuity equation while an
additional term named turbulent dispersion force has to be

  included in the momentum equation (Sardeshpande et al.,
∂  −  + t
 Uk  ε    T)
 U) 2011; Gohel et al., 2012). The Reynolds averaging along with
 ε+∇ ∇ε = C1 t ∇ U(∇ U + (∇
∂t ε k the employment of diffusion terms in the continuity equation
ε2 (to model the product of fluctuating volumetric fraction and
− C2  (20)
k the fluctuating velocity) on the one hand and the Favre averag-
ing along with the inclusion of a suitable turbulence dispersion
force in the momentum equation on the other hand, can be
Velocities and physical properties appearing in equations considered as just different ways to get consistent results. For
(19)–(20) are the “mixture” averaged properties. this reason, only Reynolds averaged cases are investigated in
the present paper.
 = r˛ ˛ + rˇ ˇ (21)
4. Numerical details
 = 1 (r˛ ˛ U
U  ˛ + rˇ ˇ U
 ˇ) (22) Central differences were employed for all diffusive terms and
 the QUICK third-order upwind scheme for the advective ones.
The segregated SIMPLEC algorithm was adopted to couple
 = r˛ ˛ + rˇ ˇ (23)
pressure and velocity.
Baffles and impeller blades were considered as thin sur-
t = r˛ t˛ + rˇ tˇ (24)
faces. No-slip boundary conditions were assumed for all the
tank boundaries with the exception of the top surface where
3.3.1.3. Continuity equations. Turbulence closure may give
free-slip conditions were employed to simulate a free surface.
rise to alternative formulations of the continuity equation
As far as the treatment of the impeller-baffle relative rota-
(Table 1). In the first and simpler case, the continuity equa-
tion is concerned, both the steady state Multiple Reference
tions are Eqs. (1) and (2): the turbulent dispersion of volumetric
Frame (MRF) by Luo et al. (1994) and the time dependent Slid-
fractions is neglected (Murthy et al., 2007; Hosseini et al., 2010;
ing Grid (SG) algorithm by Murthy et al. (1994) were adopted in
Wang et al., 2010; Liu and Barigou, 2013), i.e. the continuity
the present work.
equation does not include the terms arising from Reynolds
Tamburini et al. (2011a) carried out both MRF and SG simu-
averaging. Summarizing, this former approach includes tur-
lations on system A and found very similar results as regards
bulence closure terms only in the momentum equations. The
suspension curves and sediment height. As a consequence,
Reference model case utilizes this approach. When a homo-
the MRF technique was adopted to simulate the system A in
geneous turbulence model is used, the neglect of turbulence
view of its large computational savings.
dispersion terms in the continuity equations gives rise to the
Conversely, Tamburini et al. (2013b) have shown that differ-
treatment called Homogeneous no terms in Table 1.
ent predictions can be found by employing the SG or the MRF
In a second case, named here Homogeneous 2 terms, the
approach for the case of system B where local quantities are
homogeneous model is used for turbulence and the two conti-
predicted: as a difference from the suspension curves (inte-
nuity equations include additional terms (Zhao et al., 2013):
gral data) of system A case, the CFD prediction of local solids
concentration profiles (data relevant to system B) requires a
∂  ˛ ) − t˛ ∇ 2 r˛ = 0
 · (r˛ U
˛ (r˛ ) + ˛ ∇ (25) more accurate calculation, especially if different modelling
∂t t
approaches have to be carefully compared. In accordance with
the literature (Ochieng and Lewis, 2006; Panneerselvam et al.,
ˇ
∂
(rˇ ) + ˇ ∇  ˇ ) − tˇ ∇ 2 rˇ = 0
 · (rˇ U (26)
∂t t 2008) more accurate predictions of the solid–liquid flow field
can be provided by the transient SG simulation. Therefore, the
where  t is a turbulent Schimdt number which is assumed to transient SG approach was adopted to simulate the system B
be equal to 0.8 as suggested by the literature (Montante and investigated by Micheletti et al. (2003). A time step equal to the
Magelli, 2007). According to Montante et al. (2001), contrary time the impeller needs to sweep an azimuthal angle equal to
to what happens with dynamic phenomena, the sensitivity a cell (one cell time step) was adopted for all the SG simulations
of the results to this parameter is negligible in steady-state performed.
solid–liquid systems. Summarizing, the Homogeneous 2 terms The position of both the surface separating the two
approach includes turbulence closure terms in both the con- domains both in the SG and in the MRF framework was chosen
tinuity and in the momentum equations by means of the in accordance with the assumptions of Luo et al. (1994) at the
homogeneous k–ε model. dimensionless radial position R = 0.62 (T/2).
Finally a third variant consists of adopting the asymmetric According to the systems axial symmetry, only one half
turbulence model and, coherently, in including the dispersion of the tank was included in the computational domain and
term only in the liquid phase continuity equation (Eq. (25)), two periodic boundaries were imposed along the azimuthal
while neglecting it in equation 26 for the solids phase conti- direction. The structured grid chosen for the discretization of
nuity. This approach is dubbed Asymmetric 1 term in Table 1. system A encompasses 53,760 cells distributed as 24 × 70 × 32
The corresponding results obtained with this approach turned along the azimuthal, axial and radial direction, respectively.
out to be similar to (and actually slightly worse than) those rel- The computational grid is finer in the proximity of the
evant to the Reference Model and thus will not be reported in the impeller where the largest gradients of the flow variables are
following for the sake of brevity. expected. A similar structured grid with 74,592 computational

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
8 chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx

1.0
0.9
0.8
Exp_231micron_33.8%
0.7
0.6 Exp_231micron_16.9%

xsusp [-]
0.5 Exp_550 micron_33.8%
0.4
CFD_231micron_33.8%
0.3
CFD_231micron_16.9%
0.2
0.1 CFD_550 micron_33.8%
(a)
0.0
0 100 200 300 400 500 600 700
N [rpm]

0.25 Exp_231micron_33.8%
Exp_231micron_16.9%
0.2
Exp_550micron_33.8%
CFD_231micron_33.8%
CFD_231micron_16.9%
0.15 CFD_550micron_33.8%
h sed/H [-]

0.1

0.05

(b)
0
0 100 200 300 400

N [rpm]

Fig. 2 – Reference Model MRF simulations versus experimental data (Tamburini et al., 2011a) for system A. (a) Suspension
curves, (b) normalized sediment heights.
cells (distributed as 72 × 37 × 28 along the axial, radial and The Unsuspended Solid Criterion (USC) proposed by
azimuthal direction, respectively) was employed for system Tamburini et al. (2011a) was employed to compute the
B of Micheletti et al. (2003). As well-known in the literature, mass fraction of suspended particles. In accordance with this
the choice of adopting a completely hexahedral grid may criterion solid particles are judged as being unsuspended if
allow, for any given required accuracy, a discretization degree and only if they belong to cells where rˇ ≥ rˇ packed .
much lower than that necessary for a tetrahedral discreti- Both these algorithms were found suitable to deal with
zation of the computational domain (Scully and Frawley, 2011; solid–liquid suspensions in baffled stirred tank under par-
Tamburini et al., 2012d). tial to complete suspension regimes (Tamburini et al., 2011a,
In order to assess the grid dependence of the CFD results, 2012a,b, 2013b).
Tamburini et al. (2011a) and Tamburini (2011) employed com- Typically in MRF steady state simulations 12,000 SIMPLEC
putational grids up to 8 times finer for the two systems and iterations were found to be sufficient for variable residuals to
found no significant grid dependence. In particular the mean settle to very low values in all the investigated cases and to
value of the discrepancy between the finest grid and the grid allow the ESVC algorithm to efficiently operate.
adopted here was found to be ∼1.0% for rˇ which is the most As concerns the time dependent SG simulations of system
significant variable in solid–liquid systems. Therefore, only B, 100 full revolutions were found sufficient to reach steady
coarse grid simulations were performed for the purpose of the state conditions, in agreement with the literature (Micale et al.,
present work in view of its comparative nature and of the large 2004; Tamburini et al., 2009a, 2011a, 2013b,c). The number of
number of test cases investigated. SIMPLEC iterations per time step was set to 30 to allow resid-
The Excess Solid Volume Fraction Correction (ESVC) algorithm uals to settle before moving to the next time step.
firstly proposed by Tamburini et al. (2009a) and successively As far as the initial guess or initial condition is concerned,
optimized by Tamburini et al. (2011a) was adopted since partial the fluid was assumed to be still and the particles were con-
suspension conditions were investigated in the present work. sidered to be motionless on the vessel bottom and at their
This algorithm guarantees that the maximum packing volu- maximum packing value rˇ packed .
metric fraction rˇ packed is not exceeded in the computational
domain even at very low impeller speeds, when a sediment 5. Results and discussion
is present. It operates iteratively at the end of each SIMPLEC
iteration forcing the volumetric fractions exceeding rˇ packed 5.1. Results of the Reference Model
(solid-excess) to be distributed among the neighbouring com-
putational cells and eventually moved towards regions where The comparison between the experimental values relevant to
no solid-excess is present. system A and system B and the corresponding CFD predictions

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx 9

1.0 1.0
0.9 Experimental_400rpm 0.9 Experimental_500rpm
0.8 Experimental_600rpm 0.8 Experimental_700rpm
0.7 Reference Model_400rpm 0.7 Reference Model_500rpm
0.6 Reference Model_600rpm 0.6 Reference Model_700rpm

z/H [-]
z/H [-]

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 1.0 2.0 3.0 4.0 5.0 6.0 7.0 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0
r /r _av [-] r /r _av [-]

Fig. 3 – Local axial profiles of solid concentration (system B): Reference Model SG simulations versus experimental data
(Micheletti et al., 2003).

1.0
0.9
0.8
Experimental
0.7
xsusp [-]

0.6 Njs_Zwietering
0.5
0.4 MRF_Reference Model
0.3 MRF_DPE
0.2
0.1 MRF_PwC
0.0
0 100 200 300 400 500 600 700

N [rpm]

Fig. 4 – MRF simulations versus suspension curve experimental data (Tamburini et al., 2011a) for the case of 212–250 ␮m
glass ballottini particles with a solid loading of 33.8% (w/w): comparison of different drag force formulations.

provided by the Reference Model are shown in the following following, some modifications to the present Reference Model
Figs. 2 and 3. Experimental data include the suspension curves will be tested with the aim at improving the predictions
and the normalized sediment height as visually observed on of the experimental data. Therefore, in the following the
the lateral wall midway two subsequent baffles for the case of discussion will be focused on the influence of the modelling
system A and local axial profiles of solid concentration for the modifications described in section 3 (concerning inter-phase
case of system B. drag force and turbulence closure) on the prediction of the
As far as the suspension phenomenon is concerned, experimental data.
the discrepancy between the experimental and the CFD
data relevant to system A is acceptable. The suspension
curves are shown in Fig. 2a where some overestimations of 5.2. Influence of model changes–System A
the experimental data can be observed, especially for the
case of the largest particles. Only for the case of the low- 5.2.1. Interphase drag force modelling
concentration/small-particles case a slight underestimation is 5.2.1.1. Suspension curves. As it can be seen in Fig. 4, neither
found at about 250–300 rpm. of the proposed modifications to the interphase drag force
The sediment height is another feature of the suspension provides satisfactory results: both of them significantly over-
phenomenon and is reported in Fig. 2b. In this case a very good estimate the experimental data as well as the predictions of
agreement is found: only slight differences can be observed. the Reference Model, especially at the intermediate impeller
As it concerns the prediction of the particle distribution speeds. The present modifications appear to lead to a prema-
phenomenon, the prediction of the experimental local axial ture suspension of a large amount of solid particles.
profiles of solid concentration via the Reference Model for sys- However, both the Dense Particle Effect approach and the
tem B is reported in Fig. 3: each profile is relevant to a specific Piecewise Correlation approach are able to predict the typical
impeller speed ranging from highly incomplete (i.e. 400 rpm) sigmoidal trend of the suspension curve.
to complete (i.e. 1100 rpm > Njs , shown only in section 5.3) As regards the DPE approach, the drag force enhancement
suspension conditions. All the experimental profiles are sat- due to the dense particle effect may cause an overestimation
isfactorily followed by the CFD data, although the particle of the momentum exchange between phases thus providing
distribution degree is slightly underestimated. More precisely, a suspended solid fraction higher than the experimental one.
at 400 rpm and 500 rpm the experimental data at the very top The higher the local rˇ the higher the influence of the dense
of the vessel are not perfectly matched by the CFD predictions; particle effect term, so one may expect a larger influence at
while a somewhat larger discrepancy can be observed below low N when the particle distribution degree is very small.
the impeller at impeller speeds of 600 rpm and 700 rpm. Actually, when particles are lying packed on the bottom as
Summarizing, all the experimental data are fairly well sediment they show the same rˇ = rˇ-packed , so that the drag
predicted by the very simple Reference Model, although there force enhancement is the same in all cells where sediment is
seems to be still room for further improvements. In the present, independently of the impeller speed.

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
10 chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx

1.0
0.9
0.8
0.7 Experimental

xsusp [-]
0.6
Njs_Zwietering
0.5
0.4 MRF_Reference Model
0.3 MRF_DPE
0.2
0.1 MRF_PwC
0.0
0 100 200 300 400 500 600 700
N [rpm]

Fig. 5 – MRF simulations versus suspension curve experimental data (Tamburini et al., 2011a) for the case of 212–250 ␮m
glass ballottini particles with a solid loading of 16.9% (w/w): comparison of different drag force formulations.

When the impeller speed is very low, the drag force is not (10%, w/w). By a close inspection of Figs. 4 and 5 it is possible
sufficient to suspend many particles (even if the DPE enhance- to note that at, for example, 150 rpm the xsusp values predicted
ment is accounted for) because the slip velocity between the by the DPE and the PwC approaches for the case of the more
liquid flow and the sediment is still too low. As the agita- diluted suspension (Fig. 5) are about twice those predicted for
tion speed increases, the slip velocity increases (the liquid the case with the larger solids loading (Fig. 4). Independently of
velocity increases, while the solids velocity remains still low) the solids loading, when the impeller speed is very low, par-
and the drag force reaches its critical value for the sus- ticles are packed over the bottom exhibiting the maximum
pension of a large fraction of particles (i.e. when the curve allowed volume fraction as well as the same particle–particle
starts to rise with a large slope). In general, when a DPE interactions. As a consequence the DPE approach on the one
approach is employed the drag force reaches its critical value hand, and the PwC approach on the other hand, produces drag
at impeller speeds lower than those of the Reference Model, force enhancements which remain constant with the solids
thus yielding an overestimation of the amount of suspended loading thus predicting a similar quantity of particles to be
particles. suspended. For the more dilute suspension this quantity cor-
As far as the Piecewise Correlation (PwC) approach is con- responds to a larger fraction of suspended particles.
cerned, Fig. 4 shows that at very low impeller speeds the At higher impeller speeds, no differences between the xsusp
mass fractions of suspended solids predicted by the PwC and values predicted by the DPE and the PwC approaches are appre-
DPE methods are very similar thus suggesting that even the ciable. This occurrence is not surprising as for moderate solids
increase of the drag force induced by the adoption of the Ergun loadings, when most particles are suspended throughout the
equation is not sufficient to allow particles to suspend when vessel, rˇ is relatively low. As a consequence, the PwC approach
the slip velocity is low. At higher impeller speeds the over- switches to Eq. (15a), thus yielding drag force predictions iden-
prediction provided by the PwC approach appears to be similar tical to the DPE approach.
and slightly larger than that of the DPE model. As it is shown in Fig. 6, the two approaches provide over-
This behaviour is confirmed by Fig. 5 where the compar- predictions larger than those of the Reference Model also for the
ison of the same approaches is presented for the case of a case of the solid–liquid suspension with larger particle size
solid–liquid suspension with the same geometry and particle thus confirming that taking into account the particle–particle
size, but a lower solids loading. The Ergun equation provides a interactions via the proposed drag force modifications is not
larger increasing of the interphase drag force thus leading to suitable for partial suspension conditions.
xsusp values higher than those predicted by the DPE approach. As concerns the comparison between the DPE and the
As a difference with the previous case, here a large over- PwC approaches, Fig. 6 shows that they provide practically
estimate of the experimental data is observed for the two identical results thus suggesting that for the present case
alternative approaches even at the lowest impeller speeds. the Ergun equation and the Gidaspow correction produce the
The predictions of Hosseini et al. (2010) concerning cloud same increment of the inter-phase drag force. This difference
height and homogenization degree show a similar overesti- with respect to the previous cases (where different increments
mate at the lower impeller speeds for a similar solids loading were found) may be linked to the presence of the Brucato

1.0
0.9
0.8
0.7 Experimental
xsusp [-]

0.6
Njs_Zwietering
0.5
0.4 MRF_Reference Model
0.3
MRF_DPE
0.2
0.1 MRF_PwC
0.0
0 100 200 300 400 500 600 700
N [rpm]

Fig. 6 – MRF simulations versus suspension curve experimental data (Tamburini et al., 2011a) for the case of 500–600 ␮m
glass ballottini particles with a solid loading of 33.8% (w/w): comparison of different drag force formulations.

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx 11

0.25 0.15
Experimental Experimental
MRF_Reference Model MRF_Reference Model
0.20 0.12
MRF_DPE MRF_DPE
MRF_PwC MRF_PwC
hsed/H [-]

0.15
0.09

hsed/H [-]
0.10
0.06

0.05
0.03
0.00
0 100 200 300 400 0.00
N [rpm] 0 100 200 300 400
N [rpm]
Fig. 7 – MRF simulations versus sediment height
experimental data (Tamburini et al., 2011a) for the case of Fig. 8 – MRF simulations versus sediment height
212–250 ␮m glass ballottini particles with a solid loading of experimental data (Tamburini et al., 2011a) for the case of
33.8% (w/w): comparison of different drag force 212–250 ␮m glass ballottini particles with a solid loading of
formulations. 16.9% (w/w): comparison of different drag force
formulations.

et al. drag-turbulence correlation (1998) whose effect is neg- experimental data and the CFD predictions obtained by the
ligible for the lower particle diameter case as it can be seen in Reference Model. Also, PwC yields larger underestimations than
Table 2. In accordance with this correlation, the higher the DPE.
particle diameter, the higher the enhancement of the drag In Fig. 8 (small particles, lower solids loading) a surprisingly
coefficient. This correction does not appear in the Ergun equa- different behaviour can be observed: as a difference with the
tion, thus suggesting that the gap in drag force predictions corresponding suspension curve results in Fig. 5, here all the
between the two approaches for the lower particle size cases approaches yield very similar results and predict values of hsed
may be bridged by the action of the drag coefficient enhance- in good agreement with the experiments.
ment due to the liquid free stream turbulence. This hypothesis This occurrence can be explained by considering the 3D
might be confirmed by observing in Fig. 6 that at 368 rpm and visualization of the sediment estimated by the simulations,
413 rpm the DPE approach yields xsusp slightly larger than those shown in Fig. 9. Here the computational cells where rˇ ≥
by PwC approach since the higher the impeller speed, the lower rˇ packed (sediment) are indicated in grey. Contours of the solids
the Kolmogorov length scale , the higher the drag coefficient volumetric fraction on a vertical diametrical plane are also
enhancement. reported. The DPE and the PwC results exhibit a sediment
distribution along the peripheral wall (and thus a hsed value)
5.2.1.2. Sediment height. The following Figs. 7–9 show the similar to the reference case, but differ in the way sediment
normalized sediment height for the same configurations for is suspended from the central region of the tank bottom (and
which Figs. 4–6 report suspension curves. Notably, the sedi- thus in the xsusp value).
ment height data represent only a local piece of information The impeller speed of 171 rpm showed in Fig. 9 is just an
and should be regarded as less decisive for comparison pur- example: the same behaviour is also observable at 125 rpm
poses than the more global information provided by the and 147 rpm. Such behaviour is not in accordance with
suspension curves. the extensively validated Reference Model results and with
By observing Fig. 7 (small particles, higher solids loading) experimental visualization (Tamburini, 2011) where particles
considerations similar to those presented in the former seem to suspend preferentially from the peripheral region of
sub-section for Fig. 4 can be made. Both the DPE and the the sediment. Conversely, the analysis of the 3-D sediment
PwC approaches yield an underestimation of both the visualization for the other two suspension cases showed

Fig. 9 – 3-D sediment visualization plot upon contour plots of solid volumetric fractions on a vertical diametrical plane at
171 rpm for the case of 231 ␮m ballottini particles at 16.9% (w/w): comparison of different drag force formulations.

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
12 chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx

0.25 quite lower than that of the liquid phase, thereby leading to
Experimental
compute small velocity gradients, smaller turbulence produc-
MRF_ Reference Model
0.20 tions and thus a less intensive turbulence. This fact causes
MRF_DPE the suspension of a quantity of particles lower than that of
MRF_PwC the Reference Model.
0.15
As a difference, when the additional terms are included
hsed/H [-]

in the continuity equation (Homogeneous 2 terms), very large

0.10
overestimation of xsusp are obtained, even at very low impeller
speeds (Fig. 11). This occurrence is linked to the nature of
0.05 the additional terms including the Laplacian of the volumet-
ric fraction. The action of such term is particularly strong at
0.00 the sediments interface and results in making this interface a
0 100 200 300 400 diffuse one from which solid particles are more easily dragged
N [rpm] by the liquid.
In accordance with Mersmann et al. (1998) the suspension
Fig. 10 – MRF simulations versus sediment height
of solid particles is strictly linked to two different phenomena:
experimental data (Tamburini et al., 2011a) for the case of
the bottom lifting and the avoidance of settling. Which of them
500–600 ␮m glass ballottini particles with a solid loading of
may be the one controlling the particle suspension depends on
33.8% (w/w): comparison of different drag force
particle and liquid properties, on solids hold up, on impeller
formulations.
to tank diameter ratio. By applying the “decisive criterion for
suspension” by Mersmann et al. (1998) (see following Eq. (27)) it
that the three model approaches predict similar sediment results that bottom lifting is the phenomenon controlling the
shape evolution as a function of the impeller speed. Clearly, solid suspension for all the cases investigated in the present
a feature like the shape of the sediment is the result of a work, i.e. the amount of energy necessary to allow the particles
delicate balance among different forces including pressure to lift from the tank bottom εBL is higher than that necessary
gradients, drag and turbulent stresses so that it is not surpris- to avoid the suspended particles to settling εAS (the ratio of Eq.
ing that even slight model changes may results in significant (27) results higher than 1 for all the cases studied here):
differences in the predictions.
Fig. 10 (large particles, high solids loading) is perfectly in εBL T

5/2 d 1
p
≈ 527 (27)
accordance with the corresponding Fig. 6: the DPE and the εAS D H Ar1/8 [r n 1/4
ˇ av (1 − rˇ av ) ]
PwC approaches provide the same under-predictions of the
experiments and of the data obtained by applying the Reference where Ar is Archimedes number and n is an exponent depend-
Model. ing on the value of Ar. Full details can be found in the work by
Mersmann et al. (1998).
5.2.2. Turbulence closure The inclusion of the additional terms in the continuity
5.2.2.1. Suspension curves. Fig. 11 reports experimental and equation largely aids the bottom lifting phenomenon thus
computational results concerning the suspension curves for allowing a large amount of particles to be prematurely sus-
system A with small particles and high solids loading. The pended.
adoption of the homogeneous k–ε turbulence model along The relative effectiveness of alternative approaches
with no additional terms in the continuity equation (Homo- appears to be quite independent of particle concentration and
geneous no terms) provides results which are similar to those mean diameter. By observing Fig. 12 (small particles, low solids
obtained by the Reference Model, even closer to the experimen- loading) and Fig. 13 (large particles, high solids loading) con-
tal data at the intermediate impeller speeds. The reason of this siderations similar to those relevant to Fig. 11 can be made: the
finding is closely related to the homogeneous k–ε turbulence xsusp values predicted by the Homogeneous no terms approach
model fundamentals. The two phases share the same turbu- are lower than those obtained by the Reference Model and gen-
lence kinetic energy k and the same dissipation of turbulence erally closer to the experimental data. Conversely the adoption
kinetic energy ε which are calculated by employing the mix- of the Homogeneous 2 terms approach gives rise for both sus-
ture quantities. At low to intermediate impeller speeds, when pensions to over-predictions at all the impeller speeds and
a sediment is present, the mixture velocity used by the tur- particularly at low speeds. These results are in agreement with
bulence model near the sediment–fluid interface results to be the findings by Fletcher and Brown (2009) who found that the

1.0
0.9
0.8 Experimental
0.7
xsusp [-]

0.6 Njs_Zwietering
0.5 Reference Model
0.4
0.3 MRF_Homogeneous_no terms
0.2
0.1
MRF_Homogeneous_2 terms
0.0
0 100 200 300 400 500 600 700
N [rpm]

Fig. 11 – MRF simulations versus suspension curve experimental data (Tamburini et al., 2011a) for the case of 212–250 ␮m
glass ballottini particles with a solid loading of 33.8% (w/w): comparison of different turbulence closures.

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx 13

1.0
0.9
0.8
0.7 Experimental

xsusp [-]
0.6
Njs_Zwietering
0.5
0.4 Reference Model
0.3 MRF_Homogeneous_no terms
0.2
MRF_Homogeneous_2 terms
0.1
0.0
0 100 200 300 400 500 600 700
N [rpm]

Fig. 12 – MRF simulations versus suspension curve experimental data (Tamburini et al., 2011a) for the case of 212–250 ␮m
glass ballottini particles with a solid loading of 16.9% (w/w): comparison of different turbulence closures.

0.25 0.15
Experimental Experimental
MRF_Reference Model MRF_Reference Model
0.20 0.12
MRF_Homogeneous_no terms MRF_Homogeneous_no terms
MRF_Homogeneous_2 terms
MRF_Homogeneous_2 terms
hsed/H [-]

0.15 0.09

hsed/H [-]
0.10 0.06

0.05
0.03

0.00
0 100 200 300 400 0.00
0 100 200 300 400
N [rpm]
N [rpm]
Fig. 13 – MRF simulations versus suspension curve
Fig. 14 – MRF simulations versus sediment height
experimental data (Tamburini et al., 2011a) for the case of
experimental data (Tamburini et al., 2011a) for the case of
500–600 ␮m glass ballottini particles with a solid loading of
212–250 ␮m glass ballottini particles with a solid loading of
33.8% (w/w): comparison of different turbulence closures.
33.8% (w/w): comparison of different turbulence closures.

homogeneous and the asymmetric k–ε models provide similar Homogeneous no terms predictions appear to be quite similar
results, but the homogeneous model leads to a lower particle and in a good agreement with the experimental data.
dispersion. They stated also that the inclusion of turbulent dis- The Homogeneous 2 terms approach provides very low hsed
persion (as done in the case of Homogeneous 2 terms approach) values. This result is in accordance with the large underesti-
leads to a great enhancement of particle dispersion and that mation of the amount of unsuspended particles presented in
the soundness of its adoption at low impeller speeds has to be Fig. 11.
carefully validated. As concerns the small particle, low solids loading case,
Fig. 15, by comparing the predictions of the Reference Model and
5.2.2.2. Sediment height. In Fig. 14 the comparison of the sedi- the Homogeneous no terms, it appears that these are very similar
ment heights provided by the different turbulence closures for only for impeller speeds lower than 180 rpm. This is in accor-
the case of 212–250 ␮m glass ballottini particles with a solid dance with the results shown in Fig. 12, where similar xsusp
loading of 33.8% (w/w) is shown. The Reference Model and the values can be observed for the two approaches in the same

Fig. 15 – MRF simulations versus sediment height experimental data (Tamburini et al., 2011a) for the case of 212–250 ␮m
glass ballottini particles with a solid loading of 16.9% (w/w): comparison of different turbulence closures.

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
14 chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx

0.25 Reference Model predictions generally appear better than those

Experimental
obtained with the Homogeneous no terms approach. Finally, the
MRF_Reference Model
0.20 inclusion of the turbulence closure additional terms in the
MRF_Homogeneous_no terms
continuity equations gives rise to a very large underestima-
MRF_Homogeneous_2 terms
0.15 tion of the hsed values thus suggesting that the Homogeneous 2
hsed/H [-]

terms approach may be unsuitable to manage suspension

0.10 phenomena in solid–liquid agitated systems operated under
medium-to-highly incomplete suspension conditions.
0.05
5.3. Influence of model changes—System B
0.00
0 100 200 300 400 The experimental data employed so far for quantitative vali-
N [rpm]
dation purposes are relevant to the amount of solid particles
lying on the vessel walls (bottom and lateral wall), but they do
Fig. 16 – 3-D sediment visualization plot upon contour not provide a quantitative description of the particle distribu-
plots of solid volumetric fractions on a vertical diametrical tion over the vessel.
plane at 171 rpm for the case of 231 ␮m ballottini particles Therefore, local axial profiles of solid concentrations for a
at 16.9% (w/w): comparison of different turbulence closures. similar system (the system B by Micheletti et al., 2003) were
selected from the literature for a further comparison of the
impeller speed range. As a difference with Fig. 12, at impeller different modelling approaches.
speeds higher than 180 rpm, the Reference Model predictions
are closer to the experimental data than the Homogeneous no 5.3.1. Interphase drag force modelling
terms ones. Fig. 18 reports sliding grid (SG) predictions and experimental
The Homogeneous 2 terms approach leads to hsed values solids concetration profiles for five different impeller speeds.
lower than the experiments and the other predictions. At It should be observed that, although SG results are intrinsi-
147 rpm and 171 rpm the discrepancies between the Homo- cally time-dependent, at a radial location sufficiently away
geneous 2 terms and the relevant experimental hsed are lower from the impeller tip (as in the present case where data refer to
than the corresponding discrepancies between the xsusp values R = 0.35 T, while the outer impeller tip is at R = 0.16 T) results are
observable in Fig. 12. The use of Homogeneous 2 terms approach practically time-independent so that instantaneous results
results into the suspension of most particles from the bottom are practically coincident with the time-averaged ones. When
centre (Fig. 16), in a similar way as previously shown in Fig. 9 the slip velocity between the two phases is still too low as in
for the influence of the drag model in Section 5.2.1, thus pro- the case at 400 rpm, accounting for the particle–particle inter-
viding largely different xsusp at similar hsed . As already stated actions in the drag force formulation does not provide any
for the cases of the DPE and PwC approaches, the vigorous significant difference: as a matter of fact Fig. 18 shows that at
suspension from the bottom centre is not consistent with the this impeller speed the predicted axial profiles of rˇ are very
experimental evidence for a T/2 Rushton turbine thus suggest- similar. This is in accordance with the results shown in Section
ing that the Homogeneous 2 terms is not suitable to deal with 5.2.1, e.g. Fig. 4, for the case of system A: when the slip veloc-
the case under study. ity between the two phases is still too low, enhancements of
As far as the case of a suspension with large particle size the drag force do not produce the suspension of the particles
is concerned, the comparison among the different modelling which remain on the bottom thus showing a negligible axial
approaches is shown in Fig. 17. dispersion degree.
These results are coherent with the corresponding ones for As a difference, at 500 rpm, the increased slip velocity
the suspension curve depicted in Fig. 13: the Homogeneous no along with the particle–particle interaction effects cause an
terms provides the highest hsed values and the lowest xsusp val- excessive degree of suspension to be predicted. Although
ues, on the contrary, the Homogeneous 2 terms provides the both the DPE and PwC approaches predictions are not far
lowest hsed and the largest xsusp values. from the experimental data at intermediate heights, they
Summarizing, both the Reference Model and the Homoge- do not succeed in predicting the amount of particles placed
neous no terms approach are able to provide satisfactory pre- in the proximity of the vessel bottom, where the drag
dictions of the normalized sediment height. Particularly, the force enhancement provided by both approaches lead to an

1.0
0.9
0.8
0.7
xsusp [-]

0.6 Experimental
0.5 Njs_Zwietering
0.4 Reference Model
0.3
0.2
MRF_Homogeneous_no terms
0.1 MRF_Homogeneous_2 terms
0.0
0 100 200 300 400 500 600 700
N [rpm]

Fig. 17 – MRF simulations versus sediment height experimental data (Tamburini et al., 2011a) for the case of 500–600 ␮m
glass ballottini particles with a solid loading of 33.8% (w/w): comparison of different turbulence closures.

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx 15

1.0 1.0
0.9 Experimental_400rpm 0.9 Experimental_500rpm

0.8 SG_Reference Model 0.8 SG_Reference Model

0.7 0.7 SG_DPE
SG_DPE
0.6 0.6

z/H [-]
z/H [-]

SG_PwC
0.5 SG_PwC 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
r /r _av [-] r /r _av [-]

1.0 1.0
Experimental_600rpm 0.9 Experimental_700rpm
0.9
0.8 SG_Reference Model 0.8 SG_Reference Model
0.7 0.7 SG_DPE
SG_DPE
0.6 0.6

z/H [-]
z/H [-]

SG_PwC
0.5 SG_PwC 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
r /r _av [-] r /r _av [-]

1.0
0.9 Experimental_1100rpm
0.8
SG_Reference Model
0.7
0.6 SG_DPE
z/H [-]

0.5 SG_PwC
0.4
0.3
0.2
0.1
0.0
0 1 2 3 4 5 6 7
r /r _av [-]

Fig. 18 – SG simulations versus steady state local axial profiles of particle concentrations (midway between subsequent
baffles and at R = 0.35 T) at different impeller speeds for the case of 600–710 ␮m glass ballottini particles with a solid loading
of 25% (w/w): comparison of different drag force formulations. Experiments were collected by Micheletti et al. (2003).

excessive particle suspension. These two approaches show 5.3.2. Turbulence closure
some difference between each other in the lower part of the As it can be seen in Fig. 19, at 400 rpm the results provided
tank allegedly because of Ergun’s equation effect which is a by the Reference Model and the Homogeneous no terms approach
peculiarity of the PwC approach. are practically identical and in a good agreement with the
At 600 rpm all the approaches provide the same results up experimental data. Conversely, the inclusion of the additional
to z/H = 0.7, showing the same underestimation of the experi- turbulence-based terms in the continuity equation leads to
mental data. As far as the upper part of the vessel is concerned an excessive suspension coherent with the premature parti-
(z/H > 0.7) the adoption of Gidaspow’s correction (DPE and PwC) cle suspension shown by the suspension curves results for
seems to generate a better agreement with the experimental system A using the same approach, Figs. 11, 12 and 17 as
data than the Reference Model. Note that at these low rˇ the already discussed for the suspension curves for the case of
DPE and the PwC approaches practically reduce to the same system A.
Eq. (15a). At 500 rpm the Homogeneous 2 terms approach causes a dra-
At higher impeller speed (i.e. 700 rpm and 1100 rpm) matic overestimation of the particle distribution degree. The
the particles are better distributed throughout the ves- Homogeneous no terms approach behaves slightly better than
sel thus leading to a rˇ value close to the rˇ av in the Reference Model in the lowest part of the vessel, while it
almost all computational cells. In these conditions, the appears slightly worse in the upper part.
drag force enhancement produces by the DPE and the PwC At 600 rpm and 700 rpm, unlike the Reference Model
approaches appears practically not to affect the particle and the Homogeneous no terms approach predictions, the
distribution. Homogeneous 2 terms approach does not underestimate the

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
16 chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx

1.0 1.0
0.9 Experimental_400rpm 0.9 Experimental_500rpm
0.8 Reference Model 0.8 Reference Model
0.7 SG_Homogeneous_no terms 0.7 SG_Homogeneous_no terms
0.6 SG_Homogeneous_2 terms 0.6 SG_Homogeneous_2 terms

z/H [-]
z/H [-]

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 1.0 2.0 3.0 4.0 5.0 6.0 7.0 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0
r /r _av [-] r /r _av [-]

1.0 1.0
0.9 Experimental_600rpm 0.9 Experimental_700rpm
0.8 Reference Model 0.8 Reference Model
0.7 SG_Homogeneous_no terms 0.7 SG_Homogeneous_no terms
0.6 SG_Homogeneous_2 terms 0.6 SG_Homogeneous_2 terms

z/H [-]
z/H [-]

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 1.0 2.0 3.0 4.0 5.0 6.0 7.0 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0
r /r _av [-] r /r _av [-]

1.0
0.9 Experimental_1100rpm
0.8 Reference Model
0.7 SG_Homogeneous_no terms
0.6 SG_Homogeneous_2 terms
z/H [-]

0.5
0.4
0.3
0.2
0.1
0.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0
r /r _av [-]

Fig. 19 – SG simulations versus steady state local axial profiles of particle concentrations (midway between subsequent
baffles and at R = 0.35 T) at different impeller speeds for the case of 600–710 ␮m glass ballottini particles with a solid loading
of 25% (w/w): comparison of different turbulence closures.
rˇ experimental values below the impeller plane, while it turbines were performed. Different modelling approaches
under-predicts the peak of the experimental profile near the were compared in order to assess their ability to predict the
impeller mid-plane. Also in these cases the results of the fraction of suspended particles (xsusp ), the sediment height
Reference Model and of the Homogeneous no terms approach (hsed ) and the particle distribution (rˇ vs z) at impeller speeds
are very similar: the only significant differences concern covering the whole range from partial to complete suspension
the upper region of the vessel, where the Reference Model regimes. Different formulations both of the inter-phase drag
predictions are closer to the experiments at 600 rpm, while force and of the turbulence closure were tested. Simulations
the Homogeneous no terms approach results are in better were carried out with the commercial code Ansys CFX4.4 by
agreement with the experiments at 700 rpm. adopting the Eulerian-Eulerian Multi Fluid Model.
At 1100 rpm, under complete suspension conditions, the A very simple CFD model here regarded as the Reference
Reference Model and the Homogeneous no terms approach Model along with a drag coefficient correction for background
provide the same predictions which follow well the exper- turbulence (Brucato et al., 1998) predicted with fair accuracy
imental profile. The Homogeneous 2 terms predictions, being both the suspension curves and the local axial concentration
closer to homogeneous conditions, are further from the exper- profiles, although xsusp values were somewhat overestimated
iments. when large particles were considered.
As regards the modification of the interphase drag force,
6. Conclusions the DPE and the PwC approaches were tested, which make use
of Gidaspow’s correction for dense particle effects and usually
Numerical simulations of dense solid–liquid suspensions can lead to good results when impeller speeds higher than
within flat bottomed vessels stirred by standard Rushton Njs are considered (Ochieng and Onyango, 2008; Tamburini

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx 17

et al., 2009a). Conversely, under partial suspension conditions, On the above grounds, the Reference Model (Asymmetric no
the use of such correction can lead to an enhancement of terms) and Homogeneous no terms approaches were judged
the inter-phase drag force causing in turn an over-prediction as being the most suitable to model partial suspension
of xsusp and of the particle distribution degree. This effect is conditions, with only minor differences between the two.
especially relevant at intermediate impeller speeds: in fact, In fact present results show that such two models in which
when the impeller speed is low, slip velocities are also low neither dispersion terms in the continuity equation nor
so that the dense particle correction does not provide appre- drag enhancement corrections for dense-particle effects are
ciable effects, while at high impeller speeds the solids are included, gives already satisfactory predictions, or even slight
more uniformly distributed and, again, drag corrections are overpredictions of the suspended solids fraction (suspension
of little consequence. Furthermore, for the case of the PwC curve). Therefore, it is not surprising that the inclusion of
approach, the drag force overestimation within the solid sed- either effect worsens the prediction via an indirect or direct
iment region can further increase because of the adoption of enhancement of the liquid phase’s ability to suspend the
Ergun’s equation, despite its use at high local volume frac- particles (see discussion in sections 5.2.1a and 5.2.2).
tions is suggested by the literature (CFX-4 Documentation, On overall, the results shown in the present work point
1994; Ochieng and Onyango, 2008; Holbeach and Davidson, out a difficulty in the formulation of a single model which
2009). In this regard, some cases can be found in the litera- is universally capable of reliably and accurately predicting
ture where the adoption of drag modification due to dense both the amount of unsuspended solids resting on the vessel
particle effects is found to provide unsatisfactory predictions bottom and the distribution of solid particles within the
(e.g. Wadnerkar et al., 2012). Also, it is very worth noting suspension volume.
that all the models tested in the present work include the Finally, it is worth stressing that the present conclusions
drag correction due to the background turbulence which in are based on a large number of experimental data covering
some works is judged as an alternative to the correction for a range of impeller speeds, particle sizes and concentrations,
dense particle effects (e.g. Wadnerkar et al., 2012; Ochieng and especially quantities concerning both the solids suspen-
and Onyango, 2008). Clearly, this is conceptually wrong since sion (i.e. suspension curves and normalized sediment height)
the two corrections are due to different aspects. However, it and distribution (local axial profiles of solids volume fraction)
is also true that each correction was originally developed in phenomenon.
the absence of the other and it remains to be demonstrated
whether they can be simultaneously applied in their original Acknowledgements
form. As an example, the correction of CD due to the liquid
background turbulence is based on the ration dp /: the relation This work was carried out under financial support by Mini-
which allows  to be calculated from ε (Eq. (10)) was formu- stero dell’Università e della Ricerca, PRIN-2006 contract no
lated for the case of isotropic turbulence, a hypothesis which 2006091953 004: “Experimental analysis, modelling and simula-
is questionable in the case of a system with a high solids tions of bioslurry reactors for soil remediation”.
concentration.
As far as turbulence closure is concerned, adopting the References
homogeneous k–ε model but neglecting the two additional
terms in the continuity equations (Homogeneous no terms Angst, R., Kraume, M., 2006. Experimental investigations of
approach) yields a good agreement between the predictions stirred solid/liquid systems in three different scales: particle
and the experimental data as regards the mass fraction of distribution and power consumption. Chem. Eng. Sci. 61,
2864–2870.
suspended particles and the axial profiles of solid concentra-
Armenante, P.M., Nagamine, E.U., Susanto, J., 1998.
tion, while it leads to an overestimation of the sediment height Determination of correlations to predict the minimum
(taken midway between two subsequent baffles). Conversely, agitation speed for complete solid suspension in agitated
the adoption of the homogeneous k–ε turbulence model along vessels. Can. J. Chem. Eng. 76, 413–419.
with the inclusion of the additional turbulence terms in the Barresi, A., Baldi, G., 1987. Solid dispersion in an agitated vessel.
continuity equations (Homogeneous 2 terms approach) leads Chem. Eng. Sci. 42, 2949–2956.
Bohnet, M., Niesmak, G., 1980. Distribution of solids in stirred
to large over-predictions of experimental solids volume frac-
suspensions. Germ. Chem. Eng. 3, 57–65.
tions, thus suggesting that this approach is not suitable to
Bourne, J.R., Sharma, R.N., 1974. Suspension characteristics of
manage incomplete suspensions. The disappointing effect solid particles in propeller-agitated tanks. In: Proceedings of
produced by the inclusion of the two terms is rather surprising the First European Conference on Mixing and Centrifugal
since these are included to account for the products between Separation, Cambridge, BHRA, Cranfield, UK, B3, pp. 25–39.
the fluctuating volumetric fraction and the fluctuating velocity Brucato, A., Grisafi, F., Montante, G., 1998. Particle drag
arising from the Reynolds averaging in the continuity equa- coefficients in turbulent fluids. Chem. Eng. Sci. 53,
3295–3314.
tion. However, it should be observed that the extra-terms in
Brucato, A., Cipollina, A., Micale, G., Scargiali, F., Tamburini, A.,
Eqs. (25) and (26) are not the exact terms, but rather their 2010. Particle suspension in top-covered unbaffled tanks.
diffusive (Boussinesq-like) approximation, which may well be Chem. Eng. Sci. 65, 3001–3008.
inadequate in the very region where these terms are impor- CFX-4 Documentation. Ansys®, 1994.
tant, i.e. near the sediment–liquid interface, where anisotropy Clift, R., Grace, J.R., Weber, M.E., 1978. Bubbles, Drops and
is large. Therefore, the findings of the present work highlight Particles. Academic Press, New York, San Francisco, London.
the need of further studies in order to find an alternative and Di Piazza, I., Ciofalo, M., 2002. MHD free convection in a
liquid-metal filled cubic enclosure. II. Internal heating. Int. J.
better way to account for these terms. The comparison of the
Heat Mass Transfer 45, 1493–1511.
present results with ones obtained by an alternative approach Ergun, S., 1952. Fluid flow through packed columns. Chem. Eng.
based on Favre averaging along with the inclusion of the tur- Progress 48, 89–94.
bulent dispersion force in the momentum equation may be a Fan, L., Mao, Z., Wang, Y., 2005. Numerical simulation of
good starting point. turbulent solid-liquid two-phase flow and orientation of

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
18 chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx

slender particles in a stirred tank. Chem. Eng. Sci. 60, Micheletti, M., Nikiforaki, L., Lee, K.C., Yianneskis, M., 2003.
7045–7056. Particle concentration and mixing characteristics of
Feng, X., Li, X., Cheng, J., Yang, C., Mao, Z.-S., 2012. Numerical moderate-to-dense solid-liquid suspensions. Ind. Eng. Chem.
simulation of solid-liquid turbulent flow in a stirred tank with Res. 42, 6236–6249.
a two-phase explicit algebraic stress model. Chem. Eng. Sci. Montante, G., Micale, G., Magelli, F., Brucato, A., 2001.
82, 272–284. Experiments and CFD predictions of solid particle distribution
Fletcher, D.F., Brown, G.J., 2009. Numerical simulation of solid in a vessel agitated with four pitched blade turbines. Chem.
suspension via mechanical agitation: effect of the modeling Eng. Res. Des. 79, 1005–1010.
approach, turbulence model and hindered settling drag law. Montante, G., Magelli, F., 2005. Modelling of solids distribution in
Int. J. Comp. Fluid Dynam. 23, 173–187. stirred tanks: analysis of simulation strategies and
Gidaspow, D. (Ed.), 1994. Multiphase Flow and Fluidization. comparison with experimental data. Int. J. Comp. Fluid Dyn.
Academic Press, San Diego, USA. 19, 253–262.
Gohel, S., Joshi, S., Azhar, M., Horner, M., Padron, G., 2012. CFD Montante, G., Magelli, F., 2007. Mixed Solids Distribution in
modeling of solid suspension in a stirred tank: Effect of drag stirred vessels: experiments and Computational Fluid
models and turbulent dispersion on cloud height. Int. J. Chem. Dynamics simulations. Ind. Eng. Chem. Res. 46, 2885–2891.
Eng. (art. no. 956975). Montante, G., Paglianti, A., Magelli, F., 2012. Analysis of dilute
Gosman, A.D., Lekakou, C., Politis, S., Issa, R.I., Looney, M.K., 1992. solid-liquid suspensions in turbulent stirred tanks. Chem.
Multidimensional modeling of turbulent two-phase flows in Eng. Res. Des. 90, 1448–1456.
stirred vessels. AIChE J. 38 (12), 1946–1956. Murthy, J.Y., Mathur, S.R., Choudhury, D., 1994. CFD simulation of
Hinze, J.H., 1975. Turbulence. McGraw-Hill, New York. flows in stirred tank reactors using a sliding mesh technique.
Holbeach, J.W., Davidson, M.R., 2009. An Eulerian-Eulerian model IChemE Symp. Ser. 136, 341–348.
for the dispersion of a suspension of microscopic particles Murthy, B.N., Ghadge, R.S., Joshi, J.B., 2007. CFD simulations of
injected into a quiescent liquid. Eng. Appl. Comput. Fl. Mech. gas-liquid-solid stirred reactor: prediction of critical impeller
3, 84–97. speed for solid suspension. Chem. Eng. Sci. 62, 7184–7195.
Hosseini, S., Patel, D., Ein-Mozaffari, F., Mehrvar, M., 2010. Musil, L., Vlk, J., 1978. Suspending solid particles in an agitated
Study of solid-liquid mixing in agitated tanks through conical-bottom tank. Chem. Eng. Sci. 33, 1123–1131.
computational fluid dynamics modelling. Ind. Eng. Chem. Res. Nienow, A.W., 1968. Suspension of solid particles in
49, 4426–4435. turbine-agitated baffled vessels. Chem. Eng. Sci. 23, 1453–1459.
Jafari, R., Chaouhi, J., Tanguy, P.A., 2012. A comprehensive review Ochieng, A., Lewis, A.E., 2006. CFD simulation of solids off-bottom
of just suspended speed in liquid-solid and gas-liquid-solid suspension and cloud height. Hydrometallurgy 82, 1–12.
stirred tank reactors. Int. J. Chem. Reac. Eng. 10 (art. no. R1). Ochieng, A., Onyango, M.S., 2008. Drag models, solids
Jirout, T., Moravec, J., Rieger, F., Sinevic, V., Spidla, M., Sobolic, V., concentration and velocity distribution in a stirred tank.
Tihon, J., 2005. Electrochemical measurement of impeller Powder Tech. 181, 1–8.
speed for off-bottom suspension. Inzynieria chemiczna i Oldshue, J.Y., Sharma, R.N., 1992. The effect of off-bottom
procesowa 26, 485–497. distance of an impeller for the ‘Just Suspended Speed’. Njs.
Jirout, T., Rieger, F., 2011. Impeller design for mixing of AIChE Symp. Ser. 88, 72–78.
suspensions. Chem. Eng. Res. Des. 89, 1144–1151. Oldshue, J.Y. In: Fluid Mixing Technology, Chapter 5. McGraw-Hill,
Kasat, G.R., Khopkar, A.R., Ranade, V.V., Pandit, A.B., 2008. CFD New York, NY, 1983.
simulation of liquid-phase mixing in solid-liquid stirred Oshinowo, L.M., Bakker, A., 2002. CFD modeling of solids
reactor. Chem. Eng. Sci. 63, 3877–3885. suspensions in stirred tank. In: Proceedings of the Conference
Kee, R.C.S., Tan, R.B.H., 2002. CFD simulation of solids suspension on Computational Modeling of Materials, Minerals and Metals
in mixing vessels. Can. J. Chem. Eng. 80, 721–726. Processing, pp. 205–215.
Khopkar, A.R., Kasat, G.R., Pandit, A.B., Ranade, V.V., 2006. Panneerselvam, R., Savithri, S., Surender, G.D., 2008. CFD
Computational Fluid Dynamics simulation of the solid modeling of gas-liquid-solid mechanically agitated contactor.
suspension in a stirred slurry reactor. Ind. Eng. Chem. Res. 45, Chem. Eng. Res. Des. 86, 1331–1344.
4416–4428. Ren, C., Jiang, X., Wang, J., Yang, Y., Zhang, X., 2008.
Liu, L., Barigou, M., 2013. Numerical modelling of velocity field Determination of critical speed for complete solid suspension
and phase distribution in dense monodisperse solid-liquid using acoustic emission method based on multiscale analysis
suspensions under different regimes of agitation: CFD and in stirred tank. Ind. Eng. Chem. Res. 47, 5323–5327.
PEPT experiments. Chem. Eng. Sci. 101, 837–850. Rewatkar, V.B., Raghava Rao, K.S.M.S., Joshi, J.B., 1991. Critical
Ljungqvist, M., Rasmuson, A., 2001. Numerical simulation of the impeller speed for solid suspension in mechanically agitated
two-phase flow in an axially stirred vessel. Chem. Eng. Res. three-phase reactors. 1. Experimental part. Ind. Eng. Chem.
Des. 79, 533–546. Res. 30, 1770–1784.
Luo, J.Y., Issa, R.I., Gosman, A.D., 1994. Prediction of Rieger, F., 2000. Effect of particle content on agitator speed for
impeller-induced flows in mixing vessels using multiple off-bottom suspension. Chem. Eng. J. 79, 171–175.
frames of reference. IChemE Symp. Ser. 136, 549–556. Rieger, F., Ditl, P., Havelkova, O., 1988. Suspension of solid
Mersmann, A., Werner, F., Maurer, S., Bartosch, K., 1998. particles–Concentration profiles and particle layer on the
Theoretical prediction of the minimum stirrer speed in vessel bottom. In: Proceedings of the 6th European
mechanically agitated suspensions. Chem. Eng. Proc. 37, Conference on Mixing, Pavia, Italy, 24-26 May,
503–510. pp. 251–258.
Micale, G., Brucato, A., Grisafi, F., Ciofalo, M., 1999. Prediction of Sardeshpande, M.V., Sagi, A.R., Juvekar, V.A., Ranade, V.V., 2009.
flow fields in a dual-impeller stirred vessel. AIChE J. 45, Solid suspension and liquid phase mixing in solid-liquid
445–464. stirred tanks. Ind. Eng. Chem. Res. 48, 9713–9722.
Micale, G., Carrara, V., Grisafi, F., Brucato, A., 2000. Solids Sardeshpande, M.V., Juvekar, V.A., Ranade, V.V., 2011. Solid
suspension in three phase stirred tanks. Chem. Eng. Res. Des. suspension in stirred tanks: UVP measurements and CFD
78, 319–326. simulations. Can. J. Chem. Eng. 89, 1112–1121.
Micale, G., Grisafi, F., Brucato, A., 2002. Assessment of particle Scully, J., Frawley, P., 2011. Computational Fluid Dynamics
suspension conditions in stirred vessels by means of a analysis of the suspension of nonspherical particles in a
Pressure Gauge Technique. Chem. Eng. Res. Des. 80, stirred tank. Ind. Eng. Chem. Res. 50, 2331–2342.
893–902. Selima, Y.S., Fangary, Y.S., Mahmoud, N.A., 2008. Determination
Micale, G., Grisafi, F., Rizzuti, L., Brucato, A., 2004. CFD simulation of minimum speed required for solids suspension in stirred
of particle suspension height in stirred vessels. Chem. Eng. vessels using pressure measurements. Can. J. Chem. Eng. 86,
Res. Des. 82, 1204–1213. 661–666.

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020
CHERD-1399; No. of Pages 19
ARTICLE IN PRESS
chemical engineering research and design x x x ( 2 0 1 3 ) xxx–xxx 19

Tamburini, A., 2011. Suspension phenomena in solid-liquid laser sheet and image analysis technique. Chem. Eng. Sci. 87,
agitated systems. Doctoral Thesis. University of Palermo. 341–358.
Fotograf, Palermo, Italy. ISBN: 9788895272979. Tamburini, A., Brucato, A., Cipollina, A., Micale, G., Ciofalo, M.,
Tamburini, A., Cipollina, A., Micale, G., Ciofalo, M., Brucato, A., 2013b. CFD simulations of dense solid-liquid suspensions in
2009a. Dense solid–liquid off-bottom suspension dynamics: baffled stirred tanks: prediction solid particle distribution.
simulation and experiment. Chem. Eng. Res. Des. 87, Chem. Eng. J. 223, 875–890.
587–597. Tamburini, A., Cipollina, A., Micale, G., Brucato, A., Ciofalo, M.,
Tamburini, A., Gentile, L., Cipollina, A., Micale, Brucato, A., 2009b. 2013c. CFD prediction of solid particle distribution in baffled
Experimental investigation of dilute solid-liquid suspension stirred vessels under partial to complete suspension
in an unbaffled stirred vessels by a novel pulsed laser based conditions. Chem. Eng. Trans. 32, 1447–1452.
image analysis technique. Chem. Eng. Trans. 17, 531–536. Tatterson, G.B., 1991. Fluid Mixing and Gas Dispersion in Agitated
Tamburini, A., Cipollina, A., Micale, G., Brucato, A., Ciofalo, M., Tanks. McGraw-Hill, New York.
2011a. CFD Simulations of dense solid-liquid suspensions in Van der Westhuizen, A.P., Deglon, D.A., 2007. Evaluation of solids
baffled stirred tanks: predictions of suspension curves. Chem. suspension in a pilot-scale mechanical flotation cell: the
Eng. J. 178, 324–341. critical impeller speed. Min. Eng. 20, 233–240.
Tamburini, A., Cipollina, A., Micale, G., 2011b. CFD simulation of Wadnerkar, D., Utikar, R.P., Tade, M.O., Pareek, V.K., 2012. CFD
solid liquid suspensions in baffled stirred vessels below simulation of solid-liquid stirred tanks. Adv. Powder Technol.
complete suspension speed. Chem. Eng. Trans. 24, 1435–1440. 23, 445–453.
Tamburini, A., Cipollina, A., Micale, G., Brucato, A., 2011c. Dense Wang, F., Mao, Z., Shen, X., 2004. Numerical study of solid-liquid
solid-liquid suspensions in top-covered unbaffled stirred two-phase flow in stirred tanks with Rushton impeller. (II)
vessels. Chem. Eng. Trans. 24, 1441–1446. Prediction of critical impeller speed. Chin. J. Chem. Eng. 12,
Tamburini, A., Cipollina, A., Micale, G., Brucato, A., Ciofalo, M., 610–614.
2012a. CFD simulations of dense solid-liquid suspensions in Wang, L., Zhang, Y., Li, X., Zhang, Y., 2010. Experimental
baffled stirred tanks: prediction of the minimum impeller investigation and CFD simulation of liquid-solid-solid
speed for complete suspension. Chem. Eng. J. 193–194, dispersion in a stirred reactor. Chem. Eng. Sci. 65, 5559–5572.
234–255. Wang, S., Boger, D.V., Wu, J., 2012. Energy efficient solids
Tamburini, A., Brucato, A., Cipollina, A., Micale, G., Ciofalo, M., suspension in an agitated vessel-water slurry. Chem. Eng. Sci.
2012b. CFD predictions of sufficient suspension conditions in 74, 233–243.
solid-liquid agitated tanks. Int. J. Nonlinear Sci. Num. Sim. 13, Yamazaki, H., Tojo, K., Miyanami, K., 1986. Concentration profiles
427–443. of solids suspended in a stirred tank. Powder Tech. 48,
Tamburini, A., Cipollina, A., Micale, G., Brucato, A., 2012c. 205–216.
Measurements of Njs and power requirements in unbaffled Zhao, H., Zhang, T., Liu, Y., Zhang, Z., Zhao, C., 2013. Numerical
bioslurry reactors. Chem. Eng. Trans. 27, 343–348. simulations of solid-liquid stirred tank with an improved
Tamburini, A., La Barbera, G., Cipollina, A., Ciofalo, M., Micale, G., Intermig impeller. AIP Conf. Proc. 1542, 1286–1289.
2012d. CFD simulation of channels for direct and reverse Zhu, Y., Wu, J., 2002. Critical impeller speed for suspending solids
electro dialysis. Desalination Water Treatment 48, 370–389. in aerated agitation tanks. Can. J. Chem. Eng. 80, 682–687.
Tamburini, A., Cipollina, A., Micale, G., Brucato, A., 2013a. Particle Zwietering, T.N., 1958. Suspending of solid particles in liquids by
distribution in dilute solid-liquid unbaffled tanks via a novel agitators. Chem. Eng. Sci. 8, 244–253.

Please cite this article in press as: Tamburini, A., et al., Influence of drag and turbulence modelling on CFD predictions of solid liquid suspensions
in stirred vessels. Chem. Eng. Res. Des. (2013), http://dx.doi.org/10.1016/j.cherd.2013.10.020