Review On Gas-Liquid Mixing Analysis in Multiscale Stirred Vessel Using CFD-libre

Rev Chem Eng 28 (2012): 171189 2012 by Walter de Gruyter Berlin Boston. DOI 10.
1515/revce-2012-0003
Review on gas-liquid mixing analysis in multiscale stirred
vessel using CFD
Baharak Sajjadi, Abdul Aziz Abdul Raman *, Shaliza
Ibrahim and Raja Shazrin Shah Raja Ehsan Shah
Faculty of Engineering , Department of Chemical
Engineering, University of Malaya, 50603 Kuala Lumpur,
Malaysia , e-mail: azizraman@um.edu.my
* Corresponding author
Abstract
This review aims to establish common approaches and equa-
tions used in computational uid dynamics (CFD) analysis for
gas-liquid mixing operations and investigate their strengths and
weaknesses. The review concluded that with a sufcient com-
puting strength, Eulerian-Lagrangian approaches can simulate
detailed ow structures for dispersed multiphase ow with high
spatial resolution. Turbulence is an important factor in uid
dynamics, and literature conrmed that k- is the most widely
used turbulence model. However, it suffers from some inherent
shortcomings that stemmed from the assumption of isotropy of
turbulence and homogenous mixing, which is suitable for very
high Reynolds number in unbafed stirred vessels. In CFD
simulations for gas-liquid systems in stirred vessels, bubble size
distribution is the most important parameter; hence, different
techniques for formulation of bubble size equations have been
investigated. These techniques involve source and sink terms
for coalescence or breakup and provide a framework in which
the population balance method together with the coalescence
and breakup models can be unied into three-dimensional CFD
calculations. Different discretization schemes and solution algo-
rithms were also reviewed to conrm that third-order solutions
provide the least erroneous simulation results.
Keywords: computational uid dynamics (CFD); frame
grid; multiuid models; population balances; stirred vessel.
1. Introduction
Stirred vessels are widely used in the chemical, biochemical,
mineral, pharmaceutical, and biotechnological industries to
perform chemical operations such as halogenation, hydro-
genation, oxidation, crystallization, polymerization, and bio-
logical fermentation (Van den Akker 2006 , Buffo et al. 2012 ).
Generally, mixing affects around 25 % of all process industry
operations (Yeoh et al. 2004 ). The primary aim of mixing
is to increase the interfacial mass transfer rate in these ves-
sels, which is related to the interfacial area between phases,
volume fraction hold-up, bubble or droplet or particle size
distribution, dynamic equilibrium between coalescence and
breakage rate, turbulence level, uid properties of dispersed
and continuous phases, and mechanical mixing parameters
such as impeller size and model, vessel size, and ow velo-
city (Bouuyatiotis and Thonton 1967 ). As it is impossible to
cover all the mixing process parameters experimentally, com-
putational uid dynamics (CFD) has created opportunities to
visualize the mixing phenomena (Haresh et al. 2010 ). CFD is
able to predict uid ows, chemical reaction rates, mass and
heat transfer rates, and other occurrences by solving a set of
appropriate mathematical equations (Ding et al. 2010 ). CFD
also provides useful information for regions with intense or
mild turbulence zones, Reynolds stresses, vortex structures,
circulation patterns, ow behavior, and many other param-
eters (Haresh et al. 2010 ). Notwithstanding its potential, fun-
damental knowledge in multiphase ow simulations is still
lacking because of its complexities, particularly in stirred
vessels that have movable pieces. For example, CFD is still
undeveloped to assess the non-homogeneity impact and phase
property changes in stirred polymerization reactors (Haresh
et al. 2010 ). Although review papers on experimental meth-
ods and geometrical parameters are available in literature
(Mavros 2001 , Ascanio et al. 2004 ), a complete review on
CFD applications and mixing models in reactors is not yet
available (Van den Akker 2006 , Ochieng et al. 2009 ), which
is the focus of this review.
This review provides an insight into different approaches
used in simulating multiphase ow and different terms of
momentum balances applied in gas-liquid systems in stirred
vessels. The rst term is momentum and stress transfer from
the continuous to the dispersed phase, where different turbu-
lence models have been investigated in this work. The second
term is momentum transferred between the dispersed and the
continuous phase, which depends on the force balance on each
bubble and the dispersed phase distribution. Effects of drag
and non-drag forces on bubbles were investigated based on
force balance. The breakup and coalescence equations were
investigated for distribution of dispersed phase. This review
also identies different methods available for solving popula-
tion balance equations (PBEs). As an integrated solution for
various equations is required, various framework, discretiza-
tion schemes, and solution algorithms are also identied and
analyzed. In addition, CFD simulation using mass transfer
was also reviewed because results for these runs can also be
analyzed from a mass transfer perspective.
2. General approaches
Two approaches are available for multiuid modeling,
Eulerian-Eulerian (E-E, a two-uid model) and Eulerian-
Lagrangian (E-L). In the E-E multiuid method, the
Brought to you by | University of Malaya Library
Authenticated | azizraman@um.edu.my author's copy
Download Date | 11/29/13 5:55 AM
172 B. Sajjadi et al.: Use of CFD in gas-liquid mixing analysis
continuous phase and the different dispersed phases are
treated as a continuous interpenetrating media and described
in terms of their volume fractions. The ow elds are solved
for both phases, and the interactions between the phases are
explained through the source terms (Buffo et al. 2012 ). The
interfacial momentum transfer between liquid and gas in this
model includes investigations on the inuence of turbulent
uctuations on the effective momentum transfer, mass force,
Basset force, lift force, and the drag force (Van Wachem and
Almstedt 2003 ). However, the Eulerian approach suffers
from numerical diffusion from the smearing of the gas frac-
tion over the entire grid cell (Sokolichin et al. 1997 ). In the
E-L approach, the continuous phase is treated in a Eulerian
framework (using averaged equations), whereas particles
are tracked through the ow in a Lagrangian manner (Ding
et al. 2010 ). A force balance is solved for each particle, and
each particle is individually tracked in the system (Ashraf
Ali et al. 2008 ). The second Newtonian law is also solved for
each individual particle (Ashraf Ali and Pushpavanam 2011 ),
and a collision model is applied to handle particle encounters
(Van Wachem and Almstedt 2003 ). Generally, E-L models
simulate ow structures with higher spatial resolution com-
pared with E-E models and show more benets for dispersed
multiphase ows. Other benets include the following:
Reactions occurring within or on surface of individual par-
ticles, and different transport processes can be modeled in
a realistic way.
Bubble coalescence and breakup, four-way momentum ex-
change, and bubble-induced turbulence can be rigorously
modeled.
The exchange of momentum between the operating phases
due to drag, lift, and virtual mass through the interface can
be modeled more accurately in comparison with E-E.
Dispersed size distribution can be accounted without great
difculty (Buwa et al. 2006 ).
Time-averaged properties could be predicted well using
this viewpoint (Buwa et al. 2006 ).
However, the E-L approach faces three problems: false
uctuations of the continuous phase velocity result in sud-
den changes of dispersed phase local volume fraction; the
amplitude of the false velocity uctuation increases when
a small grid is applied; as the volume fraction of dispersed
phase increases, the interaction between the two phases will
increase as well, which is not accounted for by this method
due to the considerably high demand of computational cost
(Ochieng and Onyango 2010 ); thus, this approach for a three-
dimensional (3D) ow is more applicable in dilute systems
(Yeoh et al. 2005 ).
In both approaches, there are two basic equations that
should be solved, continuity and momentum balance, as given
in eqs. (1) and (2), respectively:

( ) ( )

+

k k k k k
U
t x

(1)

+ + + +

( ) ( ) -
k k k k k k k k k k k Ik
P
u u u g T F
t x x x

(2)
where the T
k
is the stress tensor from the liquid and F
IK
is
the momentum transferred through the force balance.
More details of the E-E approach and the E-L approach can
be found in the works of Kitagawa et al. (2001) and Ashraf
Ali et al. (2008) , respectively.
3. Stress tensor from the continuous phase
Eq. (3) expresses the turbulence treated through stress tensor
terms ( T
k
) in the momentum equation balance.

( )

_
_
+ +

,
,
,
2
( ) - - -
3
j
i k
k k k t k ij k i j
j i k j
u
u u
T u u
x x x x

(3)
The term of
i j
u u

introduces the Reynolds stresses.

_
+ +

,
2
( )
3
j
i
i j ij t
j i
U
U
u u k
x x

(4)
A variety of turbulence models are available by coupling
this term to other terms in the momentum balance equation,
such as Reynolds averaged Navier-Stokes (RANS), large
eddy simulation (LES), and the direct numerical simulation
(DNS).
3.1. RANs model
The denition of kinetic energy of uctuating motions is an
aspect of concern in CFD modeling of stirred vessels, which
is dened as the summation of the periodic component (due
to nonrandom oscillations caused by the cyclic passage of the
impeller blades) and the random component (due to turbulent
eddies). RANS-based models depend on the random compo-
nent of uctuating motions to reduce computational effort
by avoiding resolution of unsteady turbulent eddies directly
(Montante et al. 2001 ).
RANS equations are divided into two categories: the
Reynolds stress models (RSMs) and the eddy-viscosity
model. The two-equation eddy-viscosity models are based
on the assumption that an analogy exists between Reynolds
stress and viscous stress and the turbulent ow is isotropic.
Standard k- , renormalization group k- (RNG), and realiz-
able k- are three different appearances of two equations, the
k- family model and the k- models, where k is the turbulent
kinetic energy, is the energy dissipation rate, and is the
turbulent frequency. Table 1 shows the relevant equations in
literature. Mixture, per phase, and dispersed are the three dif-
ferent extensions of the standard k- turbulence model avail-
able for simulating turbulence in multiphase systems. This
method is not adequate to capture the tracer concentration his-
tory, which is used to calculate the liquid-phase mixing time.
In addition, anisotropy consideration and different time scales
do not lead to a signicant improvement over the standard k-
turbulence model (Kasat et al. 2008 ). Some improvements on
the performance were achieved through RNG and realizable
B. Sajjadi et al.: Use of CFD in gas-liquid mixing analysis 173
Table 1 Summary of turbulence models.
Model S

Parameters
RANS-based
models
Standard
(Rahimi and
Parvareh 2005)
1, 2,
-
k
S S
G
C C
k k

_

,
C
, S
= 0.009, C
1, S
= 1.44, C
2, S
= 1.92,
K , S
= 1,
, S
= 1.314
RNG (Rahimi
and Parvareh
2005)
1, 2,
- -
k
RNG RNG
G
C C
k k k

_

,

3 0
3
1-
1
C
+
, C
,RNG
= 0.845, C
1 ,RNG
= 1.42,
C
2 ,S
= 1.68,
k,s
=
,S
= 0.719,
0
= 4.8, = 0.012,

,
k
E

E
2
= 2 E
ij
E
ij
,

0.5
j
i
ij
j i
u
u
E
x x
_
+

,
Realizable k -
model (Rahimi
and Parvareh
2005)

2
1,Re 2,Re
- C E C
k
_

, +

,Re
max 0.43;
5
C
1
+
]
, C
1 ,Re
= 0.09, C
2 ,Re
= 1.9,

k,Re
= 1,

, Re
= 1.2,

k
E
, E
2
= 2 E
ij
E
ij
, 0.5
j
i
ij
j i
u
u
E
x x
_
+

,
k-

1
_
+ + + 1

,
1
]
* *
( )
-
j
k
j j j
u k
k k k
P k
t x x x

+

1
_
+ + 1

,
1
]
2
( )
-
j
k
j
d
j j j j
u
P
t x k
k k
x x x x

lim *
2
max ,
ij ij
t
S S
k
C

_

,

= 0.52, =
0
f

,
0
= 0.0708,
*
= 0.09, = 0.5

+

+
*
* 3
1 85
0.6, , ,
1 100 ( )
ij jk ki
S
f

0 for 0, 0.125 for 0
d d
j j j j
k k
x x x x

>

RSM

2
-
- -
j ij ij i
k ik jk
k k k
ijk ij ij ij
k
u
u
u
t x x x
C
x

_

+ +

,
+ +

j
i
ij
i j
u
u P
x x
1
+ 1

1
]

2
j
i
ij
k k
u
u
x x

( )
1
ijk i j k i jk j ik
C u u u P u P u
+ +

k
= 0.82 C

= 0.09
LES Delafosse et al.
2008

+ +

2
2
1
( ) -
Re
SGS
ij
i i
i j
j i j j
u u p
u u
t x x x x

-
SGS
ij i j i j
u u u u

1
- -2
3
SGS SGS SGS
ij kk ij T ij
S

2 2
2
SGS
ij ij
T S S
L S L S S

1
2
i j
ij
j i
u u
S
x x
_

+

,

( )
1
3
min ,
S S i
L kd C V
models (Wang et al. 2010 ). The effect of small-scale turbu-
lence is introduced in the RNG k- model through a random
forcing function in the Navier-Stokes equations. In this model,
the effect of small-scale motion from the governing equations
is removed by a modied viscosity and by expressing their
effects in terms of larger-scale motion (Hjertager et al. 2002 );
thus, the RNG model gives better results in predicting the tur-
bulent kinetic energy in the wall jet below the impeller com-
pared with standard k- , which overpredicts it (Harris et al.
1996 ). The realizable k- contains a new transport equation for
the dissipation rate ( ) and a new formulation for the turbulent
viscosity. C

(a constant in the denition of the turbulence
length scale) is not a constant in this model and is instead a
function of local strain rate and rotation of uid. Furthermore,
different source and sink terms are used in this model in the
transport equation for eddy dissipation. Both the RNG and
the realizable k- models have indicated important improve-
ments over the standard k- model where the ow features
include strong streamline curvature, rotation, and vortices
(Rahimi and Parvareh 2005 ). Despite all these modications,
the k- turbulence model provides unrealistic results in sev-
eral cases due to inaccurate turbulent kinetic energy distribu-
tion prediction especially in the regions close to the impeller,
particularly due to the isotropic nature of the k- turbulence
models and the unsteadiness and transient nature of the ow
(Zadghaffari et al. 2010 , Coroneo et al. 2011 ) within these
regions. Assumption of isotropic turbulence and homogene-
ity only produces accurate results for ows at high Reynolds
numbers (Trivellato 2011 ). In this situation, it is assumed that
the energy is injected into eddy motions on a scale in the order
the size of eddies and then transferred to smaller length scales
until it is dissipated at the smallest stable viscous scale (the
Kolmogorov scale) (Trivellato 2011 ), but this assumption is
not practical in real systems. Bujalski et al. (2002) reported
the calculation of power number through the integration of
turbulent energy dissipation rate over the whole vessel, which
was about 20 % lower than the experimental value. This prob-
lem became more severe in regions where the ow is highly
anisotropic and in systems containing high solid loading
(Ochieng and Onyango 2010 ); thus, the k- model cannot
give accurate results for the tangential velocity components
and in describing curvilinear and highly swirling ows. The
RSM is another turbulence model that solves the transport
equations for all Reynolds stresses, represents the full effect
of turbulence in the momentum equations, and removes the
assumption of the isotropic eddy viscosity. However, RSM
has shortcomings due to numerical difculties and high com-
putational effort and suffers from poor convergence behav-
ior. Additionally, both transient and steady-state results
using RSM have underpredicted turbulence kinetic energy
(Montante et al. 2001 , Murthy and Joshi 2008 ). Hence, a
compromise algebraic stress model (ASM) was developed to
overcome these problems.
In the ASM model, the convection and diffusion terms are
derived from a local equilibrium condition, and the Reynolds
differential equations are converted into a set of algebraic
equations. However, there is no diffusion or damping factors
in the equations, and the expressions are implicit for Reynolds
stress components; hence, this model is not applicably strong
and cannot perform for complex 3D ows. The explicit alge-
braic stress model (EASM) is another version of ASM that was
developed using a tensor polynomial expansion theory. In this
method, the Reynolds stresses was described as an explicit
algebraic correlation of turbulence characteristic quantities,
mean strain rate tensor, and rotation rate tensor. In com-
parison, the EASM produces numerical stability and lower
computational effort. Additionally, the EASM also performs
well in comparison with the DNS results as explained later
in this work. Feng et al. (2012) conducted a simulation for
complex 3D turbulent ows in a stirred vessel and suggested
the EASM as a suitable alternative model to the standard k-
model for simulating anisotropic turbulent ows in agitated
vessels. As k- model equations cannot be integrated to the
wall, k- have been derived, where the turbulence energy dis-
sipation rate is replaced by the turbulence eddy frequency, ,
allowing detailed behavior inside the boundary layer to be
calculated and performs well near walls. However, it behaves
poorly away from walls due to oversensitivity to free stream
conditions. Shear stress transport (SST) (Menter 1994 ) is a
hybrid model that combines the best features of k- and k-
models with a turbulence production limiter in the calcula-
tion of eddy viscosity. Blending between the k- model in
the near-wall region and the k- model in the free stream
using a smooth function gives a marginal improvement in the
region closer to the wall. However, it is still unable to resolve
any details of unsteady turbulent structures in the bulk uid
region. The two methods fail to properly capture the chaotic
and 3D nature of trailing vortices that form downstream of
the impeller blades (Ochieng and Lewis 2006 ). A signicant
shortcoming of unsteady two-equation models is the inabil-
ity to resolve any of the turbulence details directly due to
excessive damping of turbulence (Menter 2009 ). A complete
review on the models mentioned above was done by Joshi
et al. (2011a) .
3.2. LES model
Some authors reported poor CFD results due to turbulent
kinetic energy (Jaworski et al. 2001 , Aubin et al. 2004 ).
Many authors attended to the LES method (as shown in
Table 1), which was rst published for a stirred vessel by
Eggels (1996) . It is worth mentioning that the Smagorinsky-
Lilly model is an eddy-viscosity model, which is the most
common model for predicting the dissipation rate in the LES
model (Smagorinsky 1963 ). In LES, large scales eddies are
resolved, and the small scales, which are isotropic in nature,
are modeled using subgrid scale models. The major role of
subgrid scale models is to provide proper dissipation for
the energy transferred from the large scales to small scales
eddies (Joshi et al. 2011a ).
The LES has been shown to perform well in capturing the
complexity of turbulent ow in a mixing reactor (Hartmann
et al. 2004 , Yeoh et al. 2004 , Delafosse et al. 2008 ). It is able
to overcome k- limitations for investigating unsteady behav-
ior in turbulent ow and provide details of ow eld that can-
not be obtained by RANS (Derksen et al. 1999 ). However,
it is weak in cases where rate-controlling processes occur at
smaller scales. As the dissipation rate cannot be resolved dur-
ing LES simulation, it is estimated a posteriori and strongly
depends on the subgrid scale model used in the simulations
(Delafosse et al. 2008 ). The LES is also not ne-tuned for
quick process design validation and still requires rather intense
CPU time. There has not been any validation for high-solid-
concentrations slurries using the LES, although in contrast to
RANS, it is still easier to recognize the part of uctuations
(Hartmann et al. 2004 ). Menter (1994) have developed a tur-
bulence model called scale adaptive simulation-shear stress
transport (SAS-SST), which is based on the idea of captur-
ing as much of the turbulence eld as possible and using
RANS capabilities near walls and regions of steady ow
(Hartmann et al. 2004 , Honkanen et al. 2007 ). In this model,
small-scale turbulences are dealt through Reynolds averag-
ing, whereas larger-scale turbulences are resolved directly.
This model has shown accurate prediction of the length and
intensity of trailing vortices generated by impeller blades in a
stirred vessel. However, it requires large numbers of modeled
revolutions to obtain good statistical averaging for calculating
turbulence quantities (Singh et al. 2011 ). Further details on
single and multiphase turbulent ow models are available in
the reviews by Joshi et al. (2011a,b) .
3.3. DNS model
The DNS is another option that resolves all time scales and
turbulent length where basic equations are constructed with
no turbulent and interface models. Navier-Stokes equations
are directly integrated using a very ne grid. This method is
not feasible for an industrial scale because it requires enor-
mous amounts of grid cells, particularly for bubbly ows,
because of the very short relaxation time of bubble motions in
liquid in comparison with liquid drops and the deformability
of the bubble surface. The large density ratio of bubbles to
bulk liquid makes this approach more complicated (Kitagawa
et al. 2001 ). Furthermore, it has some restrictions with small
Reynolds and Schmitt number (Chakrabarti and Hill 1997 ), as
an example, at the lower limit of a turbulent regime, it needs
very ne grids in the critical ow regions (vicinity of the
impeller) to fully and accurately resolve the ow (Hartmann
et al. 2004 ), and it is also algebraically complex and numeri-
cally inefcient (Forney and Naa 2000 ).
Although LES and DNS approaches are generally more
accurate for turbulent uid mixing (Hartmann et al. 2004 ,
Singh et al. 2011 ), the RANS approach remains the most
popular simulation method because it requires less compu-
tational effort and time. Although different specic conclu-
sions have been drawn by various researchers in terms of
local turbulent quantities as mentioned in Table 2 , generally,
comparison of RANS and LES simulations has led to the
conclusion that the latter is more superior simply because
more satisfactory agreement between measured and cal-
culated data has been reported for the LES. In summary,
the LES methodology has advantages compared with the
RANS approach because the velocity uctuations, Reynolds
stresses, and turbulent kinetic energy are resolved down to
the scale of the numerical grid. Also, the LES presents lev-
els and structure of turbulent kinetic energy in impeller dis-
charge ow better than the RANS simulation. The levels of
energy predicted by the RANS simulation were higher than
that predicted by the LES; this was also observed for the
spreading rate of energy dissipation. In comparison, tran-
sient RANS simulation can accurately present ow elds
although turbulent kinetic energy predictions are still poor,
especially in discharge ow and impeller region where most
mixing takes place. A detailed explanation is also available
in the reviews by Joshi et al. (2011a,b) .
4. Momentum transferred between dispersed
and continuous phase
Another important term in the momentum balance equation is
the momentum transferred between bubbles (as the dispersed)
and the liquid (as the continuous) phase, which is dened in
eq. (5). The value is dependent on the dispersed phase volume
fraction ( ) (which will be explained in the next section) and
the momentum transferred ( F
IK
):

( ) ( )
- 1
k Ik cell i L
n k
F V tE F t

(5)
where V is the volume of the cell, t is the time step, E is the
cumulative dissipated energy, and the last term is the summa-
tion of the drag and nondrag forces. In addition, by solving
the force balance on each bubble, the bubble velocities can be
calculated using eq. (6):

b
b G P D L VM
t
du
m F F F F F
d
+ + + +

(6)
The left side of eq. (6) represents m
b
and u
b
, which are bub-
ble mass and velocity, respectively. The right side of the equa-
tion is the summation of all forces acting on the bubble, which
is divided into two major groups: drag and nondrag forces.
4.1. Nondrag forces
In systems that include continuous and dispersed phases, the
assessment of interactions between the operational phases
and the effect of continuous phase properties on the dispersed
phase in the CFD modeling can be introduced by force bal-
ance on bubbles (Scargiali et al. 2007 ). Apart from drag force,
various nondrag forces affect the bubbles motion when mov-
ing through a uid and the motion is non-uniform. These non-
drag forces result in interphase momentum exchange beside
the drag force. They are Basset history force, lift force, and
virtual mass force. Mass force is an inertial force caused by
relative acceleration between the secondary phase and the
primary phase (Khopkar and Tanguy 2008 ). Basset force is a
viscous force caused by the development of a boundary layer
around bubbles and relative acceleration. It is relevant only
for unsteady ows. Lift force denotes the transverse force
caused by rotational strain, velocity gradients, vorticities, and
shear in the continuous phase ow eld. Thus, in the bulk
region of stirred vessels, lift force is much smaller than in the
vicinity of the impeller (Lane et al. 2002 ).
4.2. Drag force
The interaction force between phases especially in aerated
stirred vessels is mainly due to the drag force. The effects of
virtual mass, turbulent dispersion, and lift are almost negligi-
ble despite a signicant increase in convergence difculties
and computational expenses. For example, Scargiali et al.
(2007) have reported an increase of overall gas holdup from
4.36 % to 4.67 % and from 4.36 % to 4.60 % by considering
the effect of lift force and virtual mass, respectively. Similar
results have been indicated by many other authors (Bakker
et al. 1994 , Lane et al. 2002 , Kerdouss et al. 2006 ), but
depending on the gas holdup and ow regime, the role of
these forces can be important. Some authors have reported
that on-drag forces are mainly responsible for lateral migra-
tion of bubbles and distribution of gas volume fraction in
1
7
6

B
.

S
a
j
j
a
d
i

e
t

a
l
.
:

U
s
e

o
f

C
F
D

i
n

g
a
s
-
l
i
q
u
i
d

m
i
x
i
n
g

a
n
a
l
y
s
i
s
Table 2 Comparison of LES and RANS in mixing vessels.
Authors/year/system, impeller type,
speed, T, D/T, C/T, N, W/T, Z/T,
operating material, Re
Grid size, type, impeller motion
model
Flow variable Remarks
Yeoh et al. 2004, RT, 36.08 rps
100 mm, T/3, T/3, 4, T/10, T, water,
40,000
15,672, hexahedral (full), sliding/
deforming, CFD code: in-house
coding
u
1

u
2
u
3

Np k ,
Purpose: To simulate turbulent ow
RANS model: standard k-
Notes: The global turbulent energy dissipation rate ( ) across the vessel volume
was underpredicted by 45 % with the RANS model; in contrast, was well
predicted by the LES to within 15 % of the measured value
Hartmann et al. 2004, RT, 2672,
150 mm, T/3, T/3, 6, T/10, T, silicon
oil density, 1039 kg/m
3
, dynamic
viscosity, 15:9 mPa s, 7300
228,096, hexahedral (full), SM, CFD
code: CFX 5.5.1.
u
1

u
2
u
3

k ,
Purpose: To compare URANS and LES
RANS model: SST
Notes: The LES calculations agree better with the experimental results than
the RANS calculations
Yeoh et al. 2005, 100 mm, T/3, T/3,
4, T/10, T, neutrally buoyant tracer,
, 40,000
490,000
Hexahedral (full), SM, CFD code:
correlation
Mixing pattern, time
trace of normalized
concentration
Purpose: To characterize the mixing of an inert scalar and comparison of
URANS and LES
Notes: There is a reasonable agreement with experimental, with an average
deviation of 18 % using the LES
Jahoda et al. 2007 , six-blade 458,
pitched blade turbine, a standard
RT, 0.29 m, T/3, T/3, 2,T/10, 2T, tap
water, 4.66 10
4

615,000 for one impeller, 1,230,000
for two impeller hexahedral (full),
SM, MRF, CFD code: FLUENT 6.2
u
1

u
Np, Nq
Purpose: To predict liquid homogenization
Notes: The LES approach described the real ow with best agreement
Murthy and Joshi 2008, DT,
standard PBTD60, 45, and 30, HF,
4.5 rps, 0.3 m, T/3, 0.082 m, 4, T/10,
T, water,
575,000 for RSM, 1,275,567 for
LES, hexahedral, SM, CFD code:
FLUENT 6.2
u
1

u
2
u
3

k ,
Purpose: To investigate URANS and LES
RANS model: standard k- , (RSTM)
Notes: All the three turbulent models predicted the power number (NP) well,
but by integration of local turbulent energy, the dissipation rate value obtained
from the RANS-based models was underpredicted in the range of 20 25 %
Delafosse et al. 2008, RT, 150 rpm,
0.45 m, 0.15/0.45, 4, 0.15/0.45, T/10,
T, water, 56,250
1,000,000 (full), hexahedrons, SM,
CFD code: FLUENT 6.2.16
u
1

u
2
u
3

k ,
Notes: The URANS overestimated the width of the stream but gave correct
order of magnitude in terms of turbulent dissipation rate in the whole vessel
volume except in the vicinity of the impeller. Moreover, LES simulations
predict the different contributions very well; however, the URANS simulation
failed to predict the respective amounts of turbulent and periodic kinetic
energy
Zadghaffari et al. 2010, 6-blade RT,
250 rpm, 30 cm, T/3, T/3, 4, 0.1T, T,
water,
970,997 (full), hexahedral cell, SM,
CFD code: ANSYS-FLUENT 6.3)
u
1

u
2
P
k
Notes: More reliable results of the mixing process were obtained by the LES
compared with the RANS. A power number maximum deviation of 3 % and
of about 29 % were underpredicted
B
r
o
u
g
h
t

t
o

y
o
u

b
y

|

U
n
i
v
e
r
s
i
t
y

o
f

M
a
l
a
y
a

L
i
b
r
a
r
y
A
u
t
h
e
n
t
i
c
a
t
e
d

|

a
z
i
z
r
a
m
a
n
@
u
m
.
e
d
u
.
m
y

a
u
t
h
o
r
'
s

c
o
p
y
D
o
w
n
l
o
a
d

D
a
t
e

|

1
1
/
2
9
/
1
3

5
:
5
5

A
M
A
u
t
h
o
r
s
/
y
e
a
r
/
s
y
s
t
e
m
,

i
m
p
e
l
l
e
r

t
y
p
e
,

s
p
e
e
d
,

T
,

D
/
T
,

C
/
T
,

N
,

W
/
T
,

Z
/
T
,

o
p
e
r
a
t
i
n
g

m
a
t
e
r
i
a
l
,

R
e
G
r
i
d

s
i
z
e
,

t
y
p
e
,

i
m
p
e
l
l
e
r

m
o
t
i
o
n

m
o
d
e
l
F
l
o
w

v
a
r
i
a
b
l
e
R
e
m
a
r
k
s
F
e
n
g

e
t

a
l
.

2
0
1
2
,

s
t
a
n
d
a
r
d

R
T
,

2
0
0

r
p
m
,

0
.
2
7
,

T
/
3
,

T
/
3
,

4
,

T
/
1
0
,

T
,

w
a
t
e
r
,

1
0
0
,
0
0
0

t
o

1
,
4
0
0
,
0
0
0

(
f
o
r

h
a
l
f
)
,

h
e
x
a
h
e
d
r
a
l

c
e
l
l
,

i
m
p
r
o
v
e
d

I
O
,

C
F
D

c
o
d
e
:

F
O
R
T
R
A
N

c
o
d
i
n
g

u
1
u
2

k
,

P
u
r
p
o
s
e
:

T
o

c
o
m
p
a
r
e

R
A
N
S

m
o
d
e
l
s

a
n
d

L
E
S

R
A
N
S

m
o
d
e
l
:

E
A
S
M
,

A
S
M
,

R
S
M
,

s
t
a
n
d
a
r
d

k
-

N
o
t
e
s
:

T
h
e

E
A
S
M

f
a
i
l
e
d

t
o

p
r
e
d
i
c
t

t
h
e

t
u
r
b
u
l
e
n
c
e

o
n

c
o
a
r
s
e
r

g
r
i
d
s
.

G
e
n
e
r
a
l
l
y
,

t
h
e

L
E
S

b
e
t
t
e
r

p
r
e
d
i
c
t
s

t
h
e

t
u
r
b
u
l
e
n
c
e

k
i
n
e
t
i
c

e
n
e
r
g
y
,

b
u
t

t
h
e

E
A
S
M

b
e
t
t
e
r

p
r
e
d
i
c
t
s

t
h
e

r
a
n
d
o
m

v
e
l
o
c
i
t
y

u
c
t
u
a
t
i
o
n
s
.

B
o
t
h

m
o
d
e
l
s

r
e
c
e
i
v
e
d

s
i
m
i
l
a
r

r
e
s
u
l
t
s

o
f

t
h
e

r
a
d
i
a
l

a
n
d

t
a
n
g
e
n
t
i
a
l

v
e
l
o
c
i
t
i
e
s
.

T
u
r
b
u
l
e
n
c
e

k
i
n
e
t
i
c

e
n
e
r
g
y

h
a
s

b
e
e
n

g
r
e
a
t
l
y

u
n
d
e
r
e
s
t
i
m
a
t
e
d

b
y

a
l
l

t
h
e

t
u
r
b
u
l
e
n
c
e

m
o
d
e
l
s

o
f

t
h
e

R
A
N
S

f
a
m
i
l
y
the reactor (Ashraf Ali and Pushpavanam 2011 ). Different
models are present in the literature to estimate the drag force
based on bubble rise in a stagnant liquid (Lane et al. 2002 ,
Kerdouss et al. 2006 , Scargiali et al. 2007 , Moilanen et al.
2008 ). However, a bubble experiences continual accelerations
and decelerations due to turbulent eddy, and assumptions in
stagnant ows may be erroneous. The drag coefcient in the
averaged form of interfacial force represents an averaged
value, which may differ from the drag coefcient based on
rise through a stagnant liquid, thus many authors have focused
on the investigation of an improvement of drag coefcient
by turbulence (Van den Akker 2006 ). Spelt and Biesheuvel
(1997) have used numerical simulations to determine the
average gas bubble motion in homogeneous and isotropic
turbulence and have found that bubble rise velocity is nearly
half of its value in stagnant liquid. Poorte and Biesheuvel
(2002) also found that the drag force on bubbles with diam-
eter between 0.3 and 0.6 mm increases signicantly due to
reduction of bubble terminal rise velocity, which depends
on turbulence. Montante et al. (2007) have exposed a drag
law formulated in terms of the terminal velocities. Petitti
et al. (2009) have modied the model to enable application
for turbulence and depicting ow transitions between differ-
ent regimes. Another important issue that depends on drag
force is slip velocity because it affects local bubble size dis-
tribution, gas holdup, and hence, mass transfer areas. Slip
velocity can be calculated from a force balance on a bubble.
Also it is worth noting that, in gas-liquid ows, bubbles have
a tendency to form a non-spherical shape, especially those
with a diameter > 3 mm. Thus, the selection of drag model
should be taken into account in bubble deformation (Gimbun
et al. 2009 ). Furthermore, application of a suitable balance
model in different regions of stirred vessels has direct effect
on slip velocity. Some of the areas of focus include the pres-
sure gradient caused by the uid motion and the hydrostatic
pressure gradient, which is directed to the liquid surface; the
gravitational force on bubble, which acts downward, and the
radial force on the bubbles due to rotational motion of liquid
should be balanced by drag force (Laakkonen et al. 2006 );
however, most of the authors focused on slip velocity in solid-
liquid systems rather than gas-liquid systems.
4.3. Dispersed phase volume fraction
Another important factor in Eq. (5) is the dispersed phase
volume fraction. This factor is dened by the momentum
balance equation. Generally, knowledge on bubble size dis-
tribution in stirred vessels is necessary; hence, phase conti-
nuity, momentum equations, and population balance should
be solved simultaneously. The multidimensional population
balance model (PBM) accounts for these aspects simulta-
neously and can describe interactions between gas bubbles
and continuous liquid both in terms of momentum and mass
coupling. The rst efforts in this eld have been done by
assuming monodisperse bubble size distribution during
the investigation of different parameters in gas-liquid stir-
rer systems (Khopkar et al. 2006 , Scargiali et al. 2007 );
also other authors have used different PBMs to predict the (
T
a
b
l
e

2

c
o
n
t
i
n
u
e
d
)
bubble size distributions along with hydrodynamic predic-
tion in their studies (Bakker et al. 1994 , Kerdouss et al.
2006 , Laakkonen et al. 2006 , Montante et al. 2008 , Petitti
et al. 2009 , 2010).
The number density of bubbles of size at time t , n ( , t ), is
introduced via the PBE:

( )

+ +

( , ) ( , ) ( , ) - -
gi B B C C
i
n t u t n t B D B D
t x

(7)
where B
B
, D
B
, B
C
, and D
C
represent the birth rate due to
breakup of larger bubbles, the death rate due to the breakup
into smaller bubbles, the birth rate due to the coalescence of
smaller bubbles, and the death rate due to the coalescence
with other bubbles, respectively.
4.3.1. Bubble breakup and coalescence There are many
coalescence and breakage kernels developed for bubbly ow
over the last 40 years, and they are written in similar forms
except for some minor differences in assumptions or model
constants. In a gas-liquid stirred Rushton vessel, bubble
breakup occurs due to turbulent eddy collision and instability
of large bubbles.
Bubble breakup occurs when disruptive forces in the liquid
bulk are large enough to overpower bubble surface tension.
Breakage frequency, daughter bubble size distribution, and size
distribution of bubbles are the main components of the breakup
kernel (Luo and Svendsen 1996 , Lehr and Mewes 2001 ).
Breakup rate depends on the operating parameters and uid
properties (both continuous and dispersed). Generally, there
are four main mechanisms for deformation and breakup. They
include turbulent uctuation and collision, interfacial instabil-
ity, viscous shear stress, and shearing-off process. Interfacial
instability is due to density differences, which includes the
Rayleigh-Taylor instability. The other three mechanisms are
due to ow dynamic characteristics in the continuous phase.
Breakup and deformation by viscous shear stress occur when
the velocity gradient occurs around the interface (Liao and
Lucas 2009 ). Meanwhile, shearing-off is an additional mecha-
nism caused by velocity differences across the interface, which
happens when uid particle diameter increases and becomes
unstable. The shearing-off mechanism is also called erosive
breakage and is characterized by a number of small particles
sheared off from a large one (Fu and Ishii 2003 ).
The most important mechanism for breakup and deforma-
tion is collisions with eddies or turbulent pressure uctuations
of the surrounding uid. This breakup mechanism occurs
when the dynamic pressure difference around the particle
results in the oscillation amplitude. When the value reaches a
critical point, the bubble surface deforms and stretches in one
direction and nally breaks up into two or more daughter par-
ticles. Therefore, this mechanism can be characterized by a
balance between dynamic pressure
i
and its surface stress
s
.
In stirred vessels, turbulent uctuation is the most important
factor for bubble collision and breakup.
The degree of breakup by turbulent uctuation and col-
lision can be classied by ve main factors: (1) turbu-
lent kinetic energy of bubbles greater than a critical value;
(2) velocity uctuation around bubble surfaces greater than a
critical value; (3) turbulent kinetic energy of the hitting eddy
greater than a critical value; (4) inertial force of the hitting
eddy Fe ; (5) a combination of criteria.
There is a variety of models based on these factors that can be
used in CFD simulation. Prince and Blanch (1990) introduced
a model that can predict bubble size distribution (as shown in
Table 3 ). The model is based on the rst mechanism, and the
breakup rate is given by the collision rate of bubbles with tur-
bulent eddies and breakup efciency. Prince and Blanch (1990)
also considered bubble coalescence due to laminar shear and
different rise velocities. This model does not include daugh-
ter bubble size distribution after the breakage event of particle
size. Lehr et al. (2002) further developed the Prince and Blanch
model by combining the bubble coalescence due to different
rise velocities and turbulent eddies. Laakkonen et al. (2007)
later compared the two kernels and found that the Lehr model
gives excellent prediction of local bubble size in bubble col-
umns but underpredicts in gas-liquid stirred vessels.
The breakup model by Luo and Svendsen (1996) is also
based on the rst mechanism. It was developed for predict-
ing uid-particle breakup rates in a turbulent dispersion and
assumes that bubble breakup happens when turbulent eddies
hit the bubble surface. The model contains no adjustable or
unknown parameters, so no prior assumption is required. The
turbulent breakup mechanism can be modeled by the prod-
uct of breakup probability due to the eddy energy contained
and collision frequency between turbulent eddies and bubbles
(Kerdouss et al. 2008 ).
The Luo and Svendsen (1996) model is more popular
because it simultaneously investigates bubble breakup rate
and daughter size distribution. Some researchers (Bordel
et al. 2006 , Podila et al. 2007 ) compared the various versions
of the Prince and Blanch (1990) and the Luo and Svendsen
(1996) models. They found that there is no great difference
among gas holdup, mean ow, and bubble Sauter mean diam-
eter ( d
32
) although they were predicted by means of different
kernels. However, there are some differences in the predicted
bubble size distribution, which depends on Sauter mean diam-
eter. The results obtained by Luo and Svendsen (1996) were
found to generate a real number of small bubbles as well as
very large bubbles.
Bubble coalescence occurs when two or more bubbles col-
lide and the lm of liquid between them thins and ruptures.
Collision rate and coalescence efciency are components of
the coalescence process. Generally, three mechanisms can be
distinguished for the coalescence process: (1) lm rupturing
resulting in coalescence; (2) enough bubble contacting time
for the small amount of liquid between them to become thin-
ner than critical value of drainage; (3) the last but most impor-
tant, collision of two bubbles and trapping of the liquid lm
between them.
Four criteria have been proposed for the coalescence pro-
cess in a turbulent ow: turbulent uctuations in the continu-
ous phase, mean velocity gradients in the ow, interactions or
helical/zigzag trajectories, and bubble capture in an eddy.
Several models are available for these components in
the literature (Coulaloglou and Tavlarides 1977 , Prince and
B
.

S
a
j
j
a
d
i

e
t

a
l
.
:

U
s
e

o
f

C
F
D

i
n

g
a
s
-
l
i
q
u
i
d

m
i
x
i
n
g

a
n
a
l
y
s
i
s

1
7
9
Table 3 Summary of breakup models.
Breakup models Remarks Authors

_
_ _
+ +

, ,
,
1
1
]
max
min
1 2
2 2 3
2 3 1 3
2 3
2
2 3
2 2
( )
exp -
k
i I i i
k
II
c i
V c d d
k k
k
c k dk
d
1. Based on bubble collision with a turbulent eddy
2. Considering the breakup frequency by multiplying collision
3. The collision frequency is assumed equal to the probability that turbulent eddies have sufcient
energy to rupture a bubble
4. Predicted the same and monotonic behavior even at high turbulent dissipation rates
5. Problem with determining the integration limits
6. Daughter bubble size distribution based on uniform distribution model
Prince and Blanch
1990

( ) ( )
( )
_ _ +

, ,
1 3 2
1
2 11 3
min
1
( ) 0.923(1- ) exp -
i i
i
i e
E d
V d
d E d
1. Based on bubble collision with a turbulent eddy
2. Modied version of Prince and Blanch model by calculating the critical energy as the increase
in surface energy during the breakage
3. Most widely used model because it does not include any empirical or unknown parameter
Luo and Svendsen
1996

( )
( )

_ _

, ,
,max
1 3 10 3 4
,min
11 5 9 5
1 3
7 6
11 5 4 3 7 9 7 9
3.55 1-
( , )
1 1
1.5(1- ) min , -
de
i
i j e
de
e f j
f
i
j
j j i
d
V V d d
d d
V
V
V V V
1. Based on the force balance
2. The breakup probability depends on the angle the eddy hits the bubble
3. Assumed that only eddies of length scale smaller than or equal to the bubble diameter can
induce breakup
4. The daughter bubble size distribution obtained directly from the breakup frequency
Lehr and Mewes
1999

_
+

,
2
,max
1 3 4 13 3 2 3
,min
( ) 2
( , ) 1.190 exp - ( )
de
e i
i j e
de
f j e f
d d
V V d d
d d
1. Based on the force balance
2. Modied version of (Lehr and Mewes 1999)
3. Not considered the angle under which the eddy hits the bubble
4. The breakup probability depends on the kinetic energy of the eddy exceeding the critical
energy that is obtained from the force balance
Lehr et al. 2002
B
r
o
u
g
h
t

t
o

y
o
u

b
y

|

U
n
i
v
e
r
s
i
t
y

o
f

M
a
l
a
y
a

L
i
b
r
a
r
y
A
u
t
h
e
n
t
i
c
a
t
e
d

|

a
z
i
z
r
a
m
a
n
@
u
m
.
e
d
u
.
m
y

a
u
t
h
o
r
'
s

c
o
p
y
D
o
w
n
l
o
a
d

D
a
t
e

|

1
1
/
2
9
/
1
3

5
:
5
5

A
M
Table 4 Summary of coalescence models.
Author Coalescence rate Details
Prince and Blanch 1990

( )

T B LS
ij C ij ij ij ij
C F + +

( ) -
ij ij
t
ij
e

( )

3
16 ln
ij ij L o f
t r h h

2 3 1 3
ij ij
r
Coulaloglou and Tavlarides 1977

( ) ( ) ( )
+ +
2 1 2
1 3 2 3 2 3
,
4
i j I i j i j
h d d c d d d d

( )
2 2
1 2 rel t t
u u u +
Hagesaether 2002

( ) ( ) ( ) ( )
+ +
2 1 2
1 3 2 3 2 3
, ,
4
i j i j i j i j i j
h d d n n d d d d d d

( )
( ) ( )
( ) ( )

_
1
+ +
]

+ +
,
1 2
2 3
1 2
3
1 2
0.75 1 1
, exp -
0.5 1
ij ij
i j ij
G L ij
d d C We
Blanch 1990 , Luo and Svendsen 1996 , Venneker et al. 2002 ).
However, these expressions are not satisfactory for coales-
cence efciency in stirred vessel simulations.
The Prince and Blanch bubble coalescence model is widely
used because it accounts for the cumulative effect of collision
rate by mean shear in ow elds, buoyancy, and turbulent
motions (Liu et al. 2011 ). The model assumes coalescence
of two bubbles occurs in three steps: rst, bubbles collide,
trapping a layer of liquid between them; then the liquid lm
separating the bubbles becomes thinner; and nally, the lm
ruptures and bubbles coalesce.
Coulaloglou and Tavlarides (1977) assumed that there is a
contact time when bubbles collide where they either bounce
away or coalesce. If the random contact time exceeds the
coalescence time, coalescence occurs (Liu and Li 1999 ). In
this model, the collision frequency is proposed by the effec-
tive volume swept by moving bubbles per unit time. The
bubble motion is interpreted similar to random motion of
gas molecules. This assumption is questionable because uid
particle collisions are neither elastic nor rigid. The relative velo-
city between bubbles is determined by assuming that collid-
ing bubbles have the velocity of an equal-sized eddy and is
determined by the mean square root of two equivalent eddy
velocities. Inertial subrange of isotropic turbulence is always
assumed for eddy velocity calculations. This model has some
weaknesses arising from the assumption that all bubbles have
the same velocity as equal-sized eddies and are in inertial sub-
range arbitrarily. As a result, some modications have been
made. The rst modication was done by Colin and Riou
(2004) in a gas-liquid system. He explained the effect of size
ratio between eddies and bubbles and presented that if the inte-
gral length scale of turbulence l
e
are smaller than the particles,
the relative bubble motion is mainly due to the mean shear of
ow because turbulent eddies are not sufcient to move the
particles. According to Colin and Riou (2004) , turbulent colli-
sions occur only when ( b
i
< l
e
; b
j
< l
e
) and ( b
i
< l
e
; b
j
> l
e
).
Other researchers modied the model for the second time
(Hibiki and Ishii 2000a,b, Lehr et al. 2002 ) to investigate the
increasing rate of particle collision frequency by increasing
dispersed phase volume fraction. This factor has been consid-
ered by multiplying the original equation with a factor ( ).
The third modication was to consider the ratio of the mean
distance between particles to their average relative turbulent
path length. According to Wu et al. (1998) and Wang et al.
(2005) , if the turbulent path length is smaller than the dis-
tance between particles, no collision should be counted. The
effect can be described by multiplying the original collision
frequency with a decreasing factor . (The coalescence equa-
tions and mentioned corrections are summarized in Tables 4
and 5 , respectively.)
Hagesaether (2002) assumed that bubble coalescence
depends on bubble size diameter, relative bubble velocity, and
collision angle ( ). In his model, coalescence decreases with
the increase of bubble size diameter and efciency decreases
with increasing relative velocity. A critical relative velocity
must exist for coalescence; in addition, coalescence efciency
is dependent on collision angle ( ) in the moment of coales-
cence. Collision angle ( ) is dened as the angle between dif-
ferent vector positions corresponding with colliding bubbles
and the relative velocity vector (Hagesaether 2002 ). A com-
plete review of methods for coalescence breakage problems is
reported by Laakkonen et al. (2007) .
4.3.2. Population balance solution method Coupling
PBE with CFD is another challenge in modeling multiphase
ows. Bubble size equations formulation involves source terms
for coalescence or breakup and transport by convection.
Generally, different techniques exist for solving PBEs:
complete population balance model (PBM), bubble den-
sity model (BDM), methods of classes (CM), methods of
moments (MOM), standard method of moments (SMM),
the quadrature method of moments (QMOM), direct quadra-
ture method of moments (DQMOM), conditional quadrature
method of moments (CQMOM), multiple size group model
(MUSIG), and Monte Carlo method (MCM) are several
numerical approaches for solving PBEs. These models (as
shown in Table 6 ) provide a framework in which the PBM
together with the coalescence and breakup models can be uni-
ed into 3D CFD calculations.
The BDM model is a simple model based on the calcula-
tion of gas-phase volume fraction by solving scalar type trans-
port equations (Kerdouss et al. 2008 ). A coupled CFD-BDM
Table 5 Correction based on the Coulaloglou and Tavlarides model.
First correction based on u
rel
Second correction based on Third correction based on

1 3
If ( ; ),
2 1.61
i j
t
i e j e rel
d d
C
d l d l u
+ _
< <

,

(Colin and Riou 2004)

( )
1 3
If ( ; ),
1.61
t
i e j e rel i
C
d l d l u d < >

(Colin and Riou 2004)

( )
max
1 3 1 3 1 3
max max
1
For 0.8
-

<

(Wu et al. 1998)

( )
max
1 3
max
1
For 0.741
-

<

(Hibiki and Ishii 2000a,b)

( )
max
max
max
For 0.8
-

<

(Wang et al. 2005)

2
1 3 1 3
max
max 1 3
-
For 0.6 exp -

1
_
1 <

, 1
]

(Lehr et al. 2002)

1 3 1 3
max
1 3 1 3
max
.
1-exp -
-
C

1 _
1

,
1
]

(Wu et al. 1998)

6
,12
,12
exp -
b
bt
h
l
1
_
1

1 ,
]

(Wu et al. 1998)
has been used to predict the local bubble size by other authors
(Lane et al. 2002 , Kerdouss et al. 2006 , Moilanen et al. 2008 ).
In this approach, local bubble size varies in every single point
in the region, but from point to point, a nonphysical monodis-
persed distribution is considered (Lane et al. 2002 , Kerdouss
et al. 2006 , Moilanen et al. 2008 ), but its formulation suffers
from several downsides: all equations related to bubble size
have gathered together as a function of the critical energy dis-
sipation rate and the Weber number without paying attention
to the probability and rate of bubble-eddy and bubble-bubble
collisions; proper bubble breakage and coalescence kernels
are not included; the model is not fully predictive, although by
adjusting some empirical constants introduced in the model,
good predictions are accessible.
The CM is the most commonly used technique for solving
PBE. In this method, the whole domain of internal coordinate
space is subdivided into a nite set of bins. The PBE trans-
forms from an integro-differential multidimensional equation
into a set of differential equations of smaller dimension for
the number of bubbles expected within each bin. The high
computational demand is required to receive satisfactory pre-
diction of the evolution of moments for the bubble size distri-
bution is a shortcoming of MC.
MOM and SMM are techniques that require relatively low
computational cost. System evolution is calculated through
some lower-order moments of distribution and the internal
coordinate is integrated out. The equations are solved for
moments of the bubble size distributions using quadrature
approximation. However, these methods suffer from some
simplications due to the closure problem (Rong et al. 2004 ).
To overcome the closure problems, new methods were formu-
lated based on the use of a quadrature approximation. McGraw
(1997) have proposed the QMOM, which was validated and
extended by Marchisio et al. (2003) using a quadrature formula
and solution of transport equation for lower-order moments
of bubble size distribution. With QMOM, the transport equa-
tions for moments of number density function are solved by
using quadrature approximations. QMOM is able to provide
accurate predictions with relatively small numbers of quadra-
ture points with reasonable computational time, although
QMOM do have some limitations because only monovariate
problems could be treated (Petitti et al. 2010 ).
Multivariate problems require alternative approaches
such as DQMOM or CQMOM (Yuan and Fox 2011 ). With
the former, each node of the quadrature can be treated as a
distinct phase, with its own velocity, size, and composition.
Also, weights and nodes of the quadrature approximation
are calculated directly through transport equations in every
point of the computational domain. DQMOM can also be
easily combined with multiuid models. The DQMOM
is another model that was reduced by Marchisio and Fox
(2007) based on the QMOM by McGraw (1997) . The main
advantage of DQMOM is that only a few abscissas are
necessary to describe a particle distribution because of the
numerical quadrature approximation closure used (Selma
et al. 2010 ). Recently, Buffo et al. (2012) has formulated
an approach based on the DQMOM method to describe the
evolution of the gas bubble in a realistic gas-liquid stirred
vessel reactor. In their work, the initial conditions have
been highlighted to obtain meaningful predictions and also
a proper corrective term in all the tests conducted on a zero-
dimensional homogeneous system (described by Marchisio
and Fox 2005 ).
The MCM is another method that can accurately and real-
istically describe the behavior of a physical system. Each
notional bubble is randomly selected, realizing a physical
event articially involving real bubbles. However, this method
is not completely coupled with CFD for simulation of real sys-
tems because large numbers of notional bubbles considered
are required to reduce the statistical error (Buffo et al. 2012 ).
The last approach to solve PBM is the MUSIG model,
which is a framework wherein the the PBM can be incor-
porated into 3D CFD calculations. With this technique,
bubble size distributions are divided into a suitable number
of size classes for which continuity equations are solved.
Homogenous and inhomogeneous multiphase polydispersed
ows are two approaches of the MUSIG model. For inho-
mogeneous ows, the MUSIG model assumes that there
are N different velocity groups that can be treated as a
separate phase in a multiphase ow calculation. Each
velocity group is then divided into different size groups,
so N + 1 sets of coupled equations (continuity and momen-
tum) should be solved. In a homogenous ow, to decrease
the large number of equations involved, MUSIG assumes
that different size groups share the same velocity; thus, it
is only necessary to solve one set of momentum equations
for all bubbles, although continuity equations of the size
groups are solved to present the size distribution (Moilanen
et al. 2008 , Montante et al. 2008 ).
5. Mass transfer
The aim of the models mentioned is to give a representa-
tive prediction of mass transfer, which is often controlled by
the liquid side through the volumetric mass transfer coef-
cient ( k
L
a ). Based on the models mentioned above, several
researches have focused on measuring mass transfer in stirred
reactors containing gas and liquid. Table 7 summarizes the
work on gas-liquid mass transfer prediction in stirred reactors
via CFD simulations based on bubble size distribution.
The Luo and Svendsen model can be considered the
most widely used model for the bubble breakup rate and
the Prince and Blanch model is the most widely used for
Table 6 Summary of numerical approaches for solving PBM.
Model Equations Remarks
CM

( )

+
1
]
-1
, ,
1-0.5
i i k
j k
coal
i j k j k Gk i j k
x x x
B Q x x x

( )
,
1
M
coal
i Gi i k GK k
k
D Q x

( )
,
M
brl
i i k k GK i k
k i
B n x x

brl
i i Gi
D
Most commonly used technique
Can also be coupled with CFD
Requires a large number of classes to reach a satisfactory level of
accuracy
QMOM

( ) ( )

+

3
( ) 3 3
1 1
0.5 ,
N N
k
k
coal i j i j i j
i j
B L L L L

( )

( )
1 1
,
N N
k k
coal i i j i j
i j
D L L L

( ) ( )
( )
1
,
N
k
break i i i
i
B b k L a L

( )
( )
1
N
k k
break i i i
i
D a L L
Is based on the approximation of the distribution by a quadrature formula
and on the solution of transport equation for lower-order moments
Only monovariate problems could be treated in this formulation
DQMOM

( ) ( )

+
3
( ) 3 3
1 1
0.5 ,
N N
k
k
coal i j i j i j
i j
B L L L L

( )

( )
1 1
0.5 ,
N N
k k
coal i j i i j
i j
D L L L

( ) -( ) *
1
N
k k
break i i i
i
B b a

( ) *
1
N
k k
break i i i
i
D L a
Is based on the direct solution of the transport equations for weights and
abscissas of the quadrature approximation
Is directly applicable to the PBE with more than one internal coordinate
Only a few abscissas are necessary to describe a particle distribution due
to the numerical quadrature approximation closure used
MUSIG

( )
1 1
0.5
N N
k
coal jk j k
i j
B Q n n

( )
1 1
N N
k
coal i ij j
i j
D n Q n

( )
( )
1
:
N
k
break j i j
i
B g v v n

( ) k
break i i
D g n
Considers a larger number of bubble size groups to give a better repre-
sentation of the size distribution
Uses the homogenous and inhomogeneous multiphase polydispersed
ows
Both of approaches closely match with experimental method
Table 7 Summary of investigated mass transfer in stirrer reactors.
Authors/year/system,
impeller type, speed, T,
D/T, C/T, N, W/T, Z/T,
operating material, Re
Grid size,
type, impeller
motion model
Mass transfer model Remarks
Laakkonen et al. 2006,
Rushton, 150 250 rpm,
630 mm, T/3, T/3,
fully bafed, T/10, T,
155 250, water, -
160,000, hexa-
hedral (full),
sliding, SM,
CFD code:
CFX-4.4
Two-lm theory

1 4
0.301
L L
k D

,
Purpose: To nd unknown parameters in bubble breakage,
coalescence mechanistic and multicomponent gas-liquid mass
transfer
Investigated parameters: bubble size distributions
Turbulence model: standard k-
PB method: simplied discretized PB equation
Breakup model: Luo and Svendsen
Coalescence model: Coulaloglou and Tavlarides
Kerdouss et al. 2008,
45 pitched blade with
3 blades, 600 rpm,
12.5 cm, 0.6T, 2, 4/3T,
water, 3.75 10
4

1,000,000,
hexahedral
(full), MRF,
CFD code:
FLUENT 6.2
Penetration theory

1 4
2
L L
k D

,
Purpose: To predict mass transfer
Investigated parameters: Sauter diameter, local mass transfer
coefcient
PB method: CM
Coalescence model: Hagesather et al.
Moilanen et al. 2008,
Rushton (RT), phase
jet (PJ), and combi jet
(CJ), 0.64 m, T/3, T/3,
4, T/10, T, water
> 120,000
nodes, hexa-
hedral (full),
SM, CFD
code: CFX
5.7.1.
Correlation

1 4
0.3
L L
k D

,
Purpose: To investigate local variables
Investigated parameters: ow eld comparison, volumetric
density distribution, gas hold up, volumetric mass transfer rate
Turbulence model: SST
PB method: MUSIG, BND
Coalescence model: Coulaloglou and Tavlarides
Gimbun et al. 2009,
Rushton turbine
and CD-6 impeller,
365 700, 0.26, T/3,
T/3, 4, T, water
225,000 cells,
hexahedral
(full), -,
CFD code:
FLUENT
Slip velocity

2
b
L L
b
k D
d

,
Purpose: To develop a modeling approach and then use to
evaluate the local mass transfer
Investigated parameters: gas and liquid axial and radial velocity,
local Sauter mean bubble diameter, volumetric mass transfer rate
Turbulence model: realizable k-
PB method: QMOM
Breakup model: Prince and Blanch
Coalescence model: Prince and Blanch
Xia et al. 2009, down-
pumping propeller
(DPP), 6-curved-blade
disc turbine (6CBDT),
6-arrowy-blade disc
turbine (6ABDT),
300 600, 0.3 (0.467,
0.4, 0.4), 0.43, 0.1, 1.48
638,212,
744,157,
774,306, hexa-
hedral (full),
MRF, CFD
code: ANSYS
CFX 11.0
Based on Higbie theory

1 2(1 )
2
n
L L
k D

+
_

,
Purpose: To investigate uid dynamics in three bioreactors
equipped with different impeller combinations
Investigated parameters: velocity proles, average shear
stress in the bulk, oxygen mass transfer coefcient, material
concentration
Turbulence model: multiphase k-
PB method: -
Breakup model: not considered
Coalescence model: not considered
Selma et al. 2010,
a double 6-bladed
standard Rushton
impellers, 450, 0.292,
T/3, T/10, T/2, 1.5T, 2,
2T, 7.1 10
4

255,000 cells,
hexahedral
(full),
MRF, CFD
code: open
FOAM
Eddy cell

1 4 1 2
0.4
L
L
D
k

_ _

, ,
Purpose: To validate the DQMOM with the CM
Investigated parameters: Sauter mean diameter, liquid velocity
prole, gas volume prole
PB method: DQMOM, BND
Ranganathan and
Sivaraman 2011, dual
standard Rushton
turbines, 450, 0.292 m,
0.3, 0.5, 2, 0.1, 2, water
Hexahedral
(full), MRF,
CFD code:
ANSYS
CFX11
Penetration theory, slip
velocity, eddy cell, and
rigid theory

,
1 4
-
L
eddy cell
L
l
k K D

-
2
L
slip velocity L b
b
D
k
d

,
1 2
2 3 -1 6
0.6
L
b
L l
b
k D
d
Purpose: To investigate hydrodynamics and mass transfer
Investigated parameters: axial and radial gas holdup, Sauter
mean bubble diameter, local specic area
PB method: MUSIG
the coalescence rate. In the work of Sivaraman, hydrodyna-
mics and mass transfer have been simultaneously taken into
account to predict the mass transfer in gas-liquid stirred ves-
sels (Ranganathan and Sivaraman 2011 ). Oxygen transport
equations and hydrodynamic simulation has been individu-
ally done in other works. For a good prediction of mass
transfer in these systems, the gas-liquid mass transfer, PBM,
and CFD should incorporate to each other. Laakkonen et al.
(2006) obtained a multiblock model for calculating mass
transfer area by tting unknown parameters in the bubble
breakage and coalescence models against local experimental
data. The tting included simulations along with sensiti vity
and correlation analysis to reduce parameter space. They
have reported that tted models are useful tools for scale-up
and reactor design.
Kerdouss et al. (2008) used converged steady-state single
primary phase ow eld by MRF as an initial condition for
the transient two-phase calculations to avoid numerical dif-
culties. A single bubble size was used as the initial condition
with PBEs to improve the overall speed of the calculations.
Moilanen et al. (2008) simulated the effect of impel-
ler geometry on G-L mass transfer with two different PB
approaches (MUSIG and BND) and two equation turbulence
models ( k- and k- , SST). The BND and MUSIG models
predicted different trends for vessel-averaged bubble size
using different impellers; but unfortunately, the authors did
not justify the reasons, but they explained that the BND
approach is a useful and fast design tool even if it lacks the
local BSD. Generally, a good prediction of estimated local
bubble size was exhibited, based on tted model parameters
in the breakage and coalescence terms from the multiblock
study.
Gimbun et al. (2009) have focused on the development
of a modeling approach for gas-liquid stirred vessels. First,
they performed constant bubble size with the spherical shapes
along with the Schiller and Naumann (1935) drag model with
PBM; next they used a non-uniform bubble size with two dif-
ferent drag models: the hard sphere drag model of Schiller
and Naumann (1935) and Ishii and Zuber (1979) . They have
reported that by means of a nonspherical drag model (CFD-
PBM-IZ), the results improved signicantly because of the
fact that the effect of local bubble sizes on the two phase
ow is mainly via the interphase exchange coefcient, which
depends on the drag model.
Xia et al. (2009) have investigated different uid dynamics
of a bioreactor equipped with different impeller combinations
CFD simulation was used for validation of their work and
nding optimum situations of vessel domain that affect the
fermentation process.
Selma et al. (2010) have compared the DQMOM and the
CM in a rectangular bubble column and stirred vessel reac-
tor. Both methods are in agreement with experimental results,
although DQMOM is computationally more efcient to use
in comparison with CM. The authors reported that using
CM, both the 15 and the 25 classes have produced reason-
able agreement with experimental data, but the 25 classes
appeared more accurate.
Ranganathan and Sivaraman (2011) used both inhomoge-
neous and homogenous MUSIG equations to account for the
bubble size distribution and mass transfer. The polydispersed
gas phase was subdivided into 21 groups ranging from 0.1 to
10 mm for the homogenous and two velocity groups (Air1:
between 0.1 and 5 mm; Air2: between 5 and 10 mm) for the
polydispersed gas phase for the inhomogeneous MUSIG
equations. However, Air2 showed higher discrepancy in
the results of the local Sauter mean bubble diameter com-
pared to the velocity group Air1 because the bubble size in
the reactor is not in the range of 5 10 mm. In addition, they
applied different mass transfer coefcients based on various
approaches such as penetration theory, eddy cell model, slip
velocity model, and rigid-based model. They reported that
the mass transfer coefcient predicted by the penetration
theory was higher than the eddy-cell model value predicted
by the pene tration theory. Both these models have overpre-
dicted the value of the mass transfer coefcient in the impel-
ler discharge depending on the eddy dissipation rate. The slip
velocity model and the eddy cell model closely matches with
experimental data.
Generally, bubbles in stagnant zones accumulate and
coalescence and create bubbles with larger diameters; mean-
while, bubbles break up in the vicinity of the impeller due to
the high shear and the BSD is shifted to lower bubble diam-
eters. Every single bubble has its own velocity, size, and com-
position, which must be considered to correctly predict the
evolution in time and space for local and global mass transfer
ux predictions. So, by coupling accurate PBM with CFD,
this aim will be accessible. Unfortunately, no studies have
considered all these parameters together and instead used
simple mass transfer coefcient models for prediction of gas-
liquid mass transfer.
6. Impeller motion model
Subsequent to selecting a suitable equation, suitable bound-
ary conditions for impeller blades and disc surfaces pose
a challenge because simulation of these moving sections
is attached to both stationary bafes and impeller. There
are several techniques available to address this issue.
Using the impeller boundary conditions (IBC) method is
the simplest strategy, which imposes suitable turbulence
and velocity quantities on the surface swept by impel-
ler blades. However, this method requires the availability
of experimental data. Other fully predictive procedures
are available, with no requirement for experimental data
as boundary or initial conditions (Montante et al. 2001 ).
Sliding mesh (SM), multiple reference frame (MRF), and
inner-outer (IO) are three approaches for the simulation of
ow eld in modeling problems that involve both station-
ary and moving zones (Ankamma Rao and Sivashanmugam
2010 ). The rst is based on transient computation to pro-
duce time-accurate ow eld, and the other two are based
on steady-state computation producing time-average ow
eld, resulting in considerable savings of computational
demand compared with the SM approach (Ankamma Rao
Table 8 Discretization schemes.
Scheme Conservative Bounded Accuracy Transportive Remarks
CDS Yes Conditionally
bounded
Second order No Unrealistic solutions at large Peclet number
UDS,
HDS, PLS
Yes Unconditionally
bounded
First order Yes Induce false diffusion if the velocity vector is
not parallel to one of the coordinate directions
QUICK Yes Unconditionally
bounded
Third order Yes Less computationally stable
Can give small overshoot and undershoot
and Sivashanmugam 2010 ). The choice of grid character-
istics and computational domain is essential because it may
strongly affect the quality of results, especially in the MRF
framework. The coupling between both domains must be
taken into account because if the mesh is not ne enough,
there will be a gap between the small cells around the impel-
ler and the large cells at the vessel. In this case, both grids
will not coincide on the surface at the interface resulting
in a high numerical discretization error (Abu -Farah et al.
2010 ). Thus, this method is appropriate when ows at the
boundary between the inner and the outer zones are nearly
uniform. This approach also cannot predict the rate of decay
in the local maximum velocity in the wall jet (Ochieng et al.
2009 ). For more details, refer to Joshi et al. (2011a) .
7. Discretization schemes and equation solver
algorithms
Important factors that affect accuracy and precision of ow
simulations in stirred vessels are discretization schemes
and equation solver algorithms. Choices are constrained by
the computational power available and the basis of domi-
nating transport phenomena; in other words, selected solu-
tion algorithm should be physically realistic and satisfy
overall balance (conservative). Four basic rules should be
satised: when a face is common to two adjacent control
volumes, ux across it must be represented by the same
expression in discretization equations for both the control
volumes; all coefcients in governing the differential equa-
tion must always be of the same sign; the slope lineariza-
tion of source term should be negative so erroneous results
will decrease; both neighbor coefcients (with and without
derivatives) should satisfy the governing differential equa-
tion (see Table 8 ).
Central differencing scheme (CDS), upwind differencing
scheme (UDS), power law, central differencing, hybrid differ-
encing scheme (HDS), and quadratic upstream interpolation
for convective kinetics (QUICK) are the widely used discreti-
zation schemes. The CDS does not recognize strength of con-
vection or direction of ow for accuracy and stability. Either
grid spacing should be small or velocity should be very low
(Pe < 2). The UDS is a rst-order discretization scheme used
for initiating a simulation and is desirable for stable iteration
because the direction of ow is inbuilt in the formulation. The
HDS is based on the ratio of the convective ow to diffusion
(Peclet number) and depends on the ow and uid properties.
In other words, the HDS is a combination of the CDS and
the UDS. It is identical to the CDS for Pe between -2 and 2,
and outside this range, it reduces to the UDS. The Power law
scheme is also similar to the HDS, but it is more accurate.
Discretization schemes have been dealt with in very few
works. Sahu et al. (2005) compared the upwind, power law,
and hybrid as three rst-order numerical schemes. They
explained that the rst two give similar results and con-
verge more quickly, whereas the solution of upwind scheme
was more robust. For discretizations that are rst order in
terms of Taylor series, a truncation error can be minimized
by using higher-order schemes. Higher-order discretization
involves more neighbor points and reduces discretization
errors by bringing wider inuence. Brucato et al. (1998)
compared the results of the hybrid upwind scheme and
the third-order upwind discretization scheme, QUICK,
and claimed that the only difference was QUICK tends to
predict a little higher recirculation rates in the bottom and
top of the tank. Roache (1998) concluded that the effect of
inherent numerical diffusion of the rst-order methods on
solution accuracy is devastating. Aubin et al. (2004) also
indicated this result.
8. Conclusions and challenges in CFD
simulations of stirred vessels
Generally, the results of this review have been briefed in the
following:
The E-L viewpoint can simulate detailed ow structures
with a higher spatial resolution than E-E and shows more
benets for dispersed multiphase ows but needs more
computational cost.
Generally, the LES method has obtained better results
in comparison with the other approaches especially with
the RANS method, which, with respect to the measured
mixing time, the simulated one has been overestimated in
most some cases. The major weakness of the k- model
is the assumption of isotropy of turbulence in the vessel,
which results in less accurate predictions in regions of
anisotropic turbulence, despite the generally satisfactory
comparison between the experimental ndings and the nu-
merical predictions, especially in regions far away from
the impeller (Markatos 1986 ).
There are still difculties related to turbulence mod-
els in terms of ow data such as mixing time scales to
device performance. This aspect has not been thoroughly
considered by researchers. CFD still poses a problem in
processes that involves signicant ow behavior changes,
phase change, and combining of mixing where the uid
properties change during the reaction within the reactor
(such as polymerization).
Due to the complexity of the ow eld in stirrer reactors,
the current available breakup and coalescence models for
gas-liquid ows in these vessels are not adequate for many
practical applications. Bubbles or droplets are typically
polydispersed and are not necessarily spherical, which
changes the active surface and affect mass transfer. Also,
every single bubble has its own velocity, size, and compo-
sition, and all these issues must be considered along with
the interaction between the phases to predict the evolu-
tion in time and space of the dispersed phase correctly that
results in the correct local and global mass transfer ux
predictions.
For accurate prediction of mass transfer in stirrer vessels,
both the hydrodynamic parameters and mass transfer pa-
rameters should be considered; thus, coupling of PBEs
along with the bubble breakup and coalescence and CFD
is necessary. Considering the issues mentioned, further
work based on CFD must be done to better understand the
relationship between ow eld and the other parameters in
the mixing reactors.
CFD development is still a challenge, particularly with in-
dustrial-grade meshes and large hydroturbine simulation
problems. A proper interpolation algorithm to correctly
evaluate the elds is still missing. This is especially
true in the presence of meshes where the size of cells
on either side of the mixing plane varies signicantly, or
rather, where a grid-independent solution is required.
Generally, CFD development has two steps: rst, make
it run, then make it run fast. It is at the rst make it run
stage of development that there are still many aspects left
to uncover.
Nomenclature
Ar Archimedes number
( ) ( )
3 2
-
p p L L
Ar gd
1
]

c Constant (value = 0.6); this model is referred as the rigid
model and is denoted as k
L
rigid
D
L
Diffusion coefcient
d Diameter of parent bubble, m
d
e
Eddy size, m
d
j
Diameter of daughter bubble, m
d
b
Bubble diameter
E Cumulative dissipated energy, m
2
s
-3

F
bv
Volume fraction, V
i
/V
j

F Cumulative
2
distribution
g Gravitational acceleration, m s
-2

h Collision frequency, m
3
s
-1

k Wave number of eddies, m
-1

K Von Karman constant
L
s
Mixing length (LES)
m Number of daughter bubbles
n
j
Volume density distribution of energy dissipation rate
n Number of bubbles or eddies, m
-3

N Average number of eddies that arrive at the surface of a
bubble in unit time
P Viscosity ratio
P
b
Breakup efciency
Q
i
Exchange ow rate between impeller and circulation
zones, m
3
s
-1

Q
b
Exchange ow rate between bafe and circulation zones,
m
3
s
-1

S
b
and S
c
Source terms for number density generated by coales-
cence and breakup, respectively
u Velocity vector

i
u

Filtered or solved instantaneous velocity (LES)
u
i
Instantaneous velocity

i
u

Periodic velocity

i
u

Turbulent velocity

p
i
u

Mean velocity solved by URANS simulation for the
blade position p

p
i
u

Phase average velocity for the blade position P
u
s
slip velocity

1
u

Time average radial velocity, m s
-1

2
u

Time average axial velocity, m s
-1

3
u

Time average tangential velocity, m s
-1

We Weber number
Greek symbols
Volume fraction of disperse phase
Smagorinsky eddy viscosity ( = 2/9)
Daughter size distribution
Turbulent dissipation rate, m
2
s
-3

v
b
Bubble slip velocity
Cinematic viscosity

SGS
T

Subgrid turbulent viscosity (LES)
Density, kg m
-3

SGS
ij

Subgrid turbulent stress tensor (LES)

,
SGS
T v

Voke eddy viscosity

b
Impeller volume ratio

b
Bafe volume ratio

t
Turbulent viscosity, kg m
-1
s
Dynamic viscosity, Pa s

t
Impeller energy dissipation ratio

b
Bafe energy dissipation ratio
Kolmogorov length scale = ( v
3
/ )
1/4

Coalescence efciency
Turbulent frequency in k- model
Gamma function
Eddy size divided by parent bubble size
Shear rate, s
-1

Kolmogorov length scale, m
Kinematic viscosity, m
2
s
-1

Surface tension, N m
-1

Inertial turbulent stress, N m
-

2

SI
Source term for interfacial area density generated by sur-
face instability, m
-1
s
-1

Breakup frequency, s
-1

Acknowledgements
The authors are grateful to the University of Malaya High Impact
Research Grant (HIR-MOHE-D000038-16001) from the Ministry of
Higher Education Malaysia and University of Malaya Bright Spark
Unit, which nancially supported this work.
References
Abu-Farah L, Al-Qaessi F, Schonbucher A. Cyclohexane/water dis-
persion behaviour in a stirred batch vessel experimentally and
with CFD simulation. Procedia CS 2010; 1: 655 664.
Ankamma Rao D, Sivashanmugam P. Experimental and CFD simu-
lation studies on power consumption in mixing using energy sav-
ing turbine agitator. J Ind Eng Chem 2010; 16: 157 161.
Ascanio G, Castro B, Galindo E. Measurement of power consump-
tion in stirred vessels a review. Chem Eng Res Des 2004; 82:
1282 1290.
Ashraf Ali B, Pushpavanam S. Analysis of unsteady gas-liquid
ows in a rectangular tank: comparison of Euler-Eulerian and
Euler-Lagrangian simulations. Int J Multiphase Flow 2011; 37:
268 277.
Ashraf Ali B, Siva Kumar Ch, Pushpavanam S. Analysis of liquid
circulation in a rectangular tank with a gas source at a corner.
Chem Eng J 2008; 144: 442 452.
Aubin J, Fletcher DF, Xuereb C. Modeling turbulent ow in stirred
tanks with CFD: the inuence of the modeling approach, turbu-
lence model and numerical scheme. Exp Ther Fluid Sci 2004;
28: 431 445
Bakker A, Fasano JB, Myers KJ. Effects of ow pattern on the sol-
ids distribution in a stirred tank. IChemE Symp Ser 1994; 136:
1 7.
Bordel S, Mato R, Villaverde S. Modeling of the evolution with
length of bubble size distributions in bubble columns. Chem Eng
Sci 2006; 61: 3663 3673.
Bouuyatiotis BA, Thonton JD. Liquid liquid extraction studies in
stirred tanks. Part 1: droplet size and hold-up measurements in a
seven-inch diameter bafed vessel. IChemE 1967; 26; 43 51.
Brucato A, Grisa F, Montante G. Particle drag coefcients in turbu-
lent uids. Chem Eng Sci 1998; 53: 3295 3314.
Buffo A, Vanni M, Marchisio D. Multidimensional population bal-
ance model for the simulation of turbulent gas-liquid systems in
stirred tank reactors. Chem Eng Sci 2012; 70: 31 44.
Bujalski W, Jaworski Z, Nienow AW. CFD study of homogeniza-
tion with dual Rushton tubines comparison with experimental
results part II: the multiple reference frame. Trans Inst Chem
Eng 2002; 80: 97 104.
Buwa VV, Deo DS, Ranade VV. Eulerian-Lagrangian simulations of
unsteady gas-liquid ows in bubble columns. Int J Multiphase
Flow 2006; 32: 864 885.
Chakrabarti M, Hill JC. First order closure theories for series parallel
reaction in simulated homogeneous turbulence. Adv Chem Eng
1997; 43: 902.
Colin C, Riou X. Turbulence and shear-induced coalescence in gas-
liquid pipe ows, 5th IntCon Multiphase Flow, Japon 2004;
5 30.
Coroneo M, Montante G, Paglianti A, Magelli F. CFD prediction of
uid ow and mixing in stirred tanks: numerical issues about the
RANS simulations. Comp Chem Eng 2011; 35: 1959 1968.
Coulaloglou CA, Tavlarides LL. Description of interaction processes
in agitated liquid-liquid dispersions. Chem Eng Sci 1977; 32:
1289 1297.
Delafosse A, Line A, Morchain J, Guiraud P. LES and URANS sim-
ulations of hydrodynamics in mixing tank: comparison to PIV
experiments. Chem Eng Res Des 2008; 86: 1322 1330.
Derksen J, Harry E, Akker VD. Large eddy simulations on the ow
driven by a rushton turbine. AIChE J 1999; 45: 209 221.
Ding J, Wang X, Zhou X-F, Ren N-Q, Guo W-Q. CFD optimization
of continuous stirred-tank (CSTR) reactor for biohydrogen pro-
duction. Biores Technol 2010; 101: 7005 7013.
Eggels JGM. Direct and large-eddy simulation of turbulent uid ow
using Lattice-Boltzmann scheme. Int J Heat Fluid Flow 1996;
17: 307 323.
Feng X, Cheng J, Li X, Yang C, Mao Z-S. Numerical simulation of
turbulent ow in a bafed stirred tank with an explicit algebraic
stress model. Chem Eng Sci 2012; 69: 30 44.
Forney LJ, Naa N. Eddy contact model: CFD simulations of liq-
uid reactions in nearly homogeneous turbulence. Chem Eng Sci
2000; 55: 6049 6058.
Fu XY, Ishii M. Two-group interfacial area transport in vertical air-
water ow: I. Mechanistic model. Nucl Eng Des 2003; 219:
143 168.
Gimbun J, Rielly CD, Nagy ZK. Modelling of mass transfer in gas-
liquid stirred tanks agitated by Rushton turbine and CD-6 impel-
ler: a scale-up study. Chem Eng Res Des 2009; 87: 437 451.
Hagesaether L. Coalescence and break up of drops and bubbles.
PhD thesis, Norwegian University of Science and Technology,
Norway, Trondheim, 2002.
Haresh P, Farhad E-M, Ramdhane D. CFD analysis of mixing in
thermal polymerization of styrene. Comp Chem Eng 2010; 34:
421 429.
Harris CK, Roekaerts D, Rosendal FJJ, Buitendijk FGJ, Daskopoulos
P, Vreenegoor AJN, Wang H. Computational uid dynam-
ics for chemical reactor engineering. Chem Eng Sci 1996; 51:
1569 1594.
Hartmann H, Derksen JJ, Montavon C, Pearson J, Hamill IS, Van
den Akker HEA. Assessment of large eddy and RANS stirred
tank simulations by means of LDA. Chem Eng Sci 2004; 59:
2419 2432.
Hibiki T, Ishii M. One-group interfacial area transport of bubbly
ows in vertical round tubes. Int J Heat Mass Transfer 2000a;
43: 2711 2726.
Hibiki T, Ishii M. Two-group interfacial area transport equations at
bubbly-to-slug ow transition. Nucl Eng Des 2000b; 202: 39 76.
Hjertager LK, Hjertager BH, Solberg T. CFD modelling of fast
chemical reactions in turbulent liquid ows. Comp Chem Eng
2002; 26: 507 515.
Honkanen M, Vaittinen J, Saarenrinne P, Korpij rvi J. Time-resolved
stereoscopic PIV experiments for validating transient CFD simu-
lations. Proceedings of the Seventh International Symposium on
Particle Image, Rome, Italy, 2007; 1 13.
Ishii M, Zuber N. Drag coefcient and relative velocity in bubbly,
droplet or particulate ows. AIChE J 1979; 25: 843 855.
Jahoda M, Mo t e k M, Kukukov A, Macho n V. CFD modelling of
liquid homogenization in stirred tanks with one and two impel-
lers using large eddy simulation. Chem Eng Res Des 2007; 85:
616 625.
Jaworski Z, Dyster KN, Nienow AW. The effect of size location and
pumping direction of pitched blade turbine impellers on ow pat-
terns: LDA measurements and CFD predictions. Trans Inst Chem
Eng 2001; 79: 887 894.
Joshi JB, Nere NK, Mathpati CS, Rane CV, Patwardhan AW, Murthy
BN, Ranade VV. CFD simulation of stirred tanks: comparison of
turbulence models. Part I: radial ow impellers. Can J Chem Eng
2011a; 89: 23 82.
Joshi JB, Nere NK, Mathpati CS, Rane CV, Patwardhan AW, Murthy
BN, Ranade VV. CFD simulation of stirred tanks: comparison of
turbulence models. Part II: axial ow impellers, multiple impellers
and multiphase dispersions. Can J Chem Eng 2011b; 89: 754 816.
Kasat GR, Khopkar AR, Ranade VV, Pandit AB. CFD simulation of
liquid-phase mixing in solid-liquid stirred reactor. Chem Eng Sci
2008; 63: 3877 3885.
Kerdouss F, Bannari A, Proulx P. CFD modeling of gas dispersion
and bubble size in a double turbine stirred tank. Chem Eng Sci
2006; 61: 3313 3322.
Kerdouss F, Bannari A, Proulx P, Bannari R, Skrga M, Labrecque Y.
Two-phase mass transfer coefcient prediction in stirred vessel
with a CFD model. J Comp Chem Eng 2008; 32: 1943 1955.
Khopkar AR, Tanguy PA. CFD simulation of gas-liquid ows in
stirred vessel equipped with dual Rushton turbines: inuence of
parallel, merging and diverging ow congurations. Chem Eng
Sci 2008; 63: 3810 3820.
Khopkar AR, Kasat GR, Pandit AB, Ranade VV. CFD simulation of
mixing in tall gas-liquid stirred vessel: role of local ow patterns.
Chem Eng Sci 2006; 61: 2921 2929.
Kitagawa A, Murai Y, Yamamoto F. Two-way coupling of Eulerian-
Lagrangian model for dispersed multiphase ows using ltering
functions. Int J Multiphase Flow 2001; 27: 2129 2153.
Laakkonen M, Alopaeus V, Aittamaa J. Validation of bubble break-
age, coalescence and mass transfer models for gas-liquid disper-
sion in agitated vessel. Chem Eng Sci 2006; 61: 218 228.
Laakkonen M, Moilanen P, Alopaeus V, Aittamassa J. Modeling local
bubble size distributions in agitated vessels. Chem Eng Sci 2007;
62: 721 740.
Lane GL, Schwarz MP, Evans GM. Predicting gas-liquid ow in a
mechanically stirred tank. Appl Math Mod 2002; 26: 223 235.
Lehr F, Mewes D. A transport equation for the interfacial area den-
sity in two-phase ow. Second European Congress of Chemical
Engineering, Montpellier, France, 1999.
Lehr F, Mewes D. A transport equation for the interfacial area density
applied to bubble columns. Chem Eng Sci 2001; 56: 1159 1166.
Lehr F, Millies M, Mewes D. Bubble-size distributions and ow
elds in bubble columns. AIChE J 2002; 48: 2426 2443.
Liao Y, Lucas D. A literature review of theoretical models for drop
and bubble breakup in turbulent dispersions. Chem Eng Sci
2009; 64: 3389 3406.
Liu S, Li D. Drop coalescence in turbulent dispersions. Chem Eng
Sci 1999; 54: 5667 5675.
Liu Y-J, Li W, Han L-C, Cao Y, Luo H-A, Al-Dahhan M, Dudukovic
MP. -CT measurement and CFD simulation of cross section gas
holdup distribution in a gas-liquid stirred standard Rushton tank.
Chem Eng Sci 2011; 66: 3721 3731.
Luo H, Svendsen HF. Theoretical model for drop and bubble breakup
in turbulent dispersions. AIChE J 1996; 42: 1225 1233.
Marchisio DL, Fox RO. Solution of population balance equations
using the direct quadrature method of moments. Aerosol Sci
2005; 36: 43 73.
Marchisio DL, Fox RO. Multi phase reacting ows: modelling and
simulation. Int Cent Mech Sci Courses Lect 2007; 492: 44 48.
Marchisio DL, Virgil RD, Fox RO. Implementation of the quadra-
ture method of moments in CFD codes for aggregation-breakage
problems. Chem Eng Sci 2003; 58: 3337 3351.
Markatos NC. The mathematical modeling of turbulent ows. Appl
Math Model 1986; 10: 190 220.
Mavros P. Flow visualization in stirred vessels. Review of experi-
mental techniques, Trans IChemE, Part A, Chem Eng Res 2001;
79: 113127.
McGraw R. Description of aerosol dynamics by the quadrature
method of moments. Aerosol Sci Technol 1997; 27: 255 265.
Menter FR. Two-equation eddy-viscosity turbulence models for
engineering applications. AIAA J 1994; 32: 269 289.
Menter FR. Review of the shear-stress transport turbulence model
experience from an industrial perspective. Int J Comp Fluid Dyn
2009; 23: 305 316.
Moilanen P, Laakkonen M, Visuri O, Alopaeus V, Aittamaa J.
Modelling mass transfer in an aerated 0.2m
3
vessel agitated by
Rushton, phase jet and combi jet impellers. Chem Eng J 2008;
142: 95 108.
Montante G, Lee KC, Brucato A, Yianneskis M. Numerical simula-
tions of the dependency of ow pattern on impeller clearance in
stirred vessels. Chem Eng Sci 2001; 56: 3751 3770.
Montante G, Paglianti A, Magelli F. Experimental analysis and com-
putational modeling of gas-liquid stirred vessels. Chem Eng Res
Des 2007; 85: 647 653.
Montante G, Horn D, Paglianti A. Gas-liquid ow and bubble size
distribution in stirred tanks. Chem Eng Sci 2008; 63: 2107
2118.
Murthy BN, Joshi JB. Assessment of standard k-e, RSM and les
turbulence models in a bafed stirred vessel agitated by various
impeller designs. Chem Eng Sci 2008; 63: 5468 5495.
Ochieng A, Lewis AE. CFD simulation of solids off-bottom suspen-
sion and cloud height. Hydrometallurgy 2006; 82: 1 12.
Ochieng A, Onyango M. CFD simulation of solid suspension in
stirred tanks: review. Hem Ind 2010; 64: 365 374.
Ochieng A, Onyango M, Kiriamiti K. Experimental measurement
and computational uid dynamics simulation of mixing in a
stirred tank: a review. S Afr J Sci 2009; 105: 421 426.
Petitti M, Marchisio DL, Vanni M, Baldi G, Mancini N, Podenzani
F. Effect of drag modelling on the prediction of critical regime
transitions in agitated gas-liquid reactors with bubble size
distribution modeling. Multiphase Sci Technol 2009; 21: 95
106.
Petitti M, Nasuti A, Marchisio DL, Vanni M, Baldi G, Mancini N,
Podenzani F. Bubble size distribution modeling in stirred gas-
liquid reactors with QMOM augmented by a new correction
algorithm. Am Inst Chem Eng J 2010; 56: 36 53.
Podila K, Al Taweel AM, Koksal M, Troshko A, Gupta YP. CFD
simulation of gas-liquid contacting in tubular reactors. Chem
Eng Sci 2007; 62: 7151 7162.
Poorte REG, Biesheuvel A. Experiments on the motion of gas bub-
bles in turbulence generated by an active grid. J Fluid Mech
2002; 461: 127 154.
Prince MJ, Blanch HW. Bubble coalescence and break-up in air-
sparged bubble columns. AIChE J 1990; 36: 1485 1499.
Rahimi M, Parvareh A. Experimental and CFD investigation on mix-
ing by a jet in a semi-industrial stirred tank. Chem Eng J 2005;
115: 85 92.
Ranganathan P, Sivaraman S. Investigations on hydrodynamics and
mass transfer in gas-liquid stirred reactor using computational
uid dynamics. Chem Eng Sci 2011; 66: 3108 3124.
Roache PJ. Verication of codes and calculations. AIAA J 1998; 36:
696 702.
Rong F, Daniele LM, Rodney F. Application of the direct quadra-
ture method of moments to poly disperse gas-solid uidized bed.
Powder Technol 2004; 139: 7 20.
Sahu AK, Roy R, Joshi A, Sevick-Muraca EM. Evaluation of ana-
tomical structure and non-uniform distribution of imaging agent
in near-infrared uorescence-enhanced optical tomography. Int J
Opt 2005; 13: 10182 10199.
Scargiali F, D Orazio A, Grisa F, Brucato A. Modelling and simula-
tion of gas-liquid hydrodynamics in mechanically stirred tanks.
Chem Eng Res Des 2007; 85; 637 646.
Baharak Sajjadi was born in
Iran in 1984. She graduated
with a chemical engineer-
ing degree in 2008 in Arak
University, Iran, and received
her master s degree in 2010
at the same university. She
worked on hydrodynamics and
mass transfer investigations in
airlift bioreactors with CFD.
She joined the University of
Malaya, Malaysia, in 2011 as a doctoral candidate work-
ing on biochemical processing, with particular interest in
biofuel production. Other interests include development,
validation, and optimization in unconventional geometries.
Dr. Abdul Aziz completed his
PhD in the area of three-phase
mixing. Currently, he is an
associate professor and holds
the position of deputy dean at
the Faculty of Engineering,
University of Malaya,
Malaysia. His research inter-
ests are in mixing in stirred
vessels and cleaner produc-
tion technologies. He is also active in consultancy projects
and is currently supervising 15 PhD candidates. He is a mem-
ber of a number of professional and learned societies such
as the Institution of Chemical Engineers (IChemE, UK), the
Institution of Engineers Malaysia (IEM), and the American
Chemical Society (ACS).
Prof. Shaliza Ibrahim is afli-
ated with the Department
of Civil Engineering at the
University of Malaya under the
Environmental Engineering
Programme. A chemical engi-
neer by training, her interest
in mixing research stemmed
from her PhD study back in
1988 1992 at the University
of Birmingham in the area of
multiphase mixing in stirred
vessels. She has been instrumental in keeping mixing research
going at the University of Malaya, with particular interest in
solid-liquid mixing. Her more recent projects are applied to
biological and adsorption processes in wastewater treatment.
Raja Shazrin Shah was born
in Malaysia in 1981. He
graduated with a chemical
engineering degree in 2004
in the University of Malaya,
Malaysia, and received his
master s degree in 2010 in the
same university. He worked
on applied research for
resource recovery for waste
streams coming from natural
rubber and palm oil industries in Malaysia, with particular
emphasis on membrane technologies. He joined as a doctoral
candidate in 2012 in the same university, working on hydro-
dynamic studies on multiphase systems in stirred reactors.
Schiller L, Naumann Z. A drag coefcient correlation. Z Ver Deutsch
Ing 1935; 77: 318.
Selma B, Bannari R, Proulx P. Simulation of bubbly ows: compari-
son between direct quadrature method of moments (DQMOM)
and method of classes (CM). Chem Eng Sci 2010; 65:
1925 1941.
Singh H, Fletcher DF, Nijdam JJ. An assessment of different turbu-
lence models for predicting ow in a bafed tank stirred with a
Rushton turbine. Chem Eng Sci 2011; 66: 5976 5988.
Smagorinsky J. General circulation experiments with the primitive
equations: 1. The basic experiment. Mon Wea Rev 1963; 91:
99 164.
Sokolichin A, Eigenberger G, Lapin A, Lubbert A. Dynamic numeri-
cal simulations of gas-liquid two phase ows, Euler/Euler verses
Euler/Lagrange. Chem Eng Sci 1997; 52: 611 626.
Spelt PDM, Biesheuvel A. On the motion of gas bubbles in
homogeneous isotropic turbulence. J Fluid Mech 1997; 336:
221 244.
Trivellato F. On the efciency of turbulent mixing in rotating stirrers.
Chem Eng Proc Int 2011; 50: 799 809.
Van den Akker HEA. The details of turbulent mixing process and
their simulation. Adv Chem Eng 2006; 31: 151 229.
Van Wachem BGM, Almstedt AE. Methods for multiphase computa-
tional uid dynamics. Chem Eng J 2003; 96: 81 98.
Venneker BCH, Derksen JJ, Van den Akker HEA. Population bala-
nce modeling of aerated stirred vessels based on CFD. AIChE J
2002; 48; 673 685.
Wang T, Wang J, Jin Y. Theoretical prediction of ow regime transi-
tion in bubble columns by the population balance model. Chem
Eng Sci 2005; 60: 6199 6209.
Wang L, Zhang Y, Li X, Zhang Y. Experimental investigation and
CFD simulation of liquid-solid-solid dispersion in a stirred reac-
tor. Chem Eng Sci 2010; 65: 5559 5572.
Wu Q, Kim S, Ishii M, Beus SG. One-group interfacial area trans-
port in vertical bubbly ow. Int J Heat Mass Transfer 1998; 41:
1103 1112.
Xia J-Y, Wang Y-H, Zhang S-L, Chen N, Yin P, Zhuang Y-P, Chu J.
Fluid dynamics investigation of variant impeller combinations
by simulation and fermentation experiment. Biochem Eng J
2009; 43: 252 260.
Yeoh SL, Papadakis G, Yianneskis M. Numerical simulation of tur-
bulent ow characteristics in a stirred vessel using the LES and
RANS approaches with the sliding/deforming mesh methodo-
logy. Chem Eng Res Des 2004; 82: 834 848.
Yeoh SL, Papadakis G, Yianneskis M. Determination of mixing time
and degree of homogeneity in stirred vessels with large eddy
simulation. Chem Eng Sci 2005; 60: 2293 2302.
Yuan C, Fox RO. Conditional quadrature method of moments for
kinetic equations. J Comp Phys 2011; 230: 8216 8246.
Zadghaffari R, Moghaddas JS, Revstedt J. Large-eddy simulation of
turbulent ow in an unbafed stirred tank driven by a Rushton
turbine. Comp Fluids J 2010; 39: 1183 1190.
Received March 12, 2012; accepted May 30, 2012

Review On Gas-Liquid Mixing Analysis in Multiscale Stirred Vessel Using CFD-libre

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Review On Gas-Liquid Mixing Analysis in Multiscale Stirred Vessel Using CFD-libre

Uploaded by

Copyright:

Available Formats

Rev Chem Eng 28 (2012): 171189 2012 by Walter de Gruyter Berlin Boston. DOI 10.

You might also like