Nuclear Engineering and Design 369 (2020) 110823

Contents lists available at ScienceDirect

Nuclear Engineering and Design

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

Euler-Euler large eddy simulations of the gas–liquid flow in a cylindrical T

bubble column
Mojtaba Goraki Farda, Youssef Stiribaa, , Bouchaib Gourichb, Christophe Vialc,
⁎

Francesc Xavier Graua

a
ETSEQ, Departament d’Enginyeria Mecanica, Universitat Rovira i Virgili, Av. Paisos Catalans 26, 43007 Tarragona, Spain
b
Laboratoire d’Ingénierie des Procédés et d’Environnement, Ecole Supérieure de Technologie, Université Hassan II de Casablanca, Route del Jadida, km 7, BP 8012, Oasis
Casablanca, Morocco
c
Université Clermont Auvergne, Institut Pascal, 2 Avenue Blaise Pascal, TSA 60206, CS 60026, 63178, Aubière Cedex, France

ARTICLE INFO ABSTRACT

Keywords: In this work Euler-Euler Large Eddy Simulations (LES) of dispersed turbulent gas–liquid flows in a cylindrical
Bubble column bubble column are presented. Besides, predictions are compared with experimental data from Vial et al. 2000
Euler-Euler model using laser Doppler velocimetry (LDV). Two test cases are considered where vortical-spiral and turbulent flow
Large eddy simulation regimes occur. The sub-grid scale (SGS) modelling is based on the Smagorinsky kernel with model constant
One-equation SGS model
Cs = 0.08 and the one-equation model for SGS kinetic energy. The emphasis of this work is to analyse the per­
formance of the one-equation SGS model for the prediction of bubbly flow in a three-dimensional high aspect
ratio bubble column (H / D ) of 20 and investigate the influence of the superficial gas velocity using the
OpenFOAM package. The model is compared with the Smagorinsky SGS model and the mixture k model in
terms of the axial liquid velocity, the gas hold-up and liquid velocity fluctuations. The bubble induced turbulence
and various interfacial forces including the drag, virtual mass and turbulent dispersion where incorporated in the
current model. Overall, the predictions of the liquid velocities are in good agreement with experimental mea­
surement using the one-equation SGS model and the Smagorinsky model which improve the mixture k model
in the core and near-wall regions. However, small discrepancies in the gas hold-up are observed in the bubble
plume region and the mixture k model performs much better. The numerical simulations confirm that the
energy spectra of the resolved liquid velocities in churn-turbulent regime follows the classical −5/3 law for low
frequency regions and are close to −3 for high frequencies. More details of the instantaneous local flow structure
have been obtained by the Euler-Euler LES model including large-scale structures and vortices developed in the
bubble plume edge.

1. Introduction pseudo-turbulence in the liquid phase. Several numerical studies of

these types of flows have been carried out by incorporating the tur­
Bubbly gas–liquid flows in multiphase reactors are important for bulence of the liquid phase through the Reynold-Averaged Navier-
many industrial processes, for instance in the chemical, biochemical, Stokes (RANS) model (Mudde and Simonin, 1999; Pfleger and Becker,
or environmental industries and have advantageous characteristics in 2001; Tabib and Schwarz, 2008; Olmos et al., 2001; Selma et al.,
mass and heat transfers. In bubble column reactors, the gas phase is 2010; Stiriba et al., 2017; Kouzbour et al., 2020). The RANS approach,
dispersed in the form of tiny bubbles in a continuous liquid phase typically the k model, models the effect of liquid turbulence on the
using a gas distribution device. The complex interplay between op­ mean flow scale and uses isotropic closures, but fails to reproduce
erating conditions, the gas–liquid interfacial area, bubble size, bubble relevant flow physics since bubbles induce significant turbulence of
rise velocity, turbulence in the liquid phase, and bubble–bubble in­ anisotropic nature. It has provided valuable results and insights
teractions lead to extensive range of flow regimes and complex flow on the turbulence in bubble column reactors with reasonable com­
structures. Furthermore, as the bubbles rise in the column, they induce putational costs.

Corresponding author.
⁎

E-mail addresses: [email protected] (M.G. Fard), [email protected] (Y. Stiriba), [email protected] (B. Gourich),
[email protected] (C. Vial), [email protected] (F.X. Grau).

https://doi.org/10.1016/j.nucengdes.2020.110823
Received 4 November 2019; Received in revised form 18 August 2020; Accepted 19 August 2020
Available online 17 September 2020
0029-5493/ © 2020 Elsevier B.V. All rights reserved.
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

Nomenclature Ur relative velocity, m s−1

UG superficial gas velocity, m s−1
CD drag force coefficient u' fluctuating velocity, m s−1
CL lift force coefficient uL time-averaged axial liquid velocity, m s−1
CS Smagorinsky constant W width of the rectangular column, m
CTD turbulent dispersion coefficient
CVM virtual mass force coefficient Greek symbols
Cµ, BI constant in bubble induced turbulence model
dB bubble diameter, m grid size, m
D diameter of the column, m t time step, s
g gravity acceleration, m s−2 x grid spacing in x direction, m
H height of the column, m y grid spacing in y direction, m
k turbulent kinetic energy of phase , m2 s−2 z grid spacing in z direction, m
kSGS Sub-grid scale kinetic energy of phase, m2 s−2 gas fraction of phase
M total interfacial force acting between the phase and the L, L liquid molecular viscosity, m2 s−1
other phase, N m−3 SGS sub-grid scale viscosity, m2 s−1
M D
drag force for the phase , N m−3 eff
effective viscosity, m2 s−1
ML lift force for the phase , N m−3 L,Tur shear-induced turbulent viscosity, Pa s
MTD turbulent dispersion force for the phase , N m−3 µL,BIT bubble induced viscosity, Pa s
MVM virtual mass force for the phase , N m−3 density, kg/m3
p pressure, N m−2 shear stress of phase , Pa
r radial radius, m
Subscripts
R column radius, m
ReB bubble Reynolds number
phase ( = G ; gas phase, = L ; liquid phase)
t time, s
U resolved velocity of phase , m s−1

Bubbly flow is characterised by the development of distinct flow For instance, the largest turbulence scales are comparable in size to
structures of different length scales, especially for transition and het­ those of the mean flow and depend on the reactor geometry and flow
erogeneous flow regimes. Turbulent scales varied from those of the conditions, whereas the smallest scales depend on the bubble dynamics
characteristic length of the mean flow to those of the microscopic ones. and are proportional to the bubble size. The large-scale turbulent

Table 1
Summary of previous numerical simulations of gas–liquid flow in bubble columns using LES turbulence models.
Reference Bubble column dimensions Gas distributor Bubble Superficial gas SGS model
diameter velocity

Deen et al. (2001) Rectangular column, Perforated plate, 4 mm 0.49 cm/s Smagorinsky, CS = 0.1
W = 0.15m, D = 0.15m, H = 1m 49 holes of D = 1 mm
Bove et al. (2004) Rectangular column, Perforated plate, 4 mm 0.5 cm/s VLES,
W = 0.05m, D = 0.2m, H = 0.45m 49 holes of D = 1 mm CS = 0.12
van der Hengel Rectangular column, Perforated plate, 3 mm 0.5 cm/s Smagorinsky with DBM
et al. (2005) W = 0.15m, D = 0.15m, H = 1m 49 holes of D = 1 mm
Zhang et al. Rectangular column, Perforated plate, 4 mm 0.49 cm/s Smagorinsky, CS = 0.08 0.2
(2008) W = 0.15m, D = 0.15m, H = 1m 49 holes of D = 1 mm
Niceno et al. Rectangular column, Perforated plate, 4 mm 0.5 cm/s One-equation SGS
(2011) W = 0.15m, D = 0.15m, H = 1m 49 holes of D = 1 mm
Dhotre et al. Rectangular column, Perforated plate, 4 mm 0.5 cm/s Smagorinsky CS = 0.12 , and dynamic
(2008) W = 0.15m, D = 0.15m, H = 1m 49 holes of D = 1 mm Smagorinsky
Hu and Celik Rectangular column, Flush mounted, 5 porous dicks of D = 40 1.6 mm 0.66 cm/s Smagorinsky, CS = 0.032
(2008) W = 0.08m, D = 0.15m, H = 2m mm
Darmana et al. Rectangular column, Multiple gas injection of 95 needles of 4 mm 0.7 cm/s SGS of Vreman,
(2009) W = 0.24m, D = 0.072m, H = 0.8m D = 0.51 mm CS = 0.1
Ekambara and Cylindrical column, Multipoint perforated plate, 25 holes of 6 mm 0.2 cm/s Smagorinsky, CS = 0.12
Dhotre D = 0.15 m, H = 0.9 m D = 2 mm
(2010)
Tabib and Cylindrical column, D = 0.15m, H = 1m Multipoint perforated plate, 25 holes of 3–5 mm 2 cm/s One-equation SGS
Schwarz D = 2 mm
(2011)
Ma et al. (2015a) Rectangular column, Multiple gas injection of 35 needles iMUSIG,2 0.3 and Smagorinsky, CS = 0.15
W = 0.243m, D = 0.04m, H = 1m groups 1.3 cm/s
Ma et al. (2015b) Rectangular column, Set of 8 holes in a rectangular 2 mm 0.17 cm/s Dynamic Smagorinsky
W = 0.2m, D = 0.05m, H = 0.45m configuration: 0.02m × 0.0125m
Khan et al. (2017) Cylindrical column, Multipoint perforated plate, 25 holes 5 mm 2 – 10 cm/s Smagorinsky, CS = 0.5
D = 0.15 m, H = 1 m of D = 2 mm
Liu and Li (2018) Rectangular column, Perforated plate, 4 mm 0.5 cm/s Dynamic Smagorinsky
W = 0.15m, D = 0.15m, H = 1m 49 holes of D = 1 mm
Present work Cylindrical column, Multiple orifice plate, 4.5 mm 6 and 8.4 cm/s Smagorinsky with CS = 0.08,
D = 0.1m, H = 2m 62 orifices of D = 1 mm and one-equation SGS

2
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

motions interact with the bubbles and thereby affect their motions, 2. Two fluid model and numerical setup
whereas the small scales not only dissipate the kinetic energy but can
generate energy to the largest scales and tend to be more isotropic as 2.1. The flow equations
well (Dhotre et al., 2013; Ma et al., 2015a, 2015b). The energy spectra
of the liquid fluctuations exhibits a broad range of frequency and gives The two-fluid model is built up on the spatial filtering for LES or
a power law scaling with the slope of −5/3 for low frequency regions conditional averaging for RANS of the conservation equations of mass
which is progressively replaced by −25/3 in Liu and Li (2018) and over and momentum. In this approach, both phases, the continuous liquid
than −8/3 in the works of Ma et al. (2015a) and Ma et al. (2015b) and phase and the dispersed gas phases, are modelled as two inter­
Lance and Bataille (1991) for high frequency regions. penetrating continua. In LES cases, it is assumed that the filtered
To reproduce relevant flow physics and give comprehensive insights equations are used to compute the large-scale lengths while the effect of
into two-phase flow turbulence, the LES approach has attracted great unresolved turbulent scales are modelled using a sub-grid model. In the
attention in the simulation of dispersed two-phase turbulent flows. It present work, the flow is assumed to be adiabatic, without considering
has been used in several investigations and simulations to predict the interfacial mass transfer between the air and the water phases.
multiphase flow dominated by large coherent structures or eddies in The present formulation closely follows the procedure outlined by
bubble columns, stirred tanks and many other reactors (Tabib and Weller (2005), where the mass and momentum equations for the phase
Schwarz, 2011; Dhotre et al., 2008). As in single phase flows, LES are given by
model resolves directly the interaction of the large-scale motions with
( )
bubbles, whereas the less energetic smallest motions including the in­ + ( U)=0
t (1)
teraction of the bubbles with the surrounding turbulence are re­
presented in terms of sub-grid scale closure models. The Euler-Euler LES ( U) eff
model predicts more accurately flows dominated by large coherent + ( UU)= p + g ( )+M
t
structures or eddies in bubble columns which carry most of the flow
(2)
energy (typically 90%) than the traditional RANS models and re­
presents more details of the flow structure (Ma et al., 2015a, 2015b; Here is the volume fraction of each phase, U is the phase re­
Dhotre et al., 2008). Furthermore, the k models consider isotropic solved velocity, and eff represents the effective stress tensor usually
turbulence and do not analyse the flow near the walls. Table 1 gives a decomposed into a mean viscous stress and turbulent stress tensor for
summary of previous works of gas–liquid flows in bubble column re­ the phase as
actors in a chronological manner. For instance, Zhang et al. (2008) 2 2
eff eff
investigated the Smagorinsky model with different values of the con­ = U + ( U )T ( U )I + k I
3 3 (3)
stant Cs and the dynamic Smagorinsky model. Niceno et al. (2011)
applied the one-equation SGS turbulent kinetic energy LES and sug­ where k is the turbulent kinetic energy of phase , I is the identity
gested that the sub-grid scale kinetic energy obtained from the model tensor, and eff is the effective viscosity of phase . The effective visc­
can be used to assess the SGS dispersion turbulent force. Tabib and osity of the liquid phase is obtained through the summation of the
Schwarz (2011) employed the commercial CFD package ANSYS CFX to molecular viscosity, the shear-induced turbulent viscosity, and the
analyse the inclusion of SGS turbulent dispersion (TD) force and con­ bubble-induced turbulent viscosity
cluded that the results of a coarser mesh can be improved by using a eff
lower magnitude of SGS-TD force. Liu and Li (2018) studied the scale
= L, L + L,Tur + L,BIT (4)
adaptive of LES ANSYS CFX code using a small / dB 1. These works and is formulated in the present study using two models: (a) the
have made several assumptions in the CFD modelling, reactor geometry Smagorinsky model proposed by Zhang et al. (2008), (b) and the one-
and operating conditions. Indeed, their bubble columns are operating at equation sub-grid-scale model proposed by Niceno et al. (2011).
low superficial gas velocities with non-uniform aerations or use flat The Smagorinsky model is a zero-equation turbulent LES model and
bubble column reactors. At high gas flow rates, the flow field is un­ the liquid phase shear-induced turbulent viscosity is formulated as
steady and characterised by local recirculation near the sparger and follows
different scale vortices in the core region.
In view of this, it is desirable to carry out LES in a three-dimen­ L,Tur = (CS )2 |S| (5)
sional bubble column at high inlet superficial gas velocity. The pur­ Here CS is a model constant, S is the characteristic filtered rate of the
pose of this work is therefore to employ Euler-Euler LES approach to strain and = Vol1/3 is the filtered width, where Vol is the volume of
simulate dispersed turbulent two-phase flows in a three-dimensional the computational cell. The model constant seems to be different for
cylindrical bubble column of high aspect ratio ( H / D ) of 20 with spe­ different flow situation and was chosen to be CS 0.1 according to the
cial emphasis on the performance of the one-equation SGS model and work of Zhang et al. (2008). The turbulence model corrects the SGS
the influence of the superficial gas velocity. A multiple nozzle gas turbulent viscosity by a contribution due to the bubble induced tur­
distributor is used for uniform aeration. The inlet superficial gas ve­ bulence (Zhang et al., 2008) and the model proposed by Sato and
locities, used in this work, are UG= 6.0 and 8.4 cm/s where vortical- Sekoguchi (1975) was employed
spiral and turbulent flow regimes occur, respectively. The simulations
are set up according to experimental works of Camarasa et al. (1999)
µL,BIT = L Cµ,BIT G |UG UL| (6)
and Vial et al. (2001) using LDA as well as they have been performed with its constant Cµ,BIT set to 0.6.
by using the twoPhaseEulerFoam solver implemented in the Open­ The one-equation sub-grid-scale model by Niceno et al. (2011)
FOAM v3.0.1 software package. The results achieved from the one- solves a transport equation for the unresolved kinetic energy kSGS . The
equation model SGS are compared with the Smagorinsky model with model is able to account for the effects of bubble induced turbulence
constant Cs = 0.08 and the mixture k model. The accuracy of the through an additional source term in the transport equation for kSGS in
results in comparison to experimental data are evaluated. Compre­ the continuous phase and uses the modelled SGS energy to estimate the
hensive simulations were conducted to examine the instantaneous SGS turbulent dispersion force (Niceno et al., 2011). The sub-grid ki­
flow structure and Reynolds stresses. Furthermore, the analysis of the netic energy equation is given by
energy spectra of resolved velocity and the vorticity distribution have
3/2
been addressed. kSGS
+ (kSGS U) [( +
kSGS
SGS ) kSGS ] = G C
t
L, L
(7)

3
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

where G is the production term, defined as good agreement with experimental measurements, were tested. For the
one-equation and mixture k models we use CTD = 0.6 .
G= SGS |Sij| (8)
2.2. Numerical simulation set-up
and the sub-grid viscosity is

SGS
1/2
= Ck kSGS (9) The numerical simulations were carried out in a cylindrical bubble
column with uniform aeration. The geometry of the current bubble
The model constants are C = 1.05 and Ck = 0.07 (Niceno et al.,
column reactor is the same as used by Vial et al. (2001) and Camarasa
2011).
et al. (1999) in their experiments. The height of the column is H = 2 m,
In Eq. (2), M represents the inter-phase momentum exchange be­
the diameter is D = 0.10 m, and the static liquid height is 1.35 m. The
tween phase and the other phase due to various interphase forces. The
reactor is operated with the water and air as the continuous and dis­
interfacial forces are decomposed into four contributions
persed phases at the room temperature and atmospheric pressure, re­
M = M D + M L + MVM + MTD (10) spectively, at two superficial gas velocities 6.0 cm/s and 8.4 cm/s
corresponding to transition and heterogeneous flow regimes. The gas is
where the forces on the right-hand side of equality are the drag force injected from the bottom of the column through a multiple-orifice
denoted by MD , the lift force represented by ML , the virtual mass force nozzle for uniform aeration and it allows us to study the flow regime
by MVM , and the turbulent dispersion force by MTD . There are many transition. The gas distributor is treated as a uniform mass flow rate
models for each of these forces depending on their applicability, the through the bottom boundary calculated from superficial gas velocities
flow regime and the operating conditions as discussed by (Joshi, 2001; for mass conservation with gas volume fraction of 1.0. The pressure at
Vial and Stiriba, 2013; and Ziegenhein et al., 2017). There is still no the inlet is set to zeroGradient and specified by zero gradient. At the
complete agreement on the closures or the combination to be used at outlet, the pressure is specified as the atmospheric pressure, and the gas
best. The drag force (per volume) for the liquid phase is estimated as hold up is set to inletOutlet with zero gradient for outflow and fixed
3 CD value for backward flow. The no-slip condition is applied at the walls
MDL = |Ur | Ur
4
G L
dB (11) for the velocities and Dirichlet condition for the gas hold-up. Moreover,
for the one-equation model we apply wall functions.
where CD refers to the drag force coefficient and is calculated according The numerical simulations were carried out with the open source
to the Schiller-Neumann correlation and Ur is the relative velocity. CFD package OpenFOAM library (Weller et al., 1998). The governing
Many drag model have been proposed and compared in the literature equations of continuity and momentum as well as the transport equa­
(Pourtousi et al., 2014; Tabib and Schwarz, 2008; Zhang et al., 2008; tion for kSGS are solved by the two-phase flow solver twoPhaseEu­
and Silva et al., 2012). But Tabib and Schwarz (2008) found that lerFoam available in OpenFOAM 3.0.1. The solver is based on a finite
Schiller-Naumann, Ishii-Zuber, Tomiyama, and Grace et al. using dif­ volume formulation to discretise the model equations which has shown
ferent turbulence closure (k , RNG, LES) models give the same re­ to be stable for transient calculations (Weller, 2005). The first-order
sults in a cylindrical bubble column similar to our reactor. Furthermore, bounded implicit Euler scheme is adopted for the time integration, the
the Schiller-Naumann drag model works quite well for bubbly flow in gradient terms are approximated with a linear interpolation, the con­
industrial systems since bubbles are contaminated by surfactants at the vective terms are discretized with second-order upwind scheme, and
interface and behaves like a rigid sphere (Clift et al., 1978). the diffusive terms are interpolated with the Gauss linear orthogonal
The lift force results from the movement of bubbles through a non- scheme. We employ the PIMPLE algorithm to solve the pressur­
uniform flow field due to shear or vorticity effects. The force (per vo­ e–velocity coupling where the pressure equation is solved, and the
lume) is modelled as predicted velocities are corrected by the pressure change. The pre­
MLL = G L CL Ur ×( × Ur) (12) conditioned conjugate gradient (PCG) is used for solving the discretized
pressure equation and the incomplete-Cholesky preconditioned bi-
where CL is a constant lift force. We conducted the same simulation conjugate gradient (BICCG) for the other set of linear equations. For
with different lift coefficients and the model of Tomiyama et al. (2002), more detailed discussions of all steps mentioned above (Rusche, 2002;
but no noticeable improvements in the results were observed, from Weller, 2005).
which we conclude that the lift force plays a minor role in our test Prior to the description of the computational mesh and presenting
cases. Furthermore, the steady simulations of a bubble column reactor the results, we emphasis the implications of bubble size distribution on
(Vial and Stiriba, 2013) show that the lift force was overshadowed by the model. In the present work, we assume a spherical bubble size
the turbulent dispersion force which better predicted the radial dis­ distribution of 4.5 mm according to bubble size measurements of
persion of the gas phase along the axial direction of the bubble column Camarasa et al. (1999). The same simplification was successfully used
and the experimental gas hold-up at the column center using a single- by Khan et al. (2017) to simulate their bubble column using k ,RSM,
orifice nozzle. and Smagorinsky turbulence model at high superficial gas velocities.
Liquid acceleration in the wake of the bubble is taking into account Perhaps the incorporation of bubble coalescence and break up in the
through the virtual mass force, which is modelled as LES may help in predicting the flow field in the vortical-spiral regime
DUL DUG (Khan et al., 2017). But note also that, recently, Huang et al. (2018)
MVM
L = G L CVM implemented and used variable bubble size models in modelling three-
Dt Dt (13)
dimensional large diameter bubble columns operating under churn
where CVM is the virtual mass coefficient and is taken to be 0.5 for turbulent flow regime; they concluded that the model did not lead to
individual spherical bubbles (Zhang et al., 2008; Dhotre et al., 2008). any substantial improvement relative to the single size models and
The SGS component of those forces will be neglected except the highlighted the need for improved breakup and coalescence closure
turbulent dispersion force which can be estimated using the modelled descriptions.
SGS energy in the one-equation model. The turbulent dispersion force The computational mesh was generated using the Gmsh finite ele­
proposed by Lopez de Bertodano et al. (1994) is adopted. It is modelled ment mesh generator. In order to check that the computed results are
as grid-independent, different grids with dB / = 0.75, 1.1, 1.4 and 1.875
MTD have been analysed by increasing the number of computational cells in
L = CTD Lk G (14)
the center of the column and the axial direction from 3 mm to 5 mm
Several turbulent dispersion coefficients CTD , required to obtain and stretching the mesh near the walls (Table 2). Milelli et al. (2002)

4
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

Table 2
The computational mesh and grid spacing investigated.
Mesh x × y × z (mm3) /dB Turbulence model Interfacial forces

Mesh 1 7.5 × 7.5 × 7.5 1.875 LES FD + FVM + FTD

Mesh 2 5×5×7 1.24 LES FD + FVM + FTD
Mesh 3 5×5×5 1.1 LES FD + FVM + FTD
Mesh 4 3×3×3 0.8 LES FD + FVM + FTD
Mesh 5 5×5×7 1.24 RANS FD + FVM + FTD

Fig. 2. Comparison between the simulated and experimental profiles of the

axial liquid velocity using different bubble sizes at UG = 8.4cm/s (a), and
comparison between experimental and numerical mean bubble diameter with
UG (b).

agreement with experimental data using both conditions / dB = 1.2

and / dB = 2.5, and Liu and Li (2018) used the criterion / dB 1.0 and
concluded that the grid size does not have to be larger than a single
bubble. All transient calculations are started from static conditions with
the liquid at rest and the gas is injected with a mass flow rate corre­
sponding to the experimental superficial gas velocity. We start calcu­
Fig. 1. Mesh independence analysis; comparison of the time-averaged results lations with a fixed small-time step of t = 0.0005 s for the first 20 s
for the axial liquid velocity and the different meshes investigated at then we increase it to 0.001 s to account for the transient instabilities of
UG = 8.4 cm/s (a); the ratio, , resolved kinetic energy to total kinetic energy bubbly turbulent flows. The flow was simulated for 200 s and the
(b). averaged results from t = 50 s to t = 200 s were quantitatively com­
pared with experimental data. All the simulations were performed in
established the criterion of the ratio of the bubble diameter to cut-off parallel mode on a PC cluster with 16 nodes, Intel Xeon, 2.8 GHz, 4 GB
filter size: / dB 1.5, that is to say that the mesh size must be at least RAM. The different time-averaged profiles displayed in section 3 are
50% larger than the bubble diameter for Eulerian-Eulerian simulations. given at the mid height of the bubble column (H = 1 m).
Fig. 1 shows a comparison of the axial liquid velocity. All the meshes
show very similar results and mesh 3 and 4 perform better in the near-
wall region. In this work, we have employed a medium mesh with a 3. Numerical results
filter width = 5 mm ( / dB = 1.1), which quantitively seems to give
better agreements and ensures a good compromise between the CPU 3.1. The one-equation SGS and Smagorinsky model
time and accuracy at the column center and close near the walls. As can
be seen from Fig. 1, we have checked the non-dimensional spacing x+, The resolved axial liquid velocity is presented in Fig. 1a, it can be
y+, and z+ desirable to make a large eddy simulation setup convin­ seen that there is no significant change in the prediction between the
cing. Note that for comparison, Niceno et al. (2011) used the criterion medium and the fine mesh. In order to understand that how the LES
/ dB = 1.2 and found no significant difference with the coarser one model resolves well the fluid flow in the column numerically, Pope
satisfying Milelli condition, Dhotre et al. (2008) found a good (2011) suggested to measure and check when the ratio of resolved ki­
netic energy to the total turbulent kinetic energy is greater than 80%, i.e.,

5
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

Fig. 3. Comparison between the simulated and experimental profiles of the Fig. 4. Comparison between the simulated and experimental profiles of the
axial liquid velocity at superficial gas velocity UG = 6.0cm/s (a) and local gas hold-up at superficial gas velocity UG = 6.0cm/s (a) and UG = 8.4cm/s
UG = 8.4cm/s (b). (b).

kres 1
= > 0.8, kres = uL' uL'
kSGS + kres 2 (15)

This ratio is plotted in Fig. 1b at height h = 0.7 m. We get the same

results for different height positions in a plane normal to the axial flow
direction. The ratio is around 80% with the medium grid used and the
LES resolves more flow in the core regions. Hence, the resolution of the
LES with the present mesh can be considered acceptable for analysis.
The surface bubble diameter, dB , was measured by Camarasa et al.
(1999) for different inlet superficial gas velocities (Fig. 2(b)). For het­
erogeneous hydrodynamic regimes (UG > 5 cm/s), dB is nearly equals to
4.5 mm and in reasonable agreement with our assumption. The mea­
sured bubble size distribution is also given by Camarasa et al. (1999),
where spherical bubbles were observed in the new-wall region and
large bubbles of different shapes in the core region, and exhibit a re­
latively narrow distribution around 4.5 mm. Numerically, three dif­
ferent bubble sizes 3.5, 4.5, and 5.5 mm were employed to test the
Fig. 5. Time averaged axial gas velocity at superficial gas velocity UG = 6.0cm/s
bubble size effect on the simulation results, see Fig. 2(a). The calculated and UG = 8.4cm/s .
axial liquid velocities are very similar and the results for 4.5 mm are
closest to the experimental data, this value will be used for the rest of
this work. compared with experiments, so as to realize the relative behaviour of
The axial liquid velocity profiles predicted by the mixture k different turbulence models, and the results are shown in Fig. 3. It can
model (Behzadi et al., 2004) and the one-equation SGS have been be seen that both models provide good agreements with experiments

6
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

Fig. 6. Comparison between the simulated and experimental profiles of the axial rms liquid velocity fluctuations (a) and (b), turbulent fluctuations (c), and Reynolds
shear stress (d).

Table 3 closer to the experimental data. The inclusion of the turbulent disper­
Experimental and numerical centerline axial fluctuations of the liquid velocity. sion force in the RANS model decreases the axial liquid velocity in the
Superficial gas velocity (cm/s) 6 8.4
core region and results in a comparatively flatter gas hold-up profile,
Experimental (m/s) 0.2 0.25 which can predict the profile closer to the experimental data in the core
CFD (m/s) 0.211 0.257 region. Similar RANS results were reported by Tabib and Schwarz
(2008).
Fig. 5 displays radial distribution of the time-averaged axial bubble
and that RANS model overpredicts the liquid velocity for the near-wall velocity at the mid height of the bubble column (H = 1 m). Un­
region. The reason for this over estimation may come from the fact that fortunately, experimental measurements are not available for compar­
a bubble plume moves upward in a spiral rotating manner in the center ison. The results show similar trends as those reported in Zhang et al.
with the liquid flow meantime small spherical bubbles spirally move (2008) and Dhotre et al. (2008); the bubble plume spreading in the
downward close near to the wall column, accelerating the water flow; center and a relatively steep gas velocity profile for high superficial gas
the mixture k model does not analyse this flow near the wall due to velocity which leads to less dispersed bubble plume.
the inappropriateness of standard wall functions developed basically for Fig. 6 shows profiles of the fluctuations of the liquid velocity at
single phase flow. The one-equation SGS model predicts the overall height 0.7 m. All the profiles are based on the resolved part of liquid
behaviour of the axial liquid velocity profile better than the RANS velocities. Unfortunately, experimental data on kinetic turbulent energy
model and gives good agreement with experimental measurement. For of the liquid phase are not available. In fact, Vial et al. (2001) only
the gas hold-up (Fig. 4), the two turbulence models capture the ex­ measured the rms in the axial and radial directions. Fig. 6(b) displays a
perimental profiles reasonably well. It can be however observed that comparison of liquid fluctuations in the other directions. Clearly, the
the LES model under predicts the gas hold-up at the center of the velocity fluctuations in the present bubble column reactor are aniso­
column for 0.5 x /R 0.5, where the flow is dominated by large- tropic. The time-averaged spanwise component v'v ' 1/2 increases
scale structures, whereas the RANS model performs much better. In the smoothly away from the wall and attains a maximum at the center of
near wall region where the flow is dominated by small-scale structures, the column, whereas the streamwise fluctuations u'u' 1/2 display a
the situation is different, and the gas hold up predicted by LES is much periodic trend with a lower value in the core region which attains its

7
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

with a non-uniform aeration faced the same scenario. As shown above,

the high inlet gas flow rate induces substantial turbulence both in the
core and in the wall regions which changes the trend of liquid fluc­
tuations. Furthermore, it is worth noting that values of axial liquid
fluctuations are higher than the other components and dominate in
both the core region and near the walls.
The liquid Reynolds shear stresses u'v' and v'w ' are far smaller
than the normal stresses and increase with the gas flow rate as shown in
Fig. 6(c). As mentioned in Mudde et al. (1997), the large vortical flow
structures significantly influence the Reynolds stresses since the vor­
tices span the entire width of the column. The large contribution to the
Reynolds shear stresses become larger in the vortical flow region since
the fluctuations in the vertical component dominate close to the center
region of the column where the liquid moves in a wavy manner,
whereas u'v' peaks in the central plume and in the near wall region.
The Reynolds shear stresses experience fluctuations due to the swinging
motion of the bubble plume. At higher superficial gas velocity
UG = 8.4 cm/s the intensity of the large-scale turbulence is much higher
due to the bubble motion which accelerates the liquid flow and causes
an overestimation of the liquid velocity in the central plume region.
Fig. 6(a) also shows comparisons between experiments and nu­
merically predicted vertical liquid velocity fluctuations where it can be
seen that the one-equation SGS model can reproduce the experimental
data much better than the RANS model. The time-averaged axial liquid
velocity averaged through the cross-sectional area normal to the axial
direction at height h = 0.7 m is given in Table 3. One can see the good
agreement for both gas flow rates. The effect of the turbulent dispersion
was added by incorporating the sub-grid scale turbulent dispersion
turbulent dispersion force using the SGS kinetic energy obtained from
the one-equation LES model using mesh sizes coarser than the bubble
size (4.5 mm). The axial liquid velocity profiles are practically the same
near the wall and agree best with experimental data as we increase
CTD.With coefficients (CTD ) larger than 0.6 the profiles do not improve
anymore. We found that such interfacial force improves the liquid ve­
locity profile as in Tabib and Schwarz (2011) who have shown that
even a small magnitude of turbulent dispersion SGS is enough to affect
the flow profile.
Fig. 7 shows a comparison between the Smagorinsky and one-
equation SGS models for the axial liquid velocity at UG = 8.4 cm/s. The
resolved part by the one-equation model shows better agreement for the
liquid velocity, whereas the Smagorinsky model over-predicts the ex­
perimental data in the center and captures the trend of the down-flow
circulation in the near-wall region. However, the gas hold-up is under-
estimated and becomes flatter in the core region. Zhang et al. (2008)
and Dhotre et al. (2008) compared different LES models. Their results
over-predicted experimental profiles and became steeper for high va­
lues of CS than 0.15 since the turbulent viscosity increases and damps
the bubble plume. With CS = 0.08 the CFD model provides a good so­
lution for the time-averaged axial liquid velocity. The axial liquid ve­
locity fluctuations predicted by both models are very similar to each
other. For the rest of this work we use the one-equation model to
analyse the instantaneous flow as well as the energy spectra.

3.2. Instantaneous flow

Fig. 7. Comparison between the simulated axial liquid velocity (top), the gas The instantaneous flow structure in 3D bubble columns was classi­
hold-up (center) and rms axial liquid velocity fluctuation results obtained using fied, based on visual study of (Chen et al., 1994; Lin et al., 1996; Goraki
Smagorinsky and one-equation SGS models (bottom). Fard et al., 2020), into four distinct regions: an oscillating plume ac­
companied with two staggered rows of vortices, fast bubble region,
highest value close near to the wall in a similar way to the axial fluc­ vortical flow region, and descending flow region close near the walls. A
tuations displayed in Fig. 6(a). This is probably due to the liquid two-phase flow computational model has to capture all these features
movements from upward to downward and laterally at the center and observed in the experiments. Snapshots of instantaneous liquid velo­
close to the wall column in which u'u' peaks with high magnitude near cities vector field together with the gas hold-up contour plot are shown
the wall. Ma et al. (2015b) observed the same trend in their quasi-2D in Figs. 8 and 9 for the simulations using RANS and LES models in the
bubble column and Deen et al. (2001) in a 3D bubble column reactor plane of symmetry and several cross-sections at different superficial gas
velocities, respectively. The time-averaged results are also displayed in

8
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

Fig. 8. Snapshots of instantaneous gas hold-up and liquid velocity field with RANS model (left) and LES model (center) and time averaged LES (right) at UG=6.0 cm/s
(top), and UG=8.4 cm/s (bottom).

Fig. 8. The gas injected from the bottom forms clusters of bubbles that flow regions and can be used for identification of coherent flow struc­
move upward in a wavy manner along the region of the central plume. tures.
Multiple smaller and larger vortex cells are continuously generated in There are many vorticity criteria used to identify and visualize
the vortical-spiral region and along the side of the bubble plume, which vorticity regions and characteristic three-dimensional eddy structures,
stagger on each other and change their size and position in time. The for instance, the Q criterion, 2 criterion, and criterion (see Chen
behaviour of the bubble plume and the undulation shape of the bubble et al., 2015). In this work, we chose 2 method to visualize the iso-
swarm simulated by LES are more dynamic than those obtained by surface of vortical structures coloured by the vertical liquid velocity in
RANS at the recorded instant, exhibiting more appreciable swinging the column (Fig. 10). The color shows the magnitude of the liquid axial
motion and result in more complicated bubble-induced flow structures. velocity, 2 = 2 , where the large-scale structures consist mainly of
The transient liquid field seems to be more uniform near the free sur­ plume structures meandering and oscillating. Complex vortical rings
face at UG = 6.0 cm/s for RANS; as a result, we see recirculating zones are formed in the central plume and vortical regions, adjacent to the
that push bubbles to disappear from the liquid phase. From the present descending flow region, where high velocity and velocity fluctuations
LES, the time-averaged global gas hold-up from the injector to the free- are noticed, so the liquid particles tend to spin around itself forming
surface is found to be nearly 20%. In the cross-section the flow moves in vortices. There are more vortex loops near the sparger and at
a spiral way forth and back. In the core region, higher values of the gas UG = 8.4 cm/s, suggesting that more turbulence is generated. Unlike
hold-up are obtained meaning the existence of the central bubble re­ RANS, it can be seen the high degree of randomness exhibited by LES
gion. Both the instantaneous and time-averaged snapshot show the near the center and along the sidewalls. As noticed by other authors (Hu
vortical-spiral flow region close to the wall. The instantaneous profile and Celik, 2008), it is believed that the estimated flow behaviour based
highlights that the Euler LES model has been able to capture the four on LES model, is to be closer to the real flow situation and that LES

9
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

Fig. 9. Snapshots of instantaneous gas hold-up with LES model at UG=8.4 cm/s (right), and UG=6.0 cm/s (left).

Fig. 10. Instantaneous vortical structure at time T = 200 s by 2 method coloured by the magnitude of the liquid velocity, 2= 2.0 ; at UG=6.0 cm/s (left), and
UG=8.4 cm/s (right).

resolves many more transient details of the flow (Fig. 8). 3.3. Energy spectra
Overall, the instantaneous results and the liquid velocity fluctua­
tions profiles reveal the generation of large-scale structures moving Figs. 11 and 12 show a 200 s time history plot of the resolved axial
upward in a meandering way in the bubble plume region and spiralling and streamwise liquid velocity at one point (x / R = y/ R = 0m, z = 0.7m)
downward in the near-wall region, as shown in Figs. 8 and 9, by the in the column corresponding to nearly 20,000 sample points and the
formation and motion of cluster of bubbles and the subsequent bubble corresponding energy spectra at UG = 6.0 cm/s, while Figs. 13 and 14
wake interaction. Strong vortices of different sizes are developed in the display the same information at UG = 8.4 cm/s. In Figs. 11 and 13, the
plume edge. The turbulence is anisotropic, and the liquid axial fluc­ transient behaviour is well reflected in the high-frequency oscillations
tuations are significantly larger than in the streamwise or spanwise of liquid velocity components around time averaged values, depicted in
directions dominating the turbulent kinetic energy. The one-equation red lines, due to the turbulent fluctuations. The amplitude of fluctua­
SGS model predicts accurately the axial liquid fluctuations and fails to tions increases with the inlet gas flow rate where axial component
capture the gas hold-up in the core plume region. contains more frequencies.

10
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

Fig. 11. Time history of the axial liquid velocity (a) and radial liquid velocity
(b) with one-equation SGS model at the centerline of the column, at a height of
z = 0.7 m and UG = 6.0cm/s .

The energy spectrum densities (PSD) obtained from LES with data
extracted from Figs. 11 and 13 and cover the time from 50 s to 200 s are
shown in Figs. 12 and 14. As can be seen, each spectrum displays a
broad range of frequencies with slopes of about −5/3 in the frequency
region between 1 and 10 Hz. For high frequency region (> 10 Hz) the
decay becomes faster with a slope steeper than −3 power law. Several
authors analysed the power spectrum based on Euler-Euler LES data Fig. 12. Power spectrum density of radial liquid velocity (a) and axial liquid
and obtained different slope decay in the inertial subrange region. For velocity (b) at UG = 6.0cm/s .
instance, Dhotre et al. (2008) and Ma et al. (2015a) used the BIT model
of Sato and Sekoguchi (1975) and the obtained slope was partly over
than −10/3, while Liu and Li (2018) obtained a −25/3 power laws. By authors examined a frequency that is representative of the bubbles and
comparing both predicted spectra for higher frequencies, large inlet gas its impact on the resulting spectra. In the present case, the re­
flow rate gives more dissipation and the bubbles alter the PSD sig­ presentative bubble frequency may be estimated as
nificantly (Figs. 12 and 14). Two lines with slopes −5/3 and −3 are fB = |Ur |/(2 dB ) 9 Hz, where |Ur | 25cm/s being the bubble velocity
shown in the figures. For UG = 6.0 cm/s, the PSD curve exhibits a slope and dB 4.5 mm the averaged bubble diameter, so that above this value
close to −10/3 for the radial liquid velocity and close to –11/3 for the the PSD changes the characteristic slope −5/3 to −3, which implies
axial liquid velocity, whereas for UG = 8.4 cm/s the curve clearly fol­ that there an energy input on the scale of bubble diameter (dB ) and
lows the k 3 line in the high frequency inertial and dissipation region as frequency of bubble motion.
observed experimentally in grid turbulence configurations in Lance and
Bataille (1991); Riboux et al. (2010), Martínez Mercado et al. (2010) 4. Conclusions
and Prakash et al. (2016). The PSD of the spanwise velocity component,
not presented here, exhibits the same behaviour for both superficial gas Euler-Euler large eddy simulations of dispersed turbulent gas–liquid
velocities. flow in a three-dimensional cylindrical bubble column, with high aspect
Several works investigated the fast decay in the dissipation range of ratio (H / D ) of 20 and multiple orifice gas nozzle, have been presented.
the energy spectrum and attributed it to buoyancy-generated inertia Effects of all drag forces, non-drag forces, sub-grid turbulent dispersion
force and bubble-induced viscosity effects. Ma et al. (2015a) and Ma and bubble induced turbulence are all accounted for. For the time-
et al. (2015b) compared their LES energy spectrum with the experi­ averaged axial liquid velocity and gas hold-up, it is found that the
mental spectrum of Akbar et al. (2012) and found that their resolved present model based on the one-equation SGS shows good agreement
and reliable angular bubble frequencies are far away, and that the with experimental measurement data from Camarasa et al. (1999), and
frequency information related to the bubble wake is lost. The origin of improves the axial liquid velocity of the mixture k model in the near
the −3 slope was explained by Prakash et al. (2016), wherein the wall regions and the bubble plume but small discrepancies in the gas

11
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

Fig. 13. Time history of the axial liquid velocity (a) and radial liquid velocity
(b) with one-equation SGS model at the centerline of the column, at a height of
z = 0.7 m UG = 8.4cm/s .

hold-up are observed in the core region. The mixture k model ac­
curately predicts the radial distribution of the gas hold-up. The flow
calculated by the Euler-Euler LES model is more dynamic and more
details of the instantaneous local flow structure have been obtained
including large-scale structures and vortices developed in the bubble Fig. 14. Power spectrum density of the axial liquid velocity (a) and the radial
plume edge. The one-equation model performs much better than the liquid velocity (b) at UG = 8.4cm/s .
Smagorinsky model with CS = 0.08 in the central plume and vortical
flow regions. The Smagorinsky model improves the resolved axial li­ Acknowledgements
quid velocity profile in the near-wall region.
The effect of inlet superficial gas velocities was investigated. Two This study was financially supported by the Spanish Ministry of
inlet superficial gas velocities, corresponding to transient and turbulent Economy, Industry and Competitiveness – Research National Agency
flow regimes, were chosen for simulations. It is found that the present under project DPI2016-75791-C2-1-P, FEDER funds and by Generalitat
model agrees well with experimental data for UG = 6.0 cm/s and small de Catalunya – AGAUR under project 2017 SGR 01234.
discrepancies are obtained for UG = 8.4 cm/s. The classical −5/3 law of
power spectral density of the resolved liquid velocities is obtained for References
low frequency regions and −10/3 (−3) for high frequencies at
UG = 6.0 cm/s (UG = 8.4 cm/s). The normal Reynolds stress of the re­ Akbar, M.H.M., Hayashi, K., Hosokawa, S., Tomiyama, A., 2012. Bubble tracking simu­
solved part gives very good agreement with experiment and the shear lation of bubble-induced pseudo turbulence. Multiphase Sci. Technol. 24, 197–222.
Behzadi, A., Issa, R.I., Rusche, H., 2004. Modelling of dispersed bubble and droplet flow
stresses u'w' are similar to those obtained by Ma et al. (2015a) using a at high phase fractions. Chem. Eng. Sci. 59, 759–770.
flat rectangular bubble column reactor. Finally, the present study in­ Bove, S., Solbergt, T., Hjertager, B.H., 2004. Numerical aspects of bubble column simu­
dicates that a CFD model based on Euler-Euler One-equation SGS LES lations. Int. J. Chem. Reactor Eng. 2, 1–22.
Camarasa, E., Vial, C., Poncin, S., Wild, G., Midoux, N., Bouillard, J., 1999. Influence of
reasonably predicts the hydrodynamics of two-phase flow in bubble coalescence behavior of the liquid and of gas sparging on hydrodynamics and bubble
column reactors in turbulent-churn flow regime when a multiple-orifice charactertics in a bubble column. Chem. Eng. Proc. 38 (4–6), 329–344.
nozzle is used for gas distribution. Chen, R.C., Reese, J., Fan, L.-S., 1994. Flow structure in a three-dimensional bubble
columns and three-phase fluidized bed. AIChE J. 40, 1093–1104.
Chen, Q., Zhong, Q., Qi, M., Wang, X., 2015. Comparison of vortex identification criteria
for planar velocity fields in wall turbulence. Phys. Fluids 27 (8), 2015.
Declaration of Competing Interest Clift, R., Grace, J.R., Weber, M.E., 1978. Bubbles, Drops, and Particles. Academic Press,
New York.
Darmana, D., Deen, N.G., Kuipers, J.A.M., Harteveld, W.K., Mudde, R.F., 2009. Numerical
The authors declare that they have no known competing financial study of homogeneous bubbly flow: influence of the inlet conditions to the hydro­
interests or personal relationships that could have appeared to influ­ dynamic behaviour. Int. J. Multiph. Flow 35, 1077–1099.
Deen, N.G., Hjertager, B.H., Solberg, B.H., 2001. Large eddy simulation of the gas-liquid
ence the work reported in this paper. flow in a square cross-sectioned bubble column. Chem. Eng. Sci. 56, 6341–6349.

12
M.G. Fard, et al. Nuclear Engineering and Design 369 (2020) 110823

Dhotre, M.T., Niceno, B., Smith, B.L., 2008. Large eddy simulation of a bubble column Pfleger, D., Becker, S., 2001. Modelling and simulation of the dynamic flow behavior in a
using dynamic sub-grid scale model. Chem. Eng. J. 136, 337–348. bubble column. Chem. Eng. Sci. 56, 1737–1747.
Dhotre, M.T., Deen, N.G., Niceno, B., Khan, Z., Joshi, J.B., 2013. Large eddy simulation Prakash, V.N., Mercado, J.M., van Wijngaarden, L., Mancilla, E., Tagawa, Y., Lohse, D.,
for dispersed bubbly flows: a review. Int. J. Chem. Eng. 2013 ID 343276, 22 pages. Sun, C., 2016. Energy spectra in turbulent bubbly flows. J. Fluid Mech. 791, 174–190.
Ekambara, K., Dhotre, M.T., 2010. CFD simulation of bubble column. Nuc. Eng. Des. 240, Pope, S., 2011. Turbulent Flows. Cambridge University Press.
963–969. Pourtousi, M., Sahu, J.N., Ganesan, P., 2014. Effect of interfacial forces and turbulence
Goraki Fard, M., Vernet, A., Stiriba, Y., Grau, X., 2020. Transient large-scale two-phase models on predicting flow pattern inside the bubble column. Chem. Eng. Process.
structures in a 3D bubble column reactor. Int. J. Multiph. Flow 127, 102236. Process Intensif. 75, 38–47.
Huang, Z., McClure, D.D., Barton, G.W., Fletcher, D.F., Kavanagh, J.M., 2018. Assessment Riboux, G., Risso, F., Legendre, D., 2010. Experimental characterization of the agitation
of the impact of the bubble size modelling in CFD simulations of alternative bubble generated by bubbles rising at high Reynolds number. J. Fluid Mech. 643, 509–539.
column configurations operating in the heterogenous regime. Chem. Eng. Sci. 186, Rusche, H., 2002. Computational fluid dynamics of dispersed two-phase flows at high
88–111. phase fractions. Ph.D. Thesis. Imperial College, University of London.
Hu, G., Celik, I., 2008. Eulerian-Lagrangian based large-eddy-simulation of a partially Sato, Y., Sekoguchi, K., 1975. Liquid velocity distribution in two-phase bubbly flow. Int.
aerated flat bubble column. Chem. Eng. Sci. 63, 253–271. J. Multiph. Flow 2, 79–95.
Joshi, J.B., 2001. Computational flow modelling and design of bubble column reactors. Selma, B., Bannari, R., Proulx, P., 2010. A full integration of a dispersed and interface
Chem. Eng. Sci. 56, 5893–5933. closures in the standard model of turbulence. Chem. Eng. Sci. 65, 5417–5428.
Khan, Z., Mathpati, C.S., Joshi, J.B., 2017. Comparison of turbulence models and dy­ Silva, M.K., d’Ávila, M.A., Mori, M., 2012. Study of the interfacial forces and turbulence
namics of turbulence structures in bubble column reactors: effects of sparger design models in a bubble column. Comput. Chem. Eng. 44, 34–44.
and superficial gas velocities. Chem. Eng. Sci. 164, 34–52. Stiriba, Y., Gourich, B., Vial, Ch., 2017. Numerical modelling of ferrous iron oxidation in a
Kouzbour, S., Stiriba, Y., Gourich, B., Vial, Ch., 2020. CFD simulation and analysis of split-rectangular airlift reactor. Chem. Eng. Sci. 170, 705–719.
reactive flow for dissolved manganese removal from drinking water by aeration Tabib, M.V., Schwarz, P., 2008. CFD simulation of bubble column – an analysis of in­
process using an airlift reactor. J. Water Proc. Eng. 36, 101352. terphase forces and turbulence models. Chem. Eng. Sci. 139, 589–614.
Lance, M., Bataille, J., 1991. Turbulence in the liquid phase of a uniform bubbly water-air Tabib, M.V., Schwarz, P., 2011. Quantifying sub-grid (SGS) turbulent dispersion force and
flow. J. Fluid Mech. 222, 95–118. its effect using one-equation SGS large eddy simulation (LES) model in a gas-liquid
Lin, T.-J., Reese, J., Hong, T., Fan, L.-S., 1996. Quantitative analysis and computation of and a liquid-liquid system. Chem. Eng. Sci. 68, 3071–3086.
two-dimensional bubble columns. AIChE J. 42, 301–318. Tomiyama, A., Tamai, H., Zun, I., Hosokawa, S., 2002. Transverse migration of single
Liu, Z., Li, B., 2018. Scale-adaptive analysis of Euler-Euler large eddy simulation for la­ bubbles in simple shear flows. Chem. Eng. Sci. 57, 1849–1858.
boratory scale dispersed bubbly flows. Chem. Eng. J. 338, 465–477. van der Hengel, E.I.V., Deen, N.G., Kuipers, J.A.M., 2005. Application of coalescence and
Lopez de Bertodano, M., Lahey Jr., R., Jones, O., 1994. Phase distribution in bubbly two- breakup models in a discrete bubble model for bubble column. Ind. Eng. Chem. Res.
phase flow in vertical ducts. Int. J. Multiph. Flow 20 (5), 805–818. 44, 5233–5245.
Ma, T., Ziegenhein, T., Lucas, D., Frohlich, J., 2015a. Large-eddy simulations of the gas- Vial, Ch., Lainé, R., Poncin, S., Midoux, N., Wild, G., 2001. Influence of gas distribution
liquid flow in a rectangular bubble column. Nuc. Eng. Des. 299, 146–153. and regime transition on liquid velocity and turbulence in a bubble column. Chem.
Ma, T., Ziegenhein, T., Lucas, D., Frohlich, J., 2015b. Euler-Euler large-eddy simulations Eng. Sci. 56, 1085–1093.
for dispersed turbulent bubbly flows. Int. J. Heat Fluid Flow 299, 146–153. Vial, Ch, Stiriba, Y., 2013. Characterization of bioreactors using computational fluid
Martínez Mercado, J., Chehata, D., van Gils, D.P.M., Sun, C., Lohse, D., 2010. On bubble dynamics. In: Soccol, C.R., Pandey, A., Larroche, C. (Eds.), Fermentation Processes
clustering and energy spectra in pseudo-turbulence. J. Fluid Mech. 650, 287–306. Engineering in the Food Industry. Taylor & Francis Group, pp. 121–164 978-1-4398-
Milelli, M., 2002. A numerical analysis of confined turbulent bubble plume. Diss. EH. No. 8765-3.
14799, Swiss Federal Institute of Technology, Zurich, 2007. H.G. Weller, 2005. Derivation, modelling and solution on the conditionally averaged two-
Mudde, R.F., Simonin, O., 1999. Two- and three-dimensional simulations of a bubble phase flow equations, Technical Report TR/HGW/02, OpenCFD Ltd.
plume using a two-fluid model. Chem. Eng. Sci. 54, 5061–5069. Weller, H.G., Tabor, G., Jasak, H., Fureby, D., 1998. A tensorial approach to computa­
Mudde, R.F., Groen, J.S., Van Den Akker, H.E.A., 1997. Liquid Velocity Field in a Bubble tional continuum mechanics using object-oriented techniques. Comput. Phys. 12,
Column: LDA Experiments. Chem. Eng. Sci. 52, 4217–4224. 620–631.
Niceno, B., Dhorte, M.T., Deen, N.G., 2011. One-equation sub-grid (SGS) modelling for Zhang, D., Deen, N.G., Kuipers, H.A.M., 2008. Numerical simulation of the dynamic flow
Euler-Euler large eddy simulation (EELES) of dispersed bubbly flow. Chem. Eng. Sci. behavior in a bubble column: a study of closures for turbulence and interface forces.
63 (15), 3923–3931. Chem. Eng. Sci. 61, 7593–7608.
Olmos, E., Gentric, C., Vial, Ch., Wild, G., Midoux, N., 2001. Numerical simulation of Ziegenhein, T., Rzehak, R., Ma, T., Lucas, D., 2017. Towards a unified approach for
multiphase flow in bubble column reactors. Influence of bubble coalescence and modelling uniform and non-uniform bubbly flows. Canadian J. Chem. Eng. 95,
break-up. Chem. Eng. Sci. 56, 6389–16365. 170–179.

13