Chemical Engineering Journal 317 (2017) 157–174

Contents lists available at ScienceDirect

Chemical Engineering Journal

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

CFD simulations of a bubble column with and without internals by using

OpenFOAM
V.H. Bhusare a, M.K. Dhiman c, D.V. Kalaga e, S. Roy d, J.B. Joshi a,b,⇑
a
Homi Bhabha National Institute, Anushaktinagar, Mumbai 400 094, India
b
Institute of Chemical Technology, Matunga, Mumbai 400 019, India
c
Reactor Engineering Division, Bhabha Atomic Research Center, Trombay, Mumbai 400 094, India
d
Department of Chemical Engineering, Indian Institute of Technology, Delhi 110 016, India
e
Indian Institute of Technology, Gandhinagar, Gujrat 382424, India

h i g h l i g h t s

Euler-Euler CFD model in OpenFOAM for bubble column with vertical internals.

Experimental validation with liquid velocity field with Radioactive Particle Tracking (RPT).

Good agreement with the experimental data for both bubble columns with and without internals.

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

Article history: Bubble column reactors are widely employed in various applications. Nowadays computational fluid
Received 19 August 2016 dynamics (CFD) provides the state-of-the-art capabilities of simulating the hydrodynamics in these reac-
Received in revised form 27 December 2016 tors. In the present work, bubble column is numerically simulated by using open source CFD tool,
Accepted 27 January 2017
OpenFOAM 2.3.1. Euler-Euler two-fluid model is used for the simulations. In the first part, OpenFOAM
Available online 3 February 2017
simulations are validated with experimental data from the literature over a column diameter range of
138–600 mm and the superficial gas velocity in the range of 19–169 mm/s. The velocity and holdup pat-
Keywords:
terns as well as circulation patterns in the bubble column are compared. In addition, FluentÒ simulations
Bubble column
Internals
are performed for the same conditions in order to compare the accuracy of OpenFOAM solvers. In the sec-
OpenFOAM ond part of the work, the validated OpenFOAM solver is applied to simulate the flow field in the 120 mm
TwoPhaseEulerFoam ID bubble column with and without internals. The internals consisted of (a) one vertical central rod of
Gas hold-up 36 mm diameter, (b) one central rod of (a) and four vertical additional rods of 12 mm diameter, (c)
one central rod of (a) and fourteen vertical additional rods of 12 mm diameter. The OpenFOAM simula-
tions are found to be in good agreement with the experimental data for both bubble columns with and
without internals. As regards to experiments, radial profiles of gas hold-up and the liquid velocity were
measured by using the radioactive particle technique (RPT). Such measurements were made at four axial
locations.
Ó 2017 Elsevier B.V. All rights reserved.

1. Introduction design of industrial scale bubble columns. However, such approach

The understanding of turbulent two-phase bubbly flows is
important due to the wide spread occurrence of this phenomenon
in natural and engineering systems. In the latter case, bubble col-
umns are widely used in chemical process industry. Experiments,
empirical correlations, one-dimensional convection-dispersion
models and compartment models usually form the basis for the
⇑ Corresponding author at: Homi Bhabha National Institute, Anushaktinagar,
Mumbai 400 094, India.
E-mail address: jbjoshi@gmail.com (J.B. Joshi).
remains somewhat restricted when one aims on the efficient and
reliable design and performance. In this perspective, three-
dimensional computational fluid dynamic (CFD) simulations have
proven to be useful. CFD helps to understand the complex two-
phase fluid dynamics in the bubble column through details of
mean flows (fields of three components of mean velocities and
mean gas hold-up), interphase rates of mass, energy and momen-
tum transfer and turbulence parameters (such as turbulent kinetic
energy, energy dissipation rate, Reynolds stresses, etc.). This tech-
nique has considerably grown in importance and interest during

http://dx.doi.org/10.1016/j.cej.2017.01.128
1385-8947/Ó 2017 Elsevier B.V. All rights reserved.
158 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174

Nomenclature

A anti-diffusive flux 2L mean liquid hold-up

C courant number j kappa
Ce1 , Ce2 , Ce3 Constants in epsilon equation [Eq. (13)] k parameter in Eq. (15)
CD drag coefficient l molecular viscosity, kg m1 s1
CL lift coefficient lt turbulent viscosity, kg m1 s1
Ct turbulence parameter ltm mixture turbulent viscosity, kg m1 s1
CVM virtual mass coefficient ltL liquid turbulent viscosity, kg m1 s1
CTD turbulent dispersion coefficient ltG gas turbulent viscosity, kg m1 s1
D column diameter, m m kinematic viscosity, m2 s1
dB bubble diameter, m mt eddy or turbulent diffusivity, m2 s1
FI interphase forces qG gas density, kg m3
FL lift force qL liquid density, kg m3
FrG Froude number for gas phase qm mixture density, kg m3
FX face flux or mass flux through the cell face rL surface tension of the liquid, N m1
f Flux rm turbulence parameter in km and em equation
g acceleration due to gravity, ms2 sij Reynolds stress, Pa
HD height of gas-liquid dispersion, m sP characteristic time for bubble generated turbulence in
I turbulence intensity Eq. (20)
k turbulent kinetic energy, m2 s2 / instantaneous properties such as ui ; uj ; um ; 2L ; 2G ; P
km mixture turbulent kinetic energy, m2 s2 r2 Laplacian operator
l turbulence length scale, m Dt time step, s
MoL Mortin number for liquid phase Dx single grid size, m
P pressure, Pa
QG net volumetric flow rate of gas, m3 s1 Subscripts
QL net volumetric flow rate of liquid, m3 s1 B bubble
R radius of column, m c continuous phase
r radial distance from the centerline, m d dispersed phase
Re Reynolds number fc cell face
ReG Reynolds number for gas phase G gas
Sk source term in the conservation equation for km, L liquid
m1 s2 m mixture
s surface area, m2
U0 fluctuating velocity component, ms1
Superscripts
uG gas velocity, ms1 0 fluctuating variable
uL liquid velocity, ms1 m mixture
ui ; uj three components of velocity, ms1 t turbulent
um three components of mixture velocity, ms1
V volume, m3
Abbreviations
VG superficial gas velocity, ms1
FCT flux corrected transport
x distance to first node from the wall, m
MULES multidimensional universal limiter with explicit solu-
tion
Greek letters PISO Pressure Implicit with Splitting of Operator
e turbulent energy dissipation rate, m2 s3 RNG re-normalization group
em mixture turbulent energy dissipation rate, m2 s3 SIMPLE Semi-Implicit Method for Pressure Linked Equations
2G fractional gas hold-up
VTK visualization toolkit
2L fractional liquid hold-up
2W fractional gas hold-up at wall
2G mean gas hold-up

the past two decades following exponential developments of com-
puting resources and abilities of numerical techniques. Neverthe-
less, the path of CFD modeling and simulations of bubble
columns still faces many challenges.
In this work, we have mainly focused on understanding the
behavior of two-phase bubbly flow in bubble columns with inter-
nals. The hydrodynamic studies related to the presence of internals
in bubble columns have not been sufficiently carried out and
addressed in the open literature. Larachi et al. [1] carried out useful
studies in internals-containing bubble columns by performing
two-fluid Euler continuum transient 3-D simulations by using FLU-
ENT. They simulated five pilot-scale configurations: vessels of uni-
form filling (dense and sparse), vessels of non-uniform filling with
large core and wall clearances, and equal cross-sectional hollow
vessels. Almost similar qualitative flow pattern (an upflowing
region was in the center and downflowing one nearby the wall
as that of for open bubble column) was obtained for the uniformly
arranged internals. On the contrary, non-uniform arrangements of
internals resulted into more complex flow patterns with even liq-
uid downward flow in the core region when the number of tubes in
the core region was larger than those near the wall. They further
found that the magnitude of axial velocities and turbulent kinetic
energy were reduced by the presence of internals. In addition,
some downward recirculation was revealed in the vicinity of tubes
for the sparse arrangement but not for the dense one, which was
not observed experimentally. The possible reasons given by
authors were (i) they indeed would not exist, or (ii) the spatial
scale of scrutiny of the probes was not sufficient for the downflow-
 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174
ing liquid layer to be seen, or (iii) the azimuthal averaging of the
velocity profiles, to which existing studies systematically resorted
to, smoothed out these effects.
Boutet et al. [2] carried out unsteady CFD simulations in a full
three-dimensional bubble column with internals in an Eulerian
way by using commercial code (fluent). Turbulence in the two-
phase flow was solved by using RNG k-e model and by including
bubble induced turbulence. Authors have observed no significant
effect on the liquid circulations due to the presence of internal
tubes. It was also observed to be similar to that of bubble column
without internals. As far as the gas holdup was concerned, the
presence of internal tubes leads to conditions of maximum gas
holdup in the vicinity of the internals. These high volume fractions
were the results of bubble trains (or gas channels) swirling prefer-
entially nearby the internal tubes. On the other hand, the turbu-
lence was seen to be reduced in the presence of internal tubes.
Eddies were constricted by the tube/tube and tube/outer wall dis-
tances eventually resulting into the reduced local eddy length scale
by the presence of internal tubes. It also resulted in significantly
higher turbulent energy dissipation rates. Authors have also stud-
ied the heat transfer in bubbly flows with the help of
hydrodynamic-thermal coupling and were found in good agree-
ment with the experimental results.
Guan et al. [3] simulated bubble column with internals by using
VOF method and presented the bubble dynamics in the column.
They also discussed a single bubble behavior including bubble tra-
jectories, bubble shape, bubble rise velocity, and bubble breakup
during the rise of bubbles through the compartment of internals.
The trajectories of the selected bubbles (in terms of sizes) were
found to be rectilinear and rocking and oscillations were observed
for small bubbles, which were associated with vortex shedding in
the bubble wake. As far as bubble shape is concerned, bubbles
were found to be axially elongated and bubble aspect ratio
increased with an increase in the number of internals in a particu-
lar part of the cross-sectional area. Simultaneously, bubble rise
velocity was reduced and it was related with the internals, which
covered the cross-sectional area and the pitch type. It was also
found that the internal solid walls increase the bubble break up
rate due to increase in shear and local turbulence. They also
included bubble induced turbulence.
Horvath et al. [4] performed VOF simulations in a bubble col-
umn by using OpenFOAM and compared the results with the com-
mercial solver fluent. Both the codes showed good prediction of an
average gas hold-up in the column. In addition, predictions for
bubble rise velocities by using OpenFOAM were also in good agree-
ment with the fluent solver.
Corzo et al. [5] presented a detailed description of the evapora-
tion and condensation model for subcooled flows by using Open-
FOAM. One-dimensional two-phase Euler-Euler model was
implemented in order to study the mass and heat transfer amount
phases during high heat transfer in water-steam systems. They
focused on the interfacial momentum exchange between phases
caused by drag efforts. They presented a detailed description of
the evaporation and condensation model for subcooled flows.
Numerical simulations showed a good agreement with the exper-
imental data. Michta et al. [6] implemented two-fluid adiabatic
bubbly flow model into the OpenFOAM solver and validated
against experimental data reported by Dedale. The model included
closure relationships for the interfacial momentum transfer for
bubbles moving in a turbulent continuous liquid phase. Bubble size
was predicted from the interfacial area concentration transport
equations, including the source and sink terms resulting from the
bubble coalescence and breakup. The proposed model was vali-
dated against measurements performed in a vertical upward flow
of air-water in a non-heated pipe and predictions were found to
159
be accurate except for the relative velocity between phases, which
was over-predicted.
Chen et al. [7] have studied gas hold-up profiles, liquid velocity
profiles and turbulent parameters in bubble column with air-water
and air-drakeoil system. Authors have found that both the axial liq-
uid velocity and gas hold-up are higher in air-water system than
that of in air-drakeoil system. In addition, the turbulence parame-
ters, such as turbulent stresses and eddy diffusivities were found to
be high in air-water system. They did not find much difference in
the liquid circulation velocities in the presence of internals as com-
pared to open column. On the other hand, the turbulence parame-
ters, such as turbulent stresses and eddy diffusivities were found to
be lower in the column with internals as they reduce the length
scales of the turbulence. Forret et al. [8] have studied liquid circu-
lation and mixing in the bubble columns with and without inter-
nals. Authors have observed a decrease in velocity fluctuations in
the column in the presence of internals. These reduced fluctuations
lead to increase in liquid recirculation intensity. Enhanced liquid
circulation scale significantly contributes to liquid mixing. The
authors have validated one dimensional axial dispersion model
with the experimental data and found good agreement between
them.
The foregoing discussion brings out the unexplained opportuni-
ties in the understanding of the hydrodynamics of bubble columns
with internals. Apart from the numerical studies, even experimen-
tal studies (Chen et al. [7], Forret et al. [8], Ahmed Youssef et al. [9])
carried out are also insufficient in this area. Therefore, carrying out
these studies in order to understand the effect of internals on the
hydrodynamics becomes much more essential task. Also, it is
observed that all the numerical studies carried out in the literature
are performed by using commercial tools like CFX, Fluent etc. One
of our main objectives of this work is the validation of an open-
source CFD tool OpenFOAM, which is not done until the date for
two-phase flow. Most of the cited work has been performed to
study the bubble rise phenomena by using VOF in OpenFOAM.
Though very few studies have been carried out in understanding
the source code (Rusche [10] and Jasak [11] for instance), nobody
has validated them for the wide range of column diameters, super-
ficial gas velocities and high gas phase fractions. No literature men-
tions about the detailed study of flow patterns in the bubble
column. Therefore, in this study we have (i) performed the simula-
tions by using OpenFOAM 2.3.1 for column diameters in the range
of 138 to 600 mm and the superficial gas velocity in the range of 19
to 169 mm/s. (ii) validated these simulations with the help of
experimental data reported in the published literature (iii) the val-
idated CFD model has been used for the simulation of flow patterns
in 120 mm ID bubble column with and without internals.
2. Experimental setup and measurement technique
2.1. Experimental setup
Kalaga [12] has given pertinent details of the experimental
setup and the measurement technique, namely, Radioactive Parti-
cle Tracking (RPT), using for measurement of the velocity field in
bubble column with internals. Details of the experimental column
setup has been briefly described below: Fig. 1 shows the schematic
diagram of the air-water bubble column, with RPT installation. An
acrylic column (1), having height 1.2 m and internal diameter of
120 mm was used as a cold-flow bubble column as shown in
Fig. 3. The internals were made of aluminum tubes having diame-
ter 12 mm and height 1.2 m. The internals (2) were comprised of
one central rod having outer diameter 36 mm and one concentric
bundle which is located at dimensionless radial locations of 0.43
160 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174

Fig. 1. Schematic diagram of experimental setup.

Fig. 2. Schematic of bubble column with internals.

as shown in Fig. 2. Two stainless steel spacers (5) were used to hold
the internals tightly at the top and bottom. Two different configu-
rations were used to study the effect of internals, that is, (i) bubble
column with one vertical central rod of 36 mm diameter as an
internal, and (ii) one central rod of (i) with four 12 mm diameter
rods. The top view of these configurations have been given in
Fig. 4. In all the experiments, the initial liquid height is maintained
at 655 mm. The experiment is carried out using air as the sparged
gas and tap water as the liquid phase, which was not contami-
nated. The experiments were performed at an ambient pressure
and temperature of 1 atm and 33 °C. The properties of water at
these conditions were found to be as follows: density (994.7 kg/
m3), viscosity (0.00075 Pa.s). Compressed air was passed through
air and moisture filters (2 and 3, respectively, in Fig. 1) to remove
 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174 161

between the detector and tracer particle, with the requisite correc-
tion of the solid angle subtended by the detector at the tracer par-
120 mm ticle location. A distance-count relationship was established using
a prior calibration of all detectors, which was performed by keep-
ing the particle at several known locations. Instantaneous position
of the tracer particle was calculated from the instantaneous dis-

1200 mm
tances obtained for all the detectors using a Monte Carlo simula-
655 mm tion model for c-ray emission, transmission and detection. The
time-differencing of the reconstructed instantaneous positions
yields instantaneous velocities, and the corresponding ensemble
average gives the mean velocities. Further, ‘‘turbulence” quantities
were evaluated from the instantaneous data of tracer particle fluc-
tuations (this is restricted to the lower frequencies of the turbu-
lence spectrum).
Fig. 3. Schematic of empty bubble column.

dust and oil droplets. The air to the column was sparged through
distributor and the flow rate was measured using a series of cali-
brated rotameters (5). The gas distributor was having 30 apertures,
the diameter of each aperture is 2 mm and pitch between two
apertures is 10 mm. The air volumetric flow rate is varied between
from 10 LPM to 90 LPM which corresponds to the range of air
superficial velocity from 14 mm/s to 132 mm/s for three percent-
ages of covered cross sectional area: 0% (no internals), 9% (config-
uration 1) and 11.7% (configuration 2).
2.2. Details of measurement technique
Kalaga [12] had used Radioactive Particle Tracking (RPT) tech-
nique for the experimental investigation (see, for details of the
technique, Roy et al. [13] and others). RPT uses the c-radiation
emitted by a single radioactive tracer particle to track its motion
in the multiphase reactor. The c-rays emitted by the tracer particle
were recorded by the array of strategically placed NaI Scintillation
detectors. The counts (intensity of radiation) recorded by the
detector is approximately inversely proportional to the distance
3. Governing equations
In the present work, two fluid Euler-Euler approach has been
employed. Each phase is treated as pseudo-continuum and inter-
penetrating each other. Further, each phase is represented by aver-
aged conservation equations. This averaging procedure introduces
the phase fraction into the equations, which gives the probability
of the presence of certain phase at certain point of time. The gov-
erning equations (incompressible system), in the absence of mass
transfer, for two-phase flow are:
Continuity equation:
@ @
ð2k qk Þ þ ð2k qk uk;i Þ ¼ 0 ð1Þ
@t @xi
Momentum equation:
@ @ @P
ð2k qk uk;i Þ þ uj ð2k qk uk;i Þ ¼ 2k þ 2k qk g
@t @xj @xi


@ @uk;i
þ 2k ðl þ lt Þ þ FI ð2Þ
@xj @xj
where subscript k denotes the phase (gas and liquid, i.e., k = L, G).
The hold-up for phase k is denoted by 2k . In Eq. (2), six terms
respectively represent the rate of change of momentum, convective
flux of momentum, pressure gradient, body force due to gravity,
total stress within the phase (it is addition of laminar stress and tur-
bulent stress) and interfacial momentum transfer term. In this
work, the constant drag coefficient (CD) was used in a drag model
by assuming bubbles with constant diameter. The bubble size in
the model (at inlet) was estimated from the slip velocity and the
drag correlation used. The drift flux plot is generated from the
experimental data of VG =G vs VG . The drift flux coefficient C1 is
the measure of slip velocity. The slip velocity arising out of this is
used for the calculation of bubble size in the model. The bubble
sizes estimated at various superficial gas velocity are as shown in
Table 1. According to the solution procedure employed by Joshi
[14], the CD should be adjusted in such a way that the value of drift
flux parameter C1 from CFD will be equal to that of the experimen-
tal value. In the present work, Table 5 shows the CD values used in
this simulations.
The lift coefficient plays an important role in the bubble column
fluid mechanics. A bubble moving in a liquid experiences a lift

Table 1
Estimated bubble diameters.

VG m/s dB mm
0.014 2.2
0.044 2.35
0.088 2.49
Fig. 4. Top view of the internal configurations: (A) Empty column (Configuration- 0.132 2.6
A), (B) Configuration-B, (C) Configuration-C.
162 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174

Table 2
Details of numerical simulations performed.

Set. No. Sr. No Air volumetric flow rate (LPM) Air superficial gas velocity based on empty column (m/s) Percent area covered by the internals
1 1 10 0.014 0 (no internals;
2 30 0.044 Configuration-A)
3 60 0.088
4 90 0.132
2 1 10 0.014 9 (Configuration-B)
2 30 0.044
3 60 0.088
4 90 0.132
3 1 10 0.014 11.7 (Configuration-C)
2 30 0.044
3 60 0.088
4 90 0.132

Table 3
Boundary conditions used in OpenFOAM.

Boundary field Inlet Outlet Wall

Gas phase fraction Groovy BC inletOutlet fixedValue
Gas velocity fixedValue pressureInletOutletVelocity fixedValue
Liquid velocity fixedValue fixedValue fixedValue
Pressure fixedFluxPressure fixedValue fixedFluxPressure

Table 4
Schemes used in OpenFOAM.

Modeling term Keywords for Scheme Description Scheme

Convection term divSchemes Discretization of divergence terms with an operatorr upwind
Gradient term gradSchemes Discretization of the gradient terms with an operator r default Gauss linear
Diffusive term laplacianSchemes Discretization of terms with the laplacian operator r2 default Gauss linearcorrected
Time derivative ddtShemes Discretization of first and second order time derivatives Euler
Others interpolationSchemes Point to point interpolation of the value default linear
snGradSchemes Component of gradient normal to a cell face default corrected

force due to vorticity upstream to it. As the lift force is dependent
on the cross product of slip velocity and curl of liquid velocity, it is
perpendicular to both of them. Hence, it is sometimes also referred
to as transverse lift force. The different values of CL aroused due to
the various factors resulting into a transverse force on a bubble,
like turbulent shear, pressure gradient, bubble deformation, wake
phenomena, etc. Magnus force arises due to rotation of bubble
about its axis. Most significant contribution to the vorticity comes
from the radial gradient of the axial liquid velocity ð@uL =@rÞ. There-
fore, the lift force is most significant in the radial direction in the
bubble column. This contributes to the radial component of the lift
force for bubble columns as follows:
p  where CTD is turbulent dispersion coefficient and mTB is turbulent
@uL
dB qL ðuG  uL Þ
3
FL ¼ CL
6 @r
In a typical bubble column operating in the heterogeneous
regime, the liquid circulation is present with up-flow in the central
region of the column and down-flow near the walls. The radial gra-
dient of axial liquid velocity ð@uL =@rÞ is negative. From above equa-
tion is it clear that the radial component of lift force is positive.
This pushes the bubbles away from the center if the lift coefficient
is positive. This shows that the gas hold-up is higher near the wall,
which is violation of the experimental behavior. Hence, to get accu-
rate values of average gas hold-up in the column, lift coefficient
must be chosen negative. Thus, the value of CL was chosen to get
the predicted average gas hold-up equal to an experimental value
in the column (Joshi [14]). The results are given in Table 5. The val-
ues of CL can be seen to be in the range of 0.08 to 0.23.
Bubbles in the turbulent flow field move along a path which
may be fluctuating. The fluctuations in the bubble path can dis-
perse the bubbles. The force responsible for the dispersal of bub-
bles is termed as turbulent dispersion force. This force is taken to
be perpendicular to the gradient of dispersed phase hold-up. This
force is proportional to gradient of dispersed phase hold-up and
the proportionality constant (turbulent dispersion coefficient).
One such equation given by,
p 1 mTB r2G
FTD ¼ CTD CD
2
dB qL ðuG  uL Þ
4 2 2G
kinematic viscosity of the gas phase. Drift flux parameter C 0 is a
measure of flatness of the hold-up profile. On the other hand, turbu-
lent dispersion force is responsible for bubbles to move to the wall
and hence more uniform gas hold-up. Thus, C TD is adjusted in such a
way that the value of C 0 from CFD equals to that of the experimental
value (Joshi [14]). The results are given in Table 5. The value of tur-
bulent dispersion coefficient (CTD ) can be seen to be in the range of
0.008 to 0.07. The effect of added mass on the fluid mechanics in the
bubble column is very negligible as compare to drag and lift forces.
In the present work, this fact has been verified by performing sim-
ulations with and without added mass force. The value of added
mass coefficient was chosen to between 0 to 0.5. It has been
observed that the effect of added mass was negligible. Hence, for
the present work the added mass force was not considered in the
simulations.
 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174 163

Table 5
Numerical details and the results of CFD simulations.

Configuration VG (m/s) CL CD CTD G C1 C0

4
A 0.014 0.1 3.33  10 0.05 0.0454 (0.046) 0.253 (0.245) 3.53 (3.64)
0.044 0.2 2.91  104 0.008 0.106 (0.102) 0.221 (0.245) 3.58 (3.64)
0.088 0.15 2.64  104 0.005 0.153 (0.157) 0.227 (0.245) 3.43 (3.64)
0.132 0.18 2.56  104 0.005 0.187 (0.189) 0.235 (0.245) 3.51 (3.64)
B 0.014 0.1 3.12  104 0.07 0.042 (0.046) 0.243 (0.24) 3.59 (3.63)
0.044 0.18 3.01  104 0.005 0.105 (0.107) 0.237 (0.24) 3.45 (3.63)
0.088 0.15 2.89  104 0.005 0.155 (0.159) 0.233 (0.24) 3.43 (3.63)
0.132 0.18 2.71  104 0.005 0.182 (0.189) 0.239 (0.24) 3.47 (3.63)
C 0.014 0.12 3.5  104 0.01 0.0495 (0.050) 0.221 (0.21) 3.49 (3.6)
0.044 0.23 3.31  104 0.01 0.119 (0.125) 0.217 (0.21) 3.50 (3.6)
0.088 0.2 3.12  104 0.007 0.167 (0.171) 0.212 (0.21) 3.58 (3.6)
0.132 0.08 3.01  104 0.008 0.187 (0.193) 0.201 (0.21) 3.57 (3.6)

The turbulence in the two-phase is estimated by using mixture value of km is used to find out the turbulent kinetic energy for
k-e equation given by Behzadi et al. [15], because k-e model can liquid (kL) by solving Eq. (6), while Eq. (7) is used to calculate the
offer reasonable accuracy at low computational cost. This model value of turbulent energy dissipation rate for liquid (eL ). By using
is based on the fact that both gas and liquid phases tend to fluctu- the individual values of kL and eL , the turbulent viscosity of the
ate as one entity at high phase fractions and therefore using one set liquid (ltL ) is calculated. Similar procedure has been followed in
of equations for the mixture. The mixture k-e equation is based on order to calculate the turbulent viscosity of the gas (ltG ).
the summation of the two corresponding phase-averaged (mass- Thus, the values of ltL and ltG , and hence the value of ltm , were
weighted averaging) transport equations for k and e of the two calculated and updated at each iteration. Thus above set of
phases. The set of equations (Behzadi et al. [15]) used in k-e mix- equations, i.e., Eqs. (1) to (11) is solved using open-source CFD tool
ture model is summarized as follows: OpenFOAM 2.3.1.
km -equation
@ lt 4. Numerical details
k  qm em þ Sk
ðq km Þ þ r  ðqm um km Þ ¼ r  m rkm þ Pm m
ð3Þ
@t m rm
4.1. Numerical setup
The source term Sm
k in the above equation arises from the action
of interphase forces and hence, it is contributed to interphase The column is made of acrylic having height 1.2 m and internal
energy exchange. diameter of 120 mm as shown in Fig. 3. In the present work, the
em -equation internals comprised of one central rod having outer diameter of
@ lt em 36 mm and one concentric tube bundle, which is located at dimen-
ðqm em Þ þ r  ðqm um em Þ ¼ r  m rem þ k  Ce2 qm em Þ
ðCe1 Pm sionless radial location of 0.43. Each tube had an outer diameter of
@t rm km
em 12 mm. Two different configurations were used to study the effect
þ Ce3 Sm
k ð4Þ of internals, the top view of these configurations are shown in
km
Fig. 4. In all the cases, the initial liquid height was maintained at
The mixture properties appearing in Eqs. (3) and (4) are related 655 mm as shown in Fig. 3. The simulations were performed by
to those of the continuous (liquid) and dispersed (gas) phases are using air as the sparged gas and water as the liquid phase. The
as follows: air volumetric flow rate was varied from 10 LPM to 90 LPM which
qm ¼ 2L qL þ 2G qG ð5Þ corresponds to the range of superficial air velocity of 15 mm/s to
132 mm/s for four percentages of covered cross sectional area:


q q 0% (no internals), 9% (configuration 1) and 11.7% (configuration
km ¼ 2L L þ 2G G C2t kL ð6Þ 2) as shown in Fig. 4A, B and C, respectively.
qm qm

 4.2. Boundary conditions
qL  qG 2
em ¼ 2L þ 2G C e ð7Þ
qm qm t L
4.2.1. Inlet boundary conditions

2L qL uL þ 2G qG uG C2t

um ¼ ð8Þ a. At the inlet, the gas hold-up is modelled through non-
2L qL þ 2G qG C2t uniform inlet hold-up. The need for adapting this kind of
boundary condition arises due to the elimination of actual
ð2L ltL þ 2G ltG C2t Þqm sparger design at the inlet. Using non-uniform boundary
ltm ¼ ð9Þ
2L qL þ 2G qG C2t condition instead of uniform boundary condition at the inlet
has advantages, such as: (i) Non-uniform inlet hold-up and
Pm  L  G gas velocity are actual representation of the sparger condi-
k ¼ 2L Pk þ 2G Pk ð10Þ
tions, which is not the case in uniform inlet holdup and
velocity. Hence, using non-uniform inlet boundary condi-
k ¼ Sk þ Sk ¼ FI ui ¼ AG ð22G ðCt  1Þ kL þ gL r2G  ur Þ
Sm L G 0 2
 ð11Þ
tions gives us relaxation over making complex sparger
The mixture k-e model is used in order to solve the turbulence. geometries. (ii) Using simple geometries reduces the overall
The value of mixture kinetic energy (km) and turbulent energy dis- number of grids used for the simulation and hence, compu-
sipation rate for mixture (em ) were obtained by solving conserva- tational time. The expressions for the inlet gas hold-up is
tion equations of k and e [Eqs. (3) and (4), respectively]. The given by using groovy boundary conditions (one of the
164 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174

utilities in the OpenFOAM). In order to define the gas hold- computed from difference between total and static pressure
up boundary condition at the inlet the correlation proposed where the direction is normal to the patch faces. This bound-
by Vitankar and Joshi [16] is used, which is given as follows: ary condition can be described as at velocity inletOutlet
boundary condition patches for where the pressure is spec-
mþ2
G ¼ ðG  W Þð1  gm Þ þ W ð12Þ ified and zero-gradient is applied for outflow (as defined by
m the flux) and for inflow the velocity is obtained from the
patch-face normal component of the internal-cell value.
where g = dimensionless radial distance
c. At the outlet, fixedValue pressure boundary condition was
m = the steepness parameter (Luo and Svendsen [17])
used and the value is set to the atmospheric pressure.
r d. The outlet the boundary condition for k given by using Eq.
g¼ (13) at mean flow velocity at outlet.
R
e. At the outlet, the values of e are calculated by using Eq. (14).
m ¼ 2:188  103 Re0:598
G Fr0:146
G Mo0:004
L
4.2.3. Wall boundary conditions
where
At the wall, fixedValue boundary condition was used for the
ReG ¼ DVG ðqL  qG Þ=lL fractional gas hold-up. The wall hold-up at the wall was taken as
zero. For both, the inlet gas velocity and liquid velocity, fixedValue
FrG ¼ V2G =gD boundary condition was used with the value of zero (correspond-
ing to the initial phase fraction of the water at inlet). For pressure
MoL ¼ gl4L =ðqL  qG Þr3L

START
b. The boundary condition for inlet gas velocity is given as a fix-
edValue. The radial and theta components of gas velocities
are given as zero. The axial component is given as a slip
velocity of gas at inlet based on the superficial gas velocity
and the averaged gas hold-up at the corresponding gas flow Mass conservaon loop
rates. Eq. (1) is solved by using MULES
c. Since it is a batch type of bubble column, the inlet liquid method for both the phases
velocity boundary condition is given as fixed value of zero.
d. The inlet boundary condition for pressure is fixedFluxPres-
sure. This boundary condition adjusts the pressure gradient
such that the flux on the boundary is that specified by the
Calculates and updates interphase
velocity boundary condition.
force coefficients needed to solve
e. The inlet boundary condition for turbulent kinetic energy tn=tn-1+Δt
momentum equaon (2)
(km) for air was given as the fixedValue, with the uniform
value equal to that of calculated from following equation:

PRESSURE AND VELOCITY FIELD, VOL. FRACTION

3
k¼ ðui IÞ2 ð13Þ Soluon for predicted velocity field
2
by using known values of pressure
where ui is the mean flow velocity of gas at the inlet and I is
the turbulent intensity.
CALCULATED PRESSURE FIELD

f. In case of turbulent energy dissipation rate (e), at the inlet

the values of eare calculated by using following formula: Pressure equaon loop
Solve pressure equaon based on
3=2
C3=4
l k predicted velocity field
e¼ ð14Þ
l
where Cl (=0.09) is the turbulent parameter that relates tur-
bulent viscosity with k and e, l is the turbulent length scale, Update gas and liquid velocity fields
which is equal to the column diameter for cylindrical from calculated values of pressure
columns.

4.2.2. Outlet boundary conditions

NO
a. At the outlet, inletOutlet boundary condition is used for gas CONVERGENCE?
holdup. The inletOutlet boundary condition is normally the
same as zeroGradient, but it switches to fixedValue if the
YES
velocity vector next to the boundary aims inside the domain
(backward flow). The value of that fixed value is inlet value,
which is 1 in this case. Turbulence calculaon
b. For outlet gas velocity, pressureInletOutletVelocity boundary Solves Eq. (12) and (13) for k and
condition is used. The pressureInletOutletVelocity boundary epsilon, respecvely
condition is a blend of pressureInletVelocity and inletOutlet
boundary conditions. The pressureInletVelocity, velocity is Fig. 5. PIMPLE algorithm to solve twoPhaseEulerFoam solver.
 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174
Fig. 6. Grid sensitivity analysis.
fixedFluxPressure boundary condition is used. At the wall, the
boundary condition for km is given as fixed value of zero, which
is in correspondence with the phase velocity at the wall. Near
Fig. 7. Grid size used for simulations (A) Configuration A; (B) Configuration B.
165
the wall function of type kqRWallFunction is used. This works
exactly in the same way as that of the zero gradient. Similarly,
for e also the fixed value of zero is used as a boundary condition
at the wall. Near the wall, the epsilonWallFunction is used, where
the value of the e is calculated by using following formula:
3=2
C3=4
l k
e¼ ð15Þ
jx
where j is the kappa having value of 0.41 and x is the distance to
first node from the wall.
Similar boundary conditions have been employed for liquid
phase turbulent kinetic energy also.
4.2.4. Internal field
The initial fields into the internal domain, corresponding to
initial liquid height, are set by using funkySetFields utility in
OpenFOAM. It reads fields from the file funkySetFieldDict and after
re-calculating, writes them back to the boundary file. The initial
phase fraction for gas phase is set to the values of the average
gas and liquid hold-up in a column up to initial liquid height.
Above that level, the liquid phase fraction is taken to be zero
and hence, gas phase fraction is 1. In a similar way, actual veloc-
ity based on fractional gas hold-up is set as a gas phase velocity
in the region of stable liquid height. The region where gas phase
Fig. 8. CFD validations for Hills (1974) experimental data for axial liquid velocity by
using OpenFOAM and Fluent at (A) 38 mm/s, (B) 64 mm/s and (C) 95 mm/s.
166 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174

work, which indicates a first order bounded implicit scheme, which

is sufficiently accurate due to the small time steps created by the
Courant number restriction. OpenFOAM also provides other high
resolution choices of ddtShemes such as CrankNicolson and back-
ward, which are second order accurate but are unbounded. The
unbounded behavior of these schemes may end up in non-
feasible solution of the problem. Though the backward differencing
in time is second-order accurate in time and still neglects the
temporal variation of the face values. In order to achieve this, the
backward differencing in time uses three time levels, still the
boundedness of the solution is not guaranteed. Also, in the case
of the convection term, boundedness can be obtained if the equa-
tion is discretized to first order temporal accuracy. Hence, Euler is
found to be the best option in this case.

Fig. 9. CFD validations for Hills (1974) experimental data for gas hold-up by using
OpenFOAM and Fluent at (A) 38 mm/s, (B) 64 mm/s and (C) 95 mm/s.

fraction is 1, the superficial velocity is set as a gas velocity for the

internal field. Pressure in the internal field of the column is set as
a hydrostatic pressure. All the above boundary conditions are
summarized in Table 3.

4.3. Discretization schemes

Temporal, divergence, gradient and laplacian terms in Eqs. (1)

and (2) are discretized separately in OpenFOAM by using proper
discretization schemes. The details of the schemes used are sum-
marized in Table 4 and are discussed in preceding section.

4.3.1. Temporal discretization schemes

The term ddtShemes represents the choice of time scheme. In
case of steady state simulations, choice of steadyState scheme is
considered to be appropriate. In this work, we are performing tran-
sient simulations, hence choice of transient schemes, such as, Euler,
CrankNicolson or backward is preferred. Euler is adopted in this
Fig. 10. CFD validations for Menzel (1990) experimental data for axial liquid
velocity by using OpenFOAM at (A) 12 mm/s, (B) 24 mm/s, (C) 48 mm/s and (D)
96 mm/s.
 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174 167

4.3.2. Gradient discretization schemes
The gradSchemes defines the discretization schemes for the gra-
dient terms. In this case, default Gauss linear scheme is used as a
gradSchemes. Default means that the Gauss linear scheme will be
applied to all the gradient terms. The Gauss keyword specifies
the standard finite volume discretisation of Gaussian integration.
When a Gauss discretization is used the values are interpolated
from cell centers to face centers. That is, by using Gauss Divergence
theorem volume integral is converted to surface integral as shown
in Eq. (16). These face values are then calculated by using cell cen-
ter values with linear interpolation.
Z I
X derivation and therefore no time information. In order to adopt this
r/ dV ¼ ds / ¼ sfc /fc
V @V fc
seEulerFoam is a transient solver for solving two-phase gas-liquid
flow, which uses PIMPLE algorithm. PIMPLE algorithm is a
combination of PISO and SIMPLE algorithms. The structure of
PIMPLE is based on the original PISO, but allows equation
under-relaxation to ensure the convergence of equations, like in
SIMPLE. That is, if the PISO algorithm is employed then one gets
strictly limited to the time step (where Courant number
 P ux 
C ¼ Dt ni¼1 Dxii < 1). That means that it is very expensive to solve
a real time problem in transient manner for long time especially in
very complex geometries. SIMPLE is developed for reaching the
steady-state condition relatively fast but does not contain time
ð16Þ

where subscript ‘‘fc” indicates cell face.

4.3.3. Divergence discretization schemes

The divSchemes defines the scheme used to solve convective
term in the momentum equation. The convection term in Eq. (2)
is gauss integrated as shown in Eq. (17).
Z I X
r  ðq/UÞ dV ¼ ds  ðq/UÞ ¼ sfc  ðqUÞfc /fc
V @V fc
X
¼ FX /fc ð17Þ
fc

where face flux (FX ) is the mass flux through the cell face and is cal-
culated from interpolated values of qand U. In order to solve Eq.
(17), we also require the face value of the variable calculated from
the values in the cell centers, which is obtained using the convec-
tion differencing scheme.
In this case, first order upwind is used as a divergence
scheme. Though the upwind scheme used here is only first order
accurate, it is considered to be more appropriate due to its
boundedness characteristics. While other schemes, such as,
central differencing are unbounded even though they are second
order accurate.

4.3.4. Laplacian discretization schemes

The laplacianSchemes defines the laplacian scheme, and it is
applied to the terms with the laplacian operator r2 , i.e., for
diffusional terms. In the present work, the operator is set to
default Gauss linear corrected. Here the word corrected
expresses the surface normal gradient scheme and indicates
an unbounded, conservative and second order numerical
behavior.

4.3.5. Interpolation schemes

A centered interpolation scheme is used under interpola-
tionSchemes, which is set to default linear. This indicates that the
linear interpolation scheme is used for calculating all (indicated
by default as discussed in gradSchemes) the gradient values at the
face. A snGradSchemes indicates the surface normal gradient
schemes, and evaluates the gradient normal to the face center
shared by two cells. A surface normal gradient is evaluated at a cell
face; it is the component, normal to the face, of the gradient of val-
ues at the centers of the two cells that the face connects. In this
case it is default corrected.

4.4. Solution procedure and post-processing

The simulations were performed by using twoPhaseEulerFoam

solver present in OpenFOAM 2.3.1. TwoPhaseEulerFoam is a solver
for two compressible fluids with one phase dispersed (eg: gas bub- Fig. 11. CFD validations for Menzel (1990) experimental data for gas hold-up by
bles dispersed in liquid phase) including heat transfer. TwoPha- using OpenFOAM at (A) 12 mm/s, (B) 24 mm/s, (C) 48 mm/s and (D) 96 mm/s.
168 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174
nature of SIMPLE algorithm in transient problems, relaxation factor
is introduced that limits the values of the new solution. The
relaxation factor is used in the pimple loop (the value given by
nOuterCorrectors = n) except the last one and again set to the value
assigned at the start of the simulation to ensure time consistence.
That is, the pressure-momentum coupling is calculated n times, in
which it is calculated (n-1) times with relaxation (like SIMPLE) and
the last time without relaxation. For cases that are very unstable
PIMPLE has proved to get a smooth convergence. In the OpenFOAM
code, PIMPLE algorithm is provided with two other parameters
called nCorrectors and nOuterCorrectors. The first one indicates
that the number of internal iterations to be performed for
calculating the pressure equation (internal loop for pressure),
while the later one tells us about the number of internal iterations
to be performed for calculating the pressure-momentum coupling
in single time step. The values of these parameters indicates the
number of internal iterations for respective field calculations in
single time step.
In the present work, we have gas-liquid system without
heat transfer, so we have modified the current twoPhaseEuler-
Foam solver to mytwoPhaseEulerFoam solver. The stepwise
details of PIMPLE algorithm used in mytwoPhaseEulerFoam are
shown in Fig. 5. It starts by solving the mass conservation loop,
where it solves the continuity equation for both gas and liquid
phases [Eq. (1)]. Then the code continues the solving for phase
volume fraction by using MULES (Multidimensional Universal
Limiter with Explicit Solution) method. MULES is an iterative
technique with the implementation of Flux Corrected Transport
(FCT) and is considered reliable for the boundedness of scalar
field. FCT (Zalesak [18]) constructs the net transportive flux
point by point (non-linearly) as a weighted average of a flux
computed by a low order scheme and a flux computed by a
high order scheme. The weighting is done in a manner which
insures that the high order flux is used to the greatest extent
possible without introducing ripples (overshoots and under-
shoots). The result is a family of transport algorithms capable
of resolving moving contact discontinuities over 3–4 grid
points, and shock fronts over 2 grid points, without undershoot
or overshoot. The steps in the FCT algorithm are briefly
described as follows:
Fig. 12. Comparison of axial liquid velocity profiles for Configuration (A).
L
1. Flux f in Eq. (1) is computed by using lower order scheme.
H
2. Flux f is computed by using higher order scheme.
3. Anti-diffusive flux A is defined by following equation and it is
equal and opposite of the fluxes due to artificial viscosity.
H L
A¼f f ð18Þ
4. The corrected flux is calculated by using these anti-diffusive
fluxes and Zalesak [18] weighting factor.
C L
f ¼ f þ kA ð19Þ
Calculation of the flux is dependent of the value of k. In order
to measure the effect of lower or higher order schemes in flux
calculation, value of kplays an important role. If k ¼ 0, then
the flux used for time integration has low order, on the contrary
with k ¼ 1, then the flux used has a high order. In OpenFOAM,
the value of k is calculated iteratively.
5. Solve the equation (computation of the value of phase fraction)
for given temporal scheme by using corrected flux.
Dt  C C

/nþ1
i ¼ /ni  f iþ1=2  f i1=2 ð20Þ
V
After solving for MULES algorithm, the momentum equation
[Eq. (2)] is solved for predicting, the velocities by using the pres-
sure field from previous iteration. Prior to that, all the interphase
forces are calculated and updated continuously, which is
required for solving momentum equation. These velocities are
then used to solve the equations for getting corrected pressure.
By using newly calculated pressure field, the velocities are
updated again by solving momentum equation until the conver-
gence is reached. After convergence is reached, finally, it solves
for turbulence and again same procedure is repeated for next
time steps.
The main post-processing tool provided with OpenFOAM is Par-
aview, an open-source visualization application. Paraview uses the
Visualization Toolkit (VTK) as its data processing and rendering
engine and can therefore read any data in VTK format. OpenFOAM
includes the foamToVTK utility to convert data from its native for-
mat to VTK format.
 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174
5. Results and discussion
5.1. Grid sensitivity analysis
Selection of the modeling closure, with the grid resolution, for
two-phase dispersed flow is very much important in order to
describe the underlying physics consistently. Depending upon
the selected grid, physical models of various complexities must
be employed. For instance, in a dispersed flow if the mesh is in
the micro-scale order, we can use single-fluid, interface tracking
technique to solve the problem. If the grid size is large
enough for statistical description of bubbles, the Euler-Euler
approach is used.
In this work, we are using two-fluid Euler-Euler approach,
which allows us to use coarse grid in the core of the flow. Near
the wall, it is important to capture wall effects and also to capture
minimum liquid velocity point near wall, which is important for
accurate knowledge of mass balance in the column. In order to cap-
ture these effects, the grid size used near the wall is very fine. In
this regard the grid sensitivity is studied for the bubble column
of Hills [19]. In this work, grid sizes of 200 K, 240 K, 280 K, and
320 K are used. In the corresponding mesh core region is set up
to radial distance of 0.06 m and wall effects are measured in
0.009 m section between wall and core region. In all of these
meshes, grid size in the core region was kept constant and the grid
size near the wall was varied. Fig. 6 shows the comparison of liquid
velocity profile for different mesh sizes. In Fig. 1, lines 1, 2, 3 and 4
represent the predictions of the axial liquid velocity by using grid
size of 200 K, 240 K, 280 K, and 320 K, respectively. The liquid
velocity predictions from lines 2 and 3 shows that the velocities
are under predicted and hence, there was discrepancy in the gas
phase and liquid phase mass balance (discrepancy up to 22% of
the central upward volumetric flow rate). While lines 1 and 4 rep-
resent very good agreement with the experimental data. In case of
line 1, the total mass balance was not satisfied (discrepancy up to
15% of the central upward volumetric flow rate), whereas the dis-
crepancy in gas and liquid phase mass balances was less than 5%
then the central upward volumetric flow rate in case of line 4.
Fig. 13. Comparison of gas hold-up profiles for Configuration (A).
169
Hence, the final grid size used for simulation was 320 K. Fig. 7
shows the final grid sizes used for simulations.
5.2. Code validation
As a first step, the validity of the OpenFOAM code was checked
by simulating the bubble column used by Hills [19] and Menzel
et al. [20] (both are semi-batch bubble columns, in which the liquid
is in batch condition while gas is injected continuously at the
bottom of the column) for which extensive experimental data on
phase velocities and phase fractions have been reported. In case
of Hills [19], the column is having height of 1.38 m and inner diam-
eter of 138 mm. Simulations were performed with an air injection,
at the superficial velocities of 38, 64 and 95 mm/s, in the column
filled with the water (initial water height of 0.9 m). Comparisons
for axial liquid velocities and the gas hold-up profiles between
the experimental data and the OpenFOAM predictions shown in
Figs. 8 and 9 show very good agreement (standard deviation of less
than 2% and 3%, respectively for axial liquid velocity and gas hold-
up). From Fig. 8, it is observed that the centerline velocity of the
liquid phase increases with an increase in the superficial velocity
of the gas. Fig. 9 shows the comparison for gas hold-up at different
superficial gas velocities. It is observed that there is maximum gas
hold-up in the center of the column, while gas hold-up tends to
practically zero near the wall. The centerline hold-up increases
with an increase in the superficial gas velocity.
In this work, simulations for the same column are also per-
formed by using commercial CFD tool Fluent. All the governing
equations as well as turbulence models used in both OpenFOAM
and Fluent were identical. It is very much essential to keep all
the modeling equations identical in order to compare the reliability
of the two codes. The results for the fluent simulations are also
shown in Figs. 8 and 9. The OpenFOAM simulations are compared
with Fluent and an excellent agreement can be seen between the
two predictions. The standard deviation for axial liquid velocity
and gas hold-up were found to be less than 2% in both the cases.
In both the cases, mass balances for gas and liquid phases were cal-
culated by using Eqs. (21) and (22), respectively and found to be
170 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174

Fig. 14. Comparison of axial liquid velocity profiles for Configuration (B).

very satisfactory. Thus, it is concluded that the OpenFOAM tool is 38 mm/s, 64 mm/s and 95 mm/s for which simulations were per-
able to predict the flow patterns correctly for the gas-liquid two- formed. The value of QL was found to be practical zero (less than
phase flows and hence can be used extensively for the simulation 5% then the central upward volumetric flow rate).
purpose. After getting CFD predictions, it was thought desirable In case of Menzel et al. [20], the column is having height of
to establish gas and liquid phase balance. At any cross-section, 3.45 m and inner diameter of 600 mm. The simulations were
the following equations hold: performed by using OpenFOAM for the superficial gas velocities
of 12, 24, 48 and 96 mm/s. Fig. 10 shows comparison for axial
Z R
liquid velocity profiles for Menzel et al. [20]. The CFD simulations
QG ¼ uG 2G 2prdr ð21Þ
0
were performed by using OpenFOAM and predictions were found
in a very good agreement (standard deviation of less than 3% and
Z R
4%, respectively for axial liquid velocity and gas hold-up) with
QL ¼ uL 2L 2prdr ð22Þ the experimental data. Centerline velocity increases with an
0 increase in superficial air velocity. The velocity reversal point
As per Eq. (21), the predicted values of QG were found to be con- was found between the dimensionless radial distances of
sistent with corresponding values of superficial gas velocity of 0.7–0.75 for all the cases. The values of QG, calculated by using

Fig. 15. Comparison of gas hold-up profiles for Configuration (B).

 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174 171

Fig. 16. Comparison of axial liquid velocity profiles for Configuration (C).

Eq. (21), were in consistent (standard deviation less than 6%) with
the corresponding values of gas flow rates used for simulations. In
addition, the values of QL were also calculated by using Eq. (22) and
were found to be close to zero. Fig. 11 shows the comparison
between the CFD predictions and the experimental data for the
gas hold-up and found good agreement with less than 7% standard
deviation.
5.3. CFD simulations by using OpenFOAM
In this section, CFD simulations are performed by using Open-
FOAM 2.3.1 [21] for the configurations shown in Fig. 4. Results
for the same are discussed in the forthcoming section.
5.3.1. Configuration (A)
Configuration (A) is the open column or empty bubble column
as shown in Fig. 4A with no area covered by internals. The CFD sim-
ulations are performed for varying air flowrates as shown in
Table 2. The results for CFD simulations are compared with the
experimental data obtained from Radioactive Particle Tracking
(RPT) and are shown in Figs. 12 and 13. The azimuthally averaged
axial liquid velocities (Fig. 12) and radial gas hold-up (Fig. 13) pro-
files are plotted against dimensionless radial distance. It is
observed that the OpenFOAM shows good agreement (standard
deviation of less than 3% and 2%, respectively for axial liquid veloc-
ity and gas hold-up) with the experimental data. Magnitude of
both the axial liquid velocity and the gas hold-up are maximum

Fig. 17. Comparison of gas hold-up profiles for Configuration (C).

172 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174

Fig. 18. Comparison of turbulent kinetic energy for Configuration (A).

Fig. 19. Comparison of turbulent kinetic energy for Configuration (B).

in the center of the column. Both centerline axial liquid velocity
and gas hold-up in the column increase with increase in superficial
velocity of the gas. The radial locations of flow reversal are also
adequately (with 5%) predicted. Most importantly, the overall
material balance as given by Eqs. (21) and (22) was satisfied with
7% in all the cases.
The technique of RPT has become a useful tool for the measure-
ment of velocity profiles. The gas voidage profiles can also be esti-
mated through an indirect interpretation involving the tracer
particle occurrence density distribution. The technique provides
very good accuracy in the bulk region; however, has relatively less
accuracy in the wall region. This is the reason for some disparity in
the model at higher superficial gas velocities as shown in Fig. 12.
5.3.2. Configuration (B)
Configuration (B) is the open column as shown in Fig. 4B with
one central rod of 36 mm outer diameter. The percent area covered
in this case by the presence of central internal is 9%. The CFD sim-
ulations were performed for varying air flowrates as shown in
Table 2. The axial liquid velocity and gas hold-up profiles in the
annular space between the column wall and internal tube are plot-
ted in Figs. 14 and 15, respectively. The results were found to be in
 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174
Fig. 20. Comparison of turbulent kinetic energy for Configuration (C).
very good agreement with the standard deviation of 4% for both
axial liquid velocity and gas hold-up. From the profiles of axial liq-
uid velocity, it is observed that the velocity is maximum near the
wall of internal tube. The liquid is flowing in the upward direction
in the central region (in the vicinity of internal central tube), while
there is a down flow of liquid near the column wall. Thus, the liq-
uid circulation, as observed in the case of configuration A, still per-
sists even in the presence of central rod. Thus, adding 9% of
internals (configuration - B) in the column does not affect the qual-
ity of flow patterns in the bubble column. Similar observations are
made in case of gas hold-up also. The gas hold-up is maximum at
the central region and decreases with the increase in radial dis-
tance in the presence of internals and has no significant effect on
overall gas hold-up.
5.3.3. Configuration (C)
Configuration (C) is the open column with five internal rods com-
prised of one central rod of 36 mm outer diameter and four rods with
an outer diameter of 12 mm as shown in Fig. 4C. The percent area
covered in this case by the presence of internal is 11.7%. The CFD sim-
ulations were performed by using validated CFD code for the config-
uration C for the superficial velocity of air in the range given in
Table 2. Locally, the flow patterns in the column are very complex
as shown in Fig. 16. However, overall, the formation of single circu-
lation cell over the entire length of the column (the positive value of
axial mean velocity in the center showing that the liquid moves
upward and the negative value near the wall showing the downward
flow) was observed for all the superficial gas velocities (presented in
Table 2). The liquid velocities near the wall of the internal tubes are
found out to be zero mainly because of the boundary condition
given. The velocity profiles were found to be much steeper as com-
pared to those shown in configurations A and B. The gas hold-up
(Fig. 17) and axial liquid velocity predictions were found in good
agreement (standard deviation of 4% and 3%, respectively) with
the experimental data.
5.4. Comparison of turbulent kinetic energy (k)
Figs. 18, 19 and 20 show the comparison of the turbulent kinetic
energy for configurations A, B and C, respectively. In all the cases,
173
the CFD predictions of k show reasonable deviation (standard devi-
ation of 15–200%) from the experimental data. This is because of
the isotropic assumption of the k-e model, which results in a poor
representation of the normal stress and hence turbulent kinetic
energy. Also, RPT as a technique only captured the lower frequen-
cies of the turbulence spectrum, while the k-e model arguably a
larger fraction of the frequency content. Further, the isotropic
assumption leads to a limitation to the normal stresses, that all
the normal stresses are equal to each other and hence inaccurate
predictions of k.
6. Conclusion
In this work, the open-source CFD tool OpenFOAM (v2.3.1) is
standardized and validated for the two-phase bubble columns. It
is found that the OpenFOAM simulations are in a good agreement
with the experimental data. In addition, simulations are performed
by using the commercial code Fluent in order to check the validity
and accuracy of the OpenFOAM and compared the results with that
of the OpenFOAM. OpenFOAM is found to be in a good agreement
with the Fluent predictions. It is noteworthy that the good agree-
ment of the model predictions is even more relevant, because the
velocity profiles measured from RPT involve an independent inter-
pretation method from the interpretation of holdup profiles. Thus,
for practical purposes these measurements are independent of
each other. The good agreement of the CFD predictions and the
experiment therefore holds more relevance.
It is observed that the overall flow pattern in the column (i.e.,
the formation of single circulation cell over the entire length of
the column) remains unaffected with the insertion of the internals
in the column. It was also observed that, the average gas hold-up
increases with increase in number of internals. The centerline
velocities were found to be decreased with increasing number of
internals. The reason behind this is the reduced fluctuations in
the column with internals as compared to that of the open column.
References
[1] F. Larachi, D. Desvigne, L. Donnat, D. Schweich, Simulating the effects of liquid
circulation in bubble columns with internals, Chem. Eng. Sci. 61 (2006) 4195–
4206.
174 V.H. Bhusare et al. / Chemical Engineering Journal 317 (2017) 157–174
[2] C.L. Boutet, F. Larachi, N. Dromard, O. Delsart, D. Schweinch, CFD simulation of
bubble column flows: investigations on turbulence models in RANS approach,
Chem. Eng. Sci. 64 (2009) 4399–4413.
[3] X. Guan, Z. Li, L. Wang, Y. Cheng, X. Li, CFD simulation of bubble dynamics in
bubble columns with internals, Ind. Eng. Chem. Res. 53 (2014) 16529–16538.
[4] A. Horvath, C. Jordan, M. Lukasser, C. Kuttner, A. Makaruk, M. Harasek, CFD
simulation of bubble columns using VOF model: comparison of commercial
and open source solvers with an experiment, Chem. Eng. Trans. 18 (2009).
[5] S.F. Corzo, S.M. Damian, D. Ramajo, N.M. Nigro, Numerical simulation of bubbly
two-phase flow using Eulerian-Eulerian model, Mecanica Comput. 31 (2012)
85–112.
[6] E. Michta, K. Fu, H. Anglart, K. Angela, Numerical predictions of bubbly two-
phase flows with OpenFOAM, J. Comput. Multiphase Flows 4 (2012) 351–362.
[7] J. Chen, F. Li, S. Degaleesan, P. Gupta, M.H. Al-Dahhan, M.P. Dudukovik, B.A.
Toseland, Fluid dynamic parameters in bubble columns with internals, Chem.
Eng. Sci. 54 (1999) 2187–2197.
[8] A. Forret, J.M. Schweitzer, T. Gauthier, R. Krishna, D. Schweich, Liquid
dispersion in large diameter bubble columns, with and without internals,
Can. J. Chem. Eng. 81 (2003) 360–366.
[9] A.A. Youssef, M.H. Al-Dahhan, Impact of internals on the gas holdup and
bubble properties of a bubble column, Ind. Eng. Chem. Res. 48 (2009) 8007–
8013.
[10] H. Rusche, Computational fluid dynamics of dispersed two-phase flows at high
phase fractions, (Ph.D. thesis), Department of Mechanical Engineering,
Imperial college of Science, Technology and Medicine, London, 2002.
[11] H. Jasak, Error analysis and estimation for the finite volume method with
applications to fluid flows, (Ph.D. thesis), Department of Mechanical
Engineering, Imperial college of Science, Technology and Medicine, London,
1996.
[12] D.V. Kalaga, Hydrodynamic, Mixing and Mass Transfer Characteristics of
Multiphase Reactors (Ph.D. thesis), IIT Gandhinagar, India, 2014.
[13] S. Roy, A. Kemoun, M.H. Al-Dahhan, M.P. Dudukovic, Experimental
investigation of the hydrodynamics in a liquid–solid riser, AIChE J. 51 (2005)
802–835.
[14] J.B. Joshi, Computational flow modelling and design of bubble column reactors,
Chem. Eng. Sci. 56 (2001) 5893–5933.
[15] A. Behzadi, R.I. Issa, H. Rusche, Modelling of dispersed bubble and droplet flow
at high phase fractions, Chem. Eng. Sci. 59 (2004) 759–770.
[16] V.S. Vitankar, J.B. Joshi, A comprehensive one-dimensional model for
prediction of flow pattern in bubble columns, Trans. I Chem. E 80 (2002)
499–512.
[17] H. Luo, H.F. Svendsen, Turbulent circulation in bubble columns from eddy
viscosity distributions of single-phase pipe flow, Can. J. Chem. Eng. 69 (1991)
1389–1394.
[18] S.T. Zalesak, Fully multidimensional flux-corrected transport algorithm for
fluids, J. Comput. Phys. 31 (1979) 335–362.
[19] J.H. Hills, Radial non-uniformity of velocity and voidage in a bubble column,
Trans. Inst. Chem. Eng. 52 (1974) 1–9.
[20] T. Menzel, T. Weide, O. Staudacher, U. Onken, Reynolds stress model for bubble
column reactor, Ind. Eng. Chem. Res. 29 (1990) 988–994.
[21] OpenFOAM. The open source CFD toolbox, Ver. 2.3.1. Paris, France: ESI Group:
(2014).