Research Article 1545

Vania Santos-Moreau1,*
Estimation of kLa Values in Bench-Scale
José Carlos B. Lopes2
Cláudio P. Fonte3,* Stirred Tank Reactors with Self-Inducing
Impeller by Multiphase CFD Simulations
A multiphase computational fluid dynamics (CFD) simulation methodology is de-
veloped and proposed for the estimation of the spatial distribution of kLa values
in a bench-scale reactor equipped with a self-inducing impeller. The importance
of estimating an apparent drag coefficient, which considers the effect of turbulence
on the gas bubble rising velocity, is also tackled by applying different correlations
available in literature, namely, Brucato, modified Brucato, and Pinelli correlations.
The spatial distribution of kLa values in the agitated vessel is found from the CFD
results using Danckwert’s surface renewal model. An analysis of the gas volume
fraction distribution obtained from the simulations is performed in order to
choose the most suitable drag model. The modified Brucato correction correlation
for the drag force exhibits the best agreement with experimental data.

Keywords: Computational fluid dynamics, Gas dispersion, Mass transfer, Multiphase reactor,
Stirred-tank reactor
Received: March 07, 2019; accepted: March 08, 2019
DOI: 10.1002/ceat.201900162

1 Introduction for gas suction, gas suction rate, gas holdup, gas-liquid mass
transfer, power consumption etc., is thus essential for control-
Different technologies to promote the dispersion of gas into ling its operation. Several authors studied experimentally and
liquids, such as spargers, gas ejectors, and self-inducing impel- through numerical modeling these kinds of systems for differ-
lers, are used from laboratory to industrial scale. Self-inducing ent applications, both gas-liquid or gas-solid-liquid [3–13].
impellers have hollow shafts that draw gas captured from above This study focuses on bench-scale stirred reactors equipped
the free surface down into the liquid, without the need for ad- with hollow self-inducing impellers that are used to mimic in-
ditional devices. Simultaneously, the flow turbine of the self-in- dustrial processes of the oil and gas industry. A good replica-
ducing impeller promotes the dispersion and mixing of the gas tion of the hydrodynamic and mass transfer rate conditions al-
in the agitated liquid. Its operation principle is based on the lows testing industrial processes at a smaller scale, thereby
pressure gradient generated due to rotation between inlet orifi- reducing costs associated to construction and operation. None-
ces at the top of the shaft above the liquid level and orifices theless, it is known that gas-liquid mass transfer rates depend
placed near the impeller blades immersed in the agitated liquid. on the size of the vessel, blending time, induced gas flow rate,
The pressure gradient generated between the two orifices, for power dissipation rate etc., and do not scale all in the same
sufficiently high impeller speeds (critical speed), facilitates a manner with size. Industrial-sized reactors are more limited
continuous suction of gas at the top that flows through the hol- than bench-scale reactors [14]: typical industrial mass tran sfer
low shaft and is released into the liquid. Further details on the
gas-induction mechanisms can be found in [1, 2]. If the reactor
–
1
is operated in batch mode, the gas, which is not absorbed into Dr. Vania Santos-Moreau
the liquid, escapes from the free surface to be recirculated [email protected]
IFP Energies Nouvelles, Rond-point de l’échangeur de Solaize – BP 3,
again. This makes self-inducing impellers an attractive solution
69360 Solaize, France.
when the recycling gas is costly, not abundant, or hazardous. 2
Prof. José Carlos B. Lopes
In the turbulent regime, the gas-liquid mass transfer rate is
Universidade do Porto, LA LSRE/LCM, Laboratory of Separation and
directly related to the size of the interfacial area generated be- Reaction Engineering, Faculdade de Engenharia, Rua Dr. Roberto
tween the two phases and local values of turbulence dissipation Frias s/n, 4200-465 Porto, Portugal.
in the flow. These values depend simultaneously on the impel- 3
Dr. Cláudio P. Fonte
ler geometry and dimension, its rotational speed, and on the [email protected]
physical properties of the fluids. Understanding the phenom- The University of Manchester, School of Chemical Engineering and
ena taking place and studying the effect of different reactor de- Analytical Science, Oxford Road, M13 9PL Manchester, United King-
signs and operating conditions on the critical impeller speed dom.

Chem. Eng. Technol. 2019, 42, No. 8, 1545–1554 ª 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
 Research Article 1546

coefficients, kLa, have often values around 0.1 s–1, rarely ex- tributions of kLa in the reactor that allow identifying zones of
ceeding 0.3 s–1, whereas in bench-scale stirred reactors the kLa poor mass transfer rate.
can be higher than 1 s–1 [9]. For this reason, multiphase
stirred-tank reactors equipped with self-inducing impellers are
difficult to scale since complex coupled phenomena take place. 2 Reactor Geometry
In order to obtain a bench-scale reactor that is representative
of the industrial unit, it is important to know how to adequate The studied device is a batch bench-scale cylindrical flat-bot-
the bench-scale operational conditions to the range of kLa val- tomed reactor from Top Industrie with a diameter of 58 mm
ues found during the industrial operation. and a total volume of 300 cm3 (Fig. 1 a). Gas suction and mix-
To obtain kLa values in bench-scale stirred reactors, mass ing are promoted by a stainless-steel radial flow self-inducing
transfer experiments should be performed since the majority of impeller with a diameter of 19 mm (Fig. 1 b) from Top Indus-
existing correlations in the scientific literature cannot be used. trie. The main difference between a standard Rushton turbine
Usually, these correlations were developed for industrial-sized and the used impeller lies in the fact that the blades in the latter
reactors and are based on values of power dissipation, gas flow are welded to two disks instead of one. The two disks help the
rate, and gas holdup. At bench scale, these quantities are not mechanical conception of this small hollow. In addition, the ra-
easily determined experimentally. Power dissipation from di- tio between the blade height and the impeller diameter is con-
rect electrical measurements must be done with extreme care at siderably different for the two geometries: this ratio is usually
the bench scale since friction losses can represent up to 70 % of 0.20–0.25 for Rushton turbines and 0.52 for this self-inducing
the total power and need to be quantified. More accurate tech- impeller. Four equally spaced baffles with a length of 10 mm
niques are required to overcome this issue, such as dynamome- are placed in the reactor to avoid the formation of a free-sur-
ters or torque meters, but substantially higher investment is face vortex due to the impeller rotation. To help the mechanical
necessary [15]. Different authors developed experimental conception of the baffles of this small reactor, B/T = 0.17 was
methodologies to determine kLa following the physical absorp- chosen.
tion of gas [16–19]. This method is simple to set up since most
bench-scale units are equipped with pressure sensors. Some au-
thors have applied these methods to the characterization of
bench-scale reactors [9, 17, 18, 20–23].
Alternatively, kLa values can be estimated from computa-
tional fluid dynamics (CFD) simulations. Several authors, e.g.,
Buffo et al. and Gimbun et al. [24–29] previously used this ap-
proach to estimate the spatial distribution of the gas in the agi-
tated liquid and the interfacial mass transfer rate in medium-
sized and large-scale reactors. The multiphase flow inside the
vessel was simulated following an Euler-Euler modeling ap-
proach, and kL values were estimated from the flow simulations
using Danckwerts’ surface renewal model [30]. To take into ac-
count the bubbles’ coalescence and breakup, these authors pro- Figure 1. Transparent reactor (a) and self-inducing impeller (b).
posed the use of population balance models (PBMs). The use
of PBM increases considerably the CFD model complexity and In order to promote gas suction, two inlets are located at the
requires experimental data of coalescence and breakup rates to top of the hollow shaft, above the liquid free surface, and six
be included in the model. When breakup and coalescence are outlets are positioned in the middle of the impeller, between
not the most significant phenomena, the CFD model can be each pair of blades and between the disks. The internal diame-
considerably simplified assuming spherical-shaped bubbles ter of the hollow shaft is equal to 3 mm and the inlet and outlet
with a mono-dispersed size. Many modeling studies on gas-liq- orifices have a diameter of 1.5 mm. Fig. 2 shows a schematic
uid stirred tanks have been performed in recent years, using a representation with the main dimensions of the bench-scale re-
uniform mono-dispersed bubble size giving satisfactory results actor and the self-inducing impeller. The multiphase system is
in terms of gas holdup and mean flow [31–36]. a mixture of methylcyclohaxane/hydrogen at 10 bar and 20 C.
The objective of this work is the development of a computa- The reactor is filled with methylcyclohexane, up to a height of
tional model for the prediction of gas-liquid mass transfer rates 50 mm from the bottom.
in a bench-scale stirred reactor equipped with a self-inducing
impeller. To be able to be used as an optimization tool, this
computational model should be as inexpensive as possible con- 3 Numerical Simulations
cerning hardware requirements and simulation time. The flow
was simulated with an Euler-Euler multiphase modeling, con- 3D steady-state two-phase CFD simulations have been per-
sidering a mono-dispersed bubble size throughout the tank. formed to obtain the flow field of the liquid and the gas in the
The results will allow adapting the bench-scale reactor opera- stirred tank. Details on the development of the flow model and
tional conditions to industrial gas-liquid mass transfer perfor- the adopted numerical methodologies for the flow simulations
mances. In addition, the CFD simulations provide spatial dis- are described here.

Chem. Eng. Technol. 2019, 42, No. 8, 1545–1554 ª 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
 Research Article 1547

The momentum conservation equations for each phase are

coupled through the domain pressure (shared by both phases)
and interphase momentum exchange coefficients. The momen-
tum conservation equation for each phase i = {G,L} are:

ðai ri~
ui~ g þ~
ui Þ ¼ ai p þ ti þ ai ri~ FDrag;i þ ~
FNondrag;i
(4)

where p denotes the pressure field in the vessel and ~ g the grav-
ity acceleration. ~
FDrag;i is the drag force of the gas bubbles, and
~
FNondrag;i are other interaction forces such as lift, virtual mass
forces, and turbulent dispersion forces. The stress tensor, ti , for
the phase i is given by:
 
  2
ti ¼ ai mi ~
ui þ ~ ui ÞI
uTi þ ai li  mi ð~ (5)
3

where mi is the dynamic viscosity, li is the bulk viscosity, and

Figure 2. Schematic representation and dimensions of the sim- I is the identity tensor. Gravity was set to act on the system ver-
ulated stirred-tank reactor.
tically, i.e., along the impeller axis, and downwards. The drag
force on the moving bubbles was the only considered momen-
3.1 Two-Fluid Euler-Euler Flow Modeling tum exchange term between the two phases. Other contribu-
tions to the exchange of momentum between phases, like lift
A two-fluid Euler-Euler approach was used for the develop- and virtual mass forces, have been reported in the literature to
ment of the flow model of the agitated gas-liquid mixture in be non-dominant in multiphase stirred vessels [35], and there-
the reactor. This approach describes the motion for each phase fore have not been considered in the model. The drag force was
in a macroscopic but space-resolved sense, considering the two calculated as:
fluids as two interpenetrating continua. Mass and momentum
conservation equations are solved for each of the two phases 3 a
~ FDrag;G ¼ CD rL G j~
FDrag;L ¼ ~ u ~
uL jð~
uG  ~
uL Þ (6)
individually considering interaction terms between them. In 4 db G
steady state, mass conservation for each phase i = {G,L}1) is
mathematically described by: where CD is the drag coefficient and db is the bubble mean di-
ameter.
ðai ri~
ui Þ ¼ ðGi ri ai Þ (1) The turbulent properties of the flow in the stirred tank were
simulated by solving the Reynolds-averaged form of the Nav-
where ai is the volume fraction, ri is the density, and ~ui denotes ier-Stokes (RANS) equations. This solution method was em-
the velocity field of phase i in the flow. For simplicity, the bub- ployed since it requires less computational power than more
ble path dispersion was chosen to be modeled as a diffusive advanced methods such as large eddy simulation (LES) model-
term with a dispersion coefficient Gi rather than being treated ing or direct numerical simulation (DNS). While not providing
as an interfacial momentum force in the momentum conserva- the most detailed description of all flow scales, RANS turbu-
tion equations. Further details on this simplified but estab- lence modeling is more attractive and practical in an industrial
lished approach for modeling turbulence-induced bubble dis- application context due to its lower computational require-
persion can be found in [37]. The dispersion coefficient for the ments. The turbulence in the flow was simulated with the mul-
gas phase, GG, can be estimated from the turbulent viscosity in tiphase realizable k-e model with standard wall functions for
the liquid phase, nT,L, as follows: each phase, which solves a set of turbulent kinetic energy and
turbulent energy dissipation rate transport equations for each
nT;L
GG ¼ (2) of the two fluids [38]. Previous works demonstrated that this is
ScT;VOF the most suitable turbulence model for the simulation of the
flow in this device [39]. The k-e model constants C2, C3, sk, se,
where ScT,VOF is the turbulent Schmidt number for the phases. sGL were set with default values of 1.9, 1.3, 1, 1.2, 0.75, respec-
A default value of 0.75 for ScT,VOF has been considered in the tively.
simulations. In order to retain global continuity in the system,
the dispersion term of Eq. (1) for the liquid phase must be
equal to: 3.2 Drag Coefficient Estimation in Mechanically
ðGL rL aL Þ ¼ ðGG rG aG Þ (3) Induced Turbulence

The bubble drag coefficient was calculated from the Schiller-

– Naumman [40] and Tomiyama [41] correlations. From the
1) List of symbols at the end of the paper. Schiller-Naumann correlation, CD is estimated by:

Chem. Eng. Technol. 2019, 42, No. 8, 1545–1554 ª 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
 Research Article 1548

(
24 Brucato el al. [42] proposed an empirical-based correlation
ð1 þ 0:15Re0:687 Þ if Re £ 1000
CD ¼ Re (7) obtained from experimental measurements of the settling ve-
0:44 if Re > 1000 locity of solid particles in a Taylor-Couette flow:

where Re is the bubble Reynolds number  3 !

4db
CD ¼ CD 1 þ 8:7 · 10 (11)
u ~
r j~ uL jdb lK
Re ¼ L G (8)
mL
where CD is the apparent drag coefficient under turbulent con-
The Schiller-Naumann correlation is well suited for flows re- ditions and CD is the drag coefficient of a single bubble under
gimes with non-deformed spherical bubbles. For flow regimes quiescent conditions. This correction is a function of the ratio
where the bubbles present non-negligible shape deformation between the diameter of the bubbles and the Kolmogorov scale
(ellipsoids or spherical caps), the Tomiyama correlation has  3 1=4
been proposed. In this empirical model, CD is estimated by: nL
lK ¼ (12)
    eL
24   74 8 E€o
CD ¼ max min 1 þ 0:15Re0:687 ; ; (9)
Re Re 3 E€o þ 4 for the estimation of a corrective term for the drag coefficient.
It can be argued that, since Eq. (11) has been obtained for solid
where Eö is the Eötvös number, the ratio between the buoyancy particles, it must be assumed that mechanisms of drag modifi-
forces and surface tension forces, given by: cation due to free-stream turbulence are the same for solid par-
ticles and gas bubbles. The difference between the density of
g ðrL  rG Þdb2 the liquid and the gas bubbles or solid particles on settling/ris-
E€o ¼ (10)
s ing velocities should be taken into consideration as recently
pointed out by Doroodchi et al. [43].
where s is the surface tension. Later, Lane et al. [44] proposed a modification to Brucato’s
These two drag models were tested under the same opera- correlation constant:
tional conditions, i.e., 1600 rpm, bubble mean diameter of
2 mm, same fluids etc. No significant differences were observed  3 !
6db
in the estimated value of kLa and in the gas distribution in the CD ¼ CD 1 þ 6:5 · 10 (13)
lK
tank. Consequently, the Schiller-Naumann correlation was se-
lected to continue the calculations. This choice was also sup-
ported by flow visualizations in a transparent reactor, which This modification was found to offer better results in a simu-
show mainly non-deformed spherical bubbles. While Eqs. (7) lation of the gas distribution and holdup in stirred-tank reac-
and (9) can predict quite well the drag coefficient of individual tors, which allows the conclusion that the density difference be-
bubbles in quiescent flows, they may produce large errors tween the liquid and the inclusions has a relevant impact on
under mechanically generated turbulence or non-stationary drag modification due to turbulence. However, and notwith-
conditions. standing the better results that have been obtained, the pro-
Free-stream velocity fluctuations were reported from experi- posed modification is still lacking more solid physical ground-
ments to decrease substantially the settling velocity of solid ing.
particles, in some cases to as low as 15 % of the value in a qui- More recently, Pinelli et al. [45] proposed a correlation for
escent liquid [42]. This decrease of the settling velocities of sol- the particle settling in stirred vessels that also takes into
id particles or rising velocities of bubbles may be explained by account the ratio db/lk:
constant accelerations and decelerations of the particles or bub-    2
bles in the flow due to turbulent/stochastic velocity fluctua- l
CD ¼ CD 0:4 tanh 16 K  1 þ 0:6 (14)
tions. Added mass forces acting on the particles due to the ac- db
celerations and decelerations, and the fluctuations in the
instantaneous value of the drag coefficient combined with its thereby obtaining good results in modeling the dispersion and
nonlinear dependence on the velocity may appear as an in- sedimentation of particles in baffled and unbaffled stirred-tank
creased time-averaged drag coefficient. reactors with multiple impellers.
The results from Wutz et al. [27] suggest the importance of
using a correction factor for the drag coefficient in mechani-
cally generated turbulence for the CFD simulation of multi- 3.3 Gas-Liquid Mass Transfer Rate Estimation
phase flows in stirred vessels. Their simulations show good
agreement with experimental measurements of kLa values at The spatial distribution of the gas-liquid mass transfer rate,
low agitation speeds but the start exhibits less agreement as the kLa, in the stirred reactor was obtained from the CFD simula-
turbulence intensity in the tank is increased. This can be due to tions using Danckwerts’ surface renewal model [46]. Danck-
the fact that Wutz et al. [27] did not consider any correction werts proposed that the mass transfer rate between the two
term for the effect of turbulence on the gas bubble rising veloci- phases could be related to an average surface renewal rate,
ty.

Chem. Eng. Technol. 2019, 42, No. 8, 1545–1554 ª 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
 Research Article 1549

which is resulting from the contact of the bubbles’ interface

with the turbulent eddies in the liquid phase as:
pffiffiffiffiffiffiffiffiffi
kL ¼ Dm s (15)

where s is the rate of renewal liquid at the surface of the gas

bubbles and Dm means the diffusivity of the absorbed gas in
the liquid. Lamont and Scott [47] developed Danckwerts’ as-
sumption further based on the statistical theory of turbulent
diffusion by assuming s to be inversely proportional to the Kol-
1=2
mogorov time scale, s / t1 h ¼ ðnL =eL Þ . The gas-liquid co-
efficient kL can then be estimated by:
 
pffiffiffiffiffiffiffi eL 1=4
kL ¼ C Dm (16)
nL

where C is the model constant equal to 0.4. The specific surface

area of the bubbles, a, was obtained considering the symmetric
model, which ensures that the specific area approaches the val-
ue 0 as the volume fraction of gas reaches the value 1:

6
a¼ a ð1  aG Þ (17)
db G
Figure 3. Mesh of the entire stirred tank.
The molecular diffusivity of hydrogen in methylcyclohexane
was calculated with the Wilke-Chang correlation for the condi-
tions of the experiments [48]. The boundary conditions were set as follows. No-slip and
impermeable conditions at the vessel walls were assumed for
both phases. The top surface of the vessel was set as a non-de-
3.4 Mesh, Boundary Conditions, and Numerical formable surface with constant and uniform pressure equal to
Methods the atmospheric pressure (pressure outlet). Through this sur-
face, both the gas bubbles and the liquid were allowed to leave
The geometrical domain and numerical grid for the flow simu- the domain, however, only liquid was allowed to re-enter. This
lations were generated with the software packages DesignMod- approach, although not completely satisfying in a physical
eler and Meshing, respectively, included in the ANSYS 15 suite. sense, has been reported in previous works to be a good com-
All the internal parts of the reactor were generated with the promise between accuracy and ease of solution [24, 50]. The
same geometric dimensions as the ones of the ex-
isting setup, with the exception of the thickness of
the impeller and baffles which were neglected. Re-
cent computational studies demonstrated that ac-
curate velocity profiles can be obtained computa-
tionally even when the thickness of impeller blades,
impeller disk, and baffles is neglected [49].
The geometrical domain of the entire stirred
tank was discretized with a computational confor-
mal and structured mesh obtained from 600 k hex-
ahedral elements with typical sizes ranging from
0.2 to 0.6 mm (Fig. 3 and Fig. 4). Beyond this num-
ber of elements, the numerical results of velocity
and turbulence quantities were observed to be in-
dependent from the mesh refinement. The mesh
has a higher element density in the impeller zone,
where higher spatial gradients are expected. The
necessary higher values for the mesh density at the
walls were established by ensuring that the distance
of the first volume element from any wall of the re-
actor, in terms of the dimensionless distance, y+,
are smaller than 300 and mostly in the range of
30 £ y+ £ 300. Figure 4. Detail of the mesh near the impeller.

Chem. Eng. Technol. 2019, 42, No. 8, 1545–1554 ª 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
 Research Article 1550

gas phase was set to enter the domain from the surfaces repre- for residuals smaller than 10–3, ensuring as well the conver-
senting the impeller orifices immersed in the liquid. The gas in- gence of several monitored variables: gas holdup, dissipated
let velocity was set to be normal to the inlet surfaces, with a power calculated from volume integration of turbulent dissipa-
magnitude depending on the impeller rotational velocity. tion rate, and the volume-averaged value of kLa.
The relation between the impeller rotation speed, N (in The fluids used in the simulations were methylcyclo-
rpm), and the induced gas flow rate into the liquid, QG hexane (rL = 770 kg m–3, mL = 7.32 ·10–4 Pa s–1) as the conti-
(in m3s–1) for the device of this study has been reported in the nuous phase and hydrogen as the dispersed phase
literature [39] as: (rG = 8.22 ·10–1 kg m–3, mG = 8.83 ·10–6 Pa s–1). Both fluids were
considered to be incompressible.
log QG ¼
( 10
1 if Dpimp £ 0
 2   4 Results and Discussion
b1 log10 Dpimp þ b2 log10 Dpimp þ b3 if Dpimp > 0
(18) 4.1 Multiphase Flow Simulation

where Dpimp (in Pa) is the pressure drop due to friction in the The gas dispersion in the stirred tank was modeled using differ-
impeller shaft and for b1 = –0.0645, b2 = 0.945, and b13 = –6.1, ent correlations to take into account the apparent increase of
Dpimp can estimated from: drag due to the liquid-phase turbulence generated by the im-
peller: the Brucato correlation [42], the Modified Brucato cor-
2s relation [44], and the Pinelli correlation [45]. An analysis of
Dpimp ¼ a1 N a2  (19)
do the gas volume fraction distribution obtained from the simula-
tions was performed in order to choose the most suitable drag
where s is the surface tension, do is the self-inducing impeller model (Fig. 5). Assuming no drag modification due to turbu-
orifice diameter, and the constants are a1 = 0.0721 and lence effects, i.e., using the standard Schiller-Naumann correla-
a2 = 2.24. The liquid velocity components at the inlet orifices tion, the CFD simulations predicted a high concentration of
were assumed equal to zero. gas around the stirrer, mainly between the blades and disks,
The motion of the self-inducing turbine and its interaction and in the top region of the reactor. No presence of gas was
with the stationary baffles was modeled with the multiple refer- predicted in the bottom region of the reactor.
ence frame (MRF) methodology. In the MRF method, the
equations are expressed in a reference frame that rotates with
the impeller speed and are solved in steady state [38]. This
method is often used since it is less time-demanding than the
sliding mesh (SM) method, keeping accuracy and giving satis-
factory results.
Previous works indicated that the difference between the en-
semble-averaged flow field calculated with the stationary and
time-dependent approaches was negligible [51, 52]. The mov-
ing zone in the MRF approach was defined as a cylinder sur-
rounding the impeller and the shaft, along the rotation axis,
with twice the height of the impeller blades. The cylinder radi-
us is equidistant from blade tips and baffles.
The flow-governing conservation equations were solved with
the finite-volume commercial CFD solver ANSYS Fluent 15.
The coupled pressure-based solver with a pseudo-transient al-
gorithm available in Fluent was chosen for the coupling of the
continuity and momentum conservation equations. The use of
a pseudo-transient algorithm adds an unsteady term to the
flow equations and has been reported to improve stability and
the convergence behavior of the solution [38]. The convective Figure 5. Gas volume fraction inside the reactor at 1600 rpm
terms of the flow equations were discretized with a second-or- using no drag modification, and the Pinelli, Brucato, and modi-
der scheme and the mass conservation equations with a first- fied Brucato corrections.
order scheme. The turbulent kinetic energy of the liquid phase
was initialized with a value equal to kL = 0.1 m2s–2 and the tur-
bulence energy dissipation rate in the gas phase with a value Analyzing the experimental observations of the gas distribu-
equal to eL = 10 m2s–3. tion in the reactor for the same stirring velocity of 1600 rpm in
All other variables like the velocities of both phases, vessel Fig. 6, an accumulation of gas in the bottom region of the reac-
pressure, and gas volume fraction were initialized with a value tor was observed. This indicates that the standard Schiller-Nau-
equal to zero. Steady-state solutions were achieved applying a mann correlation for spheres in a quiescent liquid is not suit-
pseudo-transient formulation and were accepted as converged able to describe accurately the gas distribution in the reactor. In

Chem. Eng. Technol. 2019, 42, No. 8, 1545–1554 ª 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
 Research Article 1551

4.2 Mass Transfer Rate

Estimation

Volume-averaged gas-liquid mass

transfer coefficients, hkL ai, pre-
dicted with the CFD model were
compared with results reported by
Braga [23]. This author investi-
gated experimentally the gas-liquid
mass transfer in a bench-scale reac-
tor very similar to the simulated ge-
ometry in this work: a cylindrical,
flat-bottomed reactor equipped
with a double-disc hollow self-in-
ducing impeller, resembling to a
Ruston turbine with a diameter of
19 mm. The only difference be-
tween the simulated reactor and
Braga’s reactor is their diameter:
his reactor had a diameter of
62 mm against the 58 mm assumed
as the simulation’s geometry. In the
simulations and experiments the
same gas (hydrogen) and liquid
Figure 6. Simulated and experimental gas distribution in the reactor at different stirring rates. (methylcyclohexane) were consid-
Simulation using the modified Brucato correlation and Pinelli correlation. ered. The reactor was filled until
the same liquid height of 50 mm.
The experimental methodology fol-
the case of the Brucato correlation, higher gas concentrations lowed by Braga [23] to determine hkL ai coefficients was the
were predicted below the impeller, mainly in the axis of the method of physical absorption. This method has been de-
stirrer. This result does not seem physically grounded, and it is scribed previously by several authors [16–18] and consists in
not in agreement with the flow visualizations that show a high- measuring the total pressure variation in the agitated reactor
er gas density at the top part of the reactor and not at the bot- during batch operation.
tom (Fig. 6). For this reason, the Brucato correlation was ex- Fig. 7 illustrates the contour maps of simulated local kLa val-
cluded as well from further studies. ues in a vertical plane in the reactor, passing through the
The Pinelli and modified Brucato correlations seem to be impeller axis for the different impeller rotational velocities, as-
the most appropriate ones to simulate the bench-scale suming a spatially invariant bubble diameter of 2 mm based on
reactor. Fig. 6 displays the modeled and experimental gas distri-
bution in the reactor at different
stirring rates. Nonetheless, compar-
ing qualitatively the experimental
visualizations and the CFD simula-
tions, the Pinelli correction correla-
tion seems to overestimate the gas
dispersion in the bottom part of
the reactor as well.
Further discussion and explana-
tion of the results obtained for the
different drag modification correla-
tions would require a deeper and
more detailed study, which was
considered to be out of the scope of
this work. The authors were mostly
focused here on the selection of
the best correlation available to de-
scribe the impact of mechanically
generated turbulence on the pre- Figure 7. Contours of local kLa values in a vertical plane of the reactor passing through the axis
diction of the gas distribution and for a constant bubble diameter of 2 mm and different impeller rotational speeds using the modi-
kLa on the agitated vessel. fied Brucato correlation and Pinelli correlation.

Chem. Eng. Technol. 2019, 42, No. 8, 1545–1554 ª 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
 Research Article 1552

experimental validations, and using the modified Brucato and the approximation of a constant and spatially homogeneous
Pinelli drag coefficient corrections. The CFD results in Fig. 7 bubble size and shape. An experimental characterization of the
indicate that the mass transfer rate between the two phases is bubble size distribution, breakup, and coalescence inside the
not spatially uniform in the reactor, varying by at least four or- reactor must be performed in order to refine the model and, in
ders of magnitude in the domain. As expected, local kLa values turn, matching even further the computational predictions of
are higher in the region closer to the impeller where both gas- hkL ai with the results of Braga [23].
liquid interfacial area and turbulent energy dissipation rate are The experimental results and CFD simulations using the
higher. The results in Fig. 7 also demonstrate that in the range modified Brucato correlation were compared as well to values
from 1000 to 2000 rpm the values of kLa can increase consider- predicted by the correlation proposed by Dietrich et al. [18]
ably inside the vessel by at least one order of magnitude, with (Fig. 9), which is the correlation most frequently applied in the
the increase of N. This is due simultaneously to the increase of characterization of interfacial mass transfer of small-scale
the gas induction rate into the system and to the higher turbu- stirred reactors with self-inducing impellers:
lent energy dissipation in the domain with the rise of the im-
peller rotational velocity. ShG=L ¼ BRe1:45 0:5
imp Weimp Sc
0:5
(21)
The numerical predictions of local values of kLa were vol-
ume-averaged to be compared with the experimental measure-
ments of mean mass transport rates from Braga’s results [23].
The volume average value of the kLa was calculated as:
ZZZ
1
hkL ai ¼ kL ð~ xÞd3~
xÞað~ x (20)
Vreactor
Vreactor

where ~ x is the position vector inside the reactor volume Vreactor.

The comparison of these two values is only possible by consid-
ering the same assumptions adopted in the experimental work
of Braga [23]: the characteristic time for mixing is much lower
than that for mass exchange between the gas and the liquid
phases, i.e., the concentration of hydrogen in the liquid phase
can be considered homogeneous in the stirred tank. While
showing a similar trend and predicting values in the same or-
der of magnitude, the CFD simulations using the Pinelli modi-
fication correlation tend to overestimate the values of hkL ai
Figure 9. Volume-average value of kLa as a function of the im-
(Fig. 8).
peller rotational speed determined experimentally by Braga
[23], from CFD simulations for a bubble mean diameter of 2 mm
using the modified Brucato model and the Dietrich correlation.

where ShG=L ¼ kL aD2 =Dm is the Sherwood number,

Reimp ¼ ND2 rL =mL is the impeller Reynolds number, Weimp ¼
N 2 D3 rL =s denotes the Weber number based on the impeller
dimensions and speed, Sc ¼ mL =ðrL Dm Þ is the Schmidt num-
ber, and D means the impeller diameter. The constant B is
given by:

3 · 104 if H=T ¼ 1
B¼ (22)
1:5 · 104 if H=T ¼ 1:4

depending on the ratio between the height of the liquid in the

vessel, H, and its diameter, T. Fig. 9 indicates that the Dietrich
correlation [18] overestimates the hkL ai values. In the bench-
scale reactor used in this study, the developed model is more
Figure 8. Volume-average value of kLa as a function of the im- accurate in the estimation of mass transfer rates.
peller rotational speed determined experimentally by Braga [23]
and from the CFD simulations for a bubble diameter of 2 mm
using the modified Brucato model and Pinelli model. 5 Conclusions
CFD simulations using the modified Brucato correction A CFD Eulerian-Eulerian model of the gas-liquid flow in a
seem to be more consistent with experimental data. The devia- bench-scale stirred-tank reactor equipped with a self-inducing
tion between CFD results and experimentation can be due to impeller was developed. Danckwerts’ bubble surface renewal

Chem. Eng. Technol. 2019, 42, No. 8, 1545–1554 ª 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
 Research Article 1553

model was coupled to multiphase CFD simulations for the pre- ~

ui [m s–1] velocity field of phase i
diction of mass transfer rate coefficients at various impeller ~
uL [m s–1] velocity field of liquid phase
speeds. Different drag modification laws proposed in the litera- uG
~ [m s–1] velocity field of gas phase
ture to take into account the effect of mechanically generated
turbulence on the rising velocities of the gas bubbles were con- Greek letters
sidered in this study.
aG [–] gas volume fraction
A selection of suitable modeling parameters and correlations
aL [–] liquid volume fraction
was proposed. While being based on some simplifications, the
e [m2s–3] average turbulent dissipation rate
proposed CFD methodology is able to predict interphase mass
eL [m2s–3] average liquid turbulent dissipation
transfer rates with more accuracy when compared to the most
rate
frequent correlation used in industrial practice for character-
rL [kg m–3] liquid density
ization of mass transfer on small-scale stirred-tank reactors
p [N m–2] pressure gradient
with self-inducing gas turbines. Efforts were made to develop a ti [kg m–1s–2] liquid phase stress tensor
computationally inexpensive modeling methodology in order
G [m2s–1] VOF turbulent dispersion coefficient
to offer an attractive optimization tool for multiphase lab-scale
nt,L [m2s–1] kinematic liquid turbulent eddy
reactors.
viscosity
With the proposed CFD methodology, the estimated values
nL [m2s–1] liquid kinematic viscosity
of hkL ai show good agreement with experimental observations
mL [kg m–1s–1] liquid shear viscosity
reported in the literature. The results of this work illustrate the
lk [m–1] Kolomogrov length scale
relevance and suitability of using advanced modeling tech-
lL [kg m–1s–1] liquid bulk viscosity
niques in the study and design of multiphase chemical reactors,
Dpimp [Pa] pressure drop due to friction
especially when these devices are in a dimension range or have
s [N m–1] surface tension
a design for which only few studies exist in the literature. Addi-
tionally, it is obvious that further research on the impact of me- Abbreviations
chanically generated turbulence on the rising/settling velocities
of droplets, bubbles, or solids for the simulation of multiphase CFD computational fluid dynamics
turbulent flows is still required. PBM population balance model

The authors have declared no conflict of interest.

References
Symbols used [1] G. Q. Martin, Ind. Eng. Chem. Proc. Des. Dev. 1972, 11 (3),
397. DOI: https://doi.org/10.1021/i260043a012
a [m–1] specific surface area [2] N. A. Deshmukh, S. S. Patil, J. B. Joshi, Chem. Eng. Res. Des.
a1 [–] constant 2006, 84 (2), 124. DOI: https://doi.org/10.1205/cherd.05005
a2 [–] constant [3] J. B. Joshi, M. M. Sharma, Can. J. Chem. Eng. 1977, 55 (6),
b1 [–] constant 683. DOI: https://doi.org/10.1002/cjce.5450550609
b2 [–] constant [4] K. Saravanan, V. D. Mundale, A. W. Patwardhan, J. B. Joshi,
b3 [–] constant Ind. Eng. Chem. Res. 1996, 35 (5), 1583. DOI: https://
CD [–] drag coefficient doi.org/10.1021/ie9503353
[5] A. W. Patwardhan, J. B. Joshi, Ind. Eng. Chem. Res. 1999,
CD [–] modified drag coefficient
38 (1), 49. DOI: https://doi.org/10.1021/ie970504e
db [m] bubble diameter
[6] S. B. Sawant, J. B. Joshi, Chem. Eng. J. 1979, 18 (1), 87. DOI:
do [m] orifice diameter
https://doi.org/10.1016/0300-9467(79)80018-0
Dm [m2s–1] molecular diffusion coefficient
[7] J. B. Joshi, Chem. Eng. Commun. 2007, 5 (1–4), 109. DOI:
Eö [–] Eötvös number
https://doi.org/10.1080/00986448008935957
~FDrag [kg m–2s–2] volumetric drag forces [8] K. Saravanan, V. D. Mundale, J. B. Joshi, Ind. Eng. Chem.
~FNon-Drag [kg m–2s–2] volumetric non-drag forces Res. 1994, 33 (9), 2226. DOI: https://doi.org/10.1021/
~g [m s–2] gravity acceleration ie00033a029
I [–] identity tensor [9] M. M. P. Zieverink, M. T. Kreutzer, F. Kapteijn, J. A. Moulijn,
k [m2s–2] turbulent kinetic energy Ind. Eng. Chem. Res. 2006, 45 (13), 4574. DOI: https://
kL [s–1m] gas-liquid mass transfer doi.org/10.1021/ie060092m
kLa [s–1] gas-liquid mass transfer [10] B. N. Murthy, N. A. Deshmukh, A. W. Patwardhan, J. B.
N [rps] impeller speed Joshi, Chem. Eng. Sci. 2007, 62 (14), 3839. DOI: https://
QG [m3s–1] Induced gas flow rate doi.org/10.1016/j.ces.2007.03.043
Re [–] Particle relative Reynolds number [11] B. N. Murthy, R. S. Ghadge, J. B. Joshi, Chem. Eng. Sci. 2007,
s [s–1] Fractional rate of surface–element 62 (24), 7184. DOI: https://doi.org/10.1016/j.ces.2007.07.005
replacement
ScT,VOF [–] Schmidt VOF turbulent number

Chem. Eng. Technol. 2019, 42, No. 8, 1545–1554 ª 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com
 Research Article 1554

[12] B. N. Murthy, R. B. Kasundra, J. B. Joshi, Chem. Eng. J. 2008, [31] K. E. Morud, B. H. Hjertager, Chem. Eng. Sci. 1996, 51 (2),
141 (1–3), 332. DOI: https://doi.org/10.1016/j.cej.2008. 233. DOI: https://doi.org/10.1016/0009-2509(95)00270-7
01.040 [32] A. R. Khopkar, V. V. Ranade, AIChE J. 2006, 52 (5), 1654.
[13] S. E. Forrester, C. D. Rielly, K. J. Carpenter, Chem. Eng. Sci. DOI: https://doi.org/10.1002/aic.10762
1998, 53 (4), 603. DOI: https://doi.org/10.1016/S0009- [33] H. Sun, Z.-S. Mao, G. Yu, Chem. Eng. Sci. 2006, 61 (12),
2509(97)00352-7 4098. DOI: https://doi.org/10.1016/j.ces.2005.12.029
[14] E. H. Stitt, Chem. Eng. J. 2002, 90 (1–2), 47. DOI: https:// [34] Z.-S. Mao, C. Yang, Ind. Eng. Chem. Res. 2006, 45 (3), 1141.
doi.org/10.1016/S1385-8947(02)00067-0 DOI: https://doi.org/10.1021/ie0503085
[15] G. Ascanio, B. Castro, E. Galindo, Chem. Eng. Res. Des. [35] F. Scargiali, A. D’Orazio, F. Grisafi, A. Brucato, Chem. Eng.
2004, 82 (9), 1282. DOI: https://doi.org/10.1205/cerd.82.9. Res. Des. 2007, 85 (5), 637. DOI: https://doi.org/10.1205/
1282.44164 cherd06243
[16] R. V. Chaudhari, R. V. Gholap, G. Emig, H. Hofmann, Can. [36] M. Jahoda, L. Tomášková, M. Moštěk, Chem. Eng. Res. Des.
J. Chem. Eng. 1987, 65 (5), 744. DOI: https://doi.org/ 2009, 87 (4), 460. DOI: https://doi.org/10.1016/j.cherd.
10.1002/cjce.5450650506 2008.12.006
[17] V. Meille, N. Pestre, P. Fongarland, C. de Bellefon, Ind. Eng. [37] A. Sokolichin, G. Eigenberger, A. Lapin, AIChE J. 2004,
Chem. Res. 2004, 43 (4), 924. DOI: https://doi.org/10.1021/ 50 (1), 24. DOI: https://doi.org/10.1002/aic.10003
ie030569j [38] ANSYS, ANSYS FLUENT.
[18] E. Dietrich, C. Mathieu, H. Delmas, J. Jenck, Chem. Eng. Sci. [39] C. P. Fonte, B. S. Pinho, V. Santos-Moreau, J. C. B. Lopes,
1992, 47 (13–14), 3597. DOI: https://doi.org/10.1016/0009- Chem. Eng. Technol. 2014, 37 (4), 571. DOI: https://doi.org/
2509(92)85075-M 10.1002/ceat.201300412
[19] A. Deimling, B. M. Karandikar, Y. T. Shah, N. L. Carr, Chem. [40] L. Schiller, Z. Naumann, Z. Ver. Dtsch. Ing. 1935, 77, 318.
Eng. J. 1984, 29 (3), 127. DOI: https://doi.org/10.1016/0300- [41] T. Takamasa, A. Tomiyama, Proc. of the 1999 NURETH-9
9467(84)85038-8 Conf., San Francisco, October 1999.
[20] K. C. Ruthiya, J. van der Schaaf, B. F. M. Kuster, J. C. Schout- [42] A. Brucato, F. Grisafi, G. Montante, Chem. Eng. Sci. 1998, 53
en, Chem. Eng. J. 2003, 96 (1–3), 55. DOI: https://doi.org/ (18), 3295. DOI: https://doi.org/10.1016/S0009-2509(98)
10.1016/j.cej.2003.08.005 00114-6
[21] M. Mitrovic, Ph. D. Thesis, Université Claude Bernard Lyon [43] E. Doroodchi, G. M. Evans, M. P. Schwarz, G. L. Lane,
1, Lyon 2001. N. Shah, A. Nguyen, Chem. Eng. J. 2008, 135 (1–2), 129.
[22] I. Pitault, P. Fongarland, D. Koepke, M. Mitrovica, D. Ronze, [44] G. Lane, M. P. Schwarz, G. M. Evans, Proc. of the 10th Eur.
M. Forissier, Chem. Eng. Sci. 2005, 60 (22), 6240. DOI: Conf. on Mixing, Delft, July 2000.
https://doi.org/10.1016/j.ces.2005.04.041 [45] D. Pinelli, M. Nocentini, F. Magelli, Chem. Eng. Commun.
[23] M. Braga, Ph. D. Thesis, Université Claude Bernard Lyon 1, 2007, 188 (1), 91. DOI: https://doi.org/10.1080/
Lyon 2013. 00986440108912898
[24] A. Buffo, M. Vanni, D. L. Marchisio, Chem. Eng. Sci. 2012, [46] P. Danckwerts, Ind. Eng. Chem. 2004, 43 (6), 1460.
70, 31. DOI: https://doi.org/10.1016/j.ces.2011.04.042 [47] J. C. Lamont, D. S. Scott, AIChE J. 1970, 16 (4), 513. DOI:
[25] J. Gimbun, C. D. Rielly, Z. K. Nagy, Chem. Eng. Res. Des. https://doi.org/10.1002/aic.690160403
2009, 87 (4), 437. DOI: https://doi.org/10.1016/j.cherd. [48] C. R. Wilke, P. Chang, AIChE J. 1955, 1 (2), 264. DOI:
2008.12.017 https://doi.org/10.1002/aic.690010222
[26] X. Li, K. Scott, W. J. Kelly, Z. Huang, Biotechnol. Bioprocess [49] M. Coroneo, G. Montante, A. Paglianti, F. Magelli, Comput.
Eng. 2018, 23 (6), 710. DOI: https://doi.org/10.1007/s12257- Chem. Eng. 2011, 35 (10), 1959. DOI: https://doi.org/
018-0063-5 10.1016/j.compchemeng.2010.12.007
[27] J. Wutz, A. Lapin, F. Siebler, J. E. Schäfer, T. Wucherpfennig, [50] D. Zhang, N. G. Deen, J. A. M. Kuipers, Chem. Eng. Sci.
M. Berger, R. Takors, Eng. Life Sci. 2016, 16 (7), 633. DOI: 2006, 61 (23), 7593. DOI: https://doi.org/10.1016/j.ces.
https://doi.org/10.1002/elsc.201500135 2006.08.053
[28] H. Azargoshasb, S. M. Mousavi, O. Jamialahmadi, S. A. Sho- [51] G. Montante, K. C. Lee, A. Brucato, M. Yianneskis, Chem.
jaosadati, S. B. Mousavi, Can. J. Chem. Eng. 2016, 94 (1), 20. Eng. Sci. 2001, 56 (12), 3751.
[52] J. Aubin, D. F. Fletcher, C. Xuereb, Exp. Therm. Fluid Sci.
DOI: https://doi.org/10.1002/cjce.22352
[29] E. Askari, G. S.-P. Lemieux, P. Proulx, Can. J. Chem. Eng. 2004, 28 (5), 431. DOI: https://doi.org/10.1016/j.exptherm
2019, in press. DOI: https://doi.org/10.1002/cjce.23470 flusci.2003.04.001
[30] P. V. Danckwerts, Ind. Eng. Chem. 1951, 43 (6), 1460. DOI:
https://doi.org/10.1021/ie50498a055

Chem. Eng. Technol. 2019, 42, No. 8, 1545–1554 ª 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.cet-journal.com